[HN Gopher] The Snapdragon X Elite's Adreno iGPU
       ___________________________________________________________________
        
       The Snapdragon X Elite's Adreno iGPU
        
       Author : pella
       Score  : 155 points
       Date   : 2024-07-04 21:33 UTC (1 days ago)
        
 (HTM) web link (chipsandcheese.com)
 (TXT) w3m dump (chipsandcheese.com)
        
       | dagmx wrote:
       | It's been interesting seeing the difference in architecture play
       | out in benchmarks.
       | 
       | For context, there was a lot of hullabaloo a while ago when the
       | Adreno 730 was posting super impressive benchmarks, outpacing
       | Apple's GPU and putting up a good fight against AMD and NVIDIA's
       | lower/mid range cards.
       | 
       | Since then, with the Snapdragon X, there's been a bit of a
       | deflation which has shown the lead flip dramatically when
       | targeting more modern graphics loads. The Adreno now ranks behind
       | the others when it comes to benchmarks that reflect desktop
       | gaming, including being behind Apple's GPU.
       | 
       | It'll be interesting to see how Qualcomm moves forward with newer
       | GPU architectures. Whether they'll sacrifice their mobile lead in
       | the pursuit of gaining ground for higher end gaming.
        
         | swatcoder wrote:
         | > Whether they'll sacrifice their mobile lead in the pursuit of
         | gaining ground for higher end gaming.
         | 
         | It's hard to imagine why they'd distract themselves with that,
         | except perhaps with a token small-run demo for pure brand
         | marketing purposes.
         | 
         | Because of Apple's strict vertical integration, there's _so
         | much_ market for them as the de facto manufacturer delivering
         | parts to pretty much every competitor making products that want
         | a high performance /power ratio.
        
           | dagmx wrote:
           | Well it depends which space they want to be taken seriously
           | in. Currently the 741 is very poor when compared to any dGPU
           | or Apple. It only favourably compares to iGPUs.
           | 
           | I believe they have three options
           | 
           | 1. Say it's meant to be like an iGPU, and work on supporting
           | dGPUs to complement it.
           | 
           | 2. Say that they want to compete with dGPUs/Apple and risk
           | losing their mobile crown. Which is what Apple did in
           | exchange for one design across all products.
           | 
           | 3. Say they want to have a split product portfolio. A more
           | desktop focused GPU for Snapdragon X with a more mobile
           | centric one for 8xx
        
             | silisili wrote:
             | I think it's going to be 3, but a split between mobile and
             | laptop/desktop without any concern for competing with
             | dGPUs. It makes no sense at all for them to.
             | 
             | If they can give good enough, on par or better with current
             | iGPUs, with a lower power usage and potentially even
             | fanless, they're going to sell a billion of them. They'll
             | be in every Chromebook in the world.
        
               | hypercube33 wrote:
               | They aren't gunning for Chromebook deployments...these
               | are currently in business laptop models and AMD may have
               | already beaten them in all fronts on some of these per
               | dollar other than ultralight and video playback duration.
               | Lenovo has a A model that can do 16 hours and light
               | gaming. more importantly it runs x86 apps full speed.
               | 
               | I agree these likely will take over the sub $1000 market
               | if given the chance but they are shooting at $1500-2000
        
               | silisili wrote:
               | Interesting. Seems like a huge missed opportunity there,
               | as those outsell everything else by quite a margin.
               | 
               | Perhaps they'll expand into both ends like they do
               | phones, eg the 4xx on the low end and 8xx on the high
               | end.
        
               | RussianCow wrote:
               | Presumably the margins on Chromebook are terrible
               | compared to those of mid to high end laptops. I don't
               | blame them for wanting to start with the higher margin
               | market and eventually work down.
        
               | dagmx wrote:
               | They aren't fighting for Chromebook territory with this
               | though.
               | 
               | All their comparisons are to the MacBook Air and mid-high
               | end windows laptops because that's the market they're
               | gunning for. These are meant to be $1k-2k devices.
        
               | pjmlp wrote:
               | Chromebooks are basically the US school systems, Google
               | employees, and little else.
        
