Intel Compute Runtime 24.45 vs. AMD ROCm 6.3 vs. NVIDIA R565 Linux GPU Compute Benchmarks

[phoronix]

Phoronix: Intel Compute Runtime 24.45 vs. AMD ROCm 6.3 vs. NVIDIA R565 Linux GPU Compute Benchmarks

Complementing yesterday's fresh Linux gaming benchmarks of mid-range Intel Arc Graphics "Alchemist" vs. NVIDIA GeForce RTX 40 vs. AMD Radeon RX 7000 series cards ahead of the upcoming Battlemage availability, today's article is providing a fresh look at the latest Intel Compute Runtime performance for Level Zero / OpenCL on current-gen Intel discrete graphics compared to mid-range AMD Radeon GPUs on ROCm 6.3 and similar NVIDIA GeForce RTX 40 Ada graphics cards on the R565 driver.

https://www.phoronix.com/review/linux-opencl-eoy2024

[NeoMorpheus]

Michael, didnt AMD sent you a 7900 XTX or XT a little while ago?

If not, what is your email?

[Michael]

Michael, didnt AMD sent you a 7900 XTX or XT a little while ago?

Yes I have the XT and XTX. But this comparison is mid-range focused as noted in the article given the preparations for Battlemage... Later on will do a full stack comparison for all vendors, mostly time limited at the moment.

[Daktyl198]

The fact that Intel is winning ANY of these benchmarks against AMD is just sad, and doesn't bode well for AMD with Intel's new cards releasing soon. Wish AMD would take some of that new found money they have and hire actually competent software engineers to re-architecture their whole stack because the current one is not worth saving.

[blackiwid]

The fact that Intel is winning ANY of these benchmarks against AMD is just sad, and doesn't bode well for AMD with Intel's new cards releasing soon. Wish AMD would take some of that new found money they have and hire actually competent software engineers to re-architecture their whole stack because the current one is not worth saving.

So is AMD, you just didn't hear from it because Intel did a paper launch and you can then get it send to you in 1-2 months... but apparently this sad marketing trick works for you.

Btw the next gen of amd will have the compute units to do all the A.I. Upscaling I would assume that they help much with this benchmarks.

Also AMDs last gen was the worst in the last 10 years probably, Intel don't have that excuse... for them it's more a question if this headless company even produces any cards after this gen, or if they only sell 100 per country and then say "ohh we are sold out because so many people want it"...

[sophisticles]

So is AMD, you just didn't hear from it because Intel did a paper launch and you can then get it send to you in 1-2 months... but apparently this sad marketing trick works for you.

Btw the next gen of amd will have the compute units to do all the A.I. Upscaling I would assume that they help much with this benchmarks.

Also AMDs last gen was the worst in the last 10 years probably, Intel don't have that excuse... for them it's more a question if this headless company even produces any cards after this gen, or if they only sell 100 per country and then say "ohh we are sold out because so many people want it"...

Someone is looking at AMD through rose colored glasses.

[V1tol]

I wonder how rusticl on AMD (I dunno if it works on Intel) would compare to ROCM. I won't be surprised if it appears to be better (when existing functionality allows it to successfully run tests ofc) like RADV vs AMDVLK.

[Grinness]

Someone is looking at AMD through rose colored glasses.

Maybe.

Or "maybe" AMD is simply missing performant "tensor cores" in their current gen -- aka hardware matrix multiplication.
Intel has this (RT works quite well on their cards, but raster performance are lacking) since day one (archmage -- if this is the code name of their first gen dGPU)
For example Blender makes extensive use of "tensor cores" in Cycles, thus the advantage of intel and nvidia.

AMD in raster matches, and sometime outperforms NVIDIA for same 'grade' dGPU

The promise of RDNA4 is better "tensor cores" (hardware matrix multiplication) and strong high performance raster (hopefully at a decent price).

In the new year we will see if the above is correct and if it bridges the last gap.

[NeoMorpheus]

Yes I have the XT and XTX. But this comparison is mid-range focused as noted in the article given the preparations for Battlemage... Later on will do a full stack comparison for all vendors, mostly time limited at the moment.

Ah ok, sorry.

was wondering, since i noted the last couple of articles using 7800xt and under.

[Kryohi]

Or "maybe" AMD is simply missing performant "tensor cores" in their current gen -- aka hardware matrix multiplication.

Isn't any GPU literally a matrix multiplication accelerator?

I don't think tensor cores have much to do with the performance we're seeing here.