I know that Vegas Pro 13.0 is around the corner and maybe SCS will upgrade their GPU support to use with newer Kepler and ATI architectures but for the time being, the questions that is still on everyone’s mind is, “What’s the best GPU for Vegas Pro 12.0?” I think I may have found an answer.
I recently purchased a 2008 Mac Pro 2.8Ghz 8-Core Xeon with 16GB memory, 128 SSD boot, 2TB RAID 0, and ATI Radeon HD 5870 on eBay for only $740! I configured it to dual-boot with BootCamp and Windows 7 Pro 64-bit and I was really please with the performance I was getting in Vegas Pro 12.0. So much so that I thought I would perform some benchmarks against my Intel Core i7-3930K Sandy Bridge-E 3.2GHz with 16GB memory, 256GB SSD, and Quadro 4000. I was shocked by what I found. (I also posted this on the Creative COW so apologies to those of you who read booth forums. This is a summary of several of my COW posts.)
Here is my little render test: rendertestjr.veg (Vegas Pro 12.0 only)
The purpose of this test is to measure the difference between GPU acceleration for both timeline playback and rendering. The project is 15 seconds long. It has two tracks:. The lower track contains a Generated Media NTSC Color Bars that rotates 360 degrees in 15 seconds. The upper track had Generated Media Noise Texture with the Progress animated so that it would move. I added Sony Bump Map and Sony Glow (both GPU accelerated FX) to the noise texture and the composite level of the event was dropped to 60% so that the rotating colors bars would show through. Compositing random movement ensures that every frame would need to be rendered during the test. This project requires Vegas Pro 12.0 to open.
In order to test on your computer be sure to do the following:
(1) Set your RAM Preview to 200MB (the default) so that frame cacheing doesn’t skew the results.
(2) Set your preview window to Best(Full) and resize to 900x506 for measuring playback fps
(3) Don’t forget to restart Vegas Pro between turning the GPU on and off
(4) For Sony AVC use the Internet 1920x1080-30p template (turn GPU on/off as appropriate)
(5) For MainConcept AVC use the Internet HD 1080p template (turn GPU on/off as appropriate)
That’s it for the ground rules. Now for the results…
ATI Radeon HD 5870
With GPU turned OFF, that project played back at 0.5 fps on my 8-core Mac Pro with Bootcamp (btw, even on my new 6-core/12-thread Intel Core i7-3930K it only played at 0.7 fps with no GPU). With GPU turned ON it plays back at 29.97 fps. So the GPU is giving me a 60x improvement in playback (from 1/2 frame per second to 30 frames per second). That's pretty amazing! :-)
Here are the timings that I recorded for the Radeon:
Apparently, Sony AVC does not take advantage of OpenCL with this card. There was no appreciable improvement (just a few seconds) between CPU Only and Use OpenCL. Maybe if I make the test project longer (like one minute) but the GPU Load meter did not show the Sony encoder using the GPU at all (maybe one or two spikes).
MainConcept was another story entirely. The difference with MainConcept AVC GPU is incredible. MainConcept AVC saw a 3.6x improvement for GPU rendering and 10x overall between no GPU for timeline or render (2:40) and only 0:15 with both turned on!!! On the MainConcept AVC render my CPU was about 29% and my GPU was about 71% utilized. That is closer to what I expected.
At no time did the GPU slow things down. It's important to note that the CPU and GPU are pretty evenly matched. They are both circa 2008/2009 so they compliment each other well so I'm not sure how much having the CPU/GPU combination evenly matched adds to the performance boost. I did use the latest Catalyst drivers from AMD and not the older Apple drivers from BootCamp (which didn’t recognise a GPU with Vegas Pro).
NVIDIA Quadro 4000
Then I set out to do the same tests with my Quadro 4000. Remember, the Quadro 4000 is in my fairly new Intel Core i7-3930K Sandy Bridge-E 3.2GHz (6-core/12-threads) so the CPU Only times are better than my 2008 Mac Pro 2 x 2.8 GHz Quad Core Xeon E5462 (8-cores/8-threads) but much to my surprise, the GPU renders were actually slower on the NVIDIA Quadro 4000 than the ATI Radeon HD 5870!
