Framework's first desktop is a strange–but unique–mini ITX gaming PC
arstechnica.com
arstechnica.com
New 12-inch laptop form factor with 360 degree hinge (ie "tablet mode") and a touchscreen. No price announced, but it is aimed at students: https://arstechnica.com/gadgets/2025/02/frameworks-laptop-12...
New mainboard upgrade options for Framework 13 models: https://arstechnica.com/gadgets/2025/02/framework-gives-its-...
I was planning on ordering the 7840U version to where I'm staying in a trip to the US, but now it feels a bit of a let down to order a last-gen model since the new one might not arrive in time for my trip in mid-April.
Nowadays i don’t do much heavy computing on my personal laptop and i have an external 34” display anyway.
So yeah, a 12” would be very interesting.
Also i have fond memories of coding everyday on my 10” netbook when i was 16 :P
The 12" form factor tho..
I also have a 13.5" 3000x2000 laptop and it uses 200% scaling, fractional is blurry. Initially I was trying to use Ubuntu but the extend monitor scaling was so bad (Chromium, VS Code), just decided to stick with Windows for this device.
Even Surfaces have been using 1440p at 12'' since 2016, and 2880 x 1920 since 2018! Why would Android & Apple tablets at much smaller screen size have higher DPIs, if 1080p was perfect? Do you expect to put Android tablets closer to your face than x86 tablets for some reason?
Sigh... since when has DPI started _decreasing_ again? I refuse to accept this trend, in the same way it was stupid back in the 2000s when LCDs became a thing.
The human hardware isn't getting any better, so we must accept that there exists some upper bound beyond which improving resolution isn't a selling point for most people, especially given the necessary tradeoffs in battery life, processing power, memory usage, and input latency it entails. Now consider that this ceiling may have been hit 20 years ago, and that the continued dominance of 1920x1080 may not be because manufacturers are lazy, but because most people are happy enough with it.
The Steam Hardware Survey shows that 1920x1080 is still the majority resolution, and that's among an audience that's inordinately populated by technological enthusiasts. The fact that people are seemingly dead-set on sticking to 1920x1080 despite--as you point out--the availability of alternatives only further strengthens the argument that the majority of consumers just don't particularly value higher resolutions.
If you want to play the useless popularity game, go and check what are the resolutions on the phones and tablets with even smaller screens sold in the last 10 years (which exceeds the number of laptops by far), and even friggin' eink notepads.
"The fact" is people today would never accept sub-retina dpis even for cheap phones, and the market has clearly spoken. "The fact" is your arguments about human perception are utter bullshit (as trivially disproven as todays arguments about 60fps), and remind me of the discussions I had when forced to use 60fps 800x600 TN screens (pure hell on earth) after having used 1280 at 90hz for ages with CRTs, all in the name of "progress".
The hardware survey doesn't occur while games are running. It's recording the display setting of your desktop, which would put you and everyone like you in the 4K bucket.
> If you want to play the useless popularity game, go and check what are the resolutions on the phones and tablets with even smaller screens sold in the last 10 years
I have a rather new high-end phone. Its native resolution is 2400x1080, and that 2400 is only there because phones have a particularly long aspect ratio compared to other devices.
> "The fact" is people today would never accept sub-retina dpis even for cheap phones, and the market has clearly spoken.
1920x1080 is a wildly popular resolution. You appear to be frustrated that the market hasn't spoken in your favor.
> "The fact" is your arguments about human perception are utter bullshit
You appear to be hallucinating arguments that I haven't made.
Since the pandemic. I have a still functioning Galaxy S8 in my drawer, which shames modern phones with its 570ppi density.
If you put the phone in battery saver mode, lowering resolution is one of the things it does.
And, to give an idea of what are we talking about, even in such power saving modes these much smaller and older phones still have more pixels than a 1080p screen.
Even if you use today's prices, the cheapest iPad has almost double the resolution. No, 1080p at 12'' it is not plenty dense. You do not put this smaller thing as far from your face as a 24'' monitor.
Hopefully that's upgradeable someday in any case.
Fractional scaling isn’t great. Windows handles it the best by a large margin (it’s really quite respectable; honestly I’d just call it “not ideal” rather than “not great” if restricting to Windows), and has done for many years. macOS doesn’t even try, but rather downsamples from the next integer, which guarantees it will be atrocious, much worse than a lower-resolution integer-scaled panel for many purposes. Linux can be finicky, but is slowly getting there, though downsampling is still common. (Me, I’ve been using Linux/Sway at 1.5× for the last five years.)
You’d get much better results by increasing density by a third in each direction: 2560×1600 is 252ppi, excellent for a 2× scaling factor. (The effective resolution is thus 1280×800.)
Skimming Google, there are pretty much are no laptops 12" and higher resolution than 1080x1200 that's current nor made by Apple.
And let's not get started on 12'' Android tablets...
But yes, considering that Framework's Chromebook is/was expensive (over $1k) for the class doesn't give a lot of faith that the 12in model would be any different. Though that was equipped with a 2k screen and had a 12th gen i5, so I genuinely wonder what kind of meetings happened to ship out a premium Chromebook (a segment that does not exist)
You’d think they could make their most popular pages static for now and serve them out of the CloudFlare cache, though.
The four hinges must stay synchronized under lift force from the user, but I haven't seen that being a problem on a Let's note, so it's probably good.
Their website was hugged to death, so I can't confirm.
I wonder how writing on the touch screen feels?
A laptop tablet hybrid that I can actually repair would be great. Would use tablet mode for image editing and hand-written notes.
I've been looking at a Raspberry Pi 5 paired w/ a Wacom One Gen 2 13 inch screen or a Movink 13 --- will probably stick with that since I prefer Wacom EMR (and have a big investment in it in terms of devices and styluses).
Physically pressing down a diving board is just asking for frustration. Sometimes I miss the separate left, right, and middle click buttons under the trackpad like the old days.
I _really_ hope they launch an AMD version (perhaps with an iGPU) soon after that. That and preferably with Libreboot support. This would make it the ideal portable laptop for me and thus I'd be able to (finally!) replace my X220T.
The 13th-gen Intels had miserable battery life and heat issues under load. If you could manage that, all four USB-C ports were full Thunderbolt ports equally capable of driving displays, PD, and USB 4 throughput.
The AMD line had considerably better performance-per-watt but rougher firmware support (and early on, really broken Linux kernel support that required Fedora or other rolling kernel release distros). It also couldn't deliver the same "every port does everything" promise that the Intel boards did, with some ports not supporting displays or USB 4, which significantly reduced the value of the expansion-card model to kind of a novelty.
On the 12, if it's likely also going to have a smaller batter than the 13, going only with 13th-gen Intels means it likely will be either a further step back in battery life vs. the 13 or throttled to extend the battery.
Both chips were Q1 '23, so the timing's not a great excuse. They were in HP's 2023 EliteBook G10 840 (1334U) and 845 (Ryzen 5 PRO 7540) laptops, and the 845 was better on both single- and especially multi-thread Cinebench R23, _much_ better in GPU loads and gaming, _and also_ lasted longer on battery.
I think Framework mostly just wants to target an education market with a mainboard experience that's lower maintenance than AMD has been for them. Fewer USB-C restrictions, less firmware drama.
Still hard to get excited about it being the _only_ available option, though.
Combined with properly prepared HW taking less power in modern standby than in S3, but it requires hardware/firmware/driver/userland confluence to work well
It feels odd having a load of closed source EFI stuff and then putting linux on top of it. Sure linux injects a bunch of firmware into hardware later in the boot, but it's still progress
I've been burned by a small SBC that had poor support, but on laptops/desktops never felt limited.
But people always sound so excited to libreboot their personal computer... Am I missing out or is it just nerd cred?
Still would like an end to end open source UEFI bootchain, most groups end when they hit the 60% of stuff they need for their limited use cases : - |
The other things is bug-fixing. The way it currently works is that bugs often only get fixed for one version and other issues like that.
Getting this stuff open and having upstream heavy fixing of bugs will imporve the ecosystem.
The other things is performance. Faster boot times are quite nice.
The other things is being able to better play with more advanced boot security concepts, booting with Bluetooth/NFC verification and so on.
