Ramsete2
Jr. Member
Offline
Activity: 87
Merit: 5
|
|
October 12, 2021, 11:26:32 AM |
|
Reducing mem clock on 3070 did the trick. I found this annoying bug: starting the miner, on some cards, it takes some time before showing the average hashrate. Seeing some cards at 0MHs could trigger watchdog and reboot rig. https://ibb.co/vXJ3D6PIs it possible to fix that?
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
|
October 12, 2021, 11:35:06 AM |
|
Reducing mem clock on 3070 did the trick. I found this annoying bug: starting the miner, on some cards, it takes some time before showing the average hashrate. Seeing some cards at 0MHs could trigger watchdog and reboot rig.
It won't trigger the watchdog. When you increase the intensity gpu samples take longer time to complete and the average is only calculated if you have enough samples. We could let the average be calcuted of the existing samples but then the speed will trottle in the beginning.
|
|
|
|
somaton
Jr. Member
Offline
Activity: 207
Merit: 6
|
|
October 12, 2021, 11:36:09 AM |
|
Only with core-lock 1.10 and newer will be as fast as 1.09. What i dont like in 1.09 - high cpu usage.
Core lock? Increase the Temp limit so the cards don't trottle. Nope. Temperatures are around 50c, running trex for almost one year with same settings, if i run tbm 1.09 or older, hashrate is a little bit higher compared to trex. Now, exit 1.09 and run any newer versions - you have about 15-20 hashes less without touching anything. But as i showed before, gpu clocks are jumping like crazy. Usually i keep them ~1450 for 3060 ti and 3070, limited by tdp and they are in 1400-1500 range with 1.09 or older. With newer versions, gpu clock start jumping from 1400 to 1800 and hashrates is lower, specially when clocks are aroung 1700-1800. If i lock gpus core clock with nvidia-smi, then everything is fine again and hashrate is same as 1.09 and a little bit higher compared to trex. I'll make new screenshots later again if you want. I still run one 3060 for tests with newest version with core-lock and it is fine, and one more with 3 gpus with 1.09 --xintensity 64. because it is slower with any newer version. Your miner does not support core-locking and i dont like to lock cores by hand for every gpu in that rig (3060 rev 1, 3060 ti, 3070).
|
|
|
|
Ramsete2
Jr. Member
Offline
Activity: 87
Merit: 5
|
|
October 12, 2021, 11:45:41 AM |
|
Another problem I found with my rig with just AMD RX6800 cards:
│[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:45.318] GPU0 XIntensity set to 352 │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:45.625] GPU6 DAG generated in 7.67 seconds (4702206kb) │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:45.625] GPU6 XIntensity set to 352 │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:45.852] GPU5 Dag buffer verified ok! │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:46.045] GPU4 Dag buffer verified ok! │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:46.421] GPU3 DAG generated in 8.46 seconds (4702206kb) │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:46.421] GPU3 XIntensity set to 352 │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:46.804] GPU0 Dag buffer verified ok! │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:46.839] GPU2 DAG generated in 8.88 seconds (4702206kb) │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:46.839] GPU2 XIntensity set to 352 │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:47.092] GPU6 Dag buffer verified ok! │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:47.215] source/sp_workers/cuda_gpu.cpp:cuda_get_num_devices():73: A cuda runtime error occured: no CUDA-capable device is detected │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:47.216] Shutting down threads │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:47.216] Stratum thread exited │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:47.322] GPU1 DAG generated in 9.37 seconds (4702206kb) │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:47.322] GPU1 XIntensity set to 352 │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:47.322] GPU1 thread exited │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:47.537] GPU5 thread exited │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:47.730] GPU4 thread exited │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:48.139] GPU3 Dag buffer verified ok! │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:48.139] GPU3 thread exited │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:48.217] GPU0 thread exited │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:48.378] GPU2 Dag buffer verified ok! │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:48.378] GPU2 thread exited │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:48.506] GPU6 thread exited │[2021-10-12 11:43:53][info][miner/0]: 0m [2021-10-12 11:43:51.216] Exiting │[2021-10-12 11:43:53][info][miner/0]: Miner exited code:-1
Solutions?
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
|
October 12, 2021, 11:48:34 AM |
|
Another problem I found with my rig with just AMD RX6800 cards: │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:45.318] GPU0 XIntensity set to 352 Solutions?
352 is too high xintensity for AMD cards [2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:47.215] source/sp_workers/cuda_gpu.cpp:cuda_get_num_devices():73: A cuda runtime error occured: no CUDA-capable device is detected
But this could be a bug. was this in v1.13?
|
|
|
|
Ramsete2
Jr. Member
Offline
Activity: 87
Merit: 5
|
|
October 12, 2021, 11:50:35 AM Last edit: November 08, 2021, 01:08:38 PM by mprep |
|
Another problem I found with my rig with just AMD RX6800 cards: │[2021-10-12 11:43:48][info][miner/0]: 0m [2021-10-12 11:43:45.318] GPU0 XIntensity set to 352 Solutions?
