[Fixed] Radeon R9 290 Problems
I have two XFX Radeon R9 290s that don't seem to display consistent results. First of all, the setup:
Gigabyte G1.Sniper 3
Core i7 2600K
32GB (4x8GB) DDR3-1866
2x XFX Radeon R9 290 (ENFC)
Areca ARC-1230 RAID card (PCIe x8)
PCPower & Cooling Silencer Mk III 1200W PSU
Windows 7 64-bit
Catalyst v14.1 b1.6
I've just installed a fresh copy of Windows and updated it, so there shouldn't be any previous driver issues.
One GPU will mostly perform about where it should, but occasionally has a bad benchmark run. The other GPU is consistently bad. For example, on 3DMark's Fire Strike I'll see scores 8600+ most of the time on the "good" card. The other one will get in the mid-5000 range.
One potential symptom of the problem is the lack of an audible change in fan speed. The "bad" GPUs fan never (by sound) significantly changes its speed. The good one does most of the time...this is how I usually determine mid-benchmark whether the results will be good. I've enabled overdrive and set the maximum fan speed to 100%, but that didn't change anything.
I have tried each card individually in the same slot to rule out any potential lane sharing issues, but the results are still dramatically different between the two GPUs. Crossfire performance is even worse than a single GPU.
I'll post some benchmark results when I get home this afternoon.
Any suggestions?
(cc @thrax)
Gigabyte G1.Sniper 3
Core i7 2600K
32GB (4x8GB) DDR3-1866
2x XFX Radeon R9 290 (ENFC)
Areca ARC-1230 RAID card (PCIe x8)
PCPower & Cooling Silencer Mk III 1200W PSU
Windows 7 64-bit
Catalyst v14.1 b1.6
I've just installed a fresh copy of Windows and updated it, so there shouldn't be any previous driver issues.
One GPU will mostly perform about where it should, but occasionally has a bad benchmark run. The other GPU is consistently bad. For example, on 3DMark's Fire Strike I'll see scores 8600+ most of the time on the "good" card. The other one will get in the mid-5000 range.
One potential symptom of the problem is the lack of an audible change in fan speed. The "bad" GPUs fan never (by sound) significantly changes its speed. The good one does most of the time...this is how I usually determine mid-benchmark whether the results will be good. I've enabled overdrive and set the maximum fan speed to 100%, but that didn't change anything.
I have tried each card individually in the same slot to rule out any potential lane sharing issues, but the results are still dramatically different between the two GPUs. Crossfire performance is even worse than a single GPU.
I'll post some benchmark results when I get home this afternoon.
Any suggestions?
(cc @thrax)
0
Comments
GPU-Z also has some logging tools built in. I'll have to see if it records fan speeds. I kind of suspect that might be the issue.
Still having a problem with Crossfire though. I've tried disabling frame pacing (saw that as a suggestion on another site). Didn't work. I have an AMD FX-8150 system with a 990FX board. I'll have to give it a go in that system.
1. Single R9 290 performance good
2. Dual R9 290 performance w/o Crossfire same as single
3. Dual R9 290 performance w/Crossfire much better
I'm failing on #3. I've tried each GPU in the primary PCIe slot and am getting consistent performance from each card now. Put them in Crossfire and at best I have no change in performance - in the AMD system performance dropped by about 20%. Both systems are on 1200W power supplies (AMD is on a Thermaltake ToughPower 1200W), and both PSUs have recently driven the previous 3x7970 setup. The Intel system (specs in the original post) is the one that was running the 7970s, so the motherboard should be good.
I'm trying to figure out whether it's a bad GPU or if I'm doing something stupid or some combination of both.
Things I've tried on the Intel board:
- Fresh install of Win7
- Disable frame pacing (seemed to help a couple people)
- GPU in PCIe 0 & 1 (x8/x8)
- GPU in PCIe 0 & 2 (x16/x16)
Things I've tried on the AMD board:
- Fresh install of Win7
- Disable frame pacing
- GPU in both PCIe x16 slots with no lane sharing.
Something I saw this morning which seemed a bit odd was Afterburner's report on GPU 0's utilization. It was up and down quite a bit during Bioshock Infinite. GPU 1 was at 100% as expected. I'll run it again and post an image of the graphs.
3DMark Fire Strike Extreme
On Core i7-2600K
- Single (GPU1): 4413
- Single (GPU2): 4413
- Crossfire GPU1/2 (frame pacing on): fail
- Crossfire GPU1/2 (frame pacing off): fail
Bioshock Infinite @ 1920x1200, DX11 w/DDOF (min/max/avg fps)
- Single (GPU1): 13.62/223.15/81.77
- Single (GPU2): 15.00/217.42/81.68
- Crossfire GPU1/2 (frame pacing on): 18.46/214.67/80.16
- Crossfire GPU1/2 (frame pacing off): 17.55/403.46/80.64
Batman Arkham City @ 1920x1200, 8xMSAA, DX11 on, Tessellation high, Extreme detail
- Single (GPU1): 57/142/110
- Single (GPU2): 27/143/109
- Crossfire GPU1/2 (frame pacing on): fail - screen corrupted
- Crossfire GPU1/2 (frame pacing off): fail - screen corrupted
Tomb Raider @ 1920x1200, Ultimate preset
- Single (GPU1): 52.0/89.3/68.7
- Single (GPU2): 50.5/85.6/68.3
- Crossfire GPU1/2 (frame pacing on): fail - immediate freeze
- Crossfire GPU1/2 (frame pacing off): fail - immediate freeze
The Crossfire outcome was similar regardless of GPU order.
Just for kicks, I tried OCCT for a stress test. No problems.
Couldn't complete anything in Crossfire on the AMD system - the driver reset each time within a few seconds of starting the first part of Fire Strike. GPU1 had a max usage of 25%, GPU2 was 100%. Temps were 51C at their max. Clocks were as expected. I suspect Afterburner and 3DMark don't play nicely together on either system.
Run this: https://drive.google.com/file/d/0B1XjtK2OjMgnWjBzOGpwTzBHaW8/edit?usp=sharing
Reboot.
Install older driver.
Reboot.
Report back.
Please do not skip these steps.
On Core i7-2600K
3DMark Fire Strike Extreme
- Single (GPU1): 4413
- Single (GPU2): 4413
- Crossfire GPU1/2 (frame pacing on): 7742
Bioshock Infinite @ 1920x1200, DX11 w/DDOF (min/max/avg fps)
- Single (GPU1): 13.62/223.15/81.77
- Single (GPU2): 15.00/217.42/81.68
- Crossfire GPU1/2 (frame pacing on): 13.56/294.26/133.78
Batman Arkham City @ 1920x1200, 8xMSAA, DX11 on, Tessellation high, Extreme detail
- Single (GPU1): 57/142/110
- Single (GPU2): 27/143/109
- Crossfire GPU1/2 (frame pacing on): 48/244/147
Tomb Raider @ 1920x1200, Ultimate preset
- Single (GPU1): 52.0/89.3/68.7
- Single (GPU2): 50.5/85.6/68.3
- Crossfire GPU1/2 (frame pacing on): 70.0/150.0/105.3
Much better. Thank you @thrax. That worked well.
BioShock Infinite @ 6000x1920, DX11 w/DDOF: 11.25/184.39/39.77 (min/max/avg)
Tomb Raider @ 6000x1920, Ultimate Preset, TressFX: 30.1/46.6/40.2 (min/max/avg)
Batman Arkham City @ 6000x1920, 8xMSAA, DX11 on, Tessellation high, Extreme detail: 31/72/55 (min/max/avg)
Success!