1.7.1 gfx1010 hang up

[mfu-mcosys]

It seems like something goes wrong with a amd rx5700.
After some time (few mins) without problems a random gfx1010 hangs up with eth

2020-01-24 21:14:16: [Statistics] Ethereum last 10 min - GPU 0: 50.429 MH/s, GPU 1: 50.430 MH/s, GPU 2: 48.223 MH/s, GPU 3: 48.233 MH/s, GPU 4: 48.219 MH/s, GPU 5: 48.203 MH/s, GPU 6: 50.430 MH/s, GPU 7: 48.241 MH/s, GPU 8: 48.221 MH/s, GPU 9: 35.868 MH/s, GPU 10: 32.575 MH/s, GPU 11: 50.430 MH/s. Total: 559.500 MH/s. 2020-01-24 21:14:34: <warning> GPU 3 hung up 2020-01-24 21:14:34: <info> Miner has not been restarted before 2020-01-24 21:14:34: <info> Self-restarting miner process.

DMESG:
[ 1507.095911] gmc_v10_0_process_interrupt: 10 callbacks suppressed [ 1507.095923] amdgpu 0000:0b:00.0: [gfxhub] page fault (src_id:0 ring:24 vmid:1 pasid:32782, for process nanominer pid 2288 thread nanominer pid 2424) [ 1507.096104] amdgpu 0000:0b:00.0: in page starting at address 0x00000006c899b000 from client 27 [ 1507.096205] amdgpu 0000:0b:00.0: GCVM_L2_PROTECTION_FAULT_STATUS:0x00101030 [ 1507.096287] amdgpu 0000:0b:00.0: MORE_FAULTS: 0x0 [ 1507.096346] amdgpu 0000:0b:00.0: WALKER_ERROR: 0x0 [ 1507.096406] amdgpu 0000:0b:00.0: PERMISSION_FAULTS: 0x3 [ 1507.096470] amdgpu 0000:0b:00.0: MAPPING_ERROR: 0x0 [ 1507.096531] amdgpu 0000:0b:00.0: RW: 0x0
I dont know if its a hardware problem...but ethminer works nice (1hour+).

memTweak=0 was added in the config.ini

[Grumpy-Dwarf]

Thanks for reporting, did not test RX 5700 on Linux much, last time I tried they did not work at all in Linux. Will try with updated drivers.

[mfu-mcosys]

Something new: it works without mixing cuda+opencl, if i start nanominer just with all nvidia (cuda) cards it works, if i start nanominer jsut with all amd (opencl) cards, it works. (tested on win10).
The problem seems to be the cuda-nanominer-instance... because if i start 2 instance of nanominer... only the cuda-instance restarts after a mins of working.

phoenix works since 2 days without problems on linux, with a mix of:

GPU1: GeForce RTX 2080 Ti (pcie 1), CUDA cap. 7.5, 11 GB VRAM, 68 CUs
GPU2: GeForce RTX 2080 Ti (pcie 2), CUDA cap. 7.5, 11 GB VRAM, 68 CUs
GPU3: AMD Radeon RX 5700 (pcie 5), OpenCL 2.0, 8 GB VRAM, 36 CUs
GPU4: AMD Radeon RX 5700 (pcie 11), OpenCL 2.0, 8 GB VRAM, 36 CUs
GPU5: AMD Radeon RX 5700 (pcie 14), OpenCL 2.0, 8 GB VRAM, 36 CUs
GPU6: AMD Radeon RX 5700 (pcie 17), OpenCL 2.0, 8 GB VRAM, 36 CUs
GPU7: AMD Radeon VII (pcie 20), OpenCL 2.0, 16 GB VRAM, 60 CUs
GPU8: AMD Radeon RX 5700 (pcie 23), OpenCL 2.0, 8 GB VRAM, 36 CUs
GPU9: AMD Radeon RX 5700 (pcie 26), OpenCL 2.0, 8 GB VRAM, 36 CUs
GPU10: GeForce RTX 2060 SUPER (pcie 27), CUDA cap. 7.5, 7.8 GB VRAM, 34 CUs
GPU11: GeForce RTX 2080 Ti (pcie 29), CUDA cap. 7.5, 11 GB VRAM, 68 CUs
GPU12: GeForce RTX 2080 Ti (pcie 30), CUDA cap. 7.5, 11 GB VRAM, 68 CUs

So i guess its not a HW-problem. :)

[martinberlin]

Any news on this? Did someone got it working.
I get an error message that restarts and that I have to increase netTweak

[Grumpy-Dwarf]

This was reproduced on amdgpu-pro 20.40 Linux driver. Will recheck on latest 20.45 too now.

[Grumpy-Dwarf]

On 20.45 it does not hang but crashes inside driver with same messages in kernel log and terminates.

[Grumpy-Dwarf]

I'm negatively impressed with 20.45 quality. First time I've ever seen my 5700 xt at 98 degrees with default fan settings.

[Grumpy-Dwarf]

Fixed in nanominer 3.1.4