AMD Grabs over 30% CPU Market Share as Intel Continues to Decline
wccftech.com
wccftech.com
Here's some more reading if you're interested: https://www.anandtech.com/show/17641/lighter-touch-cpu-power...
99.99% of the population doesn't know this option exists.
And 0.01% want BIGGER NUMBERS OMG 720FPS!!! and ruin power consumption worldwide for everyone else.
Which is unfortunate, because when it comes to benchmarks, understanding the context of a test setup is extraordinarily important. Are you buying that very high-end memory the test setup used and then manually adjusting timings? No? Then you're not getting those numbers. Heck, you may be losing 20% or more performance compared to the test rig on just the difference in memory. Never mind adjusting other things, ensuring the system's thermals are entirely kept in check to prevent throttling, etc.
This already happened in previous generations (especially Zen2).
AMD has to play this stupid game, unfortunately, because the press is what it is.
As you should. I worked at Intel back in the 00s when they were in the "megahertz wars". All they talked about internally was "we're winning the MHz wars!!! Yay!". When anyone mentioned power consumption, or how fucking loud the cooling fans were, or MIPS/Watt, they didn't care; all that mattered was MHz. Hitching their wagon to RAMBUS memory was part of this. Then they got pissed and indignant when consumers didn't want to spend a fortune on RAMBUS memory and bought AMD CPUs with cheap SDRAM instead.
I wrote a few microbenchmarks to explore the performance promises of rambus, and didn't find anything.
Cinebench R23 multithread:
Ryzen 7950X 230 -> 125 watts = 95%
Ryzen 7950X 230 -> 105 watts = 93%
Ryzen 7950X 230 -> 65 watts = 81%
Intel I9-13900k 253 watts -> 125 watts = 78%
Intel I9-13900k 253 watts -> 105 watts = 72%
Intel I9-13900k 253 watts -> 65 watts = 56%
Personally I'd rather get the 7900X (or non-x) and you'd keep an even larger fraction of the performance within a given power limit. It's pretty clear who has the worse core and has to push harder into high clock/terrible power usage for not much gain curve.
Also keep in mind that "performance mode" on Intel can significantly increase the default 253 watt TDP. Anandtech reviews of the i9-13900k hit 380 watts I believe.
So sure there's chip to chip variation, but generally the AMD chips are more power efficient and have much lower penalties for reducing the TDP.
I've been playing around with the 13900K's settings and so far I've managed to get it down to a stable 210W TDP(according to XTU under stress testing. That's with a 0.110V undervolt(likely could do more) and the P-core turbo multipliers tapering down to 50 with all cores active from 57 with 1 core active. I've yet to mess with the E-cores though; could probably shave off some more perf/W still.
The system idles at half the wattage, too—so it's saving a ton of energy not running at the stock 170W TDP. Not sure why AMD didn't go for efficiency crown, especially for non-halo SKUs like the 7950.
Happy to be proven wrong, because it is a brilliant idea. My machine is wildly oversized for my typical usage, and I could easily take a 10%+ performance haircut and likely would not even notice.
But if you use these chips for tasks that max out all cores for several hours per day, it may be better the simply run a chip with more cores and to run those in econ mode. A 7950X@64W may perform as well as a 7900X@200W for such compute tasks, and are probably cheaper over time.
And if you use it in your house, they generate less noise, can use a cheaper cooler and doesn't dump so much heat into your room.
With a 120Hz screen, you can run the UI and overlays at 120FPS, run cinematics at 24FPS or 30FPS and the game at 40FPS or 60FPS. All fits neatly into 120Hz.
How does a 99% static UI or overlay benefit from 120 fps?
They have it a lot harder than something like Switch.
Chinese ODMs seem to have already started making mini-pcs with the new chips, and I imagine routers will follow quickly.
Until AMD’s media engine improves, the Intel chips will still be valued by the Plex/Emby crowd.
I'd love a new router board with a 6 W CPU and some storage ports. But all I can find is media player boards with at most one ethernet.
Dream board atm would have said 6 W CPU, DC power, two ethernets, one 2.5 Gbps, and some M.2 and SATA slots. For a combined routing/storage box.
I imagine an Alder lake N version of this will appear sooner or later, and the N100/N200 should give you the 6W part you’re looking for.
https://m.aliexpress.us/item/3256804762512339.html?gatewayAd...
Not as nice as my old Atom router because of the active cooling on the CPU and the ATX power connector. The latter I can fix with a PicoPSU but the former...
I'd rather have a passive radiator on the CPU and add some airflow via a huge case fan - which is what i'm doing now for my router box. That way it's a lot more silent.
It seems like the CPU fan stays off almost all the time, and is apparently very quiet even when on.
> Oh, and I forgot to mention, I have set the fan to come on at 36 degrees and go off at 34 degrees. It's off most of the time (pseudo-passive) ;-)
> s I'll use the stock one first becase the heatsink fan is dead silent as @Camprr23 has mentioned.
Farther than I thought, it turns out.
You know, for all the stuff about Apple silicon being a game changer it doesn't really seem like the game has changed that much...
Apple silicon was only a game changer for productivity tasks, and only for people willing to jump into the Apple ecosystem. In all other cases, especially for gaming, an APU with the performance of Zen4/RDNA3 at the announced TDP doesn't exist yet. So, yes, it's a game changer.
They just need to run natively on ARM to get performance that's more than competitive on the CPU side. The GPU is no slouch but is not top of the line (yet). Seems trivial to pair the ARM CPU with an external GPU.
You mean the Steam Deck 2 or a new device that will "kill" the Steam Deck?
That said no one has streaming down perfect yet. Steam Link, AMD link, Moonlight all suck.
My sweet summer child. You should read about tacit collusion.
Seems like Apple's increasingly happy to hit lower price points and disrupt the price structure for AMD and Intel based product lines.
Similarly Intel seems to finally be getting it's act together and setting up to disrupt the AMD and Nvidia GPU market. Here's hoping. I saw a pretty decent GPU (RTX3080) for $420 recently!
Or maybe having the iGPU performance go up 4-8x so most won't need a $500+ GPU?
I'm considering an Apple studio, once refreshed with the M2 Max, even for Linux. Sure the GPU driver isn't quite there, but it's improving quickly.
