Opimizing rendering speed - more use of CPU and GPU?

Please try Handbrake as advised by Andrew.

in handbrake i got 270 fps render

in XviD4psp coder i got 127 fps

Please try Handbrake as advised by Andrew.

in handbrake i got 270 fps render

So you have an answer. This is Resolve "issue". It may be related to the way how particular app works. If Resolve needs to decode frames, then pass it to its internal engine, then send to back to GPU decoder there may be some delay there which causes lower speed.
Other app may have different design and decoded frames, go 'more directly" back to GPU encoder.
Maybe BM can improve, but encoder and NLE is not exactly he same thing (each has its own design goals).

But i know what other users have 150-200 h264 render speed even on rtx2060

So it's your new card. Some bug there with Resolve vs driver etc.
Have you tried starting with other (fast format: Cineform, DNxHR etc.) and then hardware encode to h264?
Maybe it's an issues with decode+encode case.
If you tried everything then simply send a bug report to BM.

So it's your new card. Some bug there with Resolve vs driver etc.
Have you tried starting with other (fast format: Cineform, DNxHR etc.) and then hardware encode to h264?
Maybe it's an issues with decode+encode case.
If you tried everything then simply send a bug report to BM.

i dont have DNxHR or cineform source.. but i try to find. than will try to render to h264 and send a feedback here

I get 60 FPS with AMD RX 580 and h264 29.97 fps FHD source and 160 fps in handbrake. It's obviously how Resolve works. Can you link to the source that claims 200 fps is possible with Resolve?

I get 60 FPS with AMD RX 580 and h264 29.97 fps FHD source and 160 fps in handbrake. It's obviously how Resolve works. Can you link to the source that claims 200 fps is possible with Resolve?

this
viewtopic.php?f=21&t=124900&hilit=low+cpu+utilization+render

I get 60 FPS with AMD RX 580 and h264 29.97 fps FHD source and 160 fps in handbrake. It's obviously how Resolve works. Can you link to the source that claims 200 fps is possible with Resolve?

this
viewtopic.php?f=21&t=124900&hilit=low+cpu+utilization+render

So, did you install GeForce Experience?

Try to replicate as much as you can from that guy's setup (same version of Resolve, same drivers).

P.S. new development, I set AMD encoder to "Prefer Speed" instead of "Prefer Quality" and now I get 120 fps (Fusion standalone currently rendering with 100% CPU use in the background ). Perhaps that's what people see - they set quality to something less, I don't have my Nvidia machine on soo can't check.

So, did you install GeForce Experience?

Try to replicate as much as you can from that guy's setup (same version of Resolve, same drivers).

yeah i am really didnt setup gforce experience. should try it.

also u can see it article. my 3060ti should have 80 fps at 4k render
https://www.pugetsystems.com/labs/artic ... ance-2021/

So it's your new card. Some bug there with Resolve vs driver etc.
Have you tried starting with other (fast format: Cineform, DNxHR etc.) and then hardware encode to h264?
Maybe it's an issues with decode+encode case.
If you tried everything then simply send a bug report to BM.

i dont have DNxHR or cineform source.. but i try to find. than will try to render to h264 and send a feedback here

Just render one out of Resolve

Just render one out of Resolve

oh yeah) its the way) thx

With Geforce experience still 66 fps render

If you donwnloaded Puget Benchmark you can run individual benchmarks, just connect to the TestBench_4K_NVIDIA database in /Asset/Standard, load "h264_150mbps..." project and replace the media in /Asset/Standard media. Then in Deliver page you clear render status for h.264 to h.264 timeline and render. They use Youtube preset for that. Then compare to the Puget's own scores

Not sure if this is of interest but just did a test of GH5 UHD 60P 150Mbps Vlog corrected to rec709 colourspace in a 1920x1080 60P project CPU at 15% and GPU at 98%. Encode was running at about 160fps similar speed for h265. Current NVIDIA Studio driver 460.89 Resolve Studio 17b7

Running native encode to h264 was about 55fps with CPU at 46% and GPU 28%

I've also tried it on 1080Ti (Resolve 16.2)
Few Mbits h264 source going back to h264 without any grading= 300fps (GPU decode+encode, GPU load is spread 42/48%).
Same on CPU=200fps (10 core 4GHz i9, GPU decode at 36%).

