New AMD Ryzen Threadripper Editing Machine

[amspire]

And on the subject of using CPU rendering - my 2010 era Xeons are capable of pushing over 100fps with x264. As above it's setting dependent - ultrafast preset and a bitrate of 18Mbps racks up 94fps for the sample clip.

I don't recall which Xeons you're using, but surely they can whip my X5650s handily.

None of this tested using Vegas and VFW, and with that no access to comparable settings - sorry, don't run Windows on that. If you or someone else can describe the settings I can perform a better comparison.

You are right. I just set it to ultrafast and 18MB/s in Vegas 14 and it took 3:30 minutes on a slow PhenomII system. Resulting render is 294MBytes. The Phenom is 6 core, but it is slower than a modern i3 processor. I think a dual Xeon system would be well over 4 times faster - that is sub 1 minute without a GPU!.

[Monkeh]

And on the subject of using CPU rendering - my 2010 era Xeons are capable of pushing over 100fps with x264. As above it's setting dependent - ultrafast preset and a bitrate of 18Mbps racks up 94fps for the sample clip.

I don't recall which Xeons you're using, but surely they can whip my X5650s handily.

None of this tested using Vegas and VFW, and with that no access to comparable settings - sorry, don't run Windows on that. If you or someone else can describe the settings I can perform a better comparison.

You are right. I just set it to ultrafast and 18MB/s and it took 3:30 minutes on a slow PhenomII system. Resulting render is 294MBytes. The Phenom is 6 core, but it is slower then a modern i3 processor. I think a dual Xeon system would be well over 4 times faster - that is sub 1 minute without a GPU!.

Er, yes, I was wasting some cycles. Try 40 seconds on a 2010 machine.

[EEVblog]

I tried x264 with CRF19 (my usual Hnadbrake setting) and it's well over 3 minutes, didn't even wait for it to finish.
Slightly faster with ABR 4000 setting, but still slower than even CPU based MainConcept.

[amspire]

I tried x264 with CRF19 (my usual Hnadbrake setting) and it's well over 3 minutes, didn't even wait for it to finish.

I am running x264 straight from Vegas. I haven't tried Handbrake.

[Monkeh]

X5650s, Handbrake, x264 CRF 19, running.. Ah. Crap. it's already done. 115fps, 47 seconds.

280MiB output - 29.4Mbps + audio. CRF 19 seems excessive (not aided by ultrafast preset).

Which CPU(s) are you running?

[EEVblog]

You are right. I just set it to ultrafast and 18MB/s in Vegas 14 and it took 3:30 minutes on a slow PhenomII system. Resulting render is 294MBytes. The Phenom is 6 core, but it is slower than a modern i3 processor. I think a dual Xeon system would be well over 4 times faster - that is sub 1 minute without a GPU!.

Ultrafast with 19 CRF is going to take well over 3 minutes for me.

[EEVblog]

X5650s, Handbrake, x264 CRF 19, running.. Ah. Crap. it's already done. 115fps, 47 seconds.
280MiB output - 29.4Mbps + audio. CRF 19 seems excessive (not aided by ultrafast preset).

FYI, using my Handbrake script to transcode the 180MB 2min test NVENC render MP4 file at 50fps and CRF 19 took, 35sec and did approx 180fps.

[Monkeh]

X5650s, Handbrake, x264 CRF 19, running.. Ah. Crap. it's already done. 115fps, 47 seconds.
280MiB output - 29.4Mbps + audio. CRF 19 seems excessive (not aided by ultrafast preset).

FYI, using my Handbrake script to transcode the 180MB 2min test NVENC render MP4 file at 50fps and CRF 19 took, 35sec and did approx 180fps.

Much more like it.

Would you please share your render settings for those of us without Vegas to view them, and the settings you used with x264vfw?

[amspire]

You are right. I just set it to ultrafast and 18MB/s in Vegas 14 and it took 3:30 minutes on a slow PhenomII system. Resulting render is 294MBytes. The Phenom is 6 core, but it is slower than a modern i3 processor. I think a dual Xeon system would be well over 4 times faster - that is sub 1 minute without a GPU!.

Ultrafast with 19 CRF is going to take well over 3 minutes for me.

Just trying Ultrafast CRF 19 now. From Vegas, it did 3:10 - that is faster then my previous 3:30. The resulting render has gone up to 407MBytes. There is no way my CPU can even get close to a dual Xeon. You have to be able to render from Vegas in under 1 minute.

Is there a reason for using Handbrake instead of Vegas for the renders?