When doing a quick Google search on the topic, it turned out that several of the top results were blatantly misleading in that they limited the competing hardware to match the M1's limitations. For example, the top result, a wandb article [1] claims that the M1 is competitive with the V100, yet their own data shows that they aren't even fully utilizing the V100 and that when properly utilized it obviously totally outperforms the M1.
https://wandb.ai/vanpelt/m1-benchmark/reports/Can-Apple-s-M1...
Similarly, Apple in its marketing for the M1 Ultra was extremely manipulative, bordering on outright lying, when it compared the chip to the 3090. It presented them on a "relative performance vs power" graph making it look like the chip matched the 3090 while consuming less than half the power, when what it was saying was that it's more efficient than the 3090 when you underutilize the 3090 to the point of matching the M1 Ultra's performance.
I dug around and found some M2 updates.
Seems that generally, Nvidia laptops drop to a small fraction when on battery.
Examples: 3dmark wild life. M2 max = 150 fps, 3080 ti 120 fps, 3080 Ti unplugged = 27fps.
Blender 3d rendering: m2 max = 1:04 3080ti = :29 3080ti = 3:25
Lightroom: m2 max = :39 3080 Ti =:51 unplugged = 1:29
DR18 6k braw editing: m2 max = 0:55 3080 ti = 2:40 unplugged = 15 minutes
For ML comparing 3070 vs M2 max, 3 batch sizes:
CIFAR-10
m2 max = 242, 280, 327
RTX 3070 = 240, 283, 328
Seems like the M2 max does pretty well compared to a plugged in RTX 3070 to 3080 Ti. The big bonuses is that you can use all ram (not limited to VRAM of 10-16gb) and you get the same performance even on battery.
ResNet50:
M1 amx = 155.3
M2 max = 188.8
M1 ultra = 242
mobilenetv2
M1 mmx = 415.7
M2 max = 467.9
M1 ultra = 634.4
DistilBERT
M1 mmx = 134
M2 max = 164
M1 ultra = 215
BERTLarge
M1 mmx = 20
M2 max = 25
M1 ultra = 30
CIFAR-10: 3 batch sizes
RTX3070 240, 283, 328
M1 Max 297, 358, 443
M2 Max 242, 280, 327
M1 Ultra 207, 308, 453
Densenet 121: 3 batch sizes
M1 Max 57 104 704
M2 Max 66 124 831
M1 Ultra 49 99 614Do not forget ARM.
I very much doubt that architecture is all that relevant for their advancements in power usage. Apple's chips contain a significant x86 feature without ruining battery life. Meanwhile, Qualcom is struggling to compete with Apple in both performance and efficiency despite being in the ARM space for much longer.
I'm sure if Apple could've gotten an x64 license ten years ago, they would've made their own x64 chips instead of switching to ARM. When Apple's plans started coming together, there simply were no competing architectures they could base their chips on. MIPS was practically dead already, x64 was extremely closed off, RISC-V wasn't even announced and struggles to keep up today and it wasn't even announced when Apple started selling their own chips.
Maybe they could've licensed POWER6 or an early version of POWER7? The POWER architecture isn't exactly widely used or designed to be power efficient; power management wasn't introduced until 2017 and even then it was optional.
There simply weren't any serious alternatives to licensing ARM and Apple would be stupid to develop an entirely separate CPU architecture for their desktop/laptop/tablet form factors.
ARM is from the 1980s, let's just jump right to RISC-V instead. :)
Then there are concerns like register pressure. x86 has so few general purpose registers, that values need to be stored on the stack and reloaded when needed. Some of the performance impact can be reduced by complex decoder logic, but making complex logic fast nearly always leads to high power consumption.
[1] E.g., the highly variable length of x86/x86-64 instructions puts a limit on the number of instructions that can be decoded per cycle.
It's got plenty in 64-bit mode (15 + RSP) + you can spill to SSE registers instead of the stack.
It also needs fewer registers than (most) RISCs because it has a more flexible way of specifying memory addresses (base, index, scale, offset + PC-relative) and it also has proper immediates.
> the highly variable length of x86/x86-64 instructions puts a limit on the number of instructions that can be decoded per cycle.
Not that much of a limit, actually. Yes, a parallel decoder that takes on arbitrary byte sequence and decodes them is hard to scale up. The instruction lengths can be cached, though. In fact, they used to be cached as extra bits in L1 back when the size of a wide decoder was a significant size of the CPU transistor budget. It should be possible to use that idea again to go wider.
The newer x86 CPU's also have a µop cache so no decoding is even needed for tight loops.
AMD Desktop CPU Market Share 18.6%
AMD Mobility CPU Market Share 16.4%
AMD Server CPU Market Share 17.6%
AMD Overall x86 CPU Market Share 31.3%
What's causing the bump from 16%-18% share to 31% in the avgs?
Are they including game consoles in the x86 market share?
(edit: maybe not "desktop" though)
Objectively true.
> You would be hard pressed to find one that didn't
Eh, that's less true; I own 2 ARM Chromebooks, because I wanted some non-x86 in my life, and neither was hard to find, it just took me actually looking for it.
Search Chromebook on bestbuy, exclude out of stock and sort by price, and you'll see a mix of arm and x86.
Honestly, I think the research firm's divisions are crap. WTF is "mobility" anyway? Why would this not include laptops?
A Chromebox is a desktop PC. And ChromeOS devices can run a full Debian Linux distro in a VM. What's not desktop about that?
Or is the definition of desktop computer that it can, say, run Adobe Photoshop natively?
So a "desktop PC" isn't about its feature set or where you might put it, instead it is a form factor.
My point is that the Chromebook is a mobile device: it's tiny and easy to carry around. So why is it not in the "mobility" category? And for that matter, why is every laptop computer not in this category? When did "mobile" somehow become equal to "something you can carry in one hand while also operating it"? The entire point of laptops was that they're mobile. These days, not that many people use "desktop PCs", outside of corporate offices, they use "mobile devices" (i.e. laptops).
And where do tablets fall? You can prop them up on a desk and attach a keyboard, and they act just like a laptop (you can also connect external screens etc.).
