Comments on: Direct3D 12 vs OpenGL: A quick Test

By: azsx

azsx — Sat, 16 Jan 2016 21:25:45 +0000

Where comparision with DX11! And more – my DX9!…

By: fellix

fellix — Mon, 11 Jan 2016 10:54:59 +0000

On my GTX980 Ti I’m getting lower scores with the heavier settings in both tests, compared to what you posted in the article. I too noticed, while running the scripts in GeexLab, that there is CPU load limitation on one or two cores. Since my CPU (6-core Xeon @ 3.5GHz) is of much older architecture — Westmere, compared to your Skylake, I think the IPC difference plays some role in this case.

By: Aleksandar

Aleksandar — Wed, 06 Jan 2016 13:18:40 +0000

Thank you very much for some interesting results, but I have to admit they are not very informative. Would you be so kind to add GPU working frequency instead of GPU load (or as the addition to GPU load information)? As we all know, the drivers are very aggressive in reducing working frequencies as the workload decreases. Previously, there were power states (PS) that determined the frequencies and power consumption. Nowadays, the frequency is continuously changed even through the PS. If GPU utilization is low, the test is not very relevant since there is a bottleneck or underutilization.

Second, instead of FPS, using frame execution time is more informative information. Furthermore, there is no hardware (monitor) that can refresh its screen 9000 times per second. Is it a time-span between the beginning and the end of the frame execution, or between two consecutive frames? Is it a CPU or GPU time?
I wouldn’t go into deeper analysis before getting information on GPU frequencies during the tests execution and a distinguished CPU and GPU execution time. I’ll be very grateful for any additional information.

By: Ziple

Ziple — Wed, 06 Jan 2016 08:39:47 +0000

I don’t know how you send your triangles, but one improvement would be that you send more batches. If you just send one big command with the 2 millions triangles, the GPU will not appreciate it.
For example, some time ago there has been a commit in mesa by marek olsak to split the commands into smaller pieces for AMD GPUs.
I guess it is because the AMD (and NVidia) GPUs have numerous hardware thread dispatchers that can be used in that case, but they are not able to split the commands by themselves. Maybe the Intel one can do it.

By: CatchChallenger (MMORPG libre)

CatchChallenger (MMORPG libre) — Tue, 05 Jan 2016 22:19:37 +0000

In reply to fen.

August 3, 2009: OpenGL 3.2
Direct X 12: July 29, 2015
Vulkan: Not realsed yet

By: fen

fen — Tue, 05 Jan 2016 21:08:35 +0000

would something change if you use opengl 4.5? maybe some AZDO techniques could help?