And once its open, and you can more easily work with it, and its more available, hopefully more people do more interesting work with it.
That said, its not an absolute priority for me, but its just a better long term solution. So companies that push it are a big plus.
Linux on arm is very mature, but windows on arm not completely.
That being said, other companies could very well develop and sell boards for the frameworks laptop. So much so that iirc sifive did release a risc-v laptop board to use in the frameworks laptop case.
Meanwhile properly SBSA compliant chips aren't targeted at mobile market at all and thus often lack certain features or are too energy hungry.
There are a lot of more or less weird, very dedicated, very opinionated subcultures in the PC hardware world.
wut? Gamers love to pay extra for a tiny bit of less noise. Noctua and expensive water cooled systems are widespread in the gaming community.
Entirely passive (I THINK / recall there was no pump), heat based circulation, with just TONS of radiator surface and an unwieldy design. However for that entirely custom solution they could theoretically scale up with additional size.
Still, if someone's going _that_ overboard, the heat's getting dumped somewhere.
'Industrial Chilled Water' as in a connection to an external (to the room, maybe even building) pump, and an external cooler (usually evaporative cooling, could be mega radiators in the shade too) like you see at industrial sites.
ARM has nothing to do with being fanless or not, that's just whether or not you're happy with what 15w can get you or not.
Not many people are, which is why ARM laptops with fans are just if not even more common than ones without.
Also there's a single ARM SoC that isn't garbage at this power/performance bracket on the open market, and it's embattled with legal troubles (Qualcomm vs. ARM over Snapdragon X Elite). And while the X Elite CPU is great, the GPU and software for things like video deciding are bad and break regularly
On the plus side I'm hoping that it's the existence of framework that have convinced lenovo to stop soldering components in their new laptops, that's what put me off buying one and why I went with a framework.
The pitch for the Framework laptop is that it is repairable/upgradable/modular. Something that is uncommon for laptops nowadays.
This is the opposite. Desktops are modular by default, so much is that my computer is like the Ship of Theseus, I never changed it, but upgrade to upgrade, it is a completely different machine than it once was (it started off as a 486!). This one is not.
The Framework desktop doesn't look bad, but now, I am confused about the meaning of the brand. It is as if Tesla made a diesel car.
Similar to what other commenters have expressed, it just seems like they shouldn't have built this product if they couldn't figure out the soldered RAM bit.
Consider it a low-price alternative to mac mini or nvidia’s box.
This can also be chained though not as effectively as macs for example (those have thunderbolt for interconnect)
This starts at 1099$ per the article, so the mac mini is the low-price alternative here.
Select 32
It's not exactly an out of the park win though as it's only $100-$200 cheaper and it trends poorly the farther from comparing base models you go (particularly after 64 GB, for which you need to leave the Mini family for the Studio). By the time you get to 128 GB, which you'd want for the 70B class AI models, you end up back at the original statement.
However those are the base M4 chip and not the M4 Pro. You need the M4 Pro to get competitive in GPU compute numbers for a more like to like comparison. The M4 Pro mini comes out at $1800 with the 48GB option, or $2000 for 64GB. For the same price the Framework machine gets you 128GB.
Every large AI model is heavy memory bandwidth constrained to the point my 9800X3D (with the extra L3 cache) and 128 GB/s memory attached is only 60% utilized running a 32 GB model in CPU only mode (no NPU, iGPU, or GPU offload enabled). Really small AI models can start to be compute bound but, at that point, you don't really need the 32 GB of memory anymore and probably just want a normal GPU.
Also fully spec'd at 128gb is $2000, which only gets you a 64gb mac mini
It’s also not that absurd pricing of the modules themselves but of the R&D effort to cram 4 CAMMs onto a board with sufficient signal integrity
This is also being seen in the X3D CPU's where the memory is stacked on top to reduce latency and as density was maxed in 2d space.
So to ensure 256 bits @ 8000 MHz works well AMD did it in the same package. In theory they could A) ship a halo without ram B) design a new socket, C) allow motherboard makers to pick 2 x CUDIMMs or 4x DDR5 dimms. Not sure that would buy them much market though.
As apple and others have proved, not many care about replaceable dimms, especially if it gives them 2-3x the performance and/or better perf/watt.
All that aside, I absolutely can't wait for desktops to decide to go the same route of having 4 memory channels instead of 2. Right now the only way to have >2 channels is to buy workstation/server class stuff or an APU.
BTW, careful with "channels". Todays normal desktops and laptops (besides the strix halo and apple pro/max/ultra) are 128 bits wide. But in the DDR4 -> DDR5 transition they doubled the number of channels to 4. However the strix halo biggest advantage is the extra memory bandwidth going from 128 bits wide to 256 bits wide.
Finally the technology of the ps5 and XboxX comes to laptops, tablets, and small desktops.
The Ryzen AI Max+ 395 was paper launched a month ago. What Framework announced here is just another pre-order (even batch 1 won't ship until Q3). There are dozens of other systems, including mini-PCs, announced months ago which are set to launch before this. Small laptops were actually the first 395+ products to launch, hence this month's benchmark reviews all used them. E.g. https://www.tomsguide.com/computing/laptops/gaming-laptops/a...
I would not be surprised if one will be able to get a full 128 GB 395+ mini PC for less than the cost of the Framework Desktop baseboard before the Framework Desktop actually starts arriving in hands. What you're buying here is a premium to be able to replace the shell and front USB ports.
I haven't found any other with a committed release/ship date.
This Reddit thread tends to agree: https://www.reddit.com/r/MiniPCs/comments/1ieiate/when_are_r...
Even the Asus Z13 laptops with the 395+ have no shipping date yet.
Looking at the two GMKtec systems with their rattling fans I have here, I would rather take the Framework system where I can easily swap the fan. Cooling those 100-120W will take space and air flow.
Plenty of folks on Reddit have already received shipment (not order) confirmation of the Z13. It looks like even more as of this morning. I doubt if any have actually fully received said shipments yet as it has only been a couple days since launch. It'll likely remain pretty out of stock unless you managed to get an order immediately at launch, again due to the paper launch of the 395 last month and low 395 availability until Q2.
I didn't know HP was making one too! That'll be exciting as it's easier to order that at my work than e.g. an ASUS one or what have you.
That's me! I'm that person! \o/
I guess they'd claim it is only the MAX AMD procs which force soldered RAM, but since they could as well have used a non-MAX chip (and correspondingly reduce the price) this just shows how much of this is an arbitrary, and therefore questionable, decision from Framework rather than any restriction AMD sets.
If LPCAMM already can't keep up with requirements when it is barely even out, then my guess is it won't fare well going forward.
So 1 of 2 things is probably true:
AMD is not being completely truthful with their statements that LPCAMM wasn't able to work (maybe it was just more difficult/complicated than they were willing to do, but it could work or
latency/speed requirements have already outpaced what LPCAMM can provide and soldered ram is the future.
I really hope it's the former, but it wouldn't be the first time something like the second has occurred. Apparently cache also used to be a separate, swappable component before it became integrated into the die. RAM might end up going the same way.
I understand that a new company with a new product is going to have issues. But I would have strongly preferred they spent the time and effort (and money) fixing or replacing these 1st gen mainboards rather than branching out into a very non-Framework area like desktop gaming PCs.
They provide a "repair" kit that's basically a dummy CMOS battery that hooks in to the normal power system that prevents the issue from occurring again.
Also just FYI, there was a BIOS update in June of last year (3.20).
I am aware of the issue as described in [1] and the fix in [2]. However, my support request has gone unanswered for a year, as was my second support request. In addition, I have doubts as to whether this fix (which requires soldering!) will work.
I was not aware of an updated BIOS [3], foolishly believing the output of 'fwupdmgr' after following the instructions in [4]. It looks like I'll need to find a USB stick and update via EFI shell. Thanks for the tip!
But still, I think they should do more for early customers before expanding out well beyond their core market!
1 - https://framework.kustomer.help/my-laptop-is-not-powering-on...
2 - https://guides.frame.work/Guide/RTC+Battery+Substitution+on+...
3 - https://community.frame.work/t/11th-gen-intel-core-bios-3-20...
