What's new

Resolve Benchmark Thread V2

Not sure how I missed this topic. [mention]Jan Klier [/mention] please feel free to include the film grain clips in the benchmark. Happy to have them included! [emoji851]

Agreed, any performance “optimization” modes should be disabled for benchmarking.


Sent from my iPhone using Tapatalk
 
Hi,

Some tests:

24-Core AMD Ryzen Threadripper 3960X, 3900 MHz (39 x 100)
64 GB DDR4 4xDIMM2: G Skill TridentZ F4-3200C14-16GTZ 16 Go DDR4-3200 DDR4 SDRAM (CAS 14)
Win10 20H2 October 2020 Update
MSI RTX 3090 X Trio (MS-V388)

Performance Off

Test : 4K CinemaRAW Light
Temporal NR - Better 2 Frames= 36 fps,
3x Temporal NR - Better 2 Frames= 17 fps,
OpenFX - Lens Flare + Tilt-Shift Blur + Sharpen= 43 fps
Basic Grade= RT => 59.94 FPS

Test : 4K Prores444
Temporal NR - Better 2 Frames=41 fps,
3x Temporal NR - Better 2 Frames= 17 fps,
OpenFX - Lens Flare + Tilt-Shift Blur + Sharpen=46 fps
Basic Grade= RT => 59.94 FPS

Test : 4K H264 150mbps
Temporal NR - Better 2 Frames=44 fps,
3x Temporal NR - Better 2 Frames= 18 fps,
OpenFX - Lens Flare + Tilt-Shift Blur + Sharpen= 46 fps
Basic Grade= RT => 59.94 FPS


Test : 4K RED
Temporal NR - Better 2 Frames=25 fps,
3x Temporal NR - Better 2 Frames= 14 fps,
OpenFX - Lens Flare + Tilt-Shift Blur + Sharpen=30 fps
Basic Grade ~ 48/49 FPS

Test : 4K Prores 422
Temporal NR - Better 2 Frames=41 fps,
3x Temporal NR - Better 2 Frames=17 fps,
OpenFX - Lens Flare + Tilt-Shift Blur + Sharpen=46 fps
Basic Grade= RT => 59.94 FPS
-------------------------------------------------------
Test : 8K RED to 8K
Temporal NR - Better 2 Frames= 12 fps,
3x Temporal NR - Better 2 Frames= 5 fps
OpenFX - Lens Flare + Tilt-Shift Blur + Sharpen=11 fps
Basic Grade = 22 FPS

Test 8K > UHD H265 100mbps
Temporal NR - Better 2 Frames= 38 fps,
3x Temporal NR - Better 2 Frames= 17 fps,
OpenFX - Lens Flare + Tilt-Shift Blur + Sharpen= 41fps
Basic Grade= RT => 59.94 FPS

Test : 8K RED to UHD
Temporal NR - Better 2 Frames= 23 fps,
3x Temporal NR - Better 2 Frames= 13 fps
OpenFX - Lens Flare + Tilt-Shift Blur + Sharpen=26 fps
Basic Grade = 44 FPS

Test 8K > 8K H265 100mbps
Temporal NR - Better 2 Frames= 15 fps
3x Temporal NR - Better 2 Frames= 5 fps
OpenFX - Lens Flare + Tilt-Shift Blur + Sharpen=12 fps
Basic Grade= RT => 59.94 FPS

Hope this help.
 
HP Z800 Workstation
CPU dual X5650 2.67Ghz 6 core (12core dual)
48 GB DDR2 Ram
Windows 10 64 (version 1909 OS build 18363.1198)
Single: Founders Edition 3090 -24 GB Vram

4K clip

6node - 39
18node - 30
30node - 21
66node - 13

1TRN - 27.5
2TRN - 23
4TRN - 14.5
6TRN - 8.5

8K - to come :)
 
Can I express my doubts in the relevance of Puget benchmark when it comes to GPU benchmarking? Firstly, one should ignore the cumulative score since it takes everything into account and cannot be used for benchmarking a GPU. Instead, GPU score in the specifics page should always be used.
However, what I have learned from dissecting the benchmark, it generates the GPU score by rendering to h264, to disk, and reading how long it took. But this also takes into account the efficiency of GPU video encoding (NVENC, VCE, Quicksync) which is known to be drastically faster on Nvidia GPUs. Thus a GPU may well be superior in effects compute but have a lower GPU score due to poorer video encoding. Also, if native encoder was used, the CPU performance would affect the result. So the Puget benchmark can only be relevant for the GPUs from the same manufacturer.
Therefore I feel that the standard candle benchmark, however old it may be, is still greatly relevant for GPU benchmarking in Resolve.
Let me know how you guys feel, am I on or off target here?
 
You're off target, because in the end, performance is performance.

Of course to properly compare GPUs, you have to use otherwise identical systems with identical footage and color grades, but the fact that nVidia's encoders are faster than AMD's (which may at last no longer be the case) affects our work, so it's relevant even if it doesn't favor AMD.

The secret isn't to ignore GPU features, it's to use a broader suite of test scenarios because for example, the nVidia nVenc encoder, while top notch, doesn't provide any hardware acceleration for DNxHR encoding. Or ProRes, which isn't relevant for Resolve in Windows, but is for something like Scratch or Mistika in Windows.
 
Thanks. Your point may be valid if Puget cumulative scores weren't wildly varied, same processor and same GPU can have 2x difference in, say, Fusion score, which is pretty outrageous.
I assumed many of us are not as influenced by video encoder speed since the output is generally some intermediate or master codec that doesn't use GPU encoding. I believe we want to know the GPU speed in working with video in Resolve, otherwise we wouldn't be mostly posting anything but Puget results here.
 
Not that I didn't defend the Puget benchmarks... IMO they're not broad enough, since the Puget folks are techies in a tech town with almost no film industry at all, so their access to experts in our field is limited. Before moving out of Washington I met one of their reps at NAB and tried to engage them after the conference, but had no success.

Their benchmarking makes me imagine a company like Asus attempting to market gaming computers and relying on Half Life 2 for all of its benchmarking.

Sure, the numbers would look great... for that one niche.
 
You must log in or register to reply here.
Back
Top