The Snapdragon X Elite's Adreno iGPU

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.

“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?

> 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.

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?

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.

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.

Nice! What are the comparisons with Apple’s you in the latest M-series chips?

Why is there no comparison with Apple's iGPU?

How soon can I buy a handheld console with one of these inside, and can it run God of War?

That's a mouthful of a name

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).