The Snapdragon 855's iGPU
chipsandcheese.com
chipsandcheese.com
When we're talking about new development, what are the expectations that I should have for direct procurement of higher end components (this goes for Sony/Samsung camera sensors too)? Are there typically very large minimum purchase quantities? Do I have to already have a large production output before I can even talk to them about it? Is it possible to get datasheets even if I'm not at that stage yet?
I get the feeling I'm stuck with the Jetson + Off-the-shelf camera approach until I can demonstrate the ability to mass produce and sell my existing designs, but it would be nice if I could find out more about how that is supposed to play out.
If you're hoping to commercialize and are a 1-man company, I hope you have a good USP/feature that's beyond "it's cheaper than the competition". If you do have that, I'd say just get an MVP out while keeping systems as platform agnostic as you can. Doesn't matter if you need an extra Jetson, if your product is good enough the extra cost shouldn't really deter customers.
Once you've started, hire/contract folks and streamline it out.
BTW Dev kits do exist for the snapdragon 8 gen 2/3 I think, but they're about $700 last I checked. If you're handy you could even try to reuse a rooted phone, if that's feasible for you.
The CPU included in this is based on the same Cortex-A78 cores as the latest Jetson Orin, but it should have much better performance per dollar, because the cheaper Jetson variants are intentionally crippled, with many disabled cores and very low maximum clock frequency.
The GPU is weaker than on Jetson Orin, but it should be enough for many applications. There are also a couple of AI accelerators and dual simultaneous video camera capture at high resolution.
In any case, Genio 1200 is much faster than any Pi and also than RK3588. It has a similar performance to a flagship smartphone of 2021.
In the beginning, the SOMs and SBCs with Genio 1200 were expensive, in the $400 to $500 range, so while they were a better deal than Jetsons they still had worse performance per dollar than a SOM or SBC made with an Intel N100 CPU (or its Atom equivalent).
However not long ago Radxa NIO 12L has been announced, which uses Genio 1200 in a SBC of smartphone size at a price of not much over $100, similar to the older and slower RK3588.
Therefore it appears that it has become possible to obtain Genio 1200 at a decent price.
Perhaps if you contact MediaTek sales they might be more willing to provide design information than Qualcomm and they certainly have better prices.
Moreover the Radxa NIO 12L schematics and PCB assembly map are published on their site (a practice that was standard many decades ago, including for the leading US companies, but which now sadly survives mostly in Chinese companies), you can download them and study them to evaluate the effort of designing your custom board and possibly avoid any pitfalls in your design.
https://www.modalai.com/products/voxl-2-mini?variant=4424638...
Maybe this is something you could develop on instead of building your own.
The next few years will be very interesting.
I genuinely can't figure out if Apple has a point, or is doing something quirky so they can overcharge outsiders.
Given 0 other companies are really investing resources and Nvidia is the giant, I'd guess its the later. I see too many reddit posts about disappointed CPU LLM users who wanted to run large models.
Well, Apple was already soldering the memory onto the motherboard in most of their Intel machines, so moving the memory on-package didn't really change their ability to gouge for memory upgrades much. They were already doing that.
AMD is coming out with Strix Halo soon with a 256bit memory bus, on the high end we're also seen the niche ampere platform running Arm CPUs with 576-bit buses. PS5 uses a 256-bit bus and Series X a 320-bit bus, but they're using GDDR instead of DDR which increases costs and latency to optimise for bandwidth but there's no reason you couldn't design a laptop or steam deck that did the same thing. AMD has their MI300X which is using 192gb HBM3 over a 8192-bit bus.
I don't think it's just apple going this way, and I do think that more and more of the market is going to be using this unified approach instead of the approach of having a processor and coprocessor separated over a tiny PCIe bus with separate DDR/GDDR memory pools. With portable devices especially, more and more every year I struggle to see how the architecture is justified when I see the battery life benefits of ridding yourself of all this extra hardware. LLM inferencing also creates a nice incentive to go APU because the dual processor architecture tends to result in excessive bandwidth and anemic capacity.
