eGPU compute card

Get answers to your questions about color grading, editing and finishing with DaVinci Resolve.
  • Author
  • Message
Offline

David Backovsky

  • Posts: 4
  • Joined: Mon Oct 09, 2017 7:52 pm

eGPU compute card

PostWed Oct 11, 2017 2:16 pm

I'm about to purchase a laptop with Thunderbolt 3 and an eGPU enclosure. The laptop has a very good card already, but I want to use the eGPU as a compute card for video editing and use the internal quadro card for 10-bit output. Does anybody have experience with this, is this possible?
(I know that Resolve 14 does not need a GUI card, but the eGPU still offers much superior compute power) Thanks.
Offline

Peter Chamberlain

Blackmagic Design

  • Posts: 3049
  • Joined: Wed Aug 22, 2012 7:08 am

Re: eGPU compute card

PostSat Oct 14, 2017 8:28 pm

The final monitoring video output needs to use a BMD card to the calibrated viewing monitor.
DaVinci Resolve Product Manager
Offline

Uli Plank

  • Posts: 1876
  • Joined: Fri Feb 08, 2013 2:48 am

Re: eGPU compute card

PostSun Oct 15, 2017 7:33 am

And according to the tests by Puget Systems with todays powerful GPUs you are looking at diminishing returns with more than one (they didn't test Linux):

https://www.pugetsystems.com/labs/artic ... n-Xp-1060/
Offline

Andrew Kolakowski

  • Posts: 2624
  • Joined: Tue Sep 11, 2012 10:20 am

Re: eGPU compute card

PostSun Oct 15, 2017 11:41 am

Interesting- always predicted that multiple GPUs are not going scale well in Resolve, but this is very disappointing.
2 GPUs make sense in some cases, but going further seams to be total waste of money.
Offline

Uli Plank

  • Posts: 1876
  • Joined: Fri Feb 08, 2013 2:48 am

Re: eGPU compute card

PostSun Oct 15, 2017 4:01 pm

BM always said that Linux is the way to go with more than 2.
Offline

David Backovsky

  • Posts: 4
  • Joined: Mon Oct 09, 2017 7:52 pm

Re: eGPU compute card

PostSun Oct 15, 2017 4:05 pm

Peter Chamberlain wrote:The final monitoring video output needs to use a BMD card to the calibrated viewing monitor.


Thank you for your reply I thought that a quadro card's 10-bit output capability would suffice. What are the BMD solutions for laptops? Are there thunderbolt 3 cards or other alternatives?
Offline

Andrew Kolakowski

  • Posts: 2624
  • Joined: Tue Sep 11, 2012 10:20 am

Re: eGPU compute card

PostSun Oct 15, 2017 4:18 pm

Resolve GUI preview is not designed to be accurate or use 10bit surface offered by Quadro cards (10bit is supported on some Macs but accuracy is still not easy to achieve).
There are many simple/cheap and more advanced cards offered by BM, including USB3, TB2/TB3 based.
Offline

Peter Chamberlain

Blackmagic Design

  • Posts: 3049
  • Joined: Wed Aug 22, 2012 7:08 am

Re: eGPU compute card

PostMon Oct 16, 2017 2:17 am

Uli Plank wrote:And according to the tests by Puget Systems with todays powerful GPUs you are looking at diminishing returns with more than one (they didn't test Linux):

https://www.pugetsystems.com/labs/artic ... n-Xp-1060/



There are so many aspects not covered in these tests. For a GPU/multiple GPU test, eliminating all the other bottlenecks is the first step and the text clearly shows the config is not optimised for a true GPU eval.
DaVinci Resolve Product Manager
Offline

Uli Plank

  • Posts: 1876
  • Joined: Fri Feb 08, 2013 2:48 am

Re: eGPU compute card

PostMon Oct 16, 2017 7:40 am

I stand corrected, thanks Peter.
Offline

Peter Chamberlain

Blackmagic Design

  • Posts: 3049
  • Joined: Wed Aug 22, 2012 7:08 am

Re: eGPU compute card

PostMon Oct 16, 2017 9:17 am

No Problem Uli, i would not have known of the test unless you highlighted it.
DaVinci Resolve Product Manager
Offline

Andrew Kolakowski

  • Posts: 2624
  • Joined: Tue Sep 11, 2012 10:20 am

Re: eGPU compute card

PostMon Oct 16, 2017 11:25 am

Peter Chamberlain wrote:
Uli Plank wrote:And according to the tests by Puget Systems with todays powerful GPUs you are looking at diminishing returns with more than one (they didn't test Linux):

https://www.pugetsystems.com/labs/artic ... n-Xp-1060/



There are so many aspects not covered in these tests. For a GPU/multiple GPU test, eliminating all the other bottlenecks is the first step and the text clearly shows the config is not optimised for a true GPU eval.


Possible, but their system is about best single CPU PC out there which many people consider. Many people think about multiple GPUs, but this clearly shows that it's about pointless. Also- as they said their test wasn't stress test, but more real case scenario (no one is going to use 40 nodes of blur in real project).

I don't see any big/real problems with their test machines. They are about the best you can build as for single CPU machines.
There are possible issues eg.- by using RED footage they put high push on CPU, so this could be limiting factor.
They should show CPU and GPUs loads during tests, so this would tell us more about system balance.
They could also use DNxHR etc as source- this would free CPU and let GPUs shine more. But yet again- this is real case scenario. People do build these machines for RED footage and this shows that 3xGPUs (even with fastest CPUs out there) is pointless.

Return to DaVinci Resolve

Who is online

Users browsing this forum: AndreasOberg, Andrei Andreev, Brian Salisbury and 28 guests