Here are the times for the Quadro 4000:
I'm not even sure what's going on here. First I was shocked that the Quadro 4000 could not play the timeline better than 2 fps with GPU ON when the Radeon HD was a solid 29.97fps at Best(Full) all day long. This is why the render times are so poor. My Radeon HD 5870 rendered MainConcept AVC with GPU twice as fast as the Quadro 4000! (0:15 vs 0:33). I'm beginning to think that I should be using my 2008 Mac Pro as my primary Vegas Pro editing workstation because the Radeon performs better than the Quadro with Vegas Pro 12.0.
Obviously this was an artificial test. Generated Media is uncompressed in Vegas Pro so I eliminated any lag due to decoding video on the timeline. But it did isolate just the timeline GPU acceleration and rendering GPU acceleration in a project that doesn't play back smoothly without GPU on any PC.
Real-world Tests
So how about real-world rendering? I thought I would test this with Sony’s own “Red Car” test that is posted on their GPU Acceleration page. So I downloaded the 2GB project and ran it against the two computers. I should mention that Sony did their measurements with the preview set to Best(Full) at 900x506 also (which is why I used that in my previous tests) but they used different render templates so please read the Sony instructions and use the same render templates as below.
Here’s what I measured with the “Red Car” project:
What amazed me is that Sony's "Red Car" project plays back at full frame rates 29.97 Best(Full) all the way through using the Radeon HD 5870 but the Quadro 4000 could not maintain the 29.97 fps throughout the timeline like the Radeon HD 5870 did. Again quite shocking to me, because the Radeon is in a slower computer but I guess it might have more to say about Sony making better use of OpenCL with ATI cards than NVIDIA cards for timeline GPU acceleration.
Conclusions
Observation 1: GPU acceleration absolutely works! I got a 10x improvement with timeline GPU on + MainConcept AVC
Observation 2: It appears that Sony makes better use of OpenCL than CUDA, therefore ATI cards give better GPU performance than NVIDIA
Observation 3: The ATI Radeon HD 5870/6970 is about the fastest GPU that works with Sony’s requirement for older cards. (newer cards are actually slower with Vegas Pro 12.0)
From these tests I have concluded that the ATI Radeon HD 5870 is the sweet spot for GPU acceleration in Vegas Pro. Of course the 5870 is no longer available but the Radeon HD 6970 is a newer model of this card and still for sale. Nothing more powerful (i.e., 7xxx series+) will work as good.
Now I realize that the Quadro 4000 isn’t the fastest card from the Fermi line up. I’d be interested in hearing what timings others are getting with their GeForce cards. Maybe there is one that's faster than the Radeon HD 5870. I'm sure everyone would like to know.
I know we’ve been “benchmarked out” on this forum but I thought having a benchmark that we can run in Vegas Pro 12.0 and then in Vegas Pro 13.0 when it is released will tell us if Vegas Pro 13.0 has improved anything in the GPU dept.
Bottom line for me: I got a whole 8-core/16GB Mac Pro with Radeon HD 5870 for less than my $800 Quadro 4000 alone and it performs better than my current computer in real world playback and render tests with Vegas Pro 12.0. I don’t think I would ever waste my money on a Quadro card again.
~jr
I recently purchased a 2008 Mac Pro 2.8Ghz 8-Core Xeon with 16GB memory, 128 SSD boot, 2TB RAID 0, and ATI Radeon HD 5870 on eBay for only $740! I configured it to dual-boot with BootCamp and Windows 7 Pro 64-bit and I was really please with the performance I was getting in Vegas Pro 12.0. So much so that I thought I would perform some benchmarks against my Intel Core i7-3930K Sandy Bridge-E 3.2GHz with 16GB memory, 256GB SSD, and Quadro 4000. I was shocked by what I found. (I also posted this on the Creative COW so apologies to those of you who read booth forums. This is a summary of several of my COW posts.)
Here is my little render test: rendertestjr.veg (Vegas Pro 12.0 only)
The purpose of this test is to measure the difference between GPU acceleration for both timeline playback and rendering. The project is 15 seconds long. It has two tracks:. The lower track contains a Generated Media NTSC Color Bars that rotates 360 degrees in 15 seconds. The upper track had Generated Media Noise Texture with the Progress animated so that it would move. I added Sony Bump Map and Sony Glow (both GPU accelerated FX) to the noise texture and the composite level of the event was dropped to 60% so that the rotating colors bars would show through. Compositing random movement ensures that every frame would need to be rendered during the test. This project requires Vegas Pro 12.0 to open.
