search

qubic-li / client

248 stars 25 forks source link

Terrible IT/s on GPUs in last Epoch #85

Open emike09 opened 3 months ago

emike09 commented 3 months ago

Starting with Epoch 104, GPUs are getting terrible IT/s. This has been testing in both WSL and Ubuntu 22.02 native, on different systems. On my RTX 4090, VRAM usage is maximum, reporting 24,491MB used. This seems to be happening to many miners in the discord miners channel. It appears to happen to 4090s, 3080 tis, 3060ti, and possibly more. Tried on client 1.8.10 and 1.9 beta. This appears to be a per-system issue doesn't apply to all systems. One system will function just fine, while another won't. Both affected and non-affected PCs don't respond to reboots in changing whether they are mining properly or not.

qubic@hostname~$ ./qli-Client 2024-04-11 04:27:44 INFO Starting Client 1.8.10.0 2024-04-11 04:27:45 INFO Check for updates 2024-04-11 04:27:46 INFO Your Client is up to date 2024-04-11 04:27:46 INFO AI Training Starting. Your Alias is Hostname-GPU 2024-04-11 04:27:46 INFO Waiting for task (0) 2024-04-11 04:27:48 INFO Trainer: cuda version 104.0 is starting. 2024-04-11 04:27:48 INFO Trainer: Solution threshold: 43 2024-04-11 04:27:48 INFO Trainer: Accept score range [0,42] & [128,256] 2024-04-11 04:27:48 INFO Trainer: Neuron value limit: 0x1 2024-04-11 04:27:48 INFO Trainer: Mining Seed: ***** 2024-04-11 04:27:48 INFO Trainer: 1 CUDA devices are used. 2024-04-11 04:27:51 INFO E:104 | SOL: 0/0 | Try OXOE...VFHF | 0 it/s | 0 avg it/s 2024-04-11 04:27:53 INFO E:104 | SOL: 0/0 | Try OXOE...VFHF | 0 it/s | 0 avg it/s 2024-04-11 04:27:55 INFO Trainer: GPU #0: 0 it/s 2024-04-11 04:27:55 INFO E:104 | SOL: 0/0 | Try OXOE...VFHF | 0 it/s | 0 avg it/s 2024-04-11 04:27:57 INFO Trainer: GPU #0: 0 it/s

Eventually it will get up to something around: 2024-04-11 04:29:29 INFO E:104 | SOL: 0/0 | Try OXOE...VFHF | 13 it/s | 13 avg it/s

MateuszWaliczek commented 3 months ago

Same to Ryzen 9 7950X3D 59-80 it/s CPU usage 41%. Reboot miner fix issue to 130 it/s in Epoch 104 (100% cpu usage). After 10-15 min when "Try" change - same problem Both my rig that same issue

emike09 commented 3 months ago

I'm pretty sure the devs don't check the issues page here. Or at least they don't respond.

BostonBeemah commented 2 months ago

still having the same issue with my 4090 in WSL and hiveos. Tried different versions of drivers and still no luck.

emike09 commented 2 months ago

still having the same issue with my 4090 in WSL and hiveos. Tried different versions of drivers and still no luck.

I just switched GPU mining over to https://www.apool.io/ since it's a PPLNS pool. Wasn't getting any SOLs on 4090 or 3080s trying to find an entire SOL. Their miner seems to run just fine.