HD source from AX100 so XAVC-S 60P HD. About 75 fps for native encode and almost 300fps for NVIDIA encode to h264 or h265

results of purgebenchmark
Render Index: 1
Project: H264_150Mbps_4K_59.94FPS
Timeline: H.264 to H.264
Time: 54672 ms
Total Frames: 3600
FPS: 65.85
Status: COMPLETE

Render Index: 2
Project: H264_150Mbps_4K_59.94FPS
Timeline: H.264 to DNxHR HQ
Time: 163904 ms
Total Frames: 3600
FPS: 21.96
Status: COMPLETE

Render Index: 1
Project: ProRes422_4K_59.94FPS
Timeline: ProRes 422 to H.264
Time: 55994 ms
Total Frames: 3600
FPS: 64.29
Status: COMPLETE

Render Index: 2
Project: ProRes422_4K_59.94FPS
Timeline: ProRes 422 to DNxHR HQ
Time: 162390 ms
Total Frames: 3600
FPS: 22.17
Status: COMPLETE

Render Index: 1
Project: RED_4K_59.94FPS
Timeline: R3D to H.264
Time: 64484 ms
Total Frames: 3600
FPS: 55.83
Status: COMPLETE

Render Index: 2
Project: RED_4K_59.94FPS
Timeline: R3D to DNxHR HQ
Time: 173388 ms
Total Frames: 3600
FPS: 20.76
Status: COMPLETE


GPU itilization is about 70%

and the same test on ssd

Render Index: 1
Project: H264_150Mbps_4K_59.94FPS
Timeline: H.264 to H.264
Time: 54215 ms
Total Frames: 3600
FPS: 66.40
Status: COMPLETE

Render Index: 2
Project: H264_150Mbps_4K_59.94FPS
Timeline: H.264 to DNxHR HQ
Time: 75953 ms
Total Frames: 3600
FPS: 47.40
Status: COMPLETE

Render Index: 1
Project: ProRes422_4K_59.94FPS
Timeline: ProRes 422 to H.264
Time: 56532 ms
Total Frames: 3600
FPS: 63.68
Status: COMPLETE

Render Index: 2
Project: ProRes422_4K_59.94FPS
Timeline: ProRes 422 to DNxHR HQ
Time: 69952 ms
Total Frames: 3600
FPS: 51.46
Status: COMPLETE

Render Index: 1
Project: RED_4K_59.94FPS
Timeline: R3D to H.264
Time: 64517 ms
Total Frames: 3600
FPS: 55.80
Status: COMPLETE

Render Index: 2
Project: RED_4K_59.94FPS
Timeline: R3D to DNxHR HQ
Time: 76923 ms
Total Frames: 3600
FPS: 46.80
Status: COMPLETE



not bad for 4k render

this mean my GPU is fine

and problem with resolve.. will try to create a new DB or clearly uninstall DR

Creating of new DB wont help.. i am gonna sleep.. dont know what to do

Do you have hardware decode enabled? Your CPU may be struggling a bit with the media.

Do you have hardware decode enabled? Your CPU may be struggling a bit with the media.

Yeah all is checked. U dont have a problem with ur 3090?

Puget banch is stuck in th emiddle of th eproccess. it done first 6 test, and error after.. i had download only standart assets and chose standart test. what i am doing wrong

i had delete the DR config from appdata... to get clear DR. but it dont help.. still 66 fps NVIDIA render and error in pugetbench...
i give up... any suggestions?

Why don't you remove and reinstall Resolve? I suspect the Puget benchmark may not be completing due to Resolve installation. Install to C: drive if it's not there already. Also, make sure Resolve is not already running even though you have closed it. Search for it and destroy it in Task manager.

i loaded the database from standart assets of puget.
that results i got

p.s. we can see what DR dont want use my GPU for ENCODE more than 20% during 1080p render. But Handbrake or other encoders use my GPU at all power

You may have a point there, I do get 100% video encode use in Resolve. Something is stalling you before the encoding stage. Re-install Resolve, don't touch anything except prepare the project, allow Resolve to change timeline settings according to the media and deliver.

Also let us look at the compute graph at last, please (change from copy to compute).