             | HeWhoLurksLate wrote:
             | IIRC, on the LTT video on the topic they mentioned that
             | there were like 8 extra PCIe lanes that could
             | _theoretically_ get run to a dGPU
        
           | benreesman wrote:
           | I grew up in San Diego and at the time being involved with
           | technology meant living in Qualcomm's shadow in one way or
           | another.
           | 
           | So I tend to agree that being the reference mobile SoC vendor
           | outside of Cupertino is pretty on brand for their absolute
           | top priority. At Qualcomm if it doesn't make dollars it
           | doesn't make sense as we used to say.
           | 
           | And good for them! After a brief flirtation with the idea of
           | becoming a pure play CDMA patent troll they seem to have
           | gotten back to their roots and started doing engineering
           | again. It's a cool company.
        
           | a-dub wrote:
           | > It's hard to imagine why they'd distract themselves with
           | that, except perhaps with a token small-run demo for pure
           | brand marketing purposes.
           | 
           | haven't google and samsung started making their own socs?
        
           | InDubioProRubio wrote:
           | I think there is a market for a dual-world architecture, with
           | loads of dark silicon in a low battery setting (and suffering
           | performance being acceptable) and a high power mode, that is
           | able to compete with regular desktop gpu architectures.
        
         | djmips wrote:
         | The Mobile lead and gaining ground for higher end gaming are
         | aligned in many areas so you have to do both.
        
           | dagmx wrote:
           | In some areas sure, but it's really down to what you're
           | dedicating silicon to in terms of capabilities.
        
             | jayd16 wrote:
             | How so? I'm fairly confused about what you're implying. Is
             | Apple sacrificing mobile capabilities for better desktop
             | capability, for example?
        
               | dagmx wrote:
               | Yes.
               | 
               | Assuming that a GPU size and node is similar between
               | different GPUs, then different features which require
               | silicon do it at the expense of other features. It's
               | always a balancing act.
               | 
               | That's effectively the big rub between NVIDIA and AMD
               | today with raytracing + tensor support vs pure
               | raster+compute throughput.
               | 
               | Apple just went through a major GPU architecture change
               | [1]. They focused a ton on maximizing for AAA game usage
               | and followed the NVIDIA route to bias towards where they
               | think things are going. At least according to the
               | simplified architecture diagrams for both Apple graphics
               | and Adreno, Apple has more raytracing silicon than
               | Adreno.
               | 
               | It also supports stuff that doesn't require silicon but
               | does have effects on GPU design like mesh shading or
               | their new dynamic caching that improves occupancy for
               | high draw count games with large uber shaders.
               | 
               | Compared to Adreno that focused more on raw triangle
               | throughput instead, but doesn't scale as well with
               | complexity. It performs much better on mobile benchmarks
               | that fit that usage pattern, but falls behind with
               | desktop benchmarks that follow Apple's priorities.
               | 
               | [1] https://developer.apple.com/videos/play/tech-
               | talks/111375
        
               | jayd16 wrote:
               | I think the tail is wagging the dog here. Those mobile
               | workloads are tuned towards what currently works.
               | 
               | If Apple or Qualcomm pull off desktop features in a
               | mobile power envelope then the industry would happily
               | adjust.
        
               | dagmx wrote:
               | I think that assumes that most games are targeting higher
               | end graphics and are held back.
               | 
               | The majority of played mobile games would be the same as
               | they are today imho even if you could achieve 4090
               | performance in that power envelope.
        
           | hajile wrote:
           | HALF of X1's compute is F16 only which is absolutely wasted
           | silicon for most desktop games.
           | 
           | Their entire tiling setup is great for simple mobile games,
           | but (as shown in the article) is also an inefficient approach
           | for desktop games.
           | 
           | 64-wide SIMD works well in simple games and offers a FAR
           | better theoretical compute per area, but when things get more
           | complex, it's hard to keep everything filled. This is why
           | Intel is 8-wide, Nvidia is 32-wide, and AMD is
           | 32/32x2/64-wide (and is one reason why the second SIMD didn't
           | improve performance like the theoretical flops said it
           | should).
           | 
           | With the release of the M-series chips, Apple's GPUs stopped
           | ramping up performance as quickly on the simple mobile
           | benchmarks. This is very clear with A17 in Aztec not only
           | falling behind the SD8gen3, but the SD8gen2 too. At the same
           | time, GPU perf/watt has also lagged behind. However, when you
           | switch to something like the somewhat more complex Solar Bay,
           | the Apple GPU pulls ahead.
           | 
           | This is similar to the AMD/Nvidia swap from gaming to hard
           | compute then slowly back to gaming after they split into
           | server and gaming designs.
        