Nvidia may be the giant yet if you look at what most gamers actually run on, they run on Apple APUs, Snapdragon APUs, and AMD APUs. PCs with a discrete CPU and separate Nvidia GPUs have become a relatively smaller slice of the market despite that market segment having grown and Nvidia having a stranglehold upon it. The average game developed is not developed to run on a separate CPU and GPU monster beast - they're designed to run on a phone APU - and something like a Max processor is much more powerful than required to run the average game coming out.
They really don't? You're trying quite hard to conflate casual gaming with console and PC markets, but they obviously have very little overlap. Games that release for Nvidia and AMD systems almost never turn around and port themselves to Apple or Snapdragon platforms. I'd imagine the people calling themselves gamers aren't referring to their Candy Crush streak on iPhone.
> something like a Max processor is much more powerful than required to run the average game coming out.
...well, yeah. And then the laptop 4080 is some 2 times faster than that: https://browser.geekbench.com/opencl-benchmarks
I just see the distinction you're drawing as being arbitrary and old-fashioned and misses the huge rise of midcore gaming which is seeing tons of mobile/console/pc releases. I understand that a TRUE gamer would not be caught dead playing such games, but as more and more people end up buying APU based laptops to play their hoyaverse games, that's going to warp the market and cause the oppressed minority of TRUE gamers to buy the same products due to economies of scale.
The console/PC release cycle is just different. Some stuff is cross-platform (particularly when Apple goes out of their way to negotiate with the publisher), but most stuff is not. It's not even a Steam Deck situation where Apple is working to support games regardless; they simply don't care. Additionally, the majority of these cross-platform releases aren't quality experiences but gatcha games, gambling apps and subscription services. You're not wrong to perceive mobile gaming as a high-value market, but it's on a completely different level from other platforms regardless. If you watch a console/PC gaming showcase nowadays, you'd be lucky to find even a single game that is supported on iOS and Android.
> so why not do a cross-platform release if it's viable?
Some companies do; Valve famously went through a lot of work porting their games to MacOS, before Apple depreciated the graphics API they used and cut off 32-bit library support. By the looks of it, Valve and many others just shrug and ignore Apple's update treadmill altogether. There's no shortage of iOS games I played on my first-gen iPod that are flat-out depreciated on today's hardware. Meanwhile the games I bought on Steam in 2011 still run just fine today.
Besides, it's ultimately irrelevant because when Strix Halo comes out, it's going to have the memory bandwidth and compute performance to be able to play any "AAA" game released for consoles until consoles refresh around ~2028, which is 4 solid years of performance before new releases will really make them struggle. These APUs won't be competing with the 4080, but instead the 4060, which is a more popular product anyways. Discrete GPUs are in an awkward spot where they're not going to be significantly more future proof than an APU you can buy, but will suck more power, and will likely have a higher BOM to manufacture.
If you asked TRUE gamers if gaming laptop with Nvidia GPUs were worth it a few years ago, when they were already the majority of the market, they would have laughed in your face and pointed out how they didn't play the latest AAA games good and thus TRUE gamers won't buy them and to instead buy a cheap laptop paired with a big desktop.
It's really the opposite; I think obsessing over casual markets is a mistake since casual gaming customers are incidental. These are people playing the lowest-common-denominator, ad-ridden, microtransaction-laden apps that fill out the App Store, not Halo 2 or Space Cadet Pinball. It really doesn't matter when the games came out, because the market is always separated by more than just third-party ambivalence. Apple loves this unconscious traffic, because they will buy any garbage they put in front of them. Let them be gorged on Honkai Star Rail, while Apple counts 30% of their expenses on digital vice.
Again, I think it's less of a distinction between "true" and "casual" gamers, but more what their OEM encourages them to play. When you owned feature phones, it was shitty Java applets. Now that you own an iPhone... it's basically the same thing with a shinier UI and larger buttons to enter your credit card details.
I'll just say it; Apple's runtime has to play catch-up with stuff like the Steam Deck and even modern game consoles. The current piecemeal porting attempts are pathetic compared to businesses a fraction their size. Even Nvidia got more people to port to the Shield TV, and that was a failure from the start.
It is as if the casual true dichotomy is a false one.
I love how performant mobile games are on desktop/laptop hardware assuming good ports, Star Rail and Princess Connect! Re:Dive for some examples.
This will probably go away once mobile hardware gets so powerful there's no requirement for devs to be efficient with their resource usage, as has happened with desktop/laptop software, but god damnit I'll enjoy it while it lasts.