[EEVblog]

Would you please share your render settings for those of us without Vegas to view them, and the settings you used with x264vfw?

My Handbrake batch file I use for Drag'n'drop transcoding. I have multiple ones of these for different formats and frame rates etc.

Code: [Select]

@echo off
FOR /F "tokens=*" %%G IN ('DIR "%~1" /B /S /A:-d') do (
     FOR /F "tokens=*" %%H IN ('DIR "%%G" /B /A:-d') do ("C:\Program Files\Handbrake\HandBrakeCLI" -i "%%G" -t 1 -c 1 -O -o "%%~nH-1920x1080.mp4" -f mp4 -e x264 --x264-profile high -q 19 -r 50.00 --cfr -a 1 -E faac -B 192 -6 stereo -R 48 --gain=0 --audio-fallback ffac3 -x ref=1:weightp=1:subq=2:rc-lookahead=10:trellis=0:8x8dct=0 --verbose=1

     )
)
pause

[amspire]

Here are my CRF19 Settings

[Monkeh]

Your preset is set to 'medium' in x264vfw - this will be slow. Set it to ultrafast.

[EEVblog]

Your preset is set to 'medium' in x264vfw - this will be slow. Set it to ultrafast.

Makes little difference, still takes over 3 minutes.

[amspire]

Your preset is set to 'medium' in x264vfw - this will be slow. Set it to ultrafast.

Makes little difference, still takes over 3 minutes.

Are you using Vegas 15? If so, try 14.

[Monkeh]

Your preset is set to 'medium' in x264vfw - this will be slow. Set it to ultrafast.

Makes little difference.

Makes more than a factor of two difference in Handbrake on my machine.

Either x264vfw is horribly slow (seems to be just fine for amspire?) or something is wrong.

Was your 180fps performance with one or two CPUs, and with or without HT?

To clarify: I get 115fps with two X5650s, which are hexacore CPUs at 2.9GHz, with HT on - 24 threads.

You get 180fps with...? Potentially two E5-2680 v2s at 3.1GHz with 40 threads. You should spank me more than that, and you should do it equally from x264vfw.

[amspire]

I am using the latest x264vfw_full_44_2851bm_44825.exe for the x264. I installed both the 32 and 64 bit codecs just in case Vegas 14 had a preference for one over the other.

[EEVblog]

I am using the latest x264vfw_full_44_2851bm_44825.exe for the x264. I installed both the 32 and 64 bit codecs just in case Vegas 14 had a preference for one over the other.

Mine is The same, 64bit only, Vegas 15
Vegas 14 about the same speed

[amspire]

I am using the latest x264vfw_full_44_2851bm_44825.exe for the x264. I installed both the 32 and 64 bit codecs just in case Vegas 14 had a preference for one over the other.

Mine is The same, 64bit only, Vegas 15
Vegas 14 about the same speed

The only difference I could see is I had "Field Order" "Upper Field First" set. I changed it to "None (Progressive Scan)" like you and it ran 5 seconds faster!

That is not the craziest thing. Have a look at my CPU while I am doing this: 63% to 73% !



But as the guys on late night marketing TV say, "But that's not all! There is even more!"

I thought maybe on my PC, the rendered was somehow using my GPU. I disabled GPU support in Vegas preferences. CPU usage has gone up to a constant 73% and the render time is now 2:41! So the GTX580 GPU support for the on-screen graphics in Vegas was actually interfering a little in efficient rendering. Perhaps the more modern cards are causing even more interference?

You slow Xeon speed is a real puzzle. There is just nothing fast about my PC. It does have 32G DDR3 1333MHz RAM, but it doesn't change when I start rendering. I have a SSD C: drive that is at 0% during the whole render. I am writing to, believe it or not, a 500GByte PATA drive! Yup - none of that SATA rubbish! The CPU is not over-clocked - just 2.77GHz on each of its 6 slow cores.

I will try this on the other machine with the GTX1080.

[amspire]

Tried it on a second PhenomII running Windows 10 Pro 1709 (the other one was Windows 8.1) with a GTX1080 card. 8G DDR3 1333MHz RAM

Same results:

With GPU enabled in Vegas 14 and 15 preferences:  Renders in 3:14 - CPU around 63%
With GPU disabled in Vegas: Renders in 2:40 - CPU around 74%

So same story - GPU assistance for the Vegas GUI interferes with the CPU rendering.

Dave, there might be a significant difference between out renders.

I am using the test files you sent. When I start Vegas, I get a message about the following missing files: 00009.MTS, 00012.MTS and 00020.MTS. All I have is the file 00019.MTS. If you are rendering the actual edit, that may be slower.

