fcpchris, those voltages look good to me, but I'm guessing that this is not while the card is under load. When the Hotel scene "crashes," does the entire system go down, or just the LuxMark app?
fcpchris, it is my understanding that all installed video cards need to have an EFI BIOS to avoid the "error while gathering PCI device information" in the System Information app. That has been my experience so far. Since my last post, I have installed an MSI R9 280X Gaming 3G. This seems to be closer to the reference design (1 DVI port) that is supported by the Apple drivers, and the Mac EFI BIOS is working on it. I have not yet removed the resistor to enable 5.0 GT/s PCI speed, so System Information reports it as 2.5 GT/s. Before I installed Mac EFI, System Information gave the error message. I can pretty much guarantee that your card is running at 2.5 GT/s.
As rwillett mentions, enabling 5.0 GT/s may not be necessary to get full performance out of these cards. Others have also reported that it makes little difference for most, if not all, purposes. I think I will do the mod eventually and run the benchmarks again to see what the difference is.
It is odd that the Hotel scene is crashing. I wonder if that is a clue to what is wrong. I have been able to run it with no problems on 10.11.6 with both the Sapphire and MSI cards, with or without the EFI BIOS. I am getting 2120 with the flashed MSI card. I tried the LuxBall scene and got 12189. This is somewhat faster than what you are reporting, but not as much as the difference in our BruceX results would suggest.
I don't have iStat menus installed, but I am guessing that it will show you voltage for each PCIe slot, and the two PCIe booster connectors. Watch the voltages for the slot where the card is installed and the two boosters. Mine are dropping by about 0.03 V under load.
fcpchris, I doubt it's the specific version of your card. I am running the Sapphire R9 280X Vapor-X Tri-X OC (1000 MHz, not the 1100 MHz boost version). I am guessing that your card is supposed to be as fast as mine, and possibly faster.
My card is also not flashed, and I don't believe that any of the Sapphire 280X will work with the Mac EFI BIOS added anyway (because they don't match the reference design in ports provided). It certainly didn't work on mine when I tried it. My card also doesn't have the resistor mod to get 5.0 GT/s PCI speed; I assume it's operating at 2.5 GT/s.
I doubt that the display is significant factor with that card, but I could be mistaken. I'm guessing that the size of the Final Cut window does have a significant effect on performance while editing, though.
Your theory about the power supply could be valid. Some of these high-power cards may overdraw the thin conductors used for the built-in PCIe booster circuits and six-pin cables, causing a substantial voltage drop at the card (I think I read about this somewhere). I was concerned about this, so spliced 8-pin cables directly into the power supply output cables as shown here:
This should provide the required current with less voltage drop or chance of overheating the motherboard traces. I don't know if voltage drop is causing your performance hit, though.
Have you tried running LuxMark? I would be curious to know your score from LuxMark 3's Hotel scene. You can get it here:
When the app comes up, go up to Scene and choose Hotel. I am getting 2150.
fcpchris, I am a little puzzled at your results. They seem much slower than I would expect for your GPU and OS.
With a flashed Radeon HD 7950 (925 MHz) in my Mac Pro 5,1 on 10.11.3, I was getting 21 seconds (see Post #73689). I believe that this is typical of others' results. I now have a single unflashed R9 280X at 1000 MHz and am getting 19 seconds on 10.11.6, which I think is a bit slow.
In the first case I had (2) Xeon X5570 (2.93 GHz 4-core). In the second I had (2) Xeon X5675 (3.06 GHz 6-core), but I doubt that the CPU upgrade had much effect. In both cases I wrote the ProRes 422 to a SATA SSD. Background rendering was turned off, and no footage was pre-rendered.
Because BruceX has no real source footage, and the output file is only 70 MB, I wouldn't expect much performance gain from high-speed storage in this test. I'm guessing that the difference in storage speed is more noticeable when the project contains multiple large source files. As QBE says, BruceX is much more affected by GPU speed than anything else.
Cinebench R15 Results on my Mac Pro 5,1 (2X Xeon X5570, 48GB RAM, Radeon HD 7950 Boost)
Mac OS X 10.11.3
OpenGL: 55.70 fps
CPU: 978 cb
Windows 7 Pro SP1
OpenGL: 73.11 fps
CPU: 978 cb
I'm amazed that the CPU results are exactly the same. In this case, Windows 7 still trounces OS X on OpenGL.
My further testing seems to imply that there are significant improvements in OpenCL support, or video drivers in general, on El Capitan. Here are my LuxMark (Sala) results. I am using a Radeon HD 7950 at 925 MHz.
OS X 10.10.5: 1867
OS X 10.11.3: 2145 (15% increase)
These results are repeatable. I know this isn't directly related to FCP, but it might partly explain the performance improvement.
2009 Mac Pro flashed to 5,1
(2) 2.93 GHz Xeon X5570
32GB 1333 MHz DDR3
Radeon HD 7950 Dual-X Boost (925 MHz) with EFI mod
SanDisk Extreme 120GB SATA SSD (exported master file to this drive)
Final Cut Pro X 10.2.3
BruceX Results (ProRes 422, average of three tests)
OS X 10.10.5: 42 seconds
OS X 10.11.3: 21 seconds (50% improvement!)
(H.264 is not listed as an option when exporting the BruceX project. H.264 _is_ listed when exporting a 720p project that I tried.)
Final Cut appears more responsive in general; much less jerkiness when skimming through a project. Like others here, I am curious as to what causes the improvement. It’s clearly something in 10.11, as I was running the same version of FCP on both versions of the OS.
If you haven't upgraded to El Cap yet, and it works with the rest of your setup, I think you owe it to yourself to give it a try.