Wat
This is like extremely incorrect.
Its Nvidia.
Yikes, your technoblabble at the start seemed like smart people talk but the meat and potatoes is off base.
[1] https://helplama.com/wp-content/uploads/2023/02/history-of-g...
Also, much of mobile gaming isn't exactly pushing the envelope of graphics, especially by revenue.
This seems like some way to be like 'Well tecknicaklly', to justify some absurd argument that doesnt matter to anyone.
Who's "we" because big tech has absolutely been touting GPU gains in their products for a long time now [1], driven by gaming. Top of the line iPhones can do raytracing now, and are getting AAA ports like Resident Evil.
In what world is being over half of a 185B industry a technicality?. A lot of these advancements on mobile end up trickling up to their laptop/desktop counterparts (See Apple's M-series), which matters to non-mobile gamers as well. Advancements that wouldn't have happened if the money wasn't there.
[1] https://crucialtangent.files.wordpress.com/2013/10/iphone5s-...
It's very expensive to have a bus that wide, which is why it's so rarely done. Desktop GPUs have done it in the past ( https://www.techpowerup.com/gpu-specs/?buswidth=512%20bit&so... ), but they all keep pulling back from it because it's too expensive.
Apple can do it because they can just pay for it and know they can charge for it, they aren't really competing with anyone. But the M3 Max is also a stonking huge chip - at 92bn transistors it's significantly bigger than the RTX 4090 (76bn transistors). Was a 512-bit bus really a good use of those transistors? Probably not. Will others do it? Probably also no, they need to be more efficient on silicon usage. Especially as node shrinks provide less & less benefit yet cost increasingly more.
M3 pro only used a mere 37 billion transistors with a 192-bit bus, so you can get wider than 128-bit while being economical about it. I'd love for there to be a 512-bit Strix Halo but it probably won't happen, it probably does not make business sense.
I don't know if the comparison to GPUs necessarily tracks here because the price floor of having 8 chips of GDDR is a lot higher than having 8 chips of DDR.
Similarly, ultra high end APUs like the Xeon Max are doing on-package HBM for truly monstrous memory bandwidth numbers.
The thing is that nobody else is really trying to make anything like an M3 Pro, and I don't know if they even will. It's a really weird product. Big dies are expensive, hence everyone pushing towards chiplets. A really great spot to split up dies is between compute units that are largely independent - which the CPU & GPU actually are. There's a few workloads where unified memory helps, but most don't. So splitting those apart makes a ton of sense still. Then also if you push them hard to squeeze out all the performance you can, they both get very hot - so you want them physically far apart for cooling reasons. At which point you might as well just give them their own memory which can also be further specialized for their respective needs as one is latency biased and the other bandwidth biased. And now you're just back to traditional desktop architecture - it still makes just way too much sense for the high end.
It makes sense for Apple since they focus almost exclusively on laptops and you get exactly the single mix of CPU & GPU they decide, just like it does for consoles where again they have a midrange power budget & a single CPU/GPU configuration over millions of units. But as soon as different product specializations show up and different workload demands are catered to, coupling the CPU & GPU like that kinda doesn't make sense?
Well, AMD is indeed making their own M3 Pro, the Strix Halo. Which is going to achieve 4050/4060 performance in laptops that will again be cheaper and more power efficient for a lack of a Nvidia GPU. This is also a net addition to their lineup.
AMD is a minority player in the laptop market, so it doesn't necessarily make sense for them to compete hard against Intel at every price segment, and they have clear competitive advantage and counter-positioning here against Intel/Nvidia/Apple.
The speed of light isn't changing, nor is EMI. Upgradability is directly at odds with high-speed, RF-like, timing-sensitive protocols. "Planned obsolescence!!!" - no: electrical reality.
Upgrades are not really environmentally good. They might be the best compromise, but the parts you take off still are discarded.
1) reflow the board with a steady hand
2) bin it
The overwhelming majority of customers will not choose option 1.
High-end PC gaming is still a small niche. Every new and more powerful gaming console does worse than the previous generation. VR hasn't taken off to a significant degree.
Now I’m stuck with an old CPU, and have little reason to upgrade the GPU.