As I said before, their categories are stupid.
https://www.hp.com/us-en/chrome/chromebase-all-in-one-deskto...
To the original point, it sleeps perfectly.
I purchased a MacBook Pro last year...zero remorse.
Edit: Also Macs.
No, seriously. Most peoples' computing needs and desires are satisfied with a phone, they don't actually need a laptop let alone a desktop.
It's not that x86 is losing market share, it's that Personal Computers are changing from the desktop/laptop form factor to the smartphone form factor. It's mere coincidence ARM and not x86 powers all those smartphones.
Also, if 30% of x86 market share corresponds to ~20% of "big processor" market share, that means that non-x86 cores have about 30% market share.
European companies are discovering the same thing, too. They have embraced ARM and RISC-V very strongly.
Unless Google chromebooks up their game, or the pc disappears completely I don't really see arm taking over.
I thought ARM represented like ~5% of server install base, and maybe like 10% of 2022 server sales. Similarly, I understood Chromebooks + Macbooks (the two major laptop ARM products) to be on the order of 25% of laptop sales.
These are both real and meaningful chunks of market, but I don't think they're enough to make up for the discrepancy in numbers. Perhaps they are including game consoles.
[Edit] - found reporting on 2021 numbers they clearly seem to indicate that the first three sets of numbers are x86 only (https://venturebeat.com/consumer/mercury-research-amd-closes...)
The two big extra categories would seem to be game consoles (as suggested by GP), and IoT.
The overall CPU Market Share number also includes both IoT and SoC numbers which are not included in the other numbers and in which Intel's shipments have declined substantially while AMDs SoC products (including game consoles) are making up a large number. This is what's responsible for the disparity.
They also added that the data was distributed with a helpful clarifying note the news articles are mostly omitting:
* Please keep in mind that due to the inventory corrections taking place, that the statistics and share movements reported here in the past few quarters -- and likely for the first half of 2023 -- are more reflective of the suppliers differing in the depth and timing of their inventory corrections, rather than indicating sales-out share of the PC market, which is something we probably won't know with any accuracy until late in 2023. *
I think AMD originally pitched it for laptop makers since it was leaked in 2020: https://videocardz.com/newz/amd-mobile-apus-for-2021-2022-de...
But I suspect they rejected it because they don't care about IGP performance, just as they rejected Intel's "Iris" Broadwell CPUs and the hybrid amd/intel chip.
I can easily see datacenters with these things in the future.
The higher end x86 NASes are advertised closer to "home datacentre in a box" solutions that offer VM/container runtimes and third party software markets.
So far, x86 is significantly more user friendly and performant for these use cases: the "desktop-ish performance for desktop-ish prices" range doesn't really have many hardware options (Apple certainly won't sell theirs to OEMs, and the Snapdragon 8c is a bit on the low end, and the real data centre ARM many-core monsters are too big); and the software offerings aren't quite as user friendly either.
I love my Synology and love that they are going AMD. While some argue that having video transcoding on a NAS is wrong, it’s also very handy, but the AMD options I’ve seen aren’t a patch on quicksync.
That tech is unbelievably good and might be the only bit of Intel I like.
And while their discrete GPU market share is in decline -- and since they bought ATI they never had the lead, indeed it never went above 40% https://overclock3d.net/gfx/articles/2022/11/29160939924l.jp... it seems they still gain significant benefits from having them by selling a lot of APUs, especially in consoles.
My PC has a 12400 in it which actually realistically kicks my M1 Pro MBP in the teeth on some things which is embarrassing as it's a junk ass end Lenovo desktop machine that cost <$500.
We live in interesting and exciting times.
Intel does win some benchmarks, but often at substantially higher power consumption. The AMD chips run cooler and even significantly reducing the TDP down to 65 watts often has a single digit percentage slow down.
99% of what I do is single thread bound so I really want to see some single thread power benchmarks but no one seems to bother with those.
I'm in the market for a PC with the best single thread performance out there at the moment so I've got several machines configured in pcpartpicker and spreadsheets at the moment :(
I've been tracking DDR5 motherboard (I think the perf is worth it), MicroATX (don't need fullsize), I don't need blinking lights, watercooling systems, wifi, or exotic sound systems, so I go for relatively barebones motherboards. I do want something small, quiet, 2.5G ethernet, and has dual m.2 slots.
On Newegg I found an Intel ASRock Riptide for $159 [0].
On Newegg I found an AMD ASRock Riptide for $169 [1].
I'll happily take the AMD and end up with a quieter and cooler system for the $10.00 premium. Debating between a 7700 (non-x), 7700X3D, or 7900 (non-x). Hopefully it lasts as long as my current (2015) desktop.
Do you consider a $10 premium "extortionately priced"?
[0] https://www.newegg.com/p/N82E16813162061 [1] https://www.newegg.com/p/N82E16813162083
(I'm checking one vendor here who I know has a decent returns policy when your system integration goes to hell, another common AMD problem from much experience!)
BTW, AM5 added the ability to upgrade BIOS without a CPU. So no more nightmares of having to beg a CPU, borrow a CPU from AMD, or RMA to get a compatible CPU<->BIOS setup.
Good to hear that about the CPU issue. I found that one out the hard way and had to scoot around to a friend's house and borrow his CPU for a few minutes. I bought MSI board after that as they had that built in anyway.
AMD scare me on the reliability front. The 3700X and B450 combo I had before was a dick. It would come up 50% of the times with the fans cranking and have to be powered down and back up again a couple of times for it to boot. My Intel machines never did anything like that for the last 25 years or so but the last Athlon XP I had did the same damn thing.
My current desktop is a 2015 Xeon e3-1230 which was CHEAPER then the similar i7, but allowed ECC. I do like that the AMD motherboards allow ECC, it's a bit of a mess, but better then nothing. AMD doesn't guarantee/certify it until you upgrade to the Epyc. So you end up digging through motherboard docs, certification lists, and forums. My previous desktop was some early amd64, had no reliability problems with it. Before that I think it might have been an Celeron 300A that was overclocked to 450 MHz, same silicon as their non-celeron 450 MHz CPU, so overclocking was very common then. So far I've been switching between AMD and Intel every gen, and likely on my Xeon replacement.
