OpenGL benchmark results with Intel/Intel+Render-Offload/Nvidia

[sndirsch]

Not really an issue. Just a result of running the OpenBenchmark pts/unigine benchmark. I only looked a the numbers shown on the screen during running unigine-valley. Numbers are for

1. intel config: 7-10 fps
2. Intel config + NVIDIA Prime Render Offload: 20-25 fps
3. NVIDIA only config: 20-25 fps

I couldn't spot a real difference between mode 2 and 3, which I believe is a good thing. I was testing on a Dell Precision 5510 Laptop with

• NVIDIA Quadro M1000M
• Intel Corporation HD Graphics 530 (Skylake)

NVIDIA GPU goes up from 27 to 66 degree Celsius in mode 2/3 and doesn't change in mode 1 (measured by nvidia-smi tool).

[sndirsch]

@simopil @bubbleguuum See my results above. :-)

[bubbleguuum]

I have yet to install the new version.
But I'm in the case where I use suse-prime only for its saving power feature (that is: disable the NVIDIA card entirely with bbswitch, except when I need it to work on external monitor). I don't use the GPU otherwise. In fact, If I could have ordered my Thinkpad laptop without it, I would have...

[Rush]

Good news! Although, the results are not surprising to me.. Running NVIDIA only config has to transfer gfx buffers to the Intel card as well. For full screen apps it's probably the same amount of work.

[simopil]

It's cool! I tried some games and performance between nvidia only and offloading were absolutely comparable. So with offloading and vsync disable I have a terrible tearing.
EDIT: glmark2 seems have an issue (220 pts with offloading and 1187 pts nvidia only)

[sndirsch]

@simopil What do you mean by 'pts'? The bigger number is the better?

[simopil]

yes, points. Bigger is better

[sndirsch]

Indeed I get rather poor results in offloading when running pts/glmark2 benchmark via phoronix-test-suite (even worse than intel-only), although everying looks smooth to me (but the same in intel-only mode). Numbers in 1080p:

nvidia: 2009
intel: 537
intel+NVIDIA Prime Render Offload: 522

[sndirsch]

Not really an issue. Just a result of running the OpenBenchmark pts/unigine benchmark. I only looked a the numbers shown on the screen during running unigine-valley. Numbers are for

1. intel config: 7-10 fps
2. Intel config + NVIDIA Prime Render Offload: 20-25 fps
3. NVIDIA only config: 20-25 fps

I now have exact numbers for Unigine Heaven v4.0/Valley v1.0

1. Intel config: 9 fps
2. Intel config + NVIDIA Prime Render Offload: 27 fps
3. NVIDIA only config: 27 fps

[Rush]

Just a theory but maybe PRIME Render offload transports every single frame to the Intel card? This would explain why high FPS count applications would be so much slower. Since complicated apps with low FPS count spend most of their time on rendering frames in the dedicated card, the cost is not that significant.

[sndirsch]

Indeed that would make sense. Thanks!

[sndirsch]

Finally found the time again to test with a newer driver, i.e. 550.54.14 on a Lenovo P16 Gen1 (Intel AlderLake GT1 (UHD Graphics 770) / NVIDIA GA107GLM (RTX A1000 Laptop GPU)). Results with glmark2

1. Intel config: 998
2. Intel config + NVIDIA Prime Render Offload: 1001
3. NVIDIA only config: 6792

So at least with the latest version with Offload we're no longer slower than with just using Intel GPU.

[sndirsch]

pts/Unigine - Sanctuary 2.3 with the driver/hardware/resolution setting right above.

Intel config: 60.23
Intel config + NVIDIA Prime Render Offload: 239.03
NVIDIA only config: 261.64