A Few OpenCL Benchmarks With Radeon Vega Frontier Edition On Linux

[phoronix]

Phoronix: A Few OpenCL Benchmarks With Radeon Vega Frontier Edition On Linux

Phoronix reader "Yoshi" has graciously allowed me remote access to his newly-minuted Radeon Vega Frontier Edition workhorse paired with an AMD Ryzen 7 processor...

http://www.phoronix.com/scan.php?page=news_item&px=Radeon-Vega-FE-Linux-OpenCL

[Shnatsel]

Titan X + Intel results for the same workloads: http://openbenchmarking.org/result/1...TR-NVIDIANEW41 Both cards have the same price of $1000.

TL;DR: Vega on open-source ROCm stack is only slightly behind Titan X on proprietary driver.

[Michael]

Titan X + Intel results for the same workloads: http://openbenchmarking.org/result/1...TR-NVIDIANEW41 Both cards have the same price of $1000.

TL;DR: Vega on open-source ROCm stack is only slightly behind Titan X on proprietary driver.

Keep in mind those Titan X results of mine are of Maxwell, not Pascal.

[xnor]

Vega FE is a bit above GTX 1070 level of performance for twice the price.

[starshipeleven]

Vega FE is a bit above GTX 1070 level of performance for twice the price, with immature drivers.

fixed.

[M@GOid]

fixed.

I looked at PCPers benchmarks at non gaming stuff and the results were all over the place. Sometimes it crushed the Titan, sometimes it loose badly.

So I agreed with that guy that said it was rushed out of factory to meet the H1 promise of a Vega launch. The RX ones maybe have better drivers when they launch in a month.

[cb88]

The results being all over the place is typical of AMD drivers... hopefully with then becoming open source they start improving more consistently if they can do that they they will beat the 1080 handily. At least the drivers seem to be stable if not consitently performant...

[starshipeleven]

The results being all over the place is typical of AMD drivers...

Not on windows, he is talking of Windows benchmarks https://www.pcper.com/reviews/Graphi...l-Testing-SPEC

On Windows AMD has a long history of launching cards with immature drivers that nerf them and getting tons of flak in the first reviews, but the results were usually consistent nerfs.

Here they are inconsistent, something works and it gets into spitting distance or even significantly better than Titan Xp, something is plain borked and it gets like half that, which does not make sense.

[andrei_me]

Keep in mind those Titan X results of mine are of Maxwell, not Pascal.

Also the tests with Nvidia were with CUDA instead of OpenCL

[theriddick]

I don't think driver updates is going to improve the Vega's that much to the point that they are better then 1080GTX, but we can be hopeful. Lets not forget the NVIDIA cards have a good 150mhz overclock headroom (more for their memory).