Upgrade of graphics card

Wildgiles wrote: I don't know where your getting your data jmyres but all the reviews I've seen show the GTX 680 out performing the GTX 570 on all the specs. I'm comparing the GTX 570 to the GTX 680 as they both use about the same power and use 2x6 pin power cables attached to the mother board.

That may be true in gaming, but I look at GPUs for their compute performance (CUDA, OpenCL). Common benchmarks used in reviews have little bearing on how well a card will perform with creative applications like FCPX, Nuke, Resolve, etc.  

This is compounded by the fact that up until recently, both Nvidia and AMD were up against the wall with their prior GPU architectures, as they were too power inefficient to develop any further.  This meant a complete re-design, and a momentary step backward to re-tool their designs for the future.  This "step backward" happened in the 6000 series Radeons and 600 series GTXs.  

These new cards are more power efficient, and gaming performance (high frame rates with large, complex textures) was preserved largely by upping the clock, and increasing vram, but compute performance has generally been worse than the GTX500/ Radeon 5000 series of GPUs.  Here is a popular recent test of several cards with Resolve:

www.barefeats.com/rogue04.html

Resolve is a good platform for this sort of testing because of its heavy reliance on CUDA (Nvidia) and OpenCL (AMD) for most types of processing. It does use the CPU for things like H264 and ProRes coding and decoding, but virtually every other image processing function is offloaded entirely to the GPU.  In this test you see pretty clearly that the prior generation of cards clean up when it comes to creative applications.  This is why the 5870, 570 2.5GB , and 580 have retained so much of their value, even on the used market, and especially now that they are no longer in production.

Also, with the change in architecture has come a change in the way specs are presented, especially with Nvidia.  At the bottom of several of the comparison pages on Nvidia's site you now see a disclaimer that they would love you to miss, but that they include for legal reasons:

"CUDA Parallel Processing Cores cannot be compared between GPU generations due to several important architectural differences that exisit between streaming multi-processor designs."

So, if you like to look at specs, the way CUDA cores are counted between the 500 and 600 series cards is completely different. You'd probably have to divide the new core counts by about three to account for the difference.  Actually, this was also probably done for marketing reasons, as the AMD method for counting cores has always been much higher, for instance the AMD 5870 has "1200 cores" while the equivalent GTX only had about 480 cores.  

Also, some big news happened over the weekend in that Radeon 7000 drivers have finally been included in the latest developer build of OS X 10.8.3.  This means that Mountain Lion will soon support the Radeon 7000 series of cards which includes AMD's Tahiti architecture (79xx), probably in prep for the new Mac Pro.  Tahiti is AMDs latest 28nm "Graphics Core Next" design and based on testing, it will be an absolutely thundering card for Final Cut Pro X which leverages OpenCL heavily.  For perspective, take a look at this Luxmark (OpenCL) test:

media.bestofmicro.com/I/5/331133/original/luxmark.png

The 7970 is over twice as powerful in OpenCL compute performance as the GTX 580 (which is itself 150% more powerful than the GTX 680). And if that wasn't enough, power consumption is similar to a Radeon 5870 or GTX 680. So, basically the most powerfull OpenCL card in the known universe will soon work natively in OS X, and (with a 6-to-8 pin aux power adapter) should slot right into your current Mac Pro's power envelope. Along with the 10.0.6 update, this should cap off a pretty good winter for FCPX users.

To Jmyres.

I agree with your post about the technical specs and speeds. What about the 690 then? I don't think it would be safe to put a 580 or a 7970 in a mac pro tower without an external PSU. I see in the card comparisons that they are comparing a GTX 580 with 3GB RAM to a 680 with 2GB ram how much does RAM affect performance? There hasn't been any real world testing of FCP X with GTX 680 vs AMD's 7950. I'm really interested in seeing how they compare now. I'm guessing that the 7950 is the winner.

I have never tested one myself, but aside from the amount of power they draw, its seems dual GPU cards like the 690 aren't completely recognized by apps in OS X. In my experience, for creative apps at least, once you're beyond about 2GB there are diminishing returns to adding more Vram. More than 2GB may be helpful for something like 4K stereo, or high frame rate 4K, but most people aren't working on those sorts of projects in FCPX yet.

From what Ive seen, many tests indicate the power requirements of the 7970 are similar to the GTX 680, and consistently less than the 580. If thats the case it may work in a Mac Pro without endangering your logic board. But, if that doesn't turn out to be true, I agree that the 7950 may be a lot like the GTX 570, in that it will be a nice combination of performance and cost that will fit in a Mac Pro without additional power.

Just read a report that someone has tested an XFX 7970 with the 10.8.3 Beta and it's working out of the box with no changes. There's also a report from another person that their 7970 is pulling about 50w less at full load that his 5870 did, so it looks like the 7970 is going to be a viable option.

That's really surprising that the 7970 pulls less power than the 5870, as I would think only the 7870 would. This is very promising, and if 10.8.3 release version still supports the 7970 I may buy that instead of the GTX 680.

I'd wait and see before I fly out and purchased one of those 7xxx cards right away. Card support has shown up very often in beta's, to only later not be present in final releases. But if we do get 7xxx support, that'd rock.

Another confirmation from a user over on MacRumors that the XFX 7970 (GHZ Edition) is working with 10.8.3 (12D38) in a 5,1 Mac Pro.

forums.macrumors.com/showthread.php?t=1505726

Also includes some useful details on power distribution and how the 7970 fits within the power requirements of the MP logicboard. I ordered an XFX 7970 Core Edition a few days ago and will post another update once I get it installed.

Yeah I saw that thread last night, was pretty funny the banter back and forth on if the power requirements put your computer at risk or not. That was exactly my worry with the 7970. It looks to be safe anyway since the card comes with 6pin cable and a 8pin to 6pin cable. I'm certainly intrigued, waiting to see if the release version of 10.8.3 supports the 7xxx series.

I installed an XFX 7970 3GB Core Edition I recently picked up after upgrading my Mac Pro to 10.8.2, and applying the 10.8.3 Developer Update (12D38). It works perfectly. Both displays came up right away and I did not have to sleep my machine to get the desktop to appear like some people (mostly on Hackintohes) have mentioned.

All 3GB of Vram shows up in OSX, so the 2GB limit has definitely been lifted. The card runs very cool and quiet. In some basic tests it is easily twice as powerful as the 5870 I had installed previously. Projects in DaVinci Resolve that had trouble playing back on the 5870, now play at 24fps with no problem. Assimilate Scratch and FCPX are both very smooth as well.

I'll post more as I have time to test, but so far its a really nice upgrade.