352 is too high xintensity for AMD cards But the miner exit with this error: A cuda runtime error occured: no CUDA-capable device is detected It seems it's looking for NVidia cards, not present on this rig...
But this could be a bug. was this in v1.13?
YES, ver 1.13 Tried also with --amd-only option with no success [moderator's note: consecutive posts merged]
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
|
October 12, 2021, 11:54:27 AM Last edit: November 08, 2021, 01:08:49 PM by mprep |
|
Your miner does not support core-locking and i dont like to lock cores by hand for every gpu in that rig (3060 rev 1, 3060 ti, 3070).
If it can be done with the nvidia-smi it's easy to include. We write it down as a feature request. Locking the clocks can also be done by forcing the gpu into p0 state with the nvidia profile inspector. Reboot the rig, and then set the clocks in Msi afterburner afterwards. The clocks will stay the same after reboot if you click on the box to remember your settings. So you only need to tune once, until you reinstall the driver.
It seems it's looking for NVidia cards, not present on this rig...
What does your launch config look like? (bat file) [moderator's note: consecutive posts merged]
|
|
|
|
Ramsete2
Jr. Member
Offline
Activity: 87
Merit: 5
|
|
October 12, 2021, 12:02:09 PM |
|
It seems it's looking for NVidia cards, not present on this rig...
What does your launch config look like? (bat file) --algo ethash --hostname %pool_server% %pool_protocol% %pool_port% --wallet %user% --worker_name %rig_name% --api --api-ip 127.0.0.1 --api-port %api_port% --amd-only --xintensity 352
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
|
October 12, 2021, 12:11:37 PM |
|
--amd-only
ok. thanks. Found it. Fixed in 1.14
|
|
|
|
Ramsete2
Jr. Member
Offline
Activity: 87
Merit: 5
|
|
October 12, 2021, 01:44:23 PM |
|
Please check also the average hashrates (especially on the two 3070).
│[2021-10-12 13:42:05][info][miner/0]: 139m [2021-10-12 13:42:04.327] ID BOARD HASHRATE/W HASHRATE AVERAGE SHARES │[2021-10-12 13:42:05][info][miner/0]: 139m [2021-10-12 13:42:04.327] GPU0 3080 431.51 kH/W 100.97 MH/s 89.32 MH/s 76/0/1 (100.00%) │[2021-10-12 13:42:05][info][miner/0]: 139m [2021-10-12 13:42:04.327] GPU1 3070 515.61 kH/W 64.45 MH/s 86.99 MH/s 31/0/0 (100.00%) │[2021-10-12 13:42:05][info][miner/0]: 139m [2021-10-12 13:42:04.327] GPU2 3080 455.42 kH/W 102.01 MH/s 92.62 MH/s 67/0/2 (100.00%) │[2021-10-12 13:42:05][info][miner/0]: 139m [2021-10-12 13:42:04.327] GPU3 3080 451.38 kH/W 102.01 MH/s 89.50 MH/s 58/0/1 (100.00%) │[2021-10-12 13:42:05][info][miner/0]: 139m [2021-10-12 13:42:04.327] GPU4 3070 541.79 kH/W 64.47 MH/s 83.24 MH/s 34/6/2 (85.00%) │[2021-10-12 13:42:05][info][miner/0]: 139m [2021-10-12 13:42:04.327] GPU5 3080 472.29 kH/W 102.01 MH/s 102.01 MH/s 68/0/2 (100.00%) │[2021-10-12 13:42:05][info][miner/0]: 139m [2021-10-12 13:42:04.327] 2868.01 kH/W 535.94 MH/s 543.69 MH/s 334/6/8 (98.24%)
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
|
October 12, 2021, 02:08:01 PM |
|
Your sortorder is strange seems to be an index mismatch. Cuda normally sort the best cards first in the rig, so I would expect the 3080 cards be be listed first
this is not the original output of the miner:
[2021-10-12 13:42:05][info][miner/0]:
so where did you get it from?
|
|
|
|
Ramsete2
Jr. Member
Offline
Activity: 87
Merit: 5
|
|
October 12, 2021, 02:17:58 PM |
|
Your sortorder is strange seems to be an index mismatch. Cuda normally sort the best cards first in the rig, so I would expect the 3080 cards be be listed first
this is not the original output of the miner:
[2021-10-12 13:42:05][info][miner/0]:
so where did you get it from?