[amspire]

I found some x264 benchmarks:

AMD PhenomII 1055T:    25
Dual Xeon E5 2687W:     95

The E5 2680  is less then 2% slower then the E5 2687W in general benchmarks. So it should be rendering Ultrafast x264 from Vegas at 139 frames/second or 42 seconds.

In another benchmark for relative x265 rendering speeds:

Phenom II 1055T;  x 1.0
Dual Xeon E5 2680:  x 3.606
AMD Ryzen 7 1700:  x 3.55
AMD Ryzen Threadripper 1950X:  x 4.5
i7 8700K: x 3.3

There have been a number of attempts to fix the recent Intel/Windows vulnerabilities and they often have not worked properly so far.

[EEVblog]

I found some x264 benchmarks:

AMD PhenomII 1055T:    25
Dual Xeon E5 2687W:     95

The E5 2680  is less then 2% slower then the E5 2687W in general benchmarks. So it should be rendering Ultrafast x264 from Vegas at 139 frames/second or 42 seconds.

I'm getting over 3 minutes, so 

[Howardlong]

FWIW some benchmarks for Vegas Pro 15 on the 2 minute project, using MAGIX AVC/AAC MP4 codec Internet HD 1080p 50 fps:

o 1'08" Ryzen 1800x oc 4GHz (8c/16t) NVENC GTX 1080
o 2'02" Dual Xeon E5-2696v2 (24c/48t) NVENC GTX 1070
o 2'18" i7-8700k (6c/12t) NVENC GTX 1050 ti
o 2'39" i7-8700k (6c/12t) Intel QSV (iGPU acceleration)
o 2'52" Dual Xeon E5-2696v2 (24c/48t) CPU only
o 3'14" Ryzen 1800x oc 4GHz (8c/16t) CPU only
o 4'07" i7-8700k (6c/12t) CPU only

These render times seem pretty slow to me for h.264 1080p 50fps, but I am not a regular Vegas user: doing the same two minute AVC render on PowerDirector 16 with HVENC on the GTX 1070 took only 0'23", and it was a higher bitrate (27Mbps). Even upscaling it to 2160p rendered in 1'24".

[alan242]

I just ran GPU-Z during NVENC and it's showing only a 25% GPU load which cycles down the 5% or so periodically.
I think the reason for that is the NVENC is actually not part of the GPU itself, it's en entirely different part of the silicon and it's own core. NVENC using the NVENC encoding hardware and not the CUDA cores.

According to the block diagram on nvidia's website, it looks like NVENC is a front end for the CUDA cores.
https://developer.nvidia.com/nvidia-video-codec-sdk

One thing seen in VR workloads are the Pascal cards not ramping up clock speeds during a VR session. The power management seems to think it does not need to clock the load higher based on the perceived load which in VR can lead to ASW kicking in or worse. Luckily, the management can be overriden. Some people say just setting the Power Management Mode to "Prefer maximum performance" under Manage 3D settings in the nvidia control panel fixes it for them. Others (like myself) have used Afterburner or another utility to lock the GPU at a specific voltage and core clock. The GPU will still downclock if it heats up too much, starting at the 50-60c range; but the clocks can still be made to stay well above the chip's base clock and any card factory overclock.

I also wonder if there is bottleneck is outside of the video card and that is why the GPU load is low. I do not recall if your workload has RAID disks and the input and output files are on different disks or RAID arrays. If input and output files are on the same physical disk controller, maybe there is a bottleneck there. It may even be that Vegas is coded in a way that it can't feed frames fast enough to keep things moving. Microsoft does have some Windows performance tools that may help verify things are working like they should.

[Monkeh]

I just ran GPU-Z during NVENC and it's showing only a 25% GPU load which cycles down the 5% or so periodically.
I think the reason for that is the NVENC is actually not part of the GPU itself, it's en entirely different part of the silicon and it's own core. NVENC using the NVENC encoding hardware and not the CUDA cores.

According to the block diagram on nvidia's website, it looks like NVENC is a front end for the CUDA cores.
https://developer.nvidia.com/nvidia-video-codec-sdk

No.. they just communicate through the same interface.

"NVIDIA GPUs contain one or more hardware-based decoder and encoder(s) (separate from the CUDA cores)"

[Marco]

According to the block diagram on nvidia's website, it looks like NVENC is a front end for the CUDA cores.

I assume they just use them for a lookahead ME ... they aren't going to be useful for much else. Wasting power on the SMs cores when you can do the same thing with far less power on a tiny amount of silicon doesn't make a whole lot of sense.