Long and short is that the 6-core machine with D500s seems to be beating the 12-core D700s in tests. Presumably because the CPU is playing a part in those processes, and from what Barefeats is able to surmise, FCPX is limiting multicore usage to 8 threads; meaning you get more out of the 6-core 3.4GHz's pure "per-processor" speed than you do out of the 12-core 2.7GHz.
Hi everybody - I was also looking into taking advantage of Mavericks and the classic Mac Pro with two GPU's versus one. I wanted to let everyone know about an issue I ran into: Apple changed the way Slot 4 accepts RAID controller cards in 2009, and I have a late 2010 Mac Pro in my studio at home - I use a RAID controller card made by CalDigit and it is not possible to move my CalDigit RAID controller card from Slot 3 where it works beautifully, to Slot 4. Not because the card couldn't perform in Slot 4 (both are x4 lane slots) but because Apple has designed the 2009 - 2012 Mac Pro's to only accept their own RAID controller card, which only controls internal drives - so no ports to external storage on the card).
This means that I cannot physically fit two GPU processor cards in Slot 1 and Slot 2 without removing my CalDigit RAID controller card. I thought about losing my miniSAS RAID but it is too important to me and outweighs the increase in render performance. I'm so bummed out about it but that's the block for me.
I could install a second ATI Radeon 5770 video card by removing the RAID controller card but if I'm unwilling to lose that super fast storage, what's the point in the speed benchmark test? So that's where it ends for me. I think pre-2009 Mac Pro's do not have this limitation.
Regardless of my issue with physical space due to having a RAID controller card that has to be in Slot 3, do any of you worry about the way the video cards stacked on top of each other restrict the airflow to the graphic card in Slot 1? I looked at the way my ATI Radeon 5770 cards vents air at the top with the fan and when I went to place the second card into x16 lane 2 it sat immediately above the card in slot 1 leaving almost no space for the card below in Slot 1 to vent, and also the bottom 5770 is heating up the top 5770.
Have any of you worked on editing or in any way taxing to the GPU cards for extensive time and retested to see if the speeds change after both cards begin to overheat?
thanks - makes sense. I looked at upgrading to the 5780 but the specs on the 5780 are vague to me and for the cost I didn't see a big difference - I think it even has the same amount of video dedicated RAM as the 5770 - so I don't understand what makes it better and more expensive. Is the 7970 a big step up? How can I know what the true differences are from one card to another?
The only other thought I have is this: Does a dual video card set up improve your editing versus a single card? Does it allow smoother playback etc? Maybe that rendering score is only one part in the whole equation?
For me, a second 5770 card was an inexpensive experiment as I have a 2006 Mac Pro, so it can only improve my fcpx performance. Things feel faster for me.
Unless you are in touch with someone who has the same hackintosh setup as you it will be difficult to tell whether or not their experiences will apply to your particular hardware setup.
Generally we have found that two matching GPUs will always improve performance when compared to running the same GPU by itself. Mixing and matching Nvidia and AMD GPUs does NOT provide any improvement, and there are still some cases (7970 vs dual 5770 for example) in which a single newer GPU will provide better performance than two older GPUs. Also keep in mind that this benchmark is precisely just that, a benchmark; real world performance will vary based on the different tasks you through at any given hardware combination.
If you do get a second 650ti please let us know your results. Forgive me if I missed it, but I think no one has tried two Nvidia GPUs yet on this thread.
darrylyee wrote: I've been looking for a thread like this. Thanks to Pollux for mentioning it in a facebook group.
My result: 76 seconds (ProRes)
4GHz Hackintosh 3930k
16GB RAM, GTX 650ti
Has it been determined that GPU VRAM has an affect? I'm looking at either a single 7970 or 280x or dual 7850 (w/4GB each). Or should I just add another 650ti for $100?
Looking at video ram usage while rendering it's fairly low usage. You can check it out using opengl monitor so the video ram might have little impact at least for now. What you do need is matching cards . I assume it will vary depending on what your doing.
I'm not sure how much we can trust Barefeats published results, just yet at least.
You see they had published results with a footnote saying that dual 5770's outperformed any other single AMD GPU -- in fact they went as far as to say that dual 5770's performed only slightly slower than dual 5970's!
They didn't provide data to support those statements & then later when they did publish numbers those numbers did not support their earlier statements, but the tests used between the 2 articles weren't identical.
Now I no longer see their praise for dual 5770's, it would appear to me that they silently edited out those statements. I saw no notice of retraction or mea culpa. So for now at least I'm not purchasing another GPU on their results alone. The results they've been publishing since clearly do show that the 5770's are not as good an an option as they previously had claimed.
I've seen several suggestions to enable CUDA when editing and disable it when encoding the resulting video. But this was related to some Windows software. It would be interesting to see the results of this test for the same hardware, but with 1) CUDA disabled and 2) dedicated graphics card disabled.
I think the two machine configs which will give us the rest of the picture would be a 6core/D700 and an 8core/D700. I'd love to see those results before my MacPro order goes from "Processing" to "Preparing for Shipment". If there's no performance benefit to be had (or even a penalty) by going from the 6 to 8 core, then I could be persuaded to save the $1500 cost.
I stopped at my local Apple store yesterday. On a base-model 4 core Mac Pro exporting to ProRes 422 BruceX took 27 seconds. I also made up some 4-5 layer sequences with key framed transforms, chromakey, animated titles, cc, retiming etc. A 15-16 second sequence was rendering in about 15 seconds (1080p).