         | LegitShady wrote:
         | To me it seems as if the selling points of these latest
         | snapdragon chips is high efficiency/battery life and
         | competitive performance, so given the efficiency angle it makes
         | less sense to try to make gaming machines out of them right
         | now. Maybe in the future there will be a gaming oriented
         | snapdragon less concerned about battery life.
        
         | chaorace wrote:
         | I'm not surprised the Adreno numbers didn't hold up as well as
         | the rest of the Snapdragon benchmarks. Back in 2013 the Dolphin
         | team blogged about their terrible experiences with the Adreno
         | drivers and vendor support[1]. Ten years later in 2023, the
         | same team blogged about how those same continuing issues led
         | them to completely replace the official Adreno driver with a
         | userland alternative[2].
         | 
         | As it stands today, the only credible names in ARM SOC GPUs
         | seem to be Apple (M chips) & Nvidia (Tegra chips).
         | 
         | [1]: https://dolphin-emu.org/blog/2013/09/26/dolphin-emulator-
         | and...
         | 
         | [2]: https://dolphin-emu.org/blog/2023/08/13/dolphin-progress-
         | rep...
         | 
         | Kudos to the Dolphin website developers for keeping 10+ years
         | of blogs & hyperlinks fully functional and properly tagged.
         | They always produce great reading material!
        
           | eropple wrote:
           | _> As it stands today, the only credible names in ARM SOC
           | GPUs seem to be Apple (M chips)  & Nvidia (Tegra chips)._
           | 
           | I've been out of this space for years, so my knowledge is
           | definitely stale, but have Mali GPUs fallen out?
        
             | dagmx wrote:
             | Mali has decent coverage. The pixel phones use them and so
             | do a lot of mediatek based low-mid range devices.
        
         | segasaturn wrote:
         | Is there any reason why these ARM iGPUs are so much worse than
         | iGPUs from Intel and AMD? My 11th gen Intel CPU's Xe graphics
         | completely outpaces my M1 Mac's and something like a Ryzen 5
         | 5600G destroys both.
        
           | dagmx wrote:
           | I'm curious what benchmark you're looking at. The M1 is a
           | very competitive iGPU that is competitive with an Nvidia
           | 1650.
           | 
           | Are you looking at pure gaming performance and are you sure
           | it's not going through a translation layer?
           | 
           | To quote Anandtech (https://www.anandtech.com/show/16252/mac-
           | mini-apple-m1-teste...)
           | 
           | > Overall, the M1's GPU starts off very strong here. At both
           | Normal and High settings it's well ahead of any other
           | integrated GPU, and even a discrete Radeon RX 560X. Only once
           | we get to NVIDIA's GTX 1650 and better does the M1 finally
           | run out of gas.
        
             | segasaturn wrote:
             | No benchmarks, just based on personal usage. I think I
             | found the issue after posting that comment though, which is
             | macOS's unhelpful deprecation of openGL support. The games
             | that I play on macOS used OpenGL and will probably never
             | implement Metal which is a shame. They were Apple Silicon
             | native though, no translation. Games in question are
             | Factorio and RuneScape.
        
               | dagmx wrote:
               | Ah yeah it's possible individual games do perform poorly.
               | 
               | But in a general sense the integrated GPU in the M series
               | processors is closer in competition to a low/mid discrete
               | GPU than the integrated GPUs in other brands.
        
       | smusamashah wrote:
       | That's a mouthful of a name
        
         | dagmx wrote:
         | Not really any more than any other brand.
         | 
         | The Intel Meteor Lake Arc iGPU The AMD Ryzen Radeon iGPU
         | 
         | Apple are pretty much the only major SoC company who don't
         | brand the CPU and GPU independently
        
           | bigyikes wrote:
           | "Ultimately, it comes down to taste." - Steve Jobs
        
             | pmontra wrote:
             | Apple build GPUs for their own hardware and nothing else.
             | They could even do without names, it's just another
             | inevitable component that's inside the box.
        
       | gary_0 wrote:
       | Re: the manual driver updates. Recently I put a clean Win11
       | install on an ASUS Meteor Lake laptop for someone, and Windows
       | downloaded and installed all the latest drivers automatically
       | (along with a bunch of fresh bloatware, natch). Maybe Qualcomm is
       | working with Microsoft so their drivers will get updated the same
       | way?
        
         | LegitShady wrote:
         | I assume they are given the launch of snapdragon copilot
         | laptops, and your witnessing it get drivers from windows
         | update.
        