4 - https://knowledgebase.frame.work/en_us/updating-bios-on-linu...
Also you're welcome! They have been pretty hit or miss on pushing updates to LVFS so I absolutely understand why you might not have known.
I truly regret my purchase of a Framework laptop. Truth be told, I've never, ever had to repair a laptop more than this one, and I'd prefer a product that I didn't have to repair at all (and if there is a defect in manufacture or design, that the manufacturer take all necessary steps to remedy that error). If I was attracted to their products, including their desktop offering, I would think twice before spending money with them. I really really wish I'd put the money toward a new M-series macbook.
Heck their reminder to keep the BIOS updated makes no mention of the fact that their instructions for Linux are bogus, as their UEFI blobs are not available to fwupdmgr. What's more, to get to 3.20 you have to first upgrade to 3.17, all through EFI boot. It's a really rotten experience.
==============
PLEASE READ THIS MESSAGE FULLY!
This is an automated message sent to all customers that have either been identified as having a power-related issue with an 11th Gen Intel Core Mainboard or have specifically requested either a replacement RTC Battery or the solderable RTC Battery Substitute.
If you’ve needed to perform the Mainboard Reset procedure more than once on your 11th Gen system or the laptop will not power on even after performing the reset, your RTC/CMOS battery may no longer be holding a charge. There are a couple options available that the Framework Support team can provide to resolve the issue:
1. Framework Support can provide a replacement RTC/CMOS battery to you, free of charge, regardless of the status of your warranty. Simply request this option and provide your Framework Order Number along with a clear image/photo of the serial number of your Mainboard. If you do not have access to the Order Number, please provide the email used to complete the purchase on the Framework website. To access the Mainboard serial number, please follow the steps in the guide HERE and find the serial label between the memory slots. Also, please verify your shipping information/address (only in supported countries/regions) to avoid unnecessary back and forth communications. For information on acceptable shipping addresses, please see the following article HERE . If you receive a replacement RTC/CMOS battery it is important to make sure to let it fully recharge by keeping your laptop plugged into power for at least 24 hours. We also recommend keeping your laptop plugged into power during long periods of non-use to avoid letting the RTC/CMOS battery fully discharge. Note that even a normal 5V/900mA USB-C power adapter will sufficiently trickle charge the system and RTC/CMOS battery.
2. Framework Support now has an alternative option for those with the technical skills to solder electrical components. This alternative RTC/CMOS Battery Substitution is a single solder point component which replaces the coin cell battery with a circuit that keeps the RTC subsystem powered from the main battery. You can find the step by step instructions for installing this module in this guide. PLEASE NOTE: You should NOT request this option if you or the person designated for this rework do not have sufficient electronics soldering experience. While it is only a single solder point, failing to solder this correctly can result in damage to the system that is not covered under Framework’s Limited Warranty. If you accept the associated liability, please request this option and provide a clear image/photo of the serial number of your mainboard and your Framework Order Number. If you do not have access to the Order Number, please provide the email used to complete the purchase on the Framework website. To access the Mainboard serial number, please follow the steps in the guide HERE and find the serial label between the memory slots. Also, please verify your shipping information/address (only in supported countries/regions) to avoid unnecessary back and forth communications. For information on acceptable shipping addresses, please see the following article HERE .
Finally, we also recommend keeping your system up to date with the latest firmware releases. Please see the following knowledge base article which has links to the latest BIOS/Firmware and Framework Driver Bundles for each generation of Framework Laptop.
If you have provided the required imagery and your order number, one of our staff will be with you to provide an update on your request. Thank you for your patience and we apologize for any frustration this issue may have caused. Regards, Framework Support
===============
Also I will again agree that my Batch 4 (or was it 5? I can't remember) FW13 is the most I've had to repair a laptop as well. I'm just gonna chalk that up to them technically being pre-full-production units and hope they have improved their QA in the later gens.
However I will also add that I like having the option to fix it myself (which is something I have some amount of skill/enjoyment in doing) which is certainly NOT an option you'd have with that M-series MacBook (or really any other brand of laptop)
Turns out the Samsung EVO NVMe I installed had really high power draw (I guess it's meant for performance desktops?). I replaced it with a WD Blue and now I lose negligible power while suspended.
Sadly this does not generate revenue to the shareholders.
The form factor isn't small enough to make this worth it IMO
This product is for me.
A few years ago I tried to repurpose a desktop with a bad motherboard, but it was impossible to find a replacement for the 7-year-old CPU. eBay prices were more than the original MSRP, and at that point it was cheaper to buy new parts for the oldest still-selling generation.
I'm already replacing everything except the SSD and GPU with every upgrade anyway, now it will just be the SSD but I can keep the case.
That 64gb OCed 3200 (maybe 3500) mHz DDR3 RAM for example, kept up with DDR4 speeds for most of that decade when I benchmarked every few years. Intel extreme processor, workstation motherboard, and GTX Titans also kept chugging along. I’ve been looking at some of the higher end DDR5 coming down the pipeline and those are also likely to be on the higher end of performance for a decade while the consumer parts catch up. The only problem is that the workstation motherboard market seems to have disappeared.
(This was a work computer for my consulting)
I’m on a 10500 based Lenovo thing at the moment.
My needs are not immense though.
I don't do much computing on the move, that my phone can't handle. But a desktop (that generally lasts twice as long as a laptop and costs half as much) is pretty interesting to me.
Also keep in mind that framework sold a small chassis compatible with their laptop boards, so if you upgrade your laptop main board you could rehost the old one in a small desktop chassis.
Seems like a natural expansion for framework and might well get them more customers and more sales per customer.
In SFF land?, maybe, but that's already a lot closer to laptops, which is where I expect this to loop back into regular Framework lineup
Nit: You can believe that story, but there's zero chance that at some point you didn't upgrade the case, motherboard, CPU and RAM possibly at the same time. If you replace 90% of what makes a computer, does it make sense to say you "never changed" it? At what point do you consider it "changed"?
This Framework desktop also lets you swap motherboard, CPU and RAM in one go, why does it not fit your definition?
That's why I said "you can believe your story" while I pointed out it's a very unlikely one.
The point of an analogy is to point out how our intuition tells us it's the same boat, despite having no parts in common with the initial boat.
As other comment mentioned, if you keep the sails and rebuild a new boat around it, no person in their right mind would feel that it's the same boat. It's very clearly a new boat, and the analogy breaks down.
It seems more like a way to launch a mini-desktop. They can later offer alternative mainboards that fit in that desktop (or anyother).
Yes, its a strange first product, but I think its mostly because they wanted to be early with releasing this AMD chip. Seems to me AMD was looking for partners to push this chip out for 'AI'.
Memory, CPU and GPU once piece of metal, sitting in a tiny box.
A regular PC in a regular case, it a lot more modular and upgradable.
It does seem like an interesting box, and matches against Apple Studio I would presume.
Yet customers of Apple are used to having (near) 0 user modifiable parts.
It might well have a good market, It might b a great box. It is unframeworky.
At first, Framework is "laptops that are repairable". But if you broaden what they are, they are a disruptor of direct-to-consumer computing equipment, with a core competency of repairability and upgradability.
An integrated CPU/RAM is a decrease in that measure, but it is for a valid benefit - a large improvement in performance for low-power graphics and AI software. They aren't sacrificing upgradability for aesthetic, and they continue to offer fully upgradable laptops.
I wonder if modular memory will continue to evolve and be competitive bandwidth wise with soldered.
That's the promise of CAMM2, which is supposed to enable socketed LPDDR with almost the same performance as soldered-down LPDDR. It's still pretty bleeding-edge though so it's hard to blame Framework for sticking with soldered memory for now.
I'd be more than happy to see CAMM2 in general laptops, but it will probably always be much weaker at shared GPU/CPU designs like Strix Halo, Grace, Apple's M series, etc.
The ceo kind of explains why in this video. In essence it seems to be a limitation of the chip from AMD
edit: it's not for me and I can totally just not buy one, but if one identified with their original mission and sees this as betrayal of that, it'd be hard to justify getting a framework laptop when it's their turn to upgrade.