I tend to go for Asus Prime or MSI Pro boards these days.
Are there even quality grades for boards at all? Lets say I go into the super premium high price server space and start buying POWER 10s or something, would you say that the grade on those components (the basic stuff, not talking about anything design wise or the socket or whatever, just the base component quality) is also pretty similar to budget x86 boards?
I'm basically asking if there is any reason I should care at all about motherboards outside of whatever IO/cpu socket/memory its got on it, and if that also applies outside of the consumer desktop space.
Asking because IMHO it's so easy to use multiple cores on some parts of an app I feel like everyone should do it.
Any time I tried AMD in the past (I stopped trying 10 years ago) the AMD CPUs were cheaper but there was some problem that was just enough to keep me with Intel. Weird compatibility issues when compiling packages. Huge power draw. Weird motherboards. Has this improved nowadays?
Otherwise I'm thinking one of the budget i9 parts.
AMD never fully fixed the USB dropout issues under load, AMD CPU/mobo are still extremely problematic for VR setups. There are also some weird issues around stuttering with the onboard fTPM enabled (which windows 11 wants you to use).
Better than the early days where Ryzen 1000 chips had cache defects that led to segfaults under compiling workloads.
The AGESA-BIOS model foists responsibility onto partners for distributing microcode updates and AMD tends to release buggy microcode that partners then have to fix, and they churn the microcode so fast that the fixes often go stale and break other things, but partners have to do them or people accuse them of "buggy BIOS". It's sort of the inverse problem in GPUs where AMD has to fix other people's game bugs in the drivers... partners have to fix AMD's microcode problems in their BIOSs. Commonly manifests as sensor glitches/offset problems/etc, sometimes this has led to real fun problems when those sensors are controlling core/SOC voltage or things like that.
AMD GPUs still suck for VR, the 5700XT series had huge crashing problems under the windows drivers that were never really resolved for a lot of people, and the 7000 series seems to be on course to replicate those sorts of issues. They did finally bring the 7900XTX's idle power down to only about 45W though.
The fans will vehemently deny it but yeah, you're pretty much right. AMD has gotten better but their software/drivers and user-experience still aren't as polished, either CPU or GPU. It's usually good enough it's not a problem, but you're going off the beaten path and they're not always flawless. On the other hand who knows what's going to happen with Intel with an exodus of staff already happening and likely to continue.
7900X, 7800X3D, 7900X3D, and 7950X3D are great looking parts at the right price but Intel is willing to make dealz right now to keep the cash flowing, they've got DDR4 boards to bring the platform cost down, etc. You have to evaluate the whole build and see what you get... and AMD does worse with large 128GB builds right now too. Although Intel is not flawless either to be fair.
https://www.youtube.com/watch?v=P58VqVvDjxo
Around here microcenter is willing to throw in a free 32GB RAM kit of decent quality if you buy a 7000X (not the base tier and unknown whether this will include X3D) and Newegg was offering a similar promotion at one point. Which neutralizes a lot of the platform cost difference - the DDR5 + PCIe 5.0 capable Intel boards aren't cheap either, but, a bit cheaper than AMD right now. Again, gotta look at the whole cost of the build for your own local costs.
Personally I think the DDR4 compatibility makes Intel a no-brainer for very cheap builds. You can get a PCIe 4.0 + DDR4 Intel board for the usual, expected price ranges... PCPP says starting from $79. If you're a fit for a Pentium/Celeron or a 12100/12100F that's a system cost AM5 can't compete with at the low end, you have to go back to AM4 which are slower systems.
The 12600K, 13500, and 13600K are also very good gaming-focused builds. There's nothing wrong with 7800X for gaming either, especially with the free RAM, but, Intel 13th gen does outperform it. And the e-cores are "free" at the MSRP, so Intel is simply giving you more, but, AMD is also dropping prices below MSRP lately (ymmv depending on your location).
For productivity and "server"-ish workloads, the AMD 7900X is great, particularly since AMD supports AVX-512 while Intel has disabled it on 12th/13th gen. And the 7800X3D, 7900X3D, and 7950X3D should all be very attractive high-end gaming+productivity "crossover" components if you are willing to splash out.
later update: https://www.reddit.com/r/Amd/comments/m2wqkf/updated_agesa_c...
general discussion, a couple of which are post-update:
this one is interesting because the user has traced it to specific USB controllers and notes that one controller is perfect while the other 2 have problems: https://www.reddit.com/r/Amd/comments/lrcp8q/x570_usb_dropou...
https://www.reddit.com/r/Amd/comments/p5ygve/anyone_consider...
https://www.reddit.com/r/Amd/comments/wud1xo/was_the_usb_dro...
https://www.reddit.com/r/Amd/comments/twph8r/if_youve_build_...
this is all my personal summary/analysis and not fact, but: it does not seem to be tied to a particular generation of CPU or model of chipset (and this is significant because X570 is actually completely different from every other chipset which are all OEM'd by Asmedia). The symptom is that the USB controller stops responding under load (and I think probably drops frames at that point). Since VR is a very heavy USB load (especially Oculus) this is a semi-reliable reproducer but there's nothing inherent to VR that causes the problem, I think it just seems to be a "happens under load" issue.
AMD put out an AGESA patch which seems to significantly reduce the problem. And again, I think the idea that microcode fixes the problem, and that the problem exists across both AMD OEM and Asmedia OEM chipsets, suggests it's not a hardware problem with the chipset. Actually it may be the opposite, and the problem exists with the SOC implementation of USB and the chipsets themselves are fine. Some boards use the onboard SOC to drive the USB and that sounds like maybe the problem - others drive it off the chipset's USB and maybe that works better. It's possible the CPU themselves have a hardware defect or it may be a problem with the USB software stack inside the SOC. The AMD PSP is like the Intel ME, it gets in underneath the user experience and while you'd hope that USB is not implemented there maybe it is, or maybe it's interfering (mailbox overflow?) or something like that.