           | daviddever23box wrote:
           | Yes - and it is certainly possible to export the "final", up-
           | to-date set of drivers via DISM, then build an orthogonal set
           | that you can recursively install via a single one-click
           | pnputil batch file in Audit Mode (Ctrl-Shift-F3 at the top of
           | OOBE).
           | 
           | This is the easiest way to validate benchmarks across
           | neutral, bloatware-free OS versions (at least the ones
           | supported by that SoC, anyway).
        
       | benreesman wrote:
       | "In Adreno tradition, Adreno X1's first level cache is a
       | dedicated texture cache. Compute accesses bypass the L1 and go to
       | the next level in the cache hierarchy. It's quite different from
       | current AMD, Nvidia, and Intel GPU architectures, which have a
       | general purpose first level cache with significant capacity. On
       | prior Adreno generations, the GPU-wide L2 cache would have to
       | absorb all compute accesses. Adreno X1 takes some pressure off
       | the L2 by adding 128 KB cluster caches."
       | 
       | People have been tinkering with L1 cache conditionality since the
       | L1i and L1d split in 1976 but the Qualcomm people are going hard
       | on this and the jury seems out how it's going to play.
       | 
       | The line between the L1 and the register file has been getting
       | blurrier every year for over a decade and I increasingly have a
       | heuristic around paying the most attention to L2 behavior until
       | the profiles are in but I'm admittedly engaging in alchemy.
       | 
       | Can any serious chip people as opposed to an enthusiastic novice
       | like myself weigh in on how the thinking is shaping up WRT this?
        
         | rys wrote:
         | In practice, what gets labelled as the L1 cache in a GPU
         | marketing diagram or 3rd party analysis might well not be that
         | first level of a strict cache hierarchy. That means it's hard
         | to do any kind of cross-vendor or cross-architecture comparison
         | about what they are or how they work. They're highly
         | implementation dependent.
         | 
         | In the GPUs I work on, there's not really a blurred line
         | between the actual L1 and the register file. There's not even
         | just one register file. Sometimes you also get an L3!
         | 
         | These kinds of implementation specific details are where GPUs
         | find a lot of their PPA today, but they're (arguably sadly)
         | usually quite opaque to the programmer or enthusiastic
         | architecture analyst.
        
       | perdomon wrote:
       | How soon can I buy a handheld console with one of these inside,
       | and can it run God of War?
        
         | 3abiton wrote:
         | Apparently the Minisforum V3 is aiming for that market.
         | Although not super great on battery autonomy.
        
         | izacus wrote:
         | There's plenty of consoles using AMD SoCs that perform better
         | than this and run God of War. Get one of those.
        
       | mirsadm wrote:
       | With my own use case I've noticed very poor compute shader
       | performance on the Snapdragon GPUs. Even worse the drivers are
       | completely unpredictable. The same shader will sometimes run 2x
       | slower for seemingly no good reason at all. I didn't realise
       | games these days relied so much on compute shaders. It's no
       | suprise it doesn't perform as well as it should.
        
       | jauntywundrkind wrote:
       | ARM remains a shitty backwater of unsupportable crap ass nonsense
       | being thrown over the wall.
       | 
       | Qualcomm bought Imageon from AMD in 2009. Sure, they've done some
       | work, made some things somewhat better. But hearing that the
       | graphics architecture is woefully out of date, with terrible
       | compute performance is ghastly unsurprisingly. Trying to see
       | thing thing run games is going to be a sad sad sad story. And
       | that's only 50% the translation layers (which would be amazing if
       | this were Linux and not a Windows or Android device).
        
       | rubymamis wrote:
       | Why is there no comparison with Apple's iGPU?
        
         | nomercy400 wrote:
         | Because you cannot compare between an Apple's iGPU and this
         | chip, while using the same software stack. Because you cannot
         | buy a laptop with this chip and use MacOS.
         | 
         | If they would compare it iwth an Apple iGPU, they'd be
         | comparing two things: the hardware AND the OS, which makes it
         | less clear what is contributing to your benckmark results.
        
           | nuccy wrote:
           | Generally this is a correct argument - to compare hardware
           | one needs to use the same OS/software stack. But the argument
           | works the other way around also, if there is no identical
           | software stack possible does it really matter how raw
           | hardware compares? The end user running a game or an
           | application would experience hardware+OS rather than just
           | hardware.
        
