Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
best open gl card for mac pro 1,1
- Thread starter fhenry
- Start date
- Sort by reaction score
What hardware limitations? The fact that the 1,1 is slower than a 5,1? Performance is going to depend on the game/app you are running, if you have a 30" display and are running with everything cranked up the 5870 will destroy the 5770 (and the GTX 570 or GTX 680 will be even faster than that). If you are running a CPU limited app like Cinebench, then yeah, the GPU isn't going to matter.
I'm fairly certain he is talking about the fact that the 1,1 is PCIE1.0 and the 5,1 is PCIE2.0 which means the 1,1's have half the bandwidth which could hinder a GPU from running at its full potential.
There aren't many workloads where PCIe bandwidth is the limiting factor. The answer to his question is still dependent on what he's going to be using the card for:
Heavy GPGPU work = GTX 570
Gaming = GTX 680
Light web browsing = 5770
The 5770 is slower than the 5870, which is slower than the 570/680. By how much will depend on the application.
Heavy GPGPU work = GTX 570
Gaming = GTX 680
Light web browsing = 5770
The 5770 is slower than the 5870, which is slower than the 570/680. By how much will depend on the application.
He stated "a test on bare feats" which didn't mean he was talking about the article you linked to, it is actually this one:
http://www.barefeats.com/wst10g7.html
His point is that the 5870 isn't faster than the 5770 in about half the tests (on a Mac Pro 1,1) and in the tests where the 5870 is faster, it is considerably slower than on machines with PCIE2.0. It sheds a lot of light on the fact that for Mac Pro 1,1's, it might be more economical to just stick to the 5770 rather than the 5870 since the gains in speed are slight. Now some of this is probably due to processor limitations, but look especially at the OpenCL and WOW benchmarks. All PCIE 2.0 w/ 5870 score pretty much the same except for the Mac Pro 1,1 which scores considerably less.
So the question is whether or not the bandwidth limitation is what caused the lower scores, limited CPU power or what? And if the limitation is due to Bandwidth limitations, does it have the same effect on Nvidia cards and what would be the best card to get knowing these limitations.
I would love to see the above benchmarks done on an upgraded 2006 (to something like my octo 2.66ghz). Then we would know that the issue isn't CPU bound but rather PCIE bandwidth limited.
If the test is CPU limited, then yes, the GPU won't matter. If you see data where the 5770 is identical to the 5870, then the issue is that it's CPU limited. It's not a question of PCIe bandwidth, it's simply that the app/driver can't feed enough work for the GPU for it to make a difference.
I will agree that most likely it CPU bound rather than PCIE. My point is simply, using these tests, it is impossible to know for sure since the comparison is using all Mac Pros that score in the 10,000+ in Geekbench compared to a 2006 "base" that scores around 6,000.
You can rest assured that it is not a PCI-E limitation. Even in the latest and greatest graphics cards the performance hit going from PCI-E 2.0 16x to PCI-E 1.1 16x is like ~3% except in some very rare cases (ex Flight Sim X ).
Anyway it is pretty obvious that the MP 1.1 is severely cpu limited with a HD5870 in the apps tested. That isn't to say a HD5870 wouldn't be beneficial. If you play games at higher resolution (ex 2560x1440+) you will likely notice a significant performance gain due to the much wider memory bandwidth of the HD5870.
p.s. This isn't some ATI only symptom. A faster Nvidia card than the HD5870 will also be bottlenecked.
Anyway it is pretty obvious that the MP 1.1 is severely cpu limited with a HD5870 in the apps tested. That isn't to say a HD5870 wouldn't be beneficial. If you play games at higher resolution (ex 2560x1440+) you will likely notice a significant performance gain due to the much wider memory bandwidth of the HD5870.
p.s. This isn't some ATI only symptom. A faster Nvidia card than the HD5870 will also be bottlenecked.
3% was the case around 2008 with Nvidia 8800 cards (tomshardware made a great article about it, although it was about PCIe 2.0 8x but thats identical bandwidth)
I can only imagine that with GTX580 and now GTX680, the bottleneck hits much harder then 3%.
There are much more recent comparisons than that. If you care to dig around Tom's Hardware also did a comparison with HD5870 which is much closer to the performance of a GTX570 compared to a 8800. That probably isn't even the most recent, just the one I remember.
Also, there are plenty of indirect reviews that you can use to gauge the performance loss. For example, we can use the GTX690. The GTX690 is 2xGTX680 on a single PCB. It can do PCI-E 3.0 however reviews has shown that the performance difference of running the GTX690 on PCI-E 2.0 16x and PCI-E 3.0 16x to be minimal. With that knowledge you can logically extrapolate what the performance hit of half a GTX690 aka GTX680 on a PCI-E 2.0 8x bus (PCI-E 1.1 16x) vs a PCI-E 3.0 8x bus (PCI-E 2.0 16x).
PCIe bandwidth matters in a few very specific cases, and these days it tends to be certain types of GPGPU workloads (i.e. when a ton of data has to be transferred across the bus to the GPU, and then back again once the GPGPU work is complete). It's rare to see games be limited by this with recent OSes, especially Lion and beyond.