My new DIY PC

megabit wrote on 7/7/2016, 8:46 AM

Please see specs - System #1 in my listing. Somebody could say that the M2000 GPU is not fast enough to accelerate the 8-core CPU - but all I can tell you is this:

- in VP 13, with several FX'es, my 4K 25p clips play full fps at Full/Best with the Quadro enabled, and only some 12-13 fps with CPU only.

However - even though I play them from a very fast, 3x WD Black Caviar RAID 0, the 50p 4K or UHD footage merely exceeds 27 fps at playback. CPU is not the bottleneck (not even 50% usage), neither is the GPU - so what the... ?

Opinions welcome,

Piotr

Comments

OldSmoke wrote on 7/7/2016, 9:42 AM

A couple of things come to mind:

a) nVidia Quadro M2000 is a mobile workstation card? Are you on a laptop?
b) The implementation of OpenCL on NVidia cards is not as good as on AMD cards; also a reason why I switched to an AMD card. If you need 10bit out, maybe a FirePro card or a Decklink card from BM is a better option.
c) 4K @ 50p is massive and I cant see Vegas at the current state being able to handle it at Best/Full on ANY system you can dream to build.
d) Did you switch off resampling on the events on the timeline? Makes a big difference too.
e) Is this a 32bit project?

I can playback 4K 30p 100Mbps clips from my AX100 fine as long as I use only Sony plug ins, some BCC plugins are ok too but Mercalli Stabalizer or Neatvideo will bring it down to 10-15fps.

megabit wrote on 7/7/2016, 9:53 AM

No - it's a workstation card; "M" from Maxwell (mobile-class Quadro names have the small "m" suffix - like my laptop's 4000m).

Oh, and re sampling off doesn't help ,much with 50p. This 50/60p thing is strange, as if my 25p media can play full speed even with 4-5 FX'es, there is some headroom for 50p (without any effects) to play back @ at lest some 45-40 fps, I'd think....

I chose nVidia because of Resolve, where my media just fly!

Piotr

megabit wrote on 7/7/2016, 11:00 PM

One more reflection for sharing ans seeking opinions, guys:

OldSmoke has long been preaching (in accordance with SCS's own suggestions) that AMD's OpenCL works better for timeline acceleration than nVidia's CUDA (at least with newer generations of the latter - i.e. after the Fermi architecture), while CUDA might still be a bit faster in rendering. Yet my results are totally inconsistent with that theory:

- timeline acceleration (of 4K/UHD XAVC-I, 25p footage) is enormous, reaching almost 200% (13 fps CPU only vs. stable 25p with my mediocre M2000 Quadro enabled)

- rendering times are almost exactly the same with and without engaging CUDA (only have time to check this with Sony and MainConcept FHD renders of the SCS "Mercedes" benchmark veg project).

Interesting? I bet it is... And another thing: as I already wrote, I cannot understand why - with 25p stuff rock-solid at full/best resolution playback (even with quite a serious set of color FX'es), my 50p clips crawl at half the speed even w/o any effects added. This makes me wonder:

- perhaps it's my theoretically super-fast, 3 HDD RAID 0 that is the bottleneck here, after all?

My RAID strip size is 128kB, and when formatting it in Windows I decided to use non-standard cluster size of 32kB. Could that be a culprit? What strip and cluster sizes are you guys using for your video sources disk arrays? Please share your experience asap - the earlier I decide to change my Raid parameters, the less work with backing-up/restoring my files I'll have! Thanks,

Piotr

JohnnyRoy wrote on 7/8/2016, 7:18 AM

> "OldSmoke has long been preaching (in accordance with SCS's own suggestions) that AMD's OpenCL works better for timeline acceleration than nVidia's CUDA (at least with newer generations of the latter - i.e. after the Fermi architecture), while CUDA might still be a bit faster in rendering. Yet my results are totally inconsistent with that theory:"

This is due to NVIDIA's poor implementation of OpenCL on their gaming cards mostly because games don't use OpenCL. You are using a professional NVIDIA Quadro with completely different drivers which may actually have an excellent implementation of OpenCL because many of the CAD applications that the Quadro is used for demands it.

> "- timeline acceleration (of 4K/UHD XAVC-I, 25p footage) is enormous, reaching almost 200% (13 fps CPU only vs. stable 25p with my mediocre M2000 Quadro enabled)"

That's because timeline acceleration uses OpenCL and I'm guessing that your Quadro card is pretty good at that.

> "- rendering times are almost exactly the same with and without engaging CUDA (only have time to check this with Sony and MainConcept FHD renders of the SCS "Mercedes" benchmark veg project)."

That's because the CUDA support in Vegas Pro doesn't support your card so I'm guessing that its not being used and you are getting a CPU Only render regardless of your settings.

So it makes complete sense once you understand that Vegas Pro will not use your Quadro cards for rendering but will use it for timeline playback. That's not to say that an AMD Radeon HD card wouldn't be faster for a lot less money! ;-)

> "- perhaps it's my theoretically super-fast, 3 HDD RAID 0 that is the bottleneck here, after all?"

Not likely but possible. Try using a disk speed test to see what you throughput is.

~jr

megabit wrote on 7/9/2016, 8:00 AM

"Not likely but possible. Try using a disk speed test to see what you throughput is. "

I did; it's 520 MBps read, 480 MBps write - definitely NOT a bottleneck. It must be another Vegas quirk - it doesn't like 50/60p files :(

Piotr

Edit: Something must have been wrong with my project settings the last time I tested - today my 4K 10 bit 4:2:2 clips play back at the full 50 FPS speed in VP 13.