Again it's hard to know whether the residual crashes are something the mobo companies themselves might be doing to try and fix AMD's problems... maybe a fix that worked before is now crashing something else. It's hard to say. That's the problem with shipping glitchy microcode to partners. A lot of partners literally have one or two core BIOS devs for the whole company. This came up during the crisis around Zen2/Zen3 BIOS support, and after elmor left ASUS, that basically ASUS really only had a couple guys for the whole damn company and they're not the only ones. And AMD in particular pushes a lot of workload onto their partners. Great that they're moving fast but they break stuff which ends up being someone else's problem, and they're churning code so fast they break the fixes other people are doing for them. They don't publish source, only binaries, and they make the partner's lives very difficult.
https://youtu.be/JluNkjdpxFo?t=1276
https://youtu.be/36kCBQt7YBk?t=237
https://www.overclock.net/search/1039852/?q=agesa&t=post&c[u...
If I was personally in this situation I would try a standalone USB card, and pick from the list curated for the Oculus rift even if you are not using a rift, I think they are just better USB controllers. Put it in a CPU-direct slot and not one attached to the chipset. See if that helps, go from there. Maybe the card works in a chipset lane, maybe the chipset lanes work natively, etc, once it's working try to pin down the delta that causes crashing.
But like... there is this whole gaslighting cycle with the AMD fans. You saw the same thing with the 5700XT. "I have one, there are no crashes, it's 100% stable for me, you're making it up" -> lots of users report it and the gaslighting intensifies (HUB finally ran a poll on 5700XT after months of denial and it turned out over 1/3 of their very technically adept and very pro-AMD userbase reported issues) -> AMD acknowledges there is actually an issue and they're working on it -> patch released and it doesn't fix the issue for everybody -> "OK so there was an issue but it's all better now, if you're still having problems it's your fault!" -> everybody moves on to the next product and if you still have problems welp sucks to be you, my 5800X3D is amazing!
Same thing happened with defective Zen2 early silicon... GamersNexus had their 3950X miss its advertised boost clocks by up to 450 MHz (they actually understated it by 100 MHz, the advertised clock was 4.7 not 4.6), basically 10% of its advertised speed. People did the "it says up to", some people went as far as making programs that just run NOPs in a loop to show that a program could actually hit the advertised clocks so it wasn't technically false advertising. AMD acknowledged an issue, released a patch that didn't help (GamersNexus' review was after the patches, they go into F10c later in the video), the fans moved back to gaslighting, and everyone moved on and forgot about it. Later silicon was higher quality I think and just didn't have a problem but AMD wanted to ship some marginal silicon that didn't quite meet spec and didn't think anybody would call them on it. And indeed they were pretty much right, the fans shrugged and moved on.
https://www.youtube.com/watch?v=M3sNUFjV7p4&t=254s
https://www.reddit.com/r/Amd/comments/cfli2n/i_discovered_ho...
Issues with Intel or NVIDIA are front-page news, everyone freaked about POSCAP-vs-MLCC back at 30-series launch, the connector at the 40-series launch, some soldering defects with EVGA 1080s, the whole New Worlds thing (even though that killed some AMD cards too, everyone treated it as an NVIDIA issue), etc, and the AMD stuff is completely downplayed and shoved under the rug even though it's pretty often not fixed, or not fixed 100%, and it's pretty aggravating.
But AMD really stokes the reddit set, they were pioneers in going direct-to-consumer and stoking the fires. GN talks about it earlier in one of those videos, and LTT and Der8auer have said the same thing. Like, it's calmed down a lot over the last couple years but during 2017-2021 it was bad.
https://youtu.be/JluNkjdpxFo?t=259
https://www.youtube.com/watch?v=x03FyPQ3a3E
https://www.youtube.com/watch?v=rscDxMrZmGU&t=1304s (this is the LTT video that got him set up with a bribe from AMD to bury his Vega FE review, the threadripper preview and the holocube, if anyone remembers that little bit of pocket-sand!) https://youtu.be/oVxVIkw0xd8?t=261
And honestly the gaslighting aspect of that really has not gone away, which I still find incredibly annoying as far as internet discourse. Say there's a problem and people get defensive and will happily generalize their own experiences to tell you yours aren't real. Just say "AMD has driver problems" and someone will pop out of the woodwork to tell you they've been good for 10 years now, as if 5700XT and Vega and Fury X just all didn't happen. Sometimes the drivers/firmware/BIOS are great, sometimes they're not, and if they're not they don't always get fixed, or don't get fixed fully for everyone.
And then yeah. Shamefully short-sighted to use the $5 tool.
In my experience, the biggest electricity hog isn't gonna be your CPU, its gonna be your display(s), if you use any. I don't use any external monitors for my work setup.
M1 are silly efficient but not 10x magical: https://www.anandtech.com/show/17024/apple-m1-max-performanc...
So no 10x, but clearly 2.5x.
Hardly comparable.
Apple has published the power figures for some of their systems, and nowhere it says that they draw or dissipate 15W [0]
Costs more to run, costs more to make, runs hotter and louder, by "smash" we're usually talking single digit performance gulfs, and performance is worse in a fairly broad amount of applications and where Intel's performance IS better it often is in an application where you would never be able to tell the difference without measurement tools.
The 12400 "kicking the m1 in the teeth" is likely due more to software optimisation than the 12400 actually being meaningfully better.
Your like of Intel's products as a consumer isn't really a reflection of their competitiveness. The only reason you're willing to buy their products at all is that Intel is cutting their margins to the bone whereas AMD is making stacks on every sale. Intel has had the best single core performance for eons and that hasn't saved them from their decline.
To be clear I'm no loyalist. I went from an AMD 3700X to a 2020 M1 Mac Mini originally because that spanked it. Then to an M1 Pro which was regrettably no real improvement. Then to a 12400 because I had one as a comparison point and it was faster. Now likely a 13700K. Next time, it might be AMD, might be an M3. Who knows?!
I'm mostly just glad I can jump around and pick whatever is on the market rather than being locked into one product or platform now.
The next chance I am giving intel is relatively far into the future. Likely in 5-6 years at best given how many stability issues I have had.
Do note that this is pure anecdata, n=1 and likely suffering from a degree of confirmation bias.
PopOS has had a surprisingly high number of instability issues and random freezes.
I considered the possibility that the build is the problem, but it’s the same nvidia LTS 22.02 build that I used with the 5800x.
Windows runs fine.
If I didn’t rely on cuda, I’d probably use github’s code spaces for programming and just stay in windows, or run a VM.
