Steam Client Stable vs. Beta Tests With Vulkan On AMDGPU-PRO

[phoronix]

Phoronix: Steam Client Stable vs. Beta Tests With Vulkan On AMDGPU-PRO

Here's the impact of benchmarking Dota 2 with Vulkan using the stable Steam client versus the Steam client beta...

http://www.phoronix.com/scan.php?page=news_item&px=Steam-Client-Beta-VK-AMDGPU

[xeekei]

At 1920x1080 it appears that AMD cards hit the CPU limit sooner than Nvidia, which seems weird.

At 4K the cards perform very similarly to DX11 benchmarks. You'd expect a bit better with Vulkan. However, maybe Dota 2 doesn't use async compute.

[chuckula]

Interesting benchmarks. Obviously there is still a lot of tuning going on.
Did the Nvidia cards have any noticeable performance changes when running the Beta version?

[boxie]

I think it is almost time to start including the windows performance benchmarks more often. Be interesting to watch the performance gap close

[Michael]

I think it is almost time to start including the windows performance benchmarks more often. Be interesting to watch the performance gap close

If only I had more subscribers/non-ad-blockers to justify such testing... I don't keep any Windows installs around on any routine basis.

[chuckula]

I don't keep any Windows installs around on any routine basis.

A sound policy.

[dang_valve]

Interesting benchmarks. Obviously there is still a lot of tuning going on.
Did the Nvidia cards have any noticeable performance changes when running the Beta version?

The Steam Overlay performance issue does not impact NVIDIA as far as we are aware. What was happening is that memory was incorrectly being allocated in the 256MB device local visible heap on AMD (NVIDIA does not have a similar heap). This would cause over subscription resulting in the performance drop. It turned out Dota 2 Vulkan also initially had the same bug, but I fixed that too. Now the device local visible heap is only used for dynamic VB/IB/CBs and I believe internally the AMD driver uses it for the swapchain images.

[schmidtbag]

I've noticed from previous tests that the R9 290 seems to overall be right where it should be. It's performance is supposed to pretty balanced with the GTX 970 (better in some tests, worse in others), which at least the 2nd graph shows here.

[chuckula]

The Steam Overlay performance issue does not impact NVIDIA as far as we are aware. What was happening is that memory was incorrectly being allocated in the 256MB device local visible heap on AMD (NVIDIA does not have a similar heap). This would cause over subscription resulting in the performance drop. It turned out Dota 2 Vulkan also initially had the same bug, but I fixed that too. Now the device local visible heap is only used for dynamic VB/IB/CBs and I believe internally the AMD driver uses it for the swapchain images.

You raise an interesting problem with Vulkan and DX12: All this low-level power also gives developers power to shoot themselves in the foot. Valve is pretty good about tracking down and fixing the issues, but by no means are all game developers as skilled as Valve.

[juno]

If only I had more subscribers/non-ad-blockers to justify such testing... I don't keep any Windows installs around on any routine basis.

Would be nice to read about the other system specs you use for testing (CPU, RAM, clocks). Then, you could do like a broadcast and ask community members to test on their Windows, as I think a decent amount still plays on Windows. If interesting results come in, you could as well place them next to yours in such an article?



btw: seems like the steam beta client changes just made it into the final release