           | zozbot234 wrote:
           | > Because you cannot compare between an Apple's iGPU and this
           | chip, while using the same software stack.
           | 
           | Apple Silicon hardware can run Linux (with unofficial GPU
           | support as of late, although still lacking support for the
           | NPU), and official support for Linux on Snapdragon laptop
           | platforms is supposedly in the works. So we should be able to
           | do a proper comparison as soon as official support is added
           | for both platforms as part of a single mainline kernel
           | release.
        
         | mbs159 wrote:
         | They would have to run the same software, e.g. install Linux on
         | both machines.
        
         | criddell wrote:
         | If I had to bet, I would say it's because they don't beat
         | Apple.
         | 
         | If they had a benchmark result that showed a big win over
         | Apple's design, it would be at the top row of the chart.
        
         | luyu_wu wrote:
         | This is a hardware deepdive by a couple of uni students and
         | enthusiasts... Some people are interested in things that aren't
         | as shallow as fluctuating performance leads !
        
         | elabajaba wrote:
         | A lot of their testing is running custom OpenCL and Vulkan
         | code, both of which are essentially unsupported on macOS
         | (moltenvk exists, but kinda sucks and adds overhead that would
         | make the comparisons invalid anyways).
        
       | jeroenhd wrote:
       | I wonder if there's performance being left on the table because
       | of the way programs and games are designed. It's no secret
       | Qualcomm's mobile chips will run like shit when you try to use
       | desktop code on them, because they're designed differently. I
       | wonder if we're seeing aspects of that here. It would explain why
       | Qualcomm convinced their press team of impressive numbers that
       | nobody in the real world has been able to replicate.
       | 
       | There was a whole comic about design differences when porting
       | desktop style games qnd shaders to mobile (I can't find it for
       | the life of me) which was a pretty good beginner's guide to
       | porting that stuck with me.
        
         | mandarax8 wrote:
         | > There was a whole comic about design differences when porting
         | desktop style games qnd shaders to mobile
         | 
         | This one from arm? https://interactive.arm.com/story/the-arm-
         | manga-guide-to-the...
        
           | jeroenhd wrote:
           | That's the one! Guess it came directly from ARM, no wonder I
           | couldn't find it.
        
       | pjmlp wrote:
       | > DirectX 12 Ultimate: Disabled
       | 
       | That right there is already a reason not to buy this in 2024.
       | 
       | DirectX 12 Ultimate is 4 years old by now, and with DirectX 12
       | the best it can do is a 10 years old 3D API.
       | 
       | This is basically a GPU for Office work.
        
         | cubefox wrote:
         | Most games still don't use DX12 Ultimate features. Some use
         | some ray tracing, but as the article says, this is expensive
         | and should be left off for laptop devices anyway. As for mesh
         | shaders, there is currently one (1) game I know of that uses
         | them. Alan Wake part 2. I think the other features like sampler
         | feedback are also not really used in practice.
        
           | pjmlp wrote:
           | Yeah, but I don't buy hardware for what I can do today with
           | technology from years ago, rather for something that lasts 5
           | to 10 years.
        
             | Synaesthesia wrote:
             | Depends on what you want to do. This GPU is impressive for
             | a thin and light laptop with long battery life. It
             | obviously doesn't compare well to large power hungry
             | dedicated GPUs.
        
               | pjmlp wrote:
               | As mentioned on my original comment,
               | 
               | > This is basically a GPU for Office work.
        
               | cubefox wrote:
               | But why would anyone need a GPU that can run Baldur's
               | Gate 3 for office work...
        
               | pjmlp wrote:
               | That is why this is the only thing this GPU is good for.
        
             | talldayo wrote:
             | If it supports Vulkan 1.2, then it basically supports most
             | of DX12 as well. Very famously Intel's ARC GPUs had
             | terrible DirectX drivers, but good enough Vulkan support
             | that DXVK simply ran better: https://youtu.be/wktbj1dBPFY
             | 
             | As time goes on it feels like native and up-to-date DirectX
             | drivers aren't necessary, even on Windows itself. The era
             | of kowtowing to a d3d9.dll is over; the SPIR-V
             | recompilation era has begun.
        
         | theandrewbailey wrote:
         | DirectX 12 not ultimate still supports most (every?) every game
         | out there. As for "GPU for office work", that's a question left
         | up to specific in-game benchmarks.
        
       | bhouston wrote:
       | Nice! What are the comparisons with Apple's you in the latest
       | M-series chips?
        
       ___________________________________________________________________
       (page generated 2024-07-05 23:01 UTC)