In order to test on your computer be sure to do the following:
(1) Set your RAM Preview to 200MB (the default) so that frame cacheing doesn’t skew the results.
(2) Set your preview window to Best(Full) and resize to 900x506 for measuring playback fps
(3) Don’t forget to restart Vegas Pro between turning the GPU on and off
(4) For Sony AVC use the Internet 1920x1080-30p template (turn GPU on/off as appropriate)
(5) For MainConcept AVC use the Internet HD 1080p template (turn GPU on/off as appropriate)
That’s it for the ground rules. Now for the results…
ATI Radeon HD 5870
With GPU turned OFF, that project played back at 0.5 fps on my 8-core Mac Pro with Bootcamp (btw, even on my new 6-core/12-thread Intel Core i7-3930K it only played at 0.7 fps with no GPU). With GPU turned ON it plays back at 29.97 fps. So the GPU is giving me a 60x improvement in playback (from 1/2 frame per second to 30 frames per second). That's pretty amazing! :-)
Here are the timings that I recorded for the Radeon:
-------------------------------------------------------
ATI Radeon HD 5870
-------------------------------------------------------
Timeline GPU Acceleration OFF (Playback 0.5fps)
-------------------------------------------------------
Sony AVC Internet 1920x1080-30p (CPU Only) . . . 1:34
Sony AVC Internet 1920x1080-30p (OpenCL) . . . . 1:30 (0x)
MainConcept AVC Internet HD 1080p (CPU Only) . . . 2:40
MainConcept AVC Internet HD 1080p (OpenCL) . . . . 1:23 (2x)
-------------------------------------------------------
-------------------------------------------------------
Timeline GPU Acceleration ON (Playback 29.97fps)
-------------------------------------------------------
Sony AVC Internet 1920x1080-30p (CPU Only) . . . 0:31
Sony AVC Internet 1920x1080-30p (OpenCL) . . . . 0:29 (0x)
MainConcept AVC Internet HD 1080p (CPU Only) . . . 0:54
MainConcept AVC Internet HD 1080p (OpenCL) . . . . 0:15 (3.6x)
-------------------------------------------------------
Apparently, Sony AVC does not take advantage of OpenCL with this card. There was no appreciable improvement (just a few seconds) between CPU Only and Use OpenCL. Maybe if I make the test project longer (like one minute) but the GPU Load meter did not show the Sony encoder using the GPU at all (maybe one or two spikes).
MainConcept was another story entirely. The difference with MainConcept AVC GPU is incredible. MainConcept AVC saw a 3.6x improvement for GPU rendering and 10x overall between no GPU for timeline or render (2:40) and only 0:15 with both turned on!!! On the MainConcept AVC render my CPU was about 29% and my GPU was about 71% utilized. That is closer to what I expected.
At no time did the GPU slow things down. It's important to note that the CPU and GPU are pretty evenly matched. They are both circa 2008/2009 so they compliment each other well so I'm not sure how much having the CPU/GPU combination evenly matched adds to the performance boost. I did use the latest Catalyst drivers from AMD and not the older Apple drivers from BootCamp (which didn’t recognise a GPU with Vegas Pro).
NVIDIA Quadro 4000
Then I set out to do the same tests with my Quadro 4000. Remember, the Quadro 4000 is in my fairly new Intel Core i7-3930K Sandy Bridge-E 3.2GHz (6-core/12-threads) so the CPU Only times are better than my 2008 Mac Pro 2 x 2.8 GHz Quad Core Xeon E5462 (8-cores/8-threads) but much to my surprise, the GPU renders were actually slower on the NVIDIA Quadro 4000 than the ATI Radeon HD 5870!
