GTX 680M or QUADRO 5010M in a NOTEBOOK?

[Subhadip Sen]

That's interesting, 17-X is no longer listed. After a bit of digging around, it seems the Clevo 17 SB-E chassis has supply issues. It's not just Origin, but also some competitors with similar products that are showing out of stock or delisted.

No worries, 17-S Pro is still plenty capable.

Dell Precision M6700 is worth a look at as well, they have far superior display options (100% AdobeRGB 10-bit IPS!) alongwith upto Quadro K5000M if you need Quadro. Like I mentioned before, K5000M is faster than the desktop Quadro 6000! For general Premiere Pro CS6 work, 17-S Pro will be better value for money.

[Olivier Madar]

I get my clevo notebook back with the GTX680m instead of the 670M.
Premiere works with the 680. I don't know how to see the improvment. In Resolve 9, the difference between the 670m and the 680m is not big : i can read Quad HD red raw files at 19 fps for 1/4 res good and 16 fps for 1:2 res good. With the 670M, it was about 2 or 3 fps less.
If you want me to do test, just ask.
MyClevo notebook config :
- Clevo 170 EM
- Win 7 64 bits SP1
- i7 3920XM 2.9GHz
- 32 g of RAM
- 3x samsung 512 SSD
- GTX680M

Olivier

[Olivier Madar]

Neat video seems to be a good "reel" performance tester. In the preferences menu, you have a "performance" checker to evaluate the maximum numbre of frame/sec you can do, and the best combination between number of CPU, CPU alone or CPU + GPU
With the 670M i get my sustem optimized with 4 CPU + GPU 670M (6.76 frames/sec)
With the 680M, i get best performance with 5 CPU + GPU 680M (8.77 frames/sec)

20% more in denoise with is a very powerhungry task is not so bad !


Olivier

[Olivier Madar]

I tried to see what the charge is on the GPU with widget GPU observer.

With Premiere CS6 and GTX680, I don't "eat" more than 10% of the GPU power. With Resolve, I "eat" 30% of the GPU power. So, I think we need to wait for new software to be able to use the increase of the GPU power.



Olivier

[Michel Rabe]

Olivier, thank you for the test!
Have you tried rendering for example 4K footage scaled down in a 2K timeline with 'maximum render quality' checked? I think there should be a huge speed boost coming from the GPU.

[Olivier Madar]

Yes, I have tried, but I am very disapointed. With the maximum render quality checked, my rendering time gets very long (x12 the reel time) and my GPU is nearly not used (only 5 to 10% sometimes and sometines 0%). So, i thin,k the CS6 does not like the GTX680 at this time, and I need to wait for some optimization from Adobe.
All my tests are done with a 1080 timeline to a 1080p file. Maybe the GPU is only usefull with images reduction. I will do a testwith a quadHD timeline to a 1080p.


Olivier

[Richard Tran]

Olivier,
Are you testing these times for an H.264 output file?

I think CUDA only accelerates certain render outputs more than others.

[Olivier Madar]

No, i render always for the "same settings" check boxes. It provides MPEG file with maximum informations. Then i do my h264 from that file wich is my master (in the same way than i did it with ProRes)


Olivier

[Olivier Madar]

I just did a 2 rapid test :

1) with a timeline Quad HD --> 1080p h264 ipad2 preset (h264). For 5 min, it takes 1hour 25 (x17 reel time). The processor is always at 100% but the GPU goes from 0 to 10% max.
2) With the timeline Quad HD --> same settings as the sequence checked (So 4K HD in MPEG2). For 5 min it takes 7 minutes to render. Idem for the processor and GPU activity.

So, if you want to render quickly, do a sequence in the same size than your final output. I think it is good to always check the same settings as you sequence checkbox. And do trust the GPU is doing a big job for rendering (or you need a GPU validated by Adobe like the 580. Could someone test it ?).


Olivier

[Michel Rabe]

Thanks!
Yes, it should render fastest when having the timeline's resolution/framerates identical to the output.
Have you tried the workaround to 'activate' the 680M for CS6 (write the GPU's name in the CUDA cards textfile)?
Rendering to h.264 for iPad sounds like heavy compression so it should naturally take much longer than the other output I guess.