Can you share the project and media you're working with?

link for my project

https://yadi.sk/d/ykyzh_hZhboF8Q

render the 02-timeline

and graph with compute. i think something wrong.

if u got no idea i will try to uninstall at all resolve tomorrow.. maybe this will help but dunno..
i tried create clean new project usin video souce config. but problem not disappear

One example for the importance of hardware support for the codec you want to use:

Encoding 5 minutes of H.265 10 bit in UHD on a MacBook Air M1 (including some NR) takes 28 minutes. The same task on a last years MacMini with Intel's i7 and an eGPU needs over three hours! The mini has all cores (12 threads actually) running at full power, the eGPU is well under load, but not at the limit. The MB Air M1 has the GPU at full load, the CPU is half idle.

BTW, even encoding the same timeline to J2K in HD is faster on the Air by about 30%. All cores at full load on either machine (if the drives are fast enough). It's a very demanding codec.

@puhovik:

I don't have R17 installed in Win so I switched to my Hackintosh installation. The results for test2 timeline are on par with what I'd get on windows from my own h.264 to h.264 FHD workflow (cca 60 fps for two-pass - that's Prefer quality in Windows version, I guess, and 120 single-pass - Prefer Speed in Win lingo).

I am puzzled at your 0% Compute because I get like 3%. Perhaps 3060Ti so fast it doesn't register because it's under 1%??? I asked to see the Compute to see if there is some substantial GPU processing going on.

However, I noticed that you put 29.97 footage in a 30 fps timeline, it may or may not be a problem, can't conclude but most likely not.

I may fire up my Nvidia (GTX 960) machine after all sometime today. Stand by.

Anyone with 2060, 3060 or any Nvidia GPU with 7th gen NVENC/NDEC is free to join in.

P.S. IT MAY WELL BE A RESOLVE 17 THING ON WINDOWS! PLEASE INSTALL R16 AND TRY.

Just did another test of ProRes 422 UHD 60P to h265 UHD using NVIDIA encode Main10. It ran at 71 fps so just less than realtime with CPU at 30% and GPU 42%

Did another recording 30P UHD HLG on my GH5S, that is h265 4:2:0 10bit. Placed on a 3840x2160 29.97 10bit timeline in Resolve and did an encode to Cineform about 90fps encode GPU at 45%. Then used that Cineform 10bit file and encoded to h265 Main10. About 77fps GPU 66%. As you can expect the other test of source h265 to h265 test was at over 100fps ( clearly did not decode when not needed). So for my system all encodes with NVIDIA are faster than realtime. Clearly the more filters etc loaded it can of course be slowed to a standstill. But simple corrections have almost no effect on these encode times.

If you donwnloaded Puget Benchmark you can run individual benchmarks, just connect to the TestBench_4K_NVIDIA database in /Asset/Standard, load "h264_150mbps..." project and replace the media in /Asset/Standard media. Then in Deliver page you clear render status for h.264 to h.264 timeline and render. They use Youtube preset for that. Then compare to the Puget's own scores

I just tried this. For the 4K h.264 to 4K H.264 I get 95fps peak and it completes in 40 sec.
For the 4K H.264 to DNxHR I get up to 89fps peak. In both cases fps variance range is is about 5-6fps from min to max. I cant view Cuda/Compute in my task manager though Encode/Decode shows utilization.

After moving the source/benchmark to another ssd keeping the export ssd different my scores improved ever so slightly to 995/923 in extended/Std.

4K Media Score - 110
8K Media Score - 121
GPU Effects Score - 64
Fusion Score- 103

Puget banch is stuck in th emiddle of th eproccess. it done first 6 test, and error after.. i had download only standart assets and chose standart test. what i am doing wrong

The 2 minute timeout error in your screenshot is due to the DVR "Stuck on loading farlight page" hang issue. Ensure that you can launch DVR successfully before closing it and running the benchmark.

The 2 minute timeout error in your screenshot is due to the DVR "Stuck on loading farlight page" hang issue. Ensure that you can launch DVR successfully before closing it and running the benchmark.

im pretty sure i can run DVR before and after puget benchmark

add my DVR logs
https://yadi.sk/d/u0CimSJ3ouvO2A

So, here's Puget h264 to h264 testbench result with AMD FX-8320 and Nvidia GTX 960:

46 fps

and windows performance monitor:

GTX 960 Puget h264 to h264 testbench

GTX960_Puget_h264_benchmark.jpg (152.98 KiB) Viewed 3245 times

As you can see, video encode use is almost 100% and it's almost twice as fast doing it than the RX 580. I wander though why Compute use is so high this time?