Here are the times for the Quadro 4000:
-------------------------------------------------------
NVIDIA Quadro 4000
-------------------------------------------------------
Timeline GPU Acceleration OFF (Playback 0.7fps)
-------------------------------------------------------
Sony AVC Internet 1920x1080-30p (CPU Only) . . . 1:10
Sony AVC Internet 1920x1080-30p (CUDA) . . . . . 1:08 (0x)
MainConcept AVC Internet HD 1080p (CPU Only) . . . 1:17
MainConcept AVC Internet HD 1080p (CUDA) . . . . . 1:07 (0.1x)
-------------------------------------------------------
Timeline GPU Acceleration ON (Playback 2.0fps)
-------------------------------------------------------
Sony AVC Internet 1920x1080-30p (CPU Only) . . . 0:27
Sony AVC Internet 1920x1080-30p (CUDA) . . . . . 0:36 (0x)
MainConcept AVC Internet HD 1080p (CPU Only) . . . 0:40
MainConcept AVC Internet HD 1080p (CUDA) . . . . . 0:33 (0.2x)
I'm not even sure what's going on here. First I was shocked that the Quadro 4000 could not play the timeline better than 2 fps with GPU ON when the Radeon HD was a solid 29.97fps at Best(Full) all day long. This is why the render times are so poor. My Radeon HD 5870 rendered MainConcept AVC with GPU twice as fast as the Quadro 4000! (0:15 vs 0:33). I'm beginning to think that I should be using my 2008 Mac Pro as my primary Vegas Pro editing workstation because the Radeon performs better than the Quadro with Vegas Pro 12.0.
Obviously this was an artificial test. Generated Media is uncompressed in Vegas Pro so I eliminated any lag due to decoding video on the timeline. But it did isolate just the timeline GPU acceleration and rendering GPU acceleration in a project that doesn't play back smoothly without GPU on any PC.
Real-world Tests
So how about real-world rendering? I thought I would test this with Sony’s own “Red Car” test that is posted on their GPU Acceleration page. So I downloaded the 2GB project and ran it against the two computers. I should mention that Sony did their measurements with the preview set to Best(Full) at 900x506 also (which is why I used that in my previous tests) but they used different render templates so please read the Sony instructions and use the same render templates as below.
Here’s what I measured with the “Red Car” project:
------------------------------------------------
SONY “Red Car” Project
------------------------------------------------
NVIDIA Quadro 4000 w/Core i7-3930K 3.2Ghz 6c/12t
------------------------------------------------
MainConcept AVC Internet 1080-30p . . . . . 1:34 (94 seconds)
XDCAM EX HD 1920x1080-60i 35Mbps . . . . . 1:42 (102 seconds)
------------------------------------------------
ATI Radeon HD 5870 w/Xeon 2.8Ghz 8c/8t
------------------------------------------------
MainConcept AVC Internet 1080-30p . . . . . 0:57 (57 seconds)
XDCAM EX HD 1920x1080-60i 35Mbps . . . . . 1:15 (75 seconds)
What amazed me is that Sony's "Red Car" project plays back at full frame rates 29.97 Best(Full) all the way through using the Radeon HD 5870 but the Quadro 4000 could not maintain the 29.97 fps throughout the timeline like the Radeon HD 5870 did. Again quite shocking to me, because the Radeon is in a slower computer but I guess it might have more to say about Sony making better use of OpenCL with ATI cards than NVIDIA cards for timeline GPU acceleration.
Conclusions
Observation 1: GPU acceleration absolutely works! I got a 10x improvement with timeline GPU on + MainConcept AVC
Observation 2: It appears that Sony makes better use of OpenCL than CUDA, therefore ATI cards give better GPU performance than NVIDIA
Observation 3: The ATI Radeon HD 5870/6970 is about the fastest GPU that works with Sony’s requirement for older cards. (newer cards are actually slower with Vegas Pro 12.0)
From these tests I have concluded that the ATI Radeon HD 5870 is the sweet spot for GPU acceleration in Vegas Pro. Of course the 5870 is no longer available but the Radeon HD 6970 is a newer model of this card and still for sale. Nothing more powerful (i.e., 7xxx series+) will work as good.
Now I realize that the Quadro 4000 isn’t the fastest card from the Fermi line up. I’d be interested in hearing what timings others are getting with their GeForce cards. Maybe there is one that's faster than the Radeon HD 5870. I'm sure everyone would like to know.
I know we’ve been “benchmarked out” on this forum but I thought having a benchmark that we can run in Vegas Pro 12.0 and then in Vegas Pro 13.0 when it is released will tell us if Vegas Pro 13.0 has improved anything in the GPU dept.
Bottom line for me: I got a whole 8-core/16GB Mac Pro with Radeon HD 5870 for less than my $800 Quadro 4000 alone and it performs better than my current computer in real world playback and render tests with Vegas Pro 12.0. I don’t think I would ever waste my money on a Quadro card again.
~jr