RAM upgrades at reasonable prices, being able to buy the main board sans case, and supporting multiple OSes all point in that direction, without strictly being modular
The RAM is not upgradable.
FTA:
> Because the memory is non-upgradeable, we’re being deliberate in making memory pricing more reasonable than you might find with other brands.
> July 7 2022 3:26pm
> Framework’s mission is to fix consumer electronics - and we are doing that by respecting you and the planet. We have put this vision into everything we do by providing you with amazing products that are meant to last as long as possible by letting you upgrade and modify them over that lifetime to support your specific needs.
> Our mission is to reduce the amount of waste that is generated and energy that is expended and reduce our overall consumer electronics footprint, while providing a better product than you can get anywhere else. This includes using post-consumer-recycled aluminum and plastic in our products as well as recycled and recyclable packaging.
> We have just started this journey, and we are continually looking for ways to reduce our footprint and be even healthier for the environment. We’re happy to get ideas and suggestions on how we can do this. The Framework Community is a great place to share this.
You presume to have internalized Framework's core-values more than the founder/CEO? The box is not my cup of tea,but they are free to experiment.
His reaction in the livestream was along these lines when he semi-jokingly said "I'm surprised no one from the audience threw something at me"
At a larger event I would have kind of expected a "boo", but it seemed like a rather small gathering where most people knew each other. Unlike the live 12k Youtube chat, that was very surprised and disappointed at times.
Also, Framework doesn't need to make something for you to "build a PC". PCs are extremely modular already. They are solving a different problem here.
The question is how much different is a motherboard with soldered CPU vs. CPU and memory soldered. You seem to be of the opinion that it's a completely different product category, I'm of the opinion they're more the same than different. I don't remember the last time I didn't upgrade all three components at once.
> Depends on the context.
> Dollars, no. Murders, yes.
The point is it's a pretty critical component, so "just one" is doing a lot to downplay how critical a component it is. If we get rid of that one component, you're nowhere. So calling it "just one component" belies how critical a component it is.
> I don't remember the last time I didn't upgrade all three components at once.
Thank you for explaining your perspective. If they're a single component in your mind, and not modular, then no wonder we have such a disconnect.
My guess is that Framework had a unique opportunity with the AMD Ryzen AI and decided to capitalize on it to serve a fast-and-growing home market for this class of hardware, and the soldered LPDDR was a compromise considering the requirements of the CPU. If I had to choose between them offering this product with that restriction, or "sticking to their core values" and waiting for an alternative solution then I'm going to learn to live with the restriction. If the traffic queue wasn't just marketing BS, I assume many other people are also willing to live with it.
I want Framework to be a long-term successful company and that means making good use of their cash, and this gives them a safer opportunity to test a new product category. Maybe the result of this decision is an expansion of the category to include more modular options, at which point everyone wins.
One might maliciously (IMO) argue that the single motherboard in Framework products that you presumably find perfectly fine, could have been designed as several interconnected ones, that are individually replaceable. On the surface, it can seem like an unimpeachable criticism, but once you consider the cost in complexity, performance and BOM, then the "less modular" single-motherboard option becomes much more reasonable.
Framework went with soldered memory because that was the only way they could hir the memory bandwidth performance numbers they wanted for this product in the context of the segment it occupies. If you value the ability to replace/upgrade RAM over 256Gb/s, then this is not the product for you. If you think Framework shouldn't compete in this segment due to a confluence of ideological reasons and the technical limitations of slotted RAM, then the CEO disagrees with you, as do the future buyers of the compact desktop
FWIW I find the small form factor combined with the CPU and high-powered integrated GPU very appealing. I don't think I could build something with this form factor using off the shelf parts (someone correct me if I'm wrong)... it would end up needing a larger dedicated GPU.
I suspect their competition isn't actually people who build their own PCs, but people in the market for Mac Pros — they have a number of benefits over Apple here.
You can still buy the motherboard version and use your own case and PSU. That's better than not just the macs, but also many other mini PCs.
And considering the chip, it's still the most modular way to buy a Ryzen 395 for the foreseeable future.
I don't understand, but maybe someone else could explain.
Not an engineer though so please correct me. I only have a vague understanding of this.
There's none of that here. That's a concept for uncontrolled interfaces. This is a memory interface, where you either have a good signal or a flawed design. Things like ECC do exist, but those are to detect bit corruption in the memory, but a flawless communication between is still required.
At some point (if you go to high enough frequencies), the capacitance of the copper traces will become high enough (i.e. without any capacitor component connecting to a trace between a CPU and a RAM module) to prevent further frequency increases. One way to deal with that is to have shorter traces. This is exactly what CAMM memory modules do - they have shorter (total) traces than DIMM. Even shorter traces are possible if you get rid of modules completely (i.e. solder the RAM chips to the motherboard). Better yet is to place RAM and CPU cores on one chip, skipping even the motherboard traces between the CPU and RAM chips.
and games, having really speedy/low latency memory helps a lot with being competitive to some mid range dedicated GPUs
Unified RAM/VRAM is very nice for running LLMs locally, since you can get wayyyy more RAM than you typically can get VRAM on discrete GPUs. 128GB VRAM on discrete GPUs is 4x5090s — aka $8k just on GPU spend alone. This is $2k and it includes the CPU!
Of course, it'll be somewhat slower than a discrete GPU setup, but at a quarter of the cost, that's a reasonable tradeoff for most people I'd think. It should run Llama 3.1 70b (or various finetunes/LoRAs) quite easily, even with reasonably long context.
Feels like a stab in the back.
Whoever is in charge of that website gets an L.
For that matter, it makes me reconsider my hosting of things with Cloudflare. I know nothing about framework's site's configuration, but I know I don't want my site to have a waiting line like that.
edit: also, the timer went down and then went back up, so I have thoughts about this enterprise Cloudflare feature.
guess it's true what they say about hardware vs software. you gotta pick one to be good at, and the other is going to suffer for it, to varying degrees. (inb4 someone mentions Apple. Apple is a hardware company. Their software's alright but it's full of bugs and they're simply not as good as it as they are with hardware.)
E-commerce is hard. I worked at a company where we could use 1% of infra at its peak every day of the year except 15-20 days. We knew exactly when floodgates will open, and we still would suffer extra high latency or even downtime.
e-commerce is hard. that's why we get paid so well. hiring the smartest teenager that your nephew knows to setup some bullshit for $15/hr vs hiring a senior SRE at $100+/hr, when it directly leads to lost sales is a choice.
the senior part also comes after having failed to scale in production, and learning the lessons there, leading to a site that stays up on the next black Friday/cyber Monday, and stands up to various ddos attacks. (this was before Cloudflare, mind you)
I bet the waiting room makes it worse, since it cannot cache the page due to having the waiting room.
also the page has `via2.0 heroku-router' header.
https://i.ibb.co/Y4n5Qhzm/framework.webp ("Framework Desktop is a big computer made mini")
The soldered ram is particularly unacceptable. I get and believe that they could not make it work otherwise, but then they should have stopped the product instead of just adding to the e-waste.
You can get similar bandwidth with server boards that cost 5-10x as much, or with a Mac Studio that costs 2.5x as much.
The problem is that these are integrated shared-memory systems with a single RAM pool. That's nice for a lot of reasons, but GPUs need many more memory channels and larger bus widths than CPUs do in order to do work and remain fed at a reasonable power draw. It's an inherent design trade off. I don't see a CAMM style solution for GPU memory coming anytime soon except on the low end.
Could be wrong, but I don't think you can. The bandwidth limit, AFAIK, is a problem with the DDR5 spec. These soldered solutions can go faster specifically because they aren't DDR5.
The latest AMD server is 12 DIMMS wide, but has 24 channels of DDR5.
[1] https://www.tomshardware.com/pc-components/cpus/amds-beastly...
(video also features Framework's founder/CEO)
See also:
[1] https://en.wikipedia.org/wiki/Graphics_address_remapping_tab...
[2] https://www.kernel.org/doc/html/v4.19/gpu/amdgpu.html#:~:tex...
(The 32GB config is silly, though. With that little RAM, there is nothing it does better than a cheaper machine with a discrete GPU.)