I just tried this. For the 4K h.264 to 4K H.264 I get 95fps peak and it completes in 40 sec.
For the 4K H.264 to DNxHR I get up to 89fps peak. In both cases fps variance range is is about 5-6fps from min to max. I cant view Cuda/Compute in my task manager though Encode/Decode shows utilization.

After moving the source/benchmark to another ssd keeping the export ssd different my scores improved ever so slightly to 995/923 in extended/Std.

4K Media Score - 110
8K Media Score - 121
GPU Effects Score - 64
Fusion Score- 103

You take the length of the timeline in frames and subdivide with render time in seconds to get fractional fps that you can more precisely compare to Puget's. I seems 7th gen NVENC is twice as fast as 5th gen. Perhaps has also something to do with VRAM allocated to the encoder (512 vs 1024).

You take the length of the timeline in frames and subdivide with render time in seconds to get fractional fps that you can more precisely compare to Puget's. I seems 7th gen NVENC is twice as fast as 5th gen. Perhaps has also something to do with VRAM allocated to the encoder (512 vs 1024).

That requires using a stopwatch because the final results do not list time taken per test. I just retried the same benchmark without GPU decode support (unchecked) in preferences and the scores are a little higher for Fusion and VFX :

945/927 Extended/Standard
4K Media Score 104
8K Media Score 100
GPU Effects Score 68
Fusion Score 106

Why is it higher without GPU decode support? Does it free up GPU to work faster in FX and Fusion? Most CPUs have H.264 decode support anyways so the performance hit is not that great.

You take the length of the timeline in frames and subdivide with render time in seconds to get fractional fps that you can more precisely compare to Puget's. I seems 7th gen NVENC is twice as fast as 5th gen. Perhaps has also something to do with VRAM allocated to the encoder (512 vs 1024).

That requires using a stopwatch because the final results do not list time taken per test. I just retried the same benchmark without GPU decode support (unchecked) in preferences and the scores are a little higher for Fusion and VFX :

945/927 Extended/Standard
4K Media Score 104
8K Media Score 100
GPU Effects Score 68
Fusion Score 106

Why is it higher without GPU decode support? Does it free up GPU to work faster in FX and Fusion? Most CPUs have H.264 decode support anyways so the performance hit is not that great.

No, it doesn't require a stopwatch. Obviously you didn't do what I proposed i.e. manually load each benchmark into resolve and render out, at the end of which you get the info about how many seconds it took to render. Also note that actual playback in performance mode (default) can be as much as twice as fast as the final render. So many "ases", don't know how I managed that

Only Intel CPUs have h.264 decode, but it is not in effect if you disabled GPU HW decode in Resolve. Yes, I suspect in case of GPU decode and effects/VFX, the GPU needs to transfer decoded data to the main memory for editing and effects/VFX operations, while if it's decoded by the CPU it goes directly to the main memory. GPU decoding is helpful in cases where the CPU decoding takes most of CPU time, preventing it too do other CPU-only tasks, like in Fusion tab. h.264 is easy, try h.265 and fusion without GPU HW decode This is one of the gripes I have with Puget benchmark.

Yeah all is checked. U dont have a problem with ur 3090?

No. Both 1080ti and 3090 do ~330fps for me on this test.

No, it doesn't require a stopwatch. Obviously you didn't do what I proposed i.e. manually load each benchmark into resolve and render out, at the end of which you get the info about how many seconds it took to render. Also note that actual playback in performance mode (default) can be as much as twice as fast as the final render. So many "ases", don't know how I managed that

Ahh like that! Speed-reading your post many days later I got mixed up as your previous reply quoted my benchmark scores. Dumb of me hehe!

Only Intel CPUs have h.264 decode, but it is not in effect if you disabled GPU HW decode in Resolve. Yes, I suspect in case of GPU decode and effects/VFX, the GPU needs to transfer decoded data to the main memory for editing and effects/VFX operations, while if it's decoded by the CPU it goes directly to the main memory. GPU decoding is helpful in cases where the CPU decoding takes most of CPU time, preventing it too do other CPU-only tasks, like in Fusion tab. h.264 is easy, try h.265 and fusion without GPU HW decode This is one of the gripes I have with Puget benchmark.

