FCPX: AMD vs NVIDIA

[CuamckuyKot]

There is rumor that AMD graphics card beats their NVIDIA analogs in Final Cut Pro X rendering more than twice.

The best way to check this out is Alex 4D test:

http://blog.alex4d.com/2013/10/30/brucex-a-new-fcpx-benchmark/

So, my results for:
Mac Pro 4,1 X5650x2 and GTX970 is 58 seconds.

Whats about you? Wanna know especially AMD cards owners results.

 

[h9826790]

Mac Pro 4,1, W3690, HD7950 x2 (at stock 800MHz), 840 Evo (at lower optical bay via the SATA 2 port), FCPX 10.2, OSX 10.11.4 Beta 3 (15E39d) around 17s.

This is the more realistic result, because some normal apps running in the background (photos, safari, chrome, some server apps, etc.)

If I OC both 7950 a little bit to 850MHz, run the Benchmark after reboot but nothing else in the background. It's about 15s (best result).

 

[Synchro3]

Mac Pro 4,1, W3690 6-core 3.46 GHz, 48 GB 1333 RAM, GTX 770 4 GB x1, Samsung 850 Pro and Apple PCI Express SSD, OS X 10.8.5, FCPX 10.0.9 around 53s.

Does not matter if the export drive is an Apple PCI Express SSD or a Samsung 850 Pro (SATA), result is always 53 seconds.

 

[goMac]

FCPX leans a lot on GPU compute, which Nvidia is not good at. It's why Nvidia isn't in the newest Mac Pro.

Don't let the Nvidia fans know that though. Nvidia is #1 at everything! All sunshine all the time in Nvidia land!

 

[jwpoof]

Nvidia cards, by my observation are better for most things. But if you use FCPX, it's hard to beat what the nMP can deliver. My personal feeling is that if you're not using FCPX, then you're better off buying a 5.1 and loading up with NVidia. But I ran FCPX on a 5.1 for years, and it's vastly superior on the nMP. It's basically a FCPX cylinder. For other stuff, I find the nMP frustrating, but not frustrating enough to switch to windows and Premiere.

 

[pat500000]

Nvidia cards, by my observation are better for most things. But if you use FCPX, it's hard to beat what the nMP can deliver. My personal feeling is that if you're not using FCPX, then you're better off buying a 5.1 and loading up with NVidia. But I ran FCPX on a 5.1 for years, and it's vastly superior on the nMP. It's basically a FCPX cylinder. For other stuff, I find the nMP frustrating, but not frustrating enough to switch to windows and Premiere.

How is it with avid?

 

[nerdynerdynerdy]

How is it with avid?

I'd be interested in this too.

 

[mhafeez]

Here's mine;
iMac27 mid-2011, 32gb ram, 2tb raid0 ssd, gtx780m, elcapitan 10.11.3
render time = 19seconds

 

[Synchro3]

Here's mine;
iMac27 mid-2011, 32gb ram, 2tb raid0 ssd, gtx780m, elcapitan 10.11.3
render time = 19seconds

You report seriously a comparable result as with two HD7950?

Did you execute the benchmark exactly as described?

The most directly comparable iMac has:

 

[mhafeez]

First, if your iMac has a GTX 780m, it's an iMac Late 2013, not Mid 2011.

Second, you report seriously a comparable result as with two HD7950?

Did you execute the benchmark exactly as described?

The most directly comparable iMac has:
View attachment 616389

Actually mine really is a mid-2011 27" iMac. I just upgraded it's GPU with GTX 780M with 2x 1TB Samsung 840evo in RAID0 setup. I did executed the benchmark three times and I got an average of 19 seconds.

 

[Synchro3]

Actually mine really is a mid-2011 27" iMac. I just upgraded it's GPU with GTX 780M with 2x 1TB Samsung 840evo in RAID0 setup. I did executed the benchmark three times and I got an average of 19 seconds.

View attachment 616390

Did you this?:
3. In Final Cut Pro X, go to ‘Final Cut Pro Preferences…’ – in the Playback tab make sure ‘Background Render’ is off.

12. If possible do the export at least three times. Your configuration’s BruceX Score is the average export time in seconds. Before timing the next export, restart Final Cut (otherwise the exports speed up each time because X does a little caching renders to save time). -> Delete the project if Background Render was already on.

Where did you buy the GTX 780m 4 GB?

 

[h9826790]

Here's mine;
iMac27 mid-2011, 32gb ram, 2tb raid0 ssd, gtx780m, elcapitan 10.11.3
render time = 19seconds

Impossible, and not matching with others result as well. (52s for single 780m)

May I ask you to try this?

1) Please make sure background rendering is off.

2) Delete the project.

3) Fully quit FCPX.

3) Remove all the FCPX cache files in finder.

4) Empty trash

5) Open FCPX, and follow the exact step of the test.

The result should be in 5K resolution, not 1080P etc.

 

[mhafeez]

Did you this?:
3. In Final Cut Pro X, go to ‘Final Cut Pro Preferences…’ – in the Playback tab make sure ‘Background Render’ is off.

12. If possible do the export at least three times. Your configuration’s BruceX Score is the average export time in seconds. Before timing the next export, restart Final Cut (otherwise the exports speed up each time because X does a little caching renders to save time). -> Delete the project if Background Render was already on.