In the LTT video the framework CEO explains that AMD wasn't able to make LPCAMM work because of signal integrity over the bus reasons.
But 2000 dollars for up to 110GB of VRAM in Linux makes this a VERY interesting little machine, so much that the framework website has a cloudflare queue right now...
Better to think of it as a competitor to Mac Studio & Nvidia Digits, which are much less repairable by comparison. The soldered memory is an unfortunate reality of these "unified" memory systems.
The only way to get a traditional desktop with 96GB of VRAM is to spend upwards of $10K loading it up with 2-4 GPUs.
1) have ram soldered but provide also extra 1 slot for LPCAMM/LPDIMM ram. Soldered RAM would be used for GPU but external slower ram still could be used for non GPU related stuff (software) that doesn't require high bandwidth/latency.
2) I think motherboard will have 4 pads for ram and if you buy less 1-3 pads will be just having pins without chip. If they would sell memory chips and provide some tooling/documentations for specialised/certified shop they could solder those. You could still upgrade just not by yourself.
[0] https://forum.qubes-os.org/t/community-recommended-computers...
I was asking about the desktop.
Does the desktop have a discrete physical TPM chip (needed for DRTM support on Windows/Linux/Qubes)? At present, AMD's PSP firmware emulates a "mobile" fTPM that does not support SKINIT.
Would you consider a future model with AMD Ryzen AI Max "Pro" SoC, which has additional security features?
Virtualization: interrupts, vIOMMU, nested paging, GMET, SLAT
Security: memory encryption, Windows Device Guard, VBS, Secured Core
Management: DASH for rack/cluster
It also seems HP will use the PRO version of the chip.
Kudos to framework for announcing the price and anchoring the market though!
If there was a possibility to get amd dash working this would be the perfect system to use as a home server.
I'm as cynical as it gets when it comes to tech companies, but Framework is the only one that seems to be on track to actually make the world a better place. Please keep doing what you're doing, stick to the mission, and I'll be a customer for life.
Also please make a smartphone so I can finally be on an all-Framework stack! :)
#1. Any insight into the 4x PCIe slot not having the end of the slot open which would allow the user to still physically insert 8x or 16x PCIe cards?
#2. Considering a lot of people would use the desktop with ethernet, is the wifi m.2 slot keyed to only accept wifi cards or can it fit a 3rd NVMe SSD as well? (maybe an idea to also pass along to the RISC-V mainboard guys)
#3. Lastly, any chance of revisiting the possibility of upgradable RAM on future desktop models i.e. Zen6 "big APU" if CAMM modules and the like aren't so bleeding edge? (it's looking like DDR6 for example might be CAMM-only) I mean, the fact that CAMM modules lay flat would open the opportunity to have them placed directly on the backside of the CPU which would theoretically result in very short trace lengths relative to front-side CAMM (let alone DIMM).
EDIT #4. Do these Ryzen Max chips support s1/s3 sleep like traditional socketable desktops? (though it sounds like it's starting to become less supported on AM5 sadly)
BONUS. Keyboard/touchpad module(s) for laptop(s) with three (left, middle, right) separate physical mouse buttons when? :P (Heck, on the 16" laptop there's already suspicious blank spaces both above and below the touchpad where they'd fit perfectly...)
Do you plan to make computers which can actually be repaired? How exactly is the Framework Desktop any better than what Apple is doing with the Mac? I prefer to build my own machines. Why would I ever choose such a product which can't have parts replaced over something better which enables me to repair and upgrade my computer?
I ask because you were on "Buy now!" which also tackles greenwashing. I fail to see how this product is any better than all of the disposable junk sold by other companies (soldered RAM, soldered CPU, no PCI-E, no second NIC, no expandability of any kind).
Do you plan to sell products with PCI-E ports for dedicated GPUs and other devices?
Has Framework's customer support improved? Do you plan to do something about that? I've read countless posts from people who state they didn't receive a reply from Framework's customer support or that their hardware problems were never resolved. Why should a new customer trust your company?
It seems that the Framework Desktop 1st gen has a 4x PCI-E port. That's not exactly useful for a GPU. I've learned this after watching the LTT video.
Because unified memory can't be socketed due to signal integrity issues. Framework asked AMD if there's a possibility and they investigated it, but found that it was just not possible.
I recommend watching the LinusTechTips video about the new framework products. They answered all your questions.
This product goes against their principles of building products which are more environmentally friendly. They've done this for the laptops by not forcing people to buy a new laptop when their motherboard is dead or no longer fast enough for the software they run. It's also possible to replace the keyboard, the hinge, the battery, the RAM, the wifi module, the SSD, the touchpad, the case, the display and the expansion modules.
This Framework Desktop 1st gen can have the following components replaced: wifi, SSD, CPU fan, maybe the heatsink, some front panel IO modules, some decorative tiles on the front, the PSU and some parts of the case. A single broken regulator or failing memory chip forces the owner to replace the entire computer. One is forced to replace the entire thing if they have no option to get someone to find the relevant part, desolder the existing one and solder the new one on. This is also not an option for the CPU.
This means that any kind of damage forces the owner to buy another board with CPU and RAM soldered on it for about the same price as the entire thing with the case.
This Framework Desktop computer can be repaired just like most laptops with soldered RAM by replacing the entire motherboard with CPU and RAM. Why would I downgrade the desktop PC's repairability down to that of a laptop? The tradeoff isn't worth it for that price.
There will be more options by the time this becomes available.
I assume you are talking to AMD about OpenSIL, are you planning on switching to Coreboot with Linuxboot?
Also, any chance delivery to Switzerland? I can't recommend it to people if they can't order it.
1. Some kind of quantized version of DeepSeek R1?
2. Flux image generation models via popular tools like A1111 or Comfy?
You can do image generation as well! AMD has some demos of this, and they are maturing PyTorch support to broaden the software ecosystem.
Regarding LLMs, am I going to be able to run quantized models? Doesn't have to support all quantization methods, but at least one of the popular quantization methods would need to work (e.g. GGUF or EXL2). I ask this because quantized models provide the most "bang for buck" in LLMs.
Regarding image generation, in real workflows there is typically a lot of dependencies (much more than in LLMs, and much more than in toy examples for image generation). I'm wondering how many of these dependencies will in practice be limited to NVIDIA and CUDA. For example, if you grab some popular ComfyUI Flux workflow from Reddit, are you able to actually run that on Framework Desktop?
Would you consider adding keyboard options that include these keys?
For a max spec processor with ram at $2,000, this seems like a decent deal given today's market. However, this might age very fast for three reasons.
Reason 1: LPDDR6 may debut in the next year or two this could bring massive improvements to memory bandwidth and capacity for soldered on memory.
LPDDR6 vs LPDDR5 - Data bus width - 24 bits, 16 bits Burst length - 24 bits, 15 bits Memory bandwidth - Up to 38.4 GB/s, Up to 6.7 GB/s
- Camm ram may or may not be maintain signal integrity as memory bandwidth increases. Until I see it implemented for a AI use-case in a cost-effective manner, I am skeptical.
Reason 2: - It's a laptop chip with limited PCI lanes and reduced power envelope. Theoretically, a desktop chip could have better performance, more lanes, socketable (Although, I don't think I've seen a socketed CPU with soldered RAM)
Reason 3: In addition, what does hardware look like being repurposed in the future compared to alternatives?
- Unlike desktop or server counterparts which can have a higher cpu core count, PCEe/IO Expansion, this processor with its motherboard is limited on re-purposing later down the line as a server to self-host other software besides AI. I suppose could be turned into a overkill, NAS with ZFS and HBA Single Controller Card in new case.
- Buying into the framework desktop is pretty limited based on the form factor. Next generation might be able to include a 16x slot fully populated, a 10G nic. That seems about it if they're going to maintain the backward compatibility philosophy given the case form factor.
I understand that Framework’s logistics cannot match those of major retailers like Amazon or AliExpress. However, many customers rely on freight forwarders to access products from other countries. It is deeply disappointing that Framework does not allow shipments to these intermediaries, as they are a common and well-established workaround for limited international shipping. Given the widespread use of such services, excluding them seems unjustified.
How do export restrictions prevent Framework from shipping to i.e Japan?
