Uprgraded GPUs in Mac Pro 5,1 no increase in performance

[giovannylago]

So I recently upgraded to two Flashed MSI Twin Frozr R9 280x's in my 2010 Mac Pro from a single OEM 5870.

I was mainly looking to improve render and export times in FCPX, and I went with Dual R9's because the chip architecture in the R9 280x is very similar to the D700 Firepros in the nMP. I've had the cards installed for about a week now and they are running fine. (Boot screen on every restart, no kernel panics) however I did not get the speed increases I was expecting.

I have both cards installed in x16 pcie lanes in a cubix expander and I'm getting comparable render and export speeds to my single OEM Radeon 5870 in final cut pro. I'm even getting the exact same benchmark scores in Cinebench and Unigine Heaven between the R9's and the 5870.

The Cubix doesn't have a bottleneck. I installed both cards inside the Mac Pro (in the 16x lanes) and I'm getting the exact same benchmarks. What's weirder is that one R9 card is consistently getting benchmarks FPS speeds that are 10 frames/sec slower than the other R9 card and the 5870. (I booted the Mac at different times with each card installed separately to compare.)

I am getting between 60-65 fps speeds in Cinebench and Oceanwave on one R9 and my 5870, and between 50-54 fps on the slower R9 card.

I know people that are getting speeds above 90fps with the R9 280x on OSX. So strange. I thought the cards might be underclocked but I checked their clockspeed in Oceanwave and they are actually running at their "boost" speed of 1020MHz.

I have run out of trouble-shooting options. I even did a clean install of El Capitan on a brand new SSD.

Does anyone out there have R9 280x's installed in their cMP? What kind of speeds are you getting? Any idea what the issue might be?

My machine's specs are:

2 x 3.46 GHz 6 core CPU
32 GB 1333 MHz DDR3
1 TB Crucial SSD Boot Disk.

 

[Fl0r!an]

I don't think anyone is getting 90FPS in Cinebench in a cMP since it's heavily CPU limited. It's perfectly fine that you're achieving the same speed with a HD 5870 and a R9 280X in CB, since it doesn't measure GPU performance at all.

Try running Luxmark, I think it will be more meaningful, since you're interested in compute power.

 

[lowendlinux]

How are you powering them?

Those should be 8 + 6

 

[giovannylago]

I don't think anyone is getting 90FPS in Cinebench in a cMP since it's heavily CPU limited. It's perfectly fine that you're achieving the same speed with a HD 5870 and a R9 280X in CB, since it doesn't measure GPU performance at all.

Try running Luxmark, I think it will be more meaningful, since you're interested in compute power.

From my understanding, Cinebench runs two separate benchmarks, a CPU one that stresses the processor, and an OpenGL one that stresses the GPU. Unless OpenGL involves the CPU for some reason in the cMP.

I'm going to run Luxmark and compare!

 

[h9826790]

My setup is a bit lower performance then yours, dual R9 280 (or 7950 whatever it called). you can click "crossfire" in my signature and check my setup.

Anyway, for FCPX, your setup should greatly improve the rendering time. However, it won't improve export time, that's CPU's job, nothing to do with GPU (unless you send the project to compressor, which is the only way I know can encode with GPU assist).

Also, Cinebench and Oceanwave are NOT good GPU benchmark. Especially not for dual GPU setup. CineBench is the worst GPU benchmark ever, it's testing the CPU single thread performance in high performance system, not GPU. Oceanwave is severely affected by V-sync, that's why the FPS is always stuck at ~60.

If you want to test your GPUs, you better use Luxmark 3.1. HOWEVER, be careful, it will really push your GPUs, dual R9 280x without power mod may draw too much power and shutdown your Mac.

If you want a less risky benchmark, you should go for BruceX in FCPX. My setup can finish the test in 15s, your dual R9 280x should able to finish the test a bit faster.
[doublepost=1466677384][/doublepost]

From my understanding, Cinebench runs two separate benchmarks, a CPU one that stresses the processor, and an OpenGL one that stresses the GPU. Unless OpenGL involves the CPU for some reason in the cMP.

I'm going to run Luxmark and compare!

That's what they want to do, but not what it is actually doing.

That OpenGL benchmark is not even close to a proper GPU OpenGL performance test.

 

[giovannylago]

How are you powering them?

Those should be 8 + 6

The cards right now are installed in a Cubix PCIE Xpander that has it's own 1200watt power supply with 8+6 cables. With that said, when I installed one R9 card directly into the Mac Pro when trouble shooting, I powered it off of the mother board with a standard for mac mini 6pin to 6pin connector, and a custom mini 6pin to 8 cable that was included with the card when I purchased it flashed from eBay. There are no differences in performance whether installed in the Cubix, or directly in the Mac Pro.
[doublepost=1466678599][/doublepost]

My setup is a bit lower performance then yours, dual R9 280 (or 7950 whatever it called). you can click "crossfire" in my signature and check my setup.

Thank you so much, I clicked on "crossfire" and your post was incredibly helpful. I just ran two BruceX tests.

On a single 5870: 42 seconds.
On dual 280x's: 12 seconds.

I will also run Lexmark 3.1, I should be fine because I have both 280x's installed in a Cubix Xpander with it's own power supply. Thanks again.