Where did you buy the GTX 780m 4 GB?

I got the GTX780m from ebay by seller named "ChooseYourDestiny" something like that.

Impossible, and not matching with others result as well. (52s for single 780m)

May I ask you to try this?

1) Please make sure background rendering is off.

2) Delete the project.

3) Fully quit FCPX.

3) Remove all the FCPX cache files in finder.

4) Empty trash

5) Open FCPX, and follow the exact step of the test.

The result should be in 5K resolution, not 1080P etc.

I have done several times again and again, followed all steps accordingly especially No.3 and No.12 as mentioned and I still got an average of 19 seconds. I think it is due to Apple's newly improved Metal.

 

[h9826790]

I have done several times again and again, followed all steps accordingly especially No.3 and No.12 as mentioned and I still got an average of 19 seconds. I think it is due to Apple's newly improved Metal.

I got the GTX780m from ebay by seller named "ChooseYourDestiny" something like that.

Can one more person confirm the 780m's FCPX performance under 10.11.3?

If this is true, this card should be the most energy efficient GPU for FCPX.

Even the Maxwell card seems can't give out this performance with just around 120W TDP.

Metal is there, but seems not doing much at this moment. And most of our cards also support Metal, I am not saying that you are lying, but have to find out if this 200% improvement is due to software optimisation. Or not running the benchmark correctly.

If your 19s is a true result, then few other cards (at least some Nvidia card) should also benefit from the new driver / Metal / optimisation, etc.

 

[Synchro3]

Mac Pro 4,1, W3690 6-core 3.46 GHz, 48 GB 1333 RAM, GTX 770 4 GB x1, Samsung 850 Pro and Apple PCI Express SSD, OS X 10.8.5, FCPX 10.0.9 around 53s.

Ok, I was curious, and this is interesting: Now on OS X 10.11.3 with newest Nvidia drivers, FCP 10.2.3, with the same Mac Pro 4,1, W3690 6-core 3.46 GHz and the same GTX 770 4 GB benchmark is new 29s (!).

I think I should change to El Capitan when using FCP.

 

[h9826790]

Ok, I was curious, and this is interesting: Now on OS X 10.11.3 with newest Nvidia drivers, FCP 10.2.3, with the same Mac Pro 4,1, W3690 6-core 3.46 GHz, 48 GB 1333 RAM, and the same GTX 770 4 GB benchmark is new 29s (!).

I think I should change to El Capitan when using FCP.

It's good to hear that, but I am still doubt if the 780m can beat the 770 by 30% in FCPX.

 

[pat500000]

I got the GTX780m from ebay by seller named "ChooseYourDestiny" something like that.



I have done several times again and again, followed all steps accordingly especially No.3 and No.12 as mentioned and I still got an average of 19 seconds. I think it is due to Apple's newly improved Metal.

Can you post pictures?

 

[koyoot]

It's good to hear that, but I am still doubt if the 780m can beat the 770 by 30% in FCPX.

Desktop 770 and 780M are the same cores. The problem is: 780M has core clock of 797 MHz. 770 has core clock of 1046 MHz.

P.S. This is proof that something in Metal Works .

 

[Synchro3]

I think the Nvidia web drivers (non-beta) 346.03.05f01 for 10.11.3 are significantly optimized. Even with 29 seconds my Mac Pro 4,1/GTX 770 beats all the newer machines in the chart above.

I guess all the benchmarks in the chart have to be repeated on El Capitan.

Someone with a Mac Pro with GTX 970/980/Ti here?

 

[SoyCapitanSoyCapitan]

Nvidia users should use the same web drivers, app version and OS version when comparing. That chart is old now and shouldn't be used for comparison.

 

[h9826790]

I think the Nvidia web drivers (non-beta) 346.03.05f01 for 10.11.3 are significantly optimized. Even with 29 seconds my Mac Pro 4,1/GTX 770 beats all the newer machines in the chart above.

I guess all the benchmarks in the chart have to be repeated on El Capitan.

Agree, if dual 7950 can finish in about 15s, that means a single card can finish in about 30s, the 38s in the chart is outdated. On the Nvidia side, the result are even more outdated now.

For OP, did you run your test in 10.11.3? If not, can you try that?

 

[koyoot]

Agree, if dual 7950 can finish in about 15s, that means a single card can finish in about 30s, the 38s in the chart is outdated. On the Nvidia side, the result are even more outdated now.

For OP, did you run your test in 10.11.3? If not, can you try that?

This test does not scale linearly with multi-GPU configs.

 

[Synchro3]

A thesis: If a single GTX 770 can finish in about 30s, and dual 7950 in about 15s, does that mean that OpenCl in 10.11.3 for Nvidia cards is significantly improved and now equal to AMD cards? Or it's more a FCP optimization?

 

[koyoot]

I think we have too little in data to make any conclusions. Bah, it would be hard to make any assumptions!

 

[h9826790]

This test does not scale linearly with multi-GPU configs.

Understand that it won't be 100% linear, however, isn't it usually the efficiency will decrease when GPU number increase (100% scaling is the ideal case)? Therefore, if 2x GPU require 15s, isn't that means a single GPU usually require less then 30s? And assume 30s is actually on the conservative side (unless the CPU actually contribute a lot in the test result)? So, a single GPU should not require 38s to finish the test?