I guess my main hope is it’s modular and seamless to upgrade small frame machines in the future, so I can keep a case standard for generations.
You can use sysprep[1], though nowadays you don't even need to do that most of the time.
[1]: https://learn.microsoft.com/en-us/windows-hardware/manufactu...
The video card goes the down the same road as the sound card.
Given how expensive video cards have gotten thanks first to cryptocurrency mining and now "AI", it's only a matter of time.
It feels like the next few years would be really glorious for handheld gaming...
Hell yeah, we celebrate.
They won't with the mainstream number of memory channels, where iGPUs are severely bandwidth starved. AMD Strix Halo wants to change that, but then the cost difference to a discrete GPU gets smaller.
They might cannibalise even more of the low-end market, but I'm not sure they will make mid-range (or high-end) dGPUs obsolete.
Mobile phones have taught us another annoying lesson though: edge compute will forever be under-utilized. I would go so far as to say now a huge proportion of the mainstream audience are simply not impressed by any advances in this area, the noise is almost entirely just developers.
The writing is on the wall for me. A cheap streaming device + a modest games subscription will open the platform up to many who are currently held back by the cost of the console.
If car makers gave up with EVs 30 years ago, and never tried improving, where would we be today? Because EVs are not the ICE replacement today, does that mean it won't be in 30 years' time?
All I'm saying is that the natural outcome for gaming is cloud + subscription. Maybe not the next xbox console, but possibly the one after next.
This was the inevitable outcome of music and video. Gaming is next.
We are going to own nothing and we will love it.
That's collective action though. No one wants to do that.
- You have to buy all the games all over again
- people, that are familiar with google ways, didn't want to invest anything into stadia
- stadia controller, while nice, didn't want with anything, but stadia until recently (still doesn't work with ATV for some reason)
- Google own devices didn't support Stadia for absolutely no reason (you could have side-loaded the app, and it worked just fine)
xCloud always felt much better than Stadia when playing.
The bad part is needing to live close to a big, expensive, GPU-heavy data-center, but if mobile GPU's are getting that good, maybe in the near future we could have single servers or small racks serving a few dozen to a few hundred players, that would be much easier to co-locate, grow as needed, and improve latency.
The biggest issues is probably long load times. A suspend state to drive function would be a big boost.
Then it’s probably the oversold capacity.
Then it would be only 80% compatibility with all titles.
Cosmetic physics can eat up more, but as a gameplay mechanic, user interaction with realistic physics are too unpredictable (beyond a gimmick). Of course, we could then get actual computer sports.
Very few (if any) mobile class GPUs actually support true preemption. Rather they are more like pseudo-cooperative, with suspend checks in between work units on the GPU. Desktop GPUs only got instruction-level preemption not that long ago - Nvidia first added it with Pascal (GTX 10xx), so mobile still lacking this isn't surprising. It's a big cost to pay for a relatively niche problem.
So the "crash" was probably a watchdog firing for failing to make forward progress at a sufficient rate and also why the screen would freeze. The smallest work unit was "too big" and so it never would yield to other tasks.
And even then it's often limited to already scheduled shaders in the queue - things like the register files being statically allocated at task schedule time means you can't just "add" a task, and removing a task is expensive as you need to suspend everything, store off the (often pretty large) register state and any used shader local data (or similar), stop that task and deallocate the shared resources. It's avoided for good reason, and even if it's supported likely a rather untested buggy path.
If you run an infinite loop on even the latest Nvidia GPU (with enough instances to saturate the hardware) you can still get "hangs", as it ends up blocking things like composition until the driver kills the task. It's still nowhere near the experience CPU task preemption gives you.
Snapdragon 855 was indeed the chipset used in most flagship Android smartphones of 2019. I have one in my ASUS ZenFone.
Based on the big ARM cores that are included, it is easy to determine the manufacturing year of a flagship smartphone. The cheaper smartphones can continue to use older cores. Arm announces a core one year before it becomes used in smartphones.
Cortex-A72 => 2016
Cortex-A73 => 2017
Cortex-A75 => 2018
Cortex-A76 => 2019 (like in the tested Snapdragon 855)
Cortex-A77 => 2020
Cortex-X1 => 2021
Cortex-X2 => 2022
Cortex-X3 => 2023
Cortex-X4 => 2024