-
No performance difference with 2 GTX580 cards?
I’m doing some performance testing in Resolve with various footage using up to 2 GTX580 cards and I’m finding that I don’t see any difference in fps (at least in playback) whether I’m using 1 or 2 GTX580 cards for GPU processing.
I’m testing RED Epic, Arri Alexa RAW, and F65 RAW footage and I get the same FPS in playback using either 1 or 2 cards. I honestly thought I would be getting a much higher FPS with two cards. Is there something I need to configure? I see in the logs that Resolve sees both GPU cards.
Here are the system specs:
Mac Pro: 12 Core
10.7.3
24GB RAM
Internal RAID (500MB+ performance)
Quadro 4000 (As display GPU)
2x GTX580 (Processing GPU’s connected to external PSU)Thanks for any insight,
-ChrisElysian Pictures Production Company
RED Epic & RED One Rentals -
12 Replies
-
You’re not using the lite version, correct? As that is limited to only one GPU.
What kind of fps are you getting? The red footage is still CPU debayered, which means it’s probably your bottle neck and 2 GPUs won’t make a huge difference.
F65 is definitely GPU debayered, as is (i believe) ARRIRAW, so those should benefit from multiple GPUs. Though the 285s are rather slow & both together would equal roughly one 470.
Online Editor | Colorist | Post Super | VFX Artist | BD Author
-
No, using the full version Resolve (8.2 & 8.2.1 beta 4).
And it’s not GTX285’s, it’s GTX580’s (PC versions using the new Nvidia drivers for OSX)
For RED Epic footage I’m getting this in terms of fps (both 1 and 2 GTX 580 cards):
24fps – Quarter Good
18fps – Half Good
8fps – Half Premium-Chris
Elysian Pictures Production Company
RED Epic & RED One Rentals -
Sorry read 285 for some reason.
RED footage is CPU debayered, you wouldn’t see any improvement with multiple GPUs, as even one 580 is faster than your CPU can debayer the image.
F65 will benefit from multiple GPUs, but in regular playback the fastest you’ll get is what ever value your “Playback framerate” is set to. Probably 24. Try running the standard candle test with one GPU and then with two GPUs to see the performance difference without the issue of debayering hindering performance.
Online Editor | Colorist | Post Super | VFX Artist | BD Author
-
I get 11 fps in half premium with just one gtx 560 ti. My guess it’s your processor, not the gpu. I got a i7 3930k
Editor/Colorist, Amsterdam, The Netherlands
http://www.pepijnklijs.nl -
Juan, do you know where I could get ‘the standard candle test’ files? The link in the post somewhere here is dead…
Thanks!
Editor/Colorist, Amsterdam, The Netherlands
http://www.pepijnklijs.nl -
Okay I did some more testing, looks like basic playback and export (for any of my test file formats (R3D, Arri RAW, Arri ProRes, F65 RAW) is dependent on the CPU and disk speed. Having multiple GPU’s isn’t going to help here.
Where the multiple GPU’s did kick in is having a bunch of nodes. I would get anywhere from 2x to 5x the amount of nodes (in this test blur) at the same max fps that I would get with only 1 GPU card.
-Chris
Elysian Pictures Production Company
RED Epic & RED One Rentals -
The biggest difference you’ll see is with noise reduction, which is really GPU dependent.
Online Editor | Colorist | Post Super | VFX Artist | BD Author
-
So the problem sounds like the cards. If you have “off the shelf” firmware on the cards they will only run at 2.5 GT… you need the MacVidCards firmware mod to run a “non-official” Nvidia card on a Mac at full speed. I went through this same thing. I bought an EVGA GTX 570 2.5GB card. I installed it in my mac pro and I hade similar performance to what you are getting. I’m not sure what is different about the card, however, it shows up to the system at full link speed on my 2010 12-core mac pro… where prior to the firmware mod it did not.
Anyhow, David Pirinelli (who is the MacVidCards guy) is on this forum often. Or you can do a google search for macvidcards. He did the mod on my card and it was totally worth it. It even works with the adobe mercury engine (I had to modify some system files to get the adobe stuff to recognize my card, but it worked).
Anyhow, this is why you are only getting OK performance with these cards on a Mac. If you drop them in a windows machine it will work without the mod.
– Jonathon
-
i wonder how did you mod for mercury and if you need to connect monitors to resolve gnu for mercury.
a bit off topic but could be useful fur others too.
https://forums.creativecow.net/readpost/277/15801
Here you have post and thread about standard candle.
—
Margus
DaVinci 8.2.1 OSX 10.7.3
MacPro 5.1 2×2,93 24GB
GTX 470 / Quadro 4000
Multibridge 2 Pro -
Google for the gpusniffer hack. It’s pretty easy. And in cs6 at least, you do not need the gpu connected.
Online Editor | Colorist | Post Super | VFX Artist | BD Author
Reply to this Discussion! Login or Sign Up