0% GPU load while Chia farming C7

Quadro M4400 about 3500 C7 plots…signage points ~0.5s…all seems to be ok, no errors in log.

When I swap to tourist stuff, RTX 3060 Ti - GPU jumps up/down every 10s. 0% load with Quadro. Is this GPU so powerful that there is no load? or not supported? Anyone experienced similar farming?

2 Likes

Google tells me it supports Cuda 5.2, which is the minimum for Chia, but is the driver up to date?

Mind you it does show the harvester at the bottom of the screen shot. As a test try plotting on it.

Type “nvidia-smi” in a terminal window. Should show you what driver is installed and what version of CUDA is installed.

don’t bother with Maxwell/Pascal cards, the chia harvester is having issues with those (Search for GRResult issues on discord).
I tested with a GTX 980 and while it did work, the harvester/driver crashed ever so often and the power consumption was not great either (because it is slow at computing the proofs)

is it official client or GH?

Chia guys suggested chia plots check. GPU runs 100% for like 5 sec, and then

chia.plotting.check_plots        : ERROR    <class 'RuntimeError'>: GRResult is not GRResult_OK, received GRResult_Failed error in getting challenge qualities for plot xxxx.plot
chia.plotting.check_plots        : ERROR    	Proofs 4 / 30, 0.1333 Filepath: xxxx.plot

There is definitely something wrong with those old workstations GPUs. Driver “latest” stable in Debian.

´´´

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 525.125.06   Driver Version: 525.125.06   CUDA Version: 12.0     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Quadro M4000        On   | 00000000:01:00.0 Off |                  N/A |
| 46%   34C    P8    10W / 120W |    478MiB /  8192MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A     87150      C   chia_harvester                    474MiB |
+-----------------------------------------------------------------------------+

That is the error I was referring to, don"t use older GPU’s for farming (until CNI fixes this, but my guess this is a Nvidia driver issue). One thing you certainly need to do is to upgrade to the latest driver, as outlined in the beta doc :Bladebit Alpha/Beta Information - Google Docs

upgrading to non-debian-proven drivers results in “HW not supported”. I swapped to RTX 3060 Ti, undervolted, and it does nicely farming of 290TiB C7 with just extra +50Wh

Quadro M4440 manages only <300TiB according to bladebit_cuda simulation

While simulation for RTX 3060 Ti

 Context count                 : 1
 Thread per context instance   : 2
 Memory used                   : 333.6MiB ( 0.3GiB )
 Proofs / Challenges           : 1010 / 1000 ( 101.00% )
 Fetches / Challenges          : 631 / 1000
 Filter bits                   : 256
 Effective partials            : 22 ( 3.49% )
 Total fetch time elapsed      : 90.092 seconds
 Average plot lookup time      : 0.143 seconds
 Worst plot lookup lookup time : 1.833 seconds
 Average full proof lookup time: 1.055 seconds
 Fastest full proof lookup time: 0.608 seconds

 compression | plot count | size TB    | size PB    
------------------------------------------------
 C7          | 14344      | 1202       | 1.20

And that would be the limit of my farm after replotting

I still hopped farming could be done with CPU…sadly, computing overhead, efficiency of Chia lags behind the Gigahorse/mmx significantly.

AMD Ryzen 9 3950X 3.85GHz @ 1.15V

Context count                 : 16
 Thread per context instance   : 2
 Memory used                   : 5338.1MiB ( 5.2GiB )
 Proofs / Challenges           : 73 / 100 ( 73.00% )
 Fetches / Challenges          : 49 / 100
 Filter bits                   : 256
 Effective partials            : 2 ( 4.08% )
 Total fetch time elapsed      : 25.893 seconds
 Average plot lookup time      : 0.528 seconds
 Worst plot lookup lookup time : 32.565 seconds
 Average full proof lookup time: 23.486 seconds
 Fastest full proof lookup time: 14.406 seconds

*** Warning *** : Your worst plot lookup time of 32.565 was over the maximum set of 8.000.
 compression | plot count | size TB    | size PB    
------------------------------------------------
 C7          | 3875       | 324        | 0.32