The AMD Radeon RX Vega Launch Performance Compared To 2019 Linux Drivers

[phoronix]

Phoronix: The AMD Radeon RX Vega Launch Performance Compared To 2019 Linux Drivers

With the AMD Radeon VII graphics card shipping in two weeks as the second-generation Vega GPU at 7nm, I figured it would be an interesting time to see how far the original Radeon RX Vega 56 and RX Vega 64 graphics card performance has evolved since their launch back in August 2017. Here is a fresh look at the current Radeon RX Vega 56/64 GPU performance today using the bleeding-edge Linux graphics drivers compared to the driver state back in 2017 for OpenGL and Vulkan gaming performance.

http://www.phoronix.com/vr.php?view=27447

[IreMinMon]

The performance improvements are nice, but RX Vega 64, which is supposed to be the more powerful and more expensive card, still barely performs any better than Vega 56, in a couple of benchmarks it even performs worse. Who to blame?

[Brisse]

Who to blame?

CPU's?

[IreMinMon]

CPU's?

I don't think that's the case. The CPUs appear to be performing well enough on Windows.

[L_A_G]

IreMinMon

[xtf1]

Michael how high is the power consumption in idle and browser usage.

[tuxd3v]

Thanks Michael, for this tests..
Good news, I hope, power consumption could be on par with Windows,
But Great work for AMD, it shows AMD is been moving in the right track!!

They have a good start picture of What AMDGPU is becoming..
Frames improvements a lot, power decreased with optimisations and tunning..

[zanny]

The performance improvements are nice, but RX Vega 64, which is supposed to be the more powerful and more expensive card, still barely performs any better than Vega 56, in a couple of benchmarks it even performs worse. Who to blame?

Most of the time it performs 10-20% better which is how much more hardware it has and that makes sense.

If a game can't scale properly to as many CUs as the Vegas have, or can't properly pipeline their memory access to avoid contention, or do too many fence syncs across the GPU, then the performance can crest to a plateau limited by the bottlenecks or inefficiencies.

What we are seeing now with the broader use of DXVK is that a lot of the Linux performance woes of the last couple years are often as much the fault of games with really poorly written OpenGL renderers as it is the fault of unoptimized drivers. Many games run under DXVK are performing better than they did with their native OpenGL ports.

[cybertraveler]

Very nice. Across the board, significant improvements.

Michael: How come many of the bar graphs don't show minimum FPS. I always look for that value. I think minimum and average are both important.

I personally don't care about the "max" value. It seems like pointless data to me.

[Michael]

Very nice. Across the board, significant improvements.

Michael: How come many of the bar graphs don't show minimum FPS. I always look for that value. I think minimum and average are both important.

It's all a matter of what the game/engine expose... Just like only some of them expose frame-times. PTS is exposing what it's able to record.