When you’re a company who needs time to adapt to any change in policy (aka all of them) and a company that can’t afford fines for noncompliance (small companies like Framework) your strategy is to be as conservative as possible.
We have a very strict radio law that applies to anything that can produce radio or em waves
That includes motherboards, since they can technically emit on those frequencies.
From what I have seen, the framework laptop motherboards appear to abide by that law. However, I assume it's just expensive to figure out in the first place.
You are now in line.
Thank you for your patience.
Your estimated wait time is 7 minutes.
We are experiencing a high volume of traffic and using a virtual queue to limit
the amount of users on the website at the same time. This will ensure you have
the best possible online experience.
Clearly poorly messaged if it made you think it was a Cloudflare capacity issue!
[0] https://www.cloudflare.com/application-services/products/wai...
Here's a screenshot of https://frame.work/ca/en/blog/category/news:
https://i.imgur.com/7BcLyCX.png
There's nothing up there that couldn't be statically cached.
Otherwise, could be a good reason to use subdomains when architecting sites, say shop.framework.com.
But you probably want M.2 slots over a single x16 slot.
For gaming specifically, so many micro ATX motherboards offer Gen 5 PCIe, which can handle a proper video card, double the RAM, and the smaller cases are only slightly larger than this Framework.
So I'd say yes.
My gaming needs are pretty tame to the point that my current 3080Ti has been and remains overkill (usually 2+ year old titles @ 2560x1440), and as time has gone on I’ve come to value silence (which the FW Desktop seems good at, overcooling a laptop APU with a single fan desktop cooler) over raw power. In addition, the discrete GPU story continues to escalate to all-new levels of eye-wateringly expensive stupidity which makes me not want to buy any discrete GPU until Nvidia and AMD bring their prices back down to earth and that whole mess with the new Nvidia power connector is properly resolved, and that’s to say nothing about the unstoppable creep of GPU size, heat, and power consumption.
If I could sell my full size gaming tower and replace it with an effortlessly inaudible yet reasonably powerful air cooled SFF box, I might just do it. In all truth I could probably get by fine with this first gen Framework desktop, but it would make more sense to wait for a second or third gen where the APU's graphical power comes into the range of upper-tier RTX 3000 cards so I don’t need to use framegen as a crutch for decent framerates.
When I look around, gaming pcs are mostly about big and visible, sometimes reasonably silent, almost never compact, inconspicuous and silent.
To me there is a market for this kind of product, and it hasn’t been addressed properly yet.
Since I have so far failed in my quest, I now use GeForce Now from my Mac mini which is a good approximation of what I want.
Because the market you describe has very heavy expectations, and very exacting taste.
(And critically, the capability of building one themselves.)
Kind of but not really. Any SFF build that’s anywhere close to similar in size and capabilities to the Framework is probably going to be making considerably more noise, even with an AIO liquid cooler.
Between consoles and macs/macbooks, the writing is on the wall and cpu+cpu+unified memory is the future for performance. I will absolutely be looking at buying one of framework desktops instead of building a new PC soon.
I also saw that while it competed well with a 4070 at average frame rates, that the 1% lows were twice as fast. The speculation I saw was that the 12GB vram limit was causing issues on the Nvidia side.
Having twice the bandwidth, dynamically allocated VRAM, and very decent cores seems like a solid workhouse that it's more about better worst case performance (when you are heavily cache missing or running out of vram).
Sure some will buy a top of the line CPU from Intel or AMD, buy a larger case, and buy a RTX 5080 or 5090, for the 98% of the rest this strix halo will be plenty and will use less power, create less noise, and generate less heat.
They’ve brought some of the traditional modularity from desktop into the laptop world, and now bring us typical laptop design to the desktop world.
Keeping things in perfect balance.
For my mini PC I couldn't get the EGPU to work with Linux so I'm stuck on Windows for now... But I play a few games that are Windows only ( anti cheat) so this is for the best.
With a desktop there's a limit to what Valve can commit to. There's not a single controller firmware to support, and probably not even a consistent GPU setup to cache for. The extent of realistic support for these AMD boards is kinda fully realized at this point. Proton is, and will remain, a plug-and-play experience for AMD users that own supported hardware.
Valve as a company could shut down tomorrow, and AMD users could still use Proton to play Windows games as long as their GPU is Vulkan 1.2 compliant.
Theoretically awesome, but this might have some interesting market consequences for everyone else.
You could run R1 671B using unsloth’s quantized version that fits in <80GB. Not sure why that would be a benchmark though, there’s nothing that can run the model at full precision right now except for (very slow) server hardware.
Not if you are using the CPU. I am under the impression most inference use cases are memory bandwidth limited, not compute limited, so running on the GPU would gain you little to nothing unless the GPU has faster access to the shared memory.
256GB/sec, so roughly M4 Pro throughput.
Or a dual socket AMD Turin.
Or a grace+hopper (assuming you offload to the hopper).
Latency is dictated by the laws of physics, more bandwidth is easy, but not cheap.
If you want a swappable gpu then a APU isn’t for you
Two years ago, I switched from my 15" MacBook Pro to the 12" Framework, never looked back. What am I missing?
As it's visible from existing material the bezels on the Framework Laptop 12 will be huge and as it's a touchscreen, its readability will be worse.
It's just not catching on, because of course it isn't. Manufacturers have no incentive to let you upgrade parts on your own.
There just isn't a free lunch here, it's an inherent design tradeoff for these kinds of chips. The CPU is just along for the ride, in this case.
My completely-bullshit opinion is something like "I think it's attractive for a lot of mid-range or lower end SKUs, so you'll see it more frequently." It's attractive because you save a lot of power and components with this kind of design. It would be interesting to see this kind of design (APU) in a socket-able factor too, but who knows. It's also attractive for a certain class of high-end design too (like Grace Hopper), but I think you're going to see RAM sticks for a while yet in the mid-to-very-high-range.
This is completely bullshit because it is based more on vibes and basic technical design points than any kind of market forecasting or anything like that.
That ITX board looks like it's a stone's throw away.
It would obviously be better if they could limit it to something like `store.frame.work`, rather than putting a queue in front of their entire site…
For every other use-case? Yeah, just get a desktop.
The nvidia digits uses DGOS, same as their grace+hopper and similar enterprise/cloud products.
That was basically unobtainium and I've compromised down to a AliExpress mini-ITX motherboard with a mobile AMD CPU, a AM5 heatsink and firmware that is... flavorful, powered by a screaming TFX power supply crammed in an absurdly tight 3.8L noname case running on Proxmox (when you start reconfiguring PCI Express bridges through the serial port kernel debugger because that's just about the only device the Illumos kernel enumerated at all on what is supposed to be your main server, it's time to give up on SmartOS).
It works, but it's a little box of pure hatred and heresy that's quite far off from what I've wanted initially. It is actually an improvement over my previous main server, somehow.
I’d have bought a Beelink or similar mini PC if I wanted small size and low power along with low price. You lose some modularity compared to ITX boards but I am almost certain you spent more money and deal with more noise and maybe even more power consumption.
For me personally my homelab PC is just an ATX mid tower in the closet because those parts are dirt cheap and you can get lots of performance with essentially infinite modularity.
That makes noise (a glass panel separates the rack from the entrance), power (can't risk overheating in an enclosed closet) and space (6U and 25cm of usable depth is an exceedingly small volume for a homelab) all important factors at the same time. I also wanted at least 2.5 Gib Ethernet, as much DDR5 RAM and CPU cores as I could chuck into it and two NVMe slots. I rejected a mini PC due to noise concerns and I wanted standardized parts and modularity.
I have a AMD Ryzen 9 7940HS, 64 GiB of RAM and terabytes of NVMe in less than 4 liters of volume without sacrificing modularity. I've mounted an AMD Wraith Stealth, which is a comically oversized CPU cooling fan for a 35W TDP thermal load. It's fast, small, silent, overkill and still fits standard parts.