Is by any chance your disk encrypted? That's the only (major) reason nowadays why I hear about freezes in Linux.
It's possible that a lot of this in Pop_OS is related to wayland and gnome. The previous machine run a 5800x with a 1080ti, this runs a 12700k with a 3090ti, same distro otherwise.
The wattage difference between AMD and Intel cpus is not just a matter of feeding it more electricity. Heat dissipation, air flow if air cooling, or reservoir capacity if water cooling, and control of ambient temperatures matters a lot.
Then, one also needs adequate power delivery to deal with stability issues.
You can build a suitably quiet PC. Be Quiet power supply, case, fans and cooler. Job done. They run very quiet. I had a 125W TDP CPU in one of them and you could barely hear it under load.
I have 16 cores at 100% right now as I type this and there's no fan noise at all. I'm parsing 10,000+ php files and modifying them for an upgrade. I've done this before and the Mac makes a lot of noise.
MBP is good but it's not all there is. This is a AMD Ryzen 7 5700U with a dual fan setup. It's got performance and efficiency cores similar to the newer Apple chips.
Just throwing it out there! I hope more people support great laptops like this one.
And the Macbook Air doesn't even have a fan.
I also spent about $2,500 less than the option with 64GB of RAM from Apple.
On that note, a lot of laptops have goofy fan curves. In theory, you want the fan to move just enough air that the CPU doesn't throttle, since the rate of heat transfer is proportional to the heatsink-to-air temperature difference.
And if you factor in power consumption at EU prices?
TCO over 5 years for Apple is £3838 (outlay) + £247 (electricity) = £68.08/month
TCO over 5 years for PC is £1045 (outlay) + £405 (electricity) = £24.17/month
That's shit change really but the cost difference is immense.
Also the PC value has gone up since I bought it as I got it on a one day offer. I can shift it for £120 more than I paid for it. The mac, not so much.
The Mac depreciates in the first 18 months more than the entire PC set up and electricity costs.
honestly it's impressive that you gratuitously threw in a 1700GBP display on one side of the scale and the cost difference was still only 36% lol.
Also just because the retail price has gone up doesn't mean you can actually sell the PC for the retail price. Resale value of x86 laptops is garbage... it's better for desktops of course especially if you can retain pieces of it, but this is still a super loaded comparison and even then Apple only came out 36% more expensive according to your numbers?
I’m just reporting where I am not how I got there (which was mostly me going “ooh shiny”)
"Smashing" ST performance while consuming 150w is not an ideal situation for anyone but gamers and benchmarkers.
The AMD server is at full load, using around 150W whereas the Intel server idles at 450W doing absolutely nothing and pushes up to 700W if we load it up.
In hindsight, I regret not just buying 3 of the AMD servers for the same price. I had this fallacy that the Intel server would be magically more energy efficient, but that has turned out to be FUD and baloney.
NB Power usage measured at the wall socket with voltage and current monitored PDUs.
[0] https://www.anandtech.com/show/16594/intel-3rd-gen-xeon-scal...
And this AMD system is using apparently a peak of 3.75w per core? really?
But as the other commenter said, the actual per core usage will vary /wildly/ even under high load. You would half to run something fairly specialized such as prime95, which can run entirely from cache, to get more accurate per core wattage values under same load maximum load.
As it stands, my i9-13900K seems to have an idle “package” draw of around 10-12w and can peak at 360w under prime95 load. And that’s stock. I can make it ramp way past 360W with very little effort.
But under “normal” desktop usage it happily sits well under 100w.
So… shrugs. I would like to know more on this parent’s setup instead of “My server with literally 40% of the core count uses 1/6 of the power!”
But we may not be so lucky.
My commentary is only really about what we believed about AMD before making our purchase (internet forums like this one warn about energy use), how that wasn't really the case when we measured it on a real load, and why we 'feel' we would prefer to go the smaller and cheaper AMD server route next time.
Also everything you plug into it will eat power, some 10gbe stuff and SAS controllers/etc are not lightweights in idle because it's not expected they'll be idle. Every HDD generally pulls about 5W when idle+spinning, even SSDs are a couple watts apiece.
That said, yeah, that sounds way high for idle. ServeTheHome is a good place for stuff like this:
> motherboard: X11SPM-F
> memory: 6x 8GB RDIMM 2666Mt
> kill-a-watt power on the wall at idle: 30..35Watts.
https://forums.servethehome.com/index.php?threads/xeon-scala...
Single-socket though, but, another CPU probably doesn't add more than another 30 watts or so.
Epyc actually tends to be worse on idle power because it has to run all its infinity fabric links at all times, if it wants to access that memory, and data movement is comparatively expensive. There were some early Milan/Zen3 epyc numbers that need to be disregarded due to problems with AMD's reference implementation motherboard, but, you can see from that thread it's not uncommon for Epyc to idle at 100W for a single-socket board. You can see from that thread that the consensus is Skylake-SP generally idles better.
My own personal anecdote: I am tinkering with a dual-socket 2011-3 server right now (X10DRC-T4+) with a pair of 2697v3 and 2x4x32GB RDIMMs and quad 10gbase-t NICs, at idle I am around 75W at the wall... obviously Skylake-SP has a larger platform (it's the socket after mine) but it's not 350W bigger, guy has something wrong with his setup.
It is definitely fair though that under load the Epycs are a lot more energy-efficient. Skylake-SP is still 14nm and Intel lingered there for far too long, 7nm is more than a full node better and Milan has higher IPC, higher clocks, and more cores. Ice Lake-SP was better than Skylake, but nobody bought it because it was years late to market and AMD was already better. So you are effectively comparing 2016-ish Intel stuff against AMD stuff that is 4+ years newer... but on the other hand that's the Intel stuff that is deployed right now, and people generally want to minimize the amount of generations in their clusters to make it easier to operate VMWare vMotion etc. /shruggy
Going "I got X and Y and with no further information X is so much better than Y!" online gets so fucking old.
And I am genuinely curious. Not just being antagonistic. I am looking at some newer servers soon and have been debating Intel or AMD and I like seeing actual reallife results people have had.