It's from mmpOS shell
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
|
October 12, 2021, 02:20:37 PM Last edit: November 08, 2021, 01:08:58 PM by mprep |
|
It's from mmpOS shell
So it's a bug in mmpOs. The cuda toolkit has a switch to sort gpu's. We use the sortorder of the best gpu first in the list. (default) v1.14 beta has been released. Windows only for now.
Team Black Miner V1.14 beta v1.14 1. Added gpu watchdog, Automatic restart gpu threads on crash. 2. Fixed amd-only option if nvidia drivers where installed but no nvidia gpus enabled. 3. Faster NVIDIA. Increased the default xintensity. Decrease the --xintensity on pools that doesn't pay for stale shares. 4. Force the average hashrate to display at startup even if it has few samples. https://github.com/sp-hash/TeamBlackMiner/releases/TeamBlackMiner_1_14_cuda_11_4_beta.7z https://www.virustotal.com/gui/file/58103e66bf9957dfa32b0b5ff01f6c14ea499d936a00d1bf949a28beb5539682?nocache=1[moderator's note: consecutive posts merged]
|
|
|
|
somaton
Jr. Member
Offline
Activity: 207
Merit: 6
|
|
October 12, 2021, 02:37:21 PM |
|
Your miner does not support core-locking and i dont like to lock cores by hand for every gpu in that rig (3060 rev 1, 3060 ti, 3070).
If it can be done with the nvidia-smi it's easy to include. We write it down as a feature request. Locking the clocks can also be done by forcing the gpu into p0 state with the nvidia profile inspector. Reboot the rig, and then set the clocks in Msi afterburner afterwards. The clocks will stay the same after reboot if you click on the box to remember your settings. So you only need to tune once, until you reinstall the driver. All my gpus in all rigs are in P0 state for many years, it is something else.
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
|
October 12, 2021, 02:43:37 PM |
|
All my gpus in all rigs are in P0 state for many years, it is something else.
Here in p0 clocks are stable except when temp is to hot, or the card doesn't have enough power. Anyway v1.14 has been released. check it out. v1.14 has few detected stales on miningpoolhub on --xintensity 1200 (default) and the poolside profit is amazing.
|
|
|
|
Ramsete2
Jr. Member
Offline
Activity: 87
Merit: 5
|
|
October 12, 2021, 02:56:34 PM Last edit: October 12, 2021, 03:27:02 PM by Ramsete2 |
|
It's from mmpOS shell
So it's a bug in mmpOs. The cuda toolkit has a switch to sort gpu's. We use the sortorder of the best gpu first in the list. (default) v1.14 beta has been released. Windows only for now. I'm facing this weird problem just with a mixed rig with 4x3080 and 2x3070 With rigs full of 3080 no issues on average hashrate Hope this helps
|
|
|
|
dle378
Copper Member
Newbie
Offline
Activity: 77
Merit: 0
|
|
October 12, 2021, 03:32:41 PM |
|
All my gpus in all rigs are in P0 state for many years, it is something else.
Here in p0 clocks are stable except when temp is to hot, or the card doesn't have enough power. Anyway v1.14 has been released. check it out. v1.14 has few detected stales on miningpoolhub on --xintensity 1200 (default) and the poolside profit is amazing. For this release, do we still need to specify the xintensity value since by default it is configured to 1200?
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
|
October 12, 2021, 03:43:29 PM |
|
For this release, do we still need to specify the xintensity value since by default it is configured to 1200?
No need if you mine on miningpoolhub, 2miners, nanopool and many others. But Flexpool, ethermine, nicehash probobly need a lower --xintensity
|
|
|
|
Ramsete2
Jr. Member
Offline
Activity: 87
Merit: 5
|
|
October 12, 2021, 03:48:16 PM |
|
For this release, do we still need to specify the xintensity value since by default it is configured to 1200?
No need if you mine on miningpoolhub, 2miners, nanopool and many others. But Flexpool, ethermine, nicehash probobly need a lower --xintensity IMHO there is a lot of confusion around this xintensity, now default of 1200 is good for certain pools and just for NVidia cards (not AMD). Is not possible that the miner automatically detects GPUs and pools (at least the most famous ones), applying the right value? Then, if you still need to fine tune, you can still modify it manually.
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
|
October 12, 2021, 03:53:09 PM |
|
IMHO there is a lot of confusion around this xintensity, now default of 1200 is good for certain pools and just for NVidia cards (not AMD). Is not possible that the miner automatically detects GPUs and pools (at least the most famous ones), applying the right value? Then, if you still need to fine tune, you can still modify it manually.
yes, this is probobly what we need to do. optimal defaults. The problem is that the rig will show different local speed at different pools. mine at flexpool: 175MHASH, mine at Nanopool 180MHASH (because flexpool is rejecing all the stale work) But then again flexpool might give higher payouts than nanopool. In v1.14 on Amd the the dynamic intensity has been changed , and should perform well on most pools.
|
|
|
|
|