I do want to get rid of the ridiculously noisy TFX power supply at some point, most likely for a 48v DC ATX power supply. Unfortunately, since I like stupid ideas I want a fully 48v DC powered rack and Power-over-Ethernet throughout my apartment to get rid of as many power supplies and fans as I can. I've started accumulating parts for it (like a 48v 2.5 Gib 10 port PoE switch, PoE powered gigabit switches and PoE-to-12VDC adapters), but I'm still not sure how I'll pull it off in the end since I'm way off the beaten path.
At this point you're probably wondering why I do things that way. I'm not sure myself, but seeing what I tend to do for fun it's probably best not to ask, lest the universe stops its suspension of disbelief while I'm in the middle of performing yet another heretical act on some poor unsuspecting piece of hardware or software.
Those specs all seem totally attainable in a mini PC, although something with an equivalent chip and dual nvme slots is up in the $800-1000 range.
Maybe it would be louder? I don’t think they have a reputation for being loud though? I certainly can’t hear the one I’m using as a router, but it has a lower load profile than my main server.
I have a small apartment too but I might be lucky with my deep half height closet where I can shove equipment back there.
A former sysadmin job gave me a deep disdain for 40mm fans and modern 1U/2U off-the-shelf AC powered servers. It's a technological dead-end that is a legacy of old datacenter architectures that no longer make sense to continue. A holistic blade-based DC powered datacenter eliminates much of the inefficiencies inherent to that legacy AC design.
Now, I'm not aiming to do a high-density homelab and regardless how cool they are, an Oxide Computer rack is about three orders of magnitude out of reach across several metrics. That being said, I think a DC powered homelab can enjoy a lot of the benefits.
For example, I have one 10 Gib switch, three gigabit switches (one per room), a Synology NAS and a Android TV player. With AC that's six power supplies, each taking a dedicated outlet for less than 60W combined. With PoE I can power all of that directly from my 10 Gib switch. That's five less power supplies and also gets rid of the power strips.
Or take a 250W ATX power supply. On AC it requires transformers, full bridge rectifiers, big bulky capacitors, DC-to-DC converters and most likely a fan. On DC it requires just the DC-to-DC converters, with modern power electronics that's doable at a fraction of the volume and without the fan because the actual power supply is upstream and potentially shared across multiple equipments.
I might have an affinity for doing stupid and insane projects, but it does come from a thoughtful position. If I ever pull it off, hopefully it can show that we don't have to endure outdated and legacy designs forever even within the comfort of our homes. Alas, I'm busy doing other cursed things at the moment so I just put that angry little server in a spare room for now.
For $2000, you get 128 of system RAM, 96 of which is addressable as VRAM. Only ways of getting 96GB of VRAM in a desktop are to either:
1. Drop ~$5000 on a (very non-upgradeable) Mac Studio 2. Drop ~$20k on a dual RTX 6000 workstation
For running local LLMs, there's nothing on the market presently even remotely like this.
I have a max with 64GB RAM, which is good enough for 70b models with a 3 bit quant. Even if I had more RAM to run larger models, my GPU would be the bottleneck.
To get an M4 Max so you can have 128GB, you need a macbook pro. The cheapest macbook pro with 128GB is $4700.
M4 Max does have the benefit of more memory controllers, so it has twice as much memory bandwidth as Ryzen AI Max. But that's a lot of money to pay for it.
By the time the Framework ships the Mac Studio will have been updated to the M4 Max. Although 128GB will still probably be around $3k.
One of the issues in Europe is that each country has their own language and custom keyboard.
Would love to see how it performs. It supposedly has a memory bandwidth of 256 GB/sec which is about similar to a Threadripper Pro 7965WX with 8 memory channels. A Mac Studio M2 Ultra has 800 GB/sec of bandwidth though (which is RTX4080/RTX5080 territory) but is also about 2-3x more expensive at 128 GB of memory, not to mention the cost of upgrading internal storage.
Am happy with both but the AMD doesn’t support S3 sleep, and so I mildly regret buying it. Will probably not leave my desk often.
Also don’t believe they support ECC memory yet. Don’t think I’ll buy another PC without it.
The desktop is 15 minutes in. (After the notebook announcements)
https://www.youtube.com/live/dnCGwii7FqY?si=0ppcg_cjj8I-5P67
Edit: found it, 256 GB/s for the shared memory bandwidth
This has been the future for more than a decade, evidenced by stuff like smartphones and games consoles, and as far as I know, Nvidia is also working on their SoC, and so does Intel (which I think was the whole point of their gaming GPU push).
Even ignoring the fact that the FW 13 and FW 16 have socketed/upgradable RAM, you’ve got the chassis, the screen, the wireless card, storage, the keyboard, and more.
If you had bought the FW 13 upon release (in 2021, I think), you have already had several opportunities to upgrade the guts of the computer while keeping everything else the same. If my CPU is no longer suitable for my needs, then why should I have to also replace my display? If it were a desktop computer, that’d obviously be preposterous. And FW has made products that make it preposterous for laptops too.
On the whole, I think you may be confusing the goals of repairability and upgradability with the notion of modularity. They’re not wholly unrelated of course, but they are different, and I think it’s clear that FW accomplishes those goals.
Signed, Someone who wishes they had a FW
Just an AI hobbyist, so I don't have time or inclination to tweak everything. Given the non-NVIDIA GPU, how painful will it be to play around with new AI models on this system?
Inference is not an issue with AMD.
But they have now introduced reduced modularity and customisation in comparison with what's available in the desktop market. CPU is soldered, RAM Is soldered and the GPU is soldered.
I was looking from GMKTex and Beelink to have at least 64Gb, but this is fantastic deal and I can't wait for it.
I bet the desktop is interesting, but first I was in a 30m waiting room to access their website (what?!?) and then the button to pre order the desktop is broken. Ouch.
That's a no-brainer to replace my laptop-as-desktop in the next few years. By then, I expect this will be even better.
Arm laptops are plenty competitive with the x86 space with the snapdragons now.
They fired me a few weeks in, but then went out of business.
Most of the manufactured Framework-compatible accessories are skins, wraps, and expansion card organizers. Cooler Master dropped one mainboard case and seemingly bounced from the laptop project altogether. There are a bunch of cool DIY projects, a handful of which have been productized, all of them niche.
The community marketplace concept never materialized. The extensibility promise of the 16's input modules haven't materialized. The only third-party 13 mainboard that exists after 3.5 years is a cool but ultimately impractical RISC dev board/proof-of-concept; the idea that the Framework mainboard would become a laptop equivalent to the ITX/ATX standards in desktops just did not happen, and Framework's decision to start shipping a bunch of different mainboard formats means it never will.
It's particularly depressing to me that the only modular component that seems to be compatible across the 12, 13, 16, and Desktop seems to be the expansion cards, which are a fun concept but at the end of the day are just a form factor for USB-C port adapters.
I'm honestly excited about the 12 being a supposedly cheaper repairable option, although seeing this weird Desktop ready to go before the 12 is a boggling decision. I have no interest in spending $1k-$2k+ for a novel mini-PC using laptop components, in a mostly plastic case, with a bespoke motherboard crammed with soldered-on bespoke parts (even for good reasons!) that are designed to _not_ be repaired or replaced.
(By the way, why _doesn't_ the desktop use the Framework mainboard form factor? I'd be interested in a genuinely larger mainboard-compatible desktop case with more airflow, designed for a specialty Desktop mainboard but compatible with the laptop mainboards too.
A mini-ITX board that's less modular than a commodity mini-ITX board, in a mini-ITX case that isn't competitive with commodity mini-ITX cases, is such a weird choice in Framework's "keep using your mainboards" pitch. If they're going to ship a bespoke board with little to no added value when installed outside of their own case, why doesn't that board use _their own board format_?)
Hell, Cooler Master's MasterFrame line is a better execution of what I'd expect and want out of Framework shipping an *TX-compatible desktop case than Framework's case looks to be, and Cooler Master apparently worked on Framework's case too!)
And even then, the 12 is just another set of components that aren't cross-compatible with the 13. If they were selling a convertible 13 case, or even just a stylus/touchscreen display for the 13, I'd be buying it right now.
Even the 13's new AMD boards aren't exciting because I expect them to ship with the same or worse firmware and driver stability or compatibility issues that still haven't been solved on the 7040-series 13 mainboards a year after shipping them, not because Framework is a terrible company but because their support from AMD has apparently been a nightmare. I finally have my 13 stable and expect a new generation of AMD mainboard to just chuck it back into firmware hell.
