Comparing Resolve performance with GTX 285, 470, 480 and Quadra 6000

[Margus Voll]

Thanks.

—

Margus

https://iconstudios.eu

[Guillem Ventura]

Thanks for sharing!!

Could you post the same results with a not-modded GPU?
I mean any option aproved by Apple, I’m running a Quadro 4000 and I’m really disappointed, I was wondering whether GTX470 would perform much better.

What OS are you using?

Thanks again.

http://www.malgeniofilms.com

[Robbie Carman]

[Guillem Ventura] “Could you post the same results with a not-modded GPU?
I mean any option aproved by Apple”

Can’t do that the 470/80s are not approved cards from Apple all of them are hacked to work on Mac

Robbie Carman
—————-
Colorist and Author
Check out my new Books:
Video Made on a Mac
Apple Pro Training Series DVDSP
From Still To Motion
An Editors Guide To Adobe Premiere Pro

Twitter
Blog

[Jake Blackstone]

“I’m running a Quadro 4000 and I’m really disappointed, I was wondering whether GTX470 would perform much better”
GTX 470 exceeded the performance of Quadra 6000.

[Jake Blackstone]

“Can’t do that the 470/80s are not approved cards from Apple all of them are hacked to work on Mac”
According to Dave Pirinelli from MacVideoCards, BM purchased a few GTX 470s for their own internal tests or use with Resolve:-)

[David Pirinelli]

The Quadro cards are based on same exact Fermi core as the consumer cards. The 2 main differences are that Nvidia DISABLES much of the Double Precision Float capabilities in consumer cards and that Quadro cards typically have more RAM. In Windows, there are apps that require Double Precision, and these apps will be 2 or 3 times faster with a Quadro.

I am unaware of apps that use Double Precision Float in OSX. I would love to find some, as the Quadro 5000 and Quadro 6000 I have created would have a place to roar. So, when there are no places for a Quadro 4000 to use this one place it excels, you have the simple reality that this is for all intents and purposes, between a GTX450 and a GTX460, or thereabouts. These cards retail in the $100-200 range.

To get people to pay for Mac driver development, they only offer a SINGLE Mac card from the entire Fermi line, the Quadro 4000. The drivers are written and stable for each and every card in the series, there just aren’t any cards with EFI to boot and use the drivers. So you have an OS full of drivers for the range of wonderful cards, and very few ways to use them.

Sascha Haber compared my GTX470 to a Mac Quadro 4000 and found it to be quite a good deal faster. (for half as much money).

The Quadro 4000 is nice in that it is single slot and only needs one power connector. I think there was a GTX460 that was single slot, but I have never held one in my hands. All the Fermi cards I have gotten to work are Dual Power Plug…except the Quadro 5000 which is Dual Slot but only needs 1@ 6 pin power plug.

The Quadro 4000 is a good choice when you need low power consumption or have limited space.

The GTX470 is best possible card for use with Resolve in a standard Mac Pro without need for extra juice.

But if you have a Cubix or Cyclone with dual width slots and plenty of power available, then the GTX480 is the way to go. In such an expander, the use of QUadro 4000 would be a large waste of money, as you would basically be paying $800-$1,200 each for consumer grade cards worth about $150/each.

BTW, many people know that each and every thing I am stating here is true, they just can’t tell you. I understand why this is, but I can offer alternatives. It was an interesting technical puzzle and I solved it.

If you have an expander with Dual Width slots and plenty of power, get a GTX480. I can’t reiterate this enough. The card is basically TWICE as fast as a GTX285.

[Kevin Cannon]

Hi David,

When multiple GPUs were enabled for Resolve on the Mac, there was some discussion that OSX effectively had a ceiling of the number CUDA cores that could be put to work at any one time… Do you know anything to this effect or would you expect three 480s (inside of a Cubix Desktop 4) to have three times the power of one 480?

I’ve been working with approved set-up of three GTX285s (as you know).

KC

Prehistoric Digital
PhD Grading Suite

[David Pirinelli]

I do not have the luxury of having tested 3 @ GTX480s at once.

There is someone who does know the answer, he has been running 3 of my 480s in a Cubix or Cyclone for a couple months but he has never posted about it, that I am aware of.

EDIT: I just discovered that Neil has been posting about the triple GTX480 setup I sold him over on Reduser in the “DaVinci 8.1 Announced” thread. I am glad to see it is still working for him. He also has a triple GTX285 2 Gig setup, though I can’t say whether he has ever tested them against each other.

I am no colorist, I am usually an Art Director looking for work. (all you post guys have put me out of work…couldn’t beat you, had to join you) So I can only ask you guys to report speed results. I will say this, nobody has ever returned a GTX470 or GTX480 because they didn’t like it.

Watching Jake do the tests yesterday, the GTX480 was clearly in a different league from the GTX285, at least for the Noise Reduction testing we did.

The BMD guys might also know the answer.

[Michael Cinquin]

[jake blackstone] “I used the Noise Reduction to simulate a heavy GPU load. I used settings of 2 and 3 on the NR to see the playback speed.”
I guess this is the NR Radius settings. Could you mention the NR Threshold you used as well, so we have a point of reference ?

thanks

Michael Cinquin

Final Cut Pro – Avid Media Composer editor
DaVinci – Color – Baselight colorist
Color profiles for Color
http://www.michaelcinquin.com/tools : tools for FCP | Color | RED | subtitles | Cinema Tools | Timecode – Keycode calculator

[Jake Blackstone]

The test was designed specifically to just compare the performance of different cards. I had set the Threshold and only kept changing NR Radius settings between 2 and 3. But, frankly, I don’t think Threshold settings affects the GPU load that much. But I could be wrong. Nevertheless, as I said, it was only done to compare different cards and Threshold NR settings was always at the same setting for all tests.