None of them are rated at 2.4Ghz. The EPYC 7351P 2.40Ghz has a mere 16 cores (32 threads) and is rated at 170W TDP.
You wouldn't buy a dual socket box for 3x the money if you didn't think you'd get 3x the performance out of it, unless you really wanted to keep your server count down.
We kind of expected that scaling HW with more CPUs instead of more entire rack systems would be more energy efficient, but I don’t think it makes too much of a difference compared to other factors like CPU/vendor architecture.
FWIW our energy bill is $1000+ per month just from servers, and then several $k for various licensing on each host per year, so there was some sweet spot we were optimising for (low power, but less licensing cost via less racks). Our cost in the cloud would be $5k per month or more, so either way we still saved money this year.
* still 450W at idle isn't only due to dual CPU
* 150W for full load, really?
The idle case on the Intel server is RHEL OS running several Windows VMs that are doing nothing, and the loaded case is when we spin up a large VM with around 40 cores on that host and run something in MATLAB (essentially a similar job to what the runners would otherwise be running).
The comparison is not completely accurate, but our experience was only anecdotal.
You're measuring wall socket voltage on two different servers to make a commentary on CPU power usage, but that's including storage, power delivery efficiency, cooling solutions, etc
You didn't describe what "we load it up" means for the Intel server, or how that compares to the AMD server, or what load is for that even (you're saying you keep the AMD server pegged at 100% CPU, or you have 100% of your expected load?). Presumably you didn't get a dual socket CPU for giggles, so the load isn't trivially comparable between both systems by either measure.
If you're buying servers because of fallacies about entire brands, please leave the procurement to someone more experienced. That kind of stuff is more for building a gaming PC, server hardware is way too varied for "brand X is Y". You need to examine things at the SKU level.
I think the pendulum has begun to swing the other way. Intel is catching up and surpassing in some important areas already.
I've held positions both in AMD and Intel over the years. Right now I am long INTC.
On a practical level, there are a ridiculous amount of games, simulation, scientific and engineering software that falls into that category and asking people to re-write tens of thousands of lines of code to take advantage of a new architecture isn't happening.
That overstates how bad it is. Over time, it feels like they slipped to being a year or two behind in both fab and chip design, and it didn't become apparent until ~3 years ago. The reason it's not that bad is Intel rushed out some chips that are competitive with AMD chips, and despite their process difficulties, there are still only a handful of companies that can make these chips.
What's actually bad is that they lost Apple. It motivated a lot of places (including Microsoft) to seriously look into ARM.
Objective factors: Much more reliable overall with significantly less jank compared to AMD, especially on the software side of things. Much better power draw at idle and low use. Onboard GPU for practically every CPU, with only very explicit exceptions; great for video encoding (QuickSync!) or when shit hits the fan. Better single thread performance compared to AMD for no personally-practical loss in multithread performance.
Geopolitical factors: Less reliance on the Taiwan Singular Egg Basket(tm) problem.
Personal factors: Sentimental value, I grew up with Intel CPUs and I owe a lot of who I am today to them; I have an Intel Inside. The Intel chime is up there with the Windows 95 and 98 boot chimes for defining my 20th century.
1) pay more for talent than anyone else.
2) treat your engineering staff amazingly well
3) watch your business completely turn around in two years
You can lean hard into a certain flavor of this. But it means orienting internal incentives away from excellence of outcome. So it gets you modern Google (allegedly):
https://medium.com/@pravse/the-maze-is-in-the-mouse-980c57cf...
Quickly, proper implementation of simple principles becomes complex.
It seems /a lot/ of the people I know are moving from Intel/AMD to ARM, I suspect the move is bigger than I expect. So I'm wondering if that's chipping away at Intel/AMDs market share.
From what I've seen, poorly - Gravitons are significantly cheaper (10-30%) for similar performance, workload dependant*.
So Intel is losing chipset share, too.
Another option is if AMD decided to get into the mobile / embedded SoC market for real, and didn't want to pay an ARM license to do it.
https://github.com/riscv/riscv-p-spec/blob/master/P-ext-prop...
See section 3.2.
The instruction set is specified in a 'modular' fashion and has standards for various extensions, including compact instructions, vector extensions, etc. Compilers like GCC and LLVM support targeting these extensions.
I'm not saying AMD should make a RISC-V CPU, but if they were to hypothetically do that, they'd clearly be including a pile of the ratified extensions.
No, not really. The large specialized instructions in x86 run across many, many clock cycles. And IIRC, compilers will often generate a sequence of simple instructions rather than one complex instruction, because the CPUs are so much better at executing sequences of simple instructions.
(And yeah, as others have pointed out, of course high end RISC-V chips have multiply and divide instructions with dedicated multiply and divide hardware.)
To me, the more interesting question is what does the overall market of compute devices that can be used in a fashion similar to what AMD and Intel sell (being charitable and leaving out mobile) look like and what percentage of that market do AMD and Intel make up? Has the overall market grown? Is it growing faster overall than the laptop/desktop/server/hpc space that intel and AMD are catering to?
But I also heard somewhere that Intel, until recently, refused to buy ASML's EUV machines. Trying to go it mostly alone on EUV lithography.
Are both of these facts really true?
Seems insane that Intel would help ASML develop EUV and then watch competitors buy the machines and reap the benefits.
Now however Intel has low power cpu cores and high power cpu cores and I feel cheated. AMD has all full power cores.
- DRAM price fixing and limiting: we should probably have 4x more memory or more for the price we pay these days
- SSD price fixing and limiting: we should probably have 2x more SSD capacity per cost
- ARM in PC is still being foot-dragged. I have no faith that Intel can make an M1/M2 competitor or killer, but AMD, Samsung, and Qualcomm should be able to do it. I mean, Amazon/Aws have their own ARM chips.
- compact low power PCs like raspberry-pi type form factors (or the PC stick, or similar) are ripe for the taking. I have tower case PCs from 5-10 years ago that should be replaceable with a 3"x3" cube design, or imagine a base power supply where you take some sort of FrameworkPC-style headless notebook blades and they sit vertically on the base. The ports all face you. If it was a great multiple-4k display super-KVM, heck yes.