That Framework keeps taking VC money just to design and ship new laptop lines when their existing lines aren't stable, _and_ ship a less-modular, less-repairable novelty in the Desktop that they try to pitch as a gaming machine—when their laptop fundamentals are still admittedly shaky, and the gaming market still doesn't seem to care for or about them very much at all—just keeps eroding the confidence that this is going to work out in the end.
There is substantial x86 market demand for local inference that can compete with Apple. AMD/Framework 128GB for $2K will be competitive.
> A mini-ITX board that's less modular than a commodity mini-ITX board
LLMs need LPDIMM for high-bandwidth memory and efficient power. mITX board allows interop with a galaxy of existing mITX cases, including NAS.
> same or worse firmware and driver stability or compatibility issues that still haven't been solved on the 7040-series 13 mainboards a year after shipping them, not because Framework is a terrible company but because their support from AMD has apparently been a nightmare.
We're inching towards AMD's 2026 target for open-source firmware, which will hopefully help all AMD OEMs.
The only parts which can be customized for this product are the presence or absence of a handle, the cooler's fan, the case's side and some front tiles. That's it. The m.2 SSD and the wifi are the only components which can be replaced.
This isn't the kind of product I wanted Framework to make. I was hoping they'd make hardware which can be repaired and which has components available for it. The motherboard has all the chips and everything else soldered on it. The most expensive part of the computer needs to be replaced if a voltage regulator or some other part found on the motherboard fails. There's no cheap $ 100-200 motherboard to replace in this product. It's the same problem as with Apple's Macs.
Can someone at Framework answer this question: what do the customers do with your Framework Desktop hardware once it breaks and you no longer support it? It's e-waste. What happens when the motherboard in my computer dies? I buy only a replacement motherboard while keeping the RAM, the CPU and GPU, unlike for Framework Desktop. What happens when the GPU I have is no longer useful or supported? I buy only a new GPU.
This board doesn't even have PCI-E for a GPU. This product is only good as long as the iGPU provides the required performance for whatever application is of interest. This is a weakness the Framework 13 motherboard shares. There's no way to remove the board from its case to use it with a PCI-E x16 GPU with the right PSU.
AMD is known to abandon their customers once they release newer dGPUs and SoCs with iGPUs. This can be easily observed if you review the countless reports for crashes with amdgpu on Linux. The amdgpu driver has various bugs which lead to crashes of the GPU or of the entire machine. They're also not good at shipping CPU microcode for consumer CPUs to address hardware bugs and CVEs.
As a side note, even the Framework AI HX laptops are extremely expensive for what they offer in terms of hardware. A laptop which goes above $ 2000 without RAM, an SSD, a charger and without any adapters for those bays seems to be a good deal? That's absurd. There are laptops with 32 GB of RAM, the same CPU, better displays, a 1 TB SSD, a charger and all the required ports present on the laptop for less than $ 2000 (including taxes).
I hope someone from Framework reads this. I want repairable products which can be upgraded without replacing a monolithic part which is the entire computer.
Other noteworthy things
- their site went down hard with a queue to see the site... downright absurd
- they haven't posted the specs of the Framework 12
- there are still no actual repair centers which repair their products, no physical stores or sellers which sell Framework products outside of their site
- there have been reports of people who didn't have their hardware problems with Framework laptops addressed, even LTT addressed such issues
The sweet spot is 64GB but the M4 Pro Mini 64GB is a better deal at that RAM capacity.
But hey, they've refused to provide GPUs with lots of RAM at a cheap price so competition, y'all.
Nvidia's actual counterpart would be their DIGITS mini-PC, which has a similar big-and-slow memory architecture.
So its better than 4090.
The reason its better with a less powerful GPU is context switching.
"AMD also claims its Strix Halo APUs can deliver 2.2x more tokens per second than the RTX 4090 when running the Llama 70B LLM (Large Language Model) at 1/6th the TDP (75W)."
https://www.tomshardware.com/pc-components/cpus/amd-slides-c...
i just built a mini ITX gaming PC for a friend, and this one looks pretty good for quality/$. good enough that i wouldn't be surprised if these get snapped up and re-sold for more than the sticker price
i think it makes more sense to think of it as a high-end console, given that basically everything is soldered, though
For one thing, the price per FPS for gaming is going to be terrible. This price can get you a serious rig with much better performance.
Gaming is a very specific workload that almost entirely depends on the GPU. It is hard to even purchase a CPU that performs poorly in gaming by accident, or unless you are a hyper-specific type of gamer (like someone who plays e-sports titles and expects 300FPS for optimal reaction time, or someone who plays extremely CPU-intense games like Stellaris).
You don’t want a gaming PC with soldered GPU, even if you are using it as a “console,” even if you prioritize small size. You really want the ability to replace graphics cards because none of the other components important to gaming age out very quickly.
Here are all of the parts in my system that are older than my current graphics card (2024):
- Motherboard
- CPU
- RAM
- Power Supply
- SSD
- Case
- CPU cooler
Here is a list of those parts that are older than my previous graphics card (2020)
- RAM
- Case
- Power Supply
- Motherboard
- SSD
Here is a list of all the components that are older than the prior graphics cards before that! (Year unknown)
- Motherboard
- RAM
- Power Supply
Arguably I didn’t even need to upgrade the CPU, the cooler upgrade was necessitated by the CPU upgrade, and the case upgrade and SSD were both just personal wants and not needs.
Basically everything that matters besides my graphics card has stayed very constant but upgrading the GPU has increased my gaming performance by over 100% in the period I have described.
from what i can tell, a 4070 alone is ~$750. this thing is apparently $799 for the CPU+GPU+RAM+motherboard all soldered together ($300 more for case+PSU+SSD)
if the CPU+RAM+motherboard are costing me $50, i don't really care if i have to throw them out along with the GPU
from the specs, a mobile 4070 looks like its basically a 4060 ti, underclocked to ~60%
i haven't bought a laptop gpu since ~2008. from what i remember, they used to at least put "mobility" in the name, or throw and 'm' after the number, or something
That sucks. I’ve had better luck with Intel’s drivers for their first series of dGPU’s.
If this works with tinygrad’s AMD driver, that would then interest me.
Me & my 7900 XTX will be quite grateful if it does.
Otherwise as it stands the 128 GB configuration seems pretty niche.
EDIT: looking deeper it seems like the "Ryzen AI" is it's own thing with a different implementation than ROCm, so it could be interesting but might not help with ROCm.
The innards, however, are disappointing:
- I get the explanation for why soldered RAM was necessary, but that's still pretty darn close to a dealbreaker for me; I'm inclined to wait for a future motherboard revision without that limitation
- Only two Expansion Cards is vastly fewer than what I'd expect from a "Framework Desktop"
- The lack of a dGPU is unfortunate for a desktop
If I were to design a Framework Desktop, I'd replace the entire rear panel with nothing but Expansion Card slots. Literally as many as will fit; fucking fill it to the goddamn brim. And then throw in some USB4 headers to connect to even more Expansion Card slots on the front. I want a terrifyingly large number of Expansion Cards. More Type-C ports than any computer has any business having, and then even more. I don't just want people questioning my sanity; I want it to be known, plain as day, that I have gone certifiably batfuck insane.
I'd also expose the same Expansion Bay interface as the Framework 16, and offer a desktop-grade GPU in that form factor (presumably too thick and power-hungry for a laptop, but if a Framework 16 owner wants a laptop with a dummy thicc dumptruck of an ass and 30 minutes of battery life, then who am I to judge?).
And air-cooling? So 20th Century! Good opportunity as any to make liquid cooling a mainstream option.
And as for the space-efficiency... the existing non-swappable ports are already pretty sparse and spread out. The expansion cards ain't all that big; stacking a bunch of 'em shouldn't take up more space.
Also, was hard to see--there are more than six ports on the back. There's plenty of ports, many usbc and usba, which can be daisy chained.
This announcement is bog standard mini-ITX.
> (note the two bays for Framework's expansion cards at the bottom)