Right - I had the APUs in my head since I previously ran that - only APUs have the decode support. The with and without GPU-Decode benchmark - I found useful for a reason - The Internet is littered with claims that buying Studio version will automagically solve all performance problems whenever somebody complains about slow rendering - obviously that's not true. The real benefit of Studio license is the FX not available in Free version, besides paying for the effort of improving and maintaining such complex software. Even as a casual home user I felt a bit of guilt about sticking to only the free version due to this. I use just one paid feature even after more than a half-year of purchasing, which is the Dehaze feature and which can be also worked on by careful tweaking of curves. But at least I'm guilt-free with respect to this.
Edit: I forgot I use the paid feature - Colospace Transform, extensively, which has proved to be a life-saver compared to tweaking and pushing the dials extremely. So the investment is worth it.

I'm was not fair to dismiss AMD APUs that also support h.264, h.265 decode, but if you buy for Resolve, you don't buy AMD APUs, with Intel you have no choice if you buy mainstream CPUs.

FINALly !!! i won DVR!!! i had delete all plugins, delete DVR install new studio driver 461.40 and now my render speed via nvenc 250+ fps yeeehaaaaa

FINALly !!! i won DVR!!! i had delete all plugins, delete DVR install new studio driver 461.40 and now my render speed via nvenc 250+ fps yeeehaaaaa

Good for you! What's DVR? DaVinci Resolve? You meant to say you had defeated it? Give more precise info about what you did, please, so that others may benefit.

1) delete all plugins: saphire, boriscc, newblue(old from vegas i think), reafx... All plugins what i found in "program removing"
2) uninstall the dvr (davinci resolve)
3) cleanup my register via ccleaner
4) install new studio driver
5) reboot
6) install dvr
7) recover my configs in appdata/roaming (i saved it before deleting dvr)
8) profit

My dvr loading speed now= about 20 second, was 50+
Nvidia render of 264 to h264 1080p= 220+ fps

but once a load i get this error. this is because of deleted plugin saphire and my restoring the old config. but where is information about installated OFX, i should delete this manually

but once a load i get this error. this is because of deleted plugin saphire and my restoring the old config. but where is information about installated OFX, i should delete this manually

From google - I think it is in :
C:\ProgramData\Blackmagic Design\DaVinci Resolve\Support\OFXPluginCache.xml file directory ?

My scores improved to 1003/933 with the latest driver although the GFX and Fusion scores fall down the 4K/8K scores jump up.

With Resolve 17.2.1, I'm back to 20 FPS rendering (15 minutes render for 5 minute video). 99% use of i7 CPU, 15% use of GPU, zero on Video encode/decode.

Some have said that the Studio version helps, others have said it doesn't.

Can anyone post a comparison of their rendering of H.264 HD standard 1080 HD with the free version and the Studio version of Resolve?

I keep thinking about buying Studio even though I don't need the advanced features. Then I get hit with strange issues like the Intel Graphics driver causing a hard crash at startup last year, and more recently the audio driver SNAFU, which gives me pause.

ResolveRender2021-06-18Capture.PNG (57.23 KiB) Viewed 2624 times

If you are not bottlenecked by decode or color operations, NVidia hardware encoding will be about 330 FPS for HD.

Tim - This is going to sound crazy, because it did to me - try updating your video drivers to current.

I can't agree more. This made a HUGE difference for me as well. I specifically am using the studio driver version for NVIDIA. Make sure when updating you choose this instead of the default game driver.

@pciccone, I have updated the NVidia Studio drivers a couple times since my original post. There was no noticeable change in rendering with any update. I've also applied several updates to Resolve, which have not made any noticeable difference in rendering.

@Jack Fairley, the 330 FPS for HD figure would be fantastic. Can you confirm that you are using the Resolve Studio version, and you saw this rate as a result of purchasing the studio version? (your rendering was slower with the free version?) I don't need any of the advanced features of Studio currently, so the only reason I would purchase it is to increase my rendering speed by 10X or more.

Yes, that's using the hardware encoding unlocked by Studio. Keep in mind that there were no other bottlenecks in that test. No editing, no color grading, lightweight source material, and so on.