- AMD, Qualcomm, and Samsung should be pushing a linux desktop initiative. Microsoft and Windows are really holding back PC units with their awful OS releases. The disaster of Windows 8 has not really been fixed, and they frankly don't care. A better-than-OSX linux is out there and it is low hanging fruit.
The PC market is so ripe for major innovation. It is stuck with crappy motherboard and case form factors.
Intel has ALWAYS sucked at platform leadership. AMD (and/or Samsung and/or Qualcomm) don't have to be. I really hope if AMD hits 50% at some point they can push more platforms to create new markets and segments for their processors and graphics chips. Intel is beyond saving, they have never understood platforms, software, and their management is moribund in penny pinching and monopoly rent extraction.
Apple still treats the PC market as an afterthought. They have the moxie and hardware/software to do something, but they aren't open enough.
Qualcomm might simply be too OEM-focused to show this kind of leadership, and they'd need a major ARM PC CPU, and that is five years off.
Samsung is too samsung ecosystem focused. They are not open enough.
AMD is, for better or worse, the one with the market position and innovation ability to create new PC opportunities. They don't have the subservient history to Microsoft. They have more EU/world focus, and those markets should be interested/open to non-Microsoft computing.
And that is the path to long term market relevance. Intel may still come back, and they have pseudomonopoly powers and relationships with big business to relegate AMD. AMD needs to continue execution in x86, but do what I described: ARM/Apple M1/M2 competitor, Linux/Steamdeck desktop OS initiatives, and push more revolutionary small form factor computing platforms. It will hedge against Intel's monopoly powers and open new markets that Intel has shown utter inability to compete in historically.
And frankly, with the "Moores Law" family of macro-trends stalling more and more, the path to more CPU and GPU sales isn't just a faster better chip. It's getting chips into more packages, places, and uses that people currently have them. Form factors and packaging are going to make MORE of a difference for upgrading. Right now, if I could get some sort of fancy framework-style compact notebook-that-is-a-thin-but-upgradable-desktop computer in a neat package, that's a much more compelling buy than a machine that has, what, maybe real world 10% PERCEPTIBLE performance improvement over something from 5 years ago?
Realistically this should have been in full swing with the COVID boom. Not sure if the penny pinchers will allow such "radical" investment (the non-ARM initiatives might take, what a billion a year froma 26 billion a year company?)
> [...] AMD now has a market share of 31.3% (up from 28.5% in Q4 2021) versus Intel's 68.7% (down from 71.5%)
That's not "grabbing over 30%" in my book. It's grabbing 2.8%.
We're always having this same rehashed argument over and over again on HN. "My [CPU manufacturer] grew by 100% this year." Market share is an absolute percentage, so you should talk about it using absolutes. It doesn't make sense to talk about relative increases unless we're talking only about AMD revenue without respect to the overall market.
But even doing the percentage difference, I calculate a 10.2 percent increase.
So I'll just shut up and go face the corner with the parent poster.
"I grabbed four slices of pizza"
"dangus grabs 50% of the pizza"
Just underclock/undervolt, problem solved.
Did you by chance read an Intel i9 review? Yes, Intel's going crazy on the power front trying to match the AMD performance, which of course lead to AMD pushing it a bit as well. But the new non-X AMD CPUs are quite power efficient for the performance.
If that's still to much the M2 or M2 Pro in a mac mini is rather efficient on power and still performance quite well.
So yeah, 100W or 65W TDP is not much at all.
I have a 27" Dell U2719DC from 2 years ago, pretty average monitor, 2560x1440, includes a USB hub, and has a USB-c ports for charging other widgets.
On a bright screen, like full screen tab of HN I get 22 watts. If I have a darker screen like anything full screen in dark/night mode (white text on black background) I get 20 watts. When I touch the screen or top edge it's pretty close to room temperature.
To be fair my work configuration uses two of the 27" monitors.
I'm measuring with a kill-a-watt, which has a decent reputation for accuracy.
https://everymac.com/monitors/apple/studio_cinema/specs/appl... the 30" Cinema Display from 2004 consumed 150W. The IBM "Big Bertha" T220 in 2001 at 20" consumed 111W. So that's the era when I'd expect a 27" to consume 80W or even more.
By 2010 some 27" displays like the Sceptre X270W consumed less than 45W. https://images.anandtech.com/graphs/apple27inchcinemadisplay...
only if you disable boost. 88W PPT for AMD's "65W" CPUs at stock.
not a fan of the "PPT" terminology, it's pretty deliberately designed to mislead people and give AMD a mental "cheat factor" in these comparisons.
And no this is not something that "everyone did", in the 5960X/7700K days Intel's TDP values were expected to cover boost clocks and AVX as well, it's only 8700K/9900K/etc where Intel started to play fast and loose with their TDP figures. Which was after the introduction of Ryzen and the "PPT" terminology... in those days AMD pulled more than Intel's chips and they needed a way to hide it. Then they took the lead on efficiency.
kinda like TSMC naming their chips "7nm" and that resulting in huge mindshare problems for Intel even though 10ESF/Intel 7 (third-gen 10nm) was actually better than N7. Or Samung 8nm not even being as good as first-gen Intel 10nm, lol. So Intel had to rename their nodes too. A few bad actors ruin it for everyone.
88W is the number that matters for your load power measurement comparison - again, unless you disable boost.
The "X" chips are also clocked higher... 7950X would be 170W with boost disabled and runs 220W at stock.
There is some new inefficiency on AMD builds now that the Core Chiplet Die (CCD) and IO Die (IOD) are separate units, not monolithic. DDR5 also taked more power. The fancier X670 chipset (really, two chipset dies) also takes power & could very much use some tuning, but it's also doing a beastly amount of IO work, in some cases for good cause (more low power capabilities/better idle would be appreciated though! One of AMD's first in house chipsets in a long time).
We've barely started exploring how well behaved we can make the just launched 65w ryzen parts... but it wouldnt be shocking to me if there's not really any advantage, if they literally end up being nearly the same part, rebadged & preconfigured to lower power. Is there even really going to be any binning to have selected these? You know maybe not!
Out of the box most chips come tuned for speed, but ask them to be a little nixer, turn on eco mode or power the power point, they are stunningingly efficient. Alas idle could use some tuning; staying under 40w idle is harder than it ought to be.