add some notes on troubleshooting GPU usage

This commit is contained in:
whyrusleeping 2019-12-17 11:59:14 -08:00
parent 866715b7c4
commit faad2485ee
3 changed files with 23 additions and 2 deletions

View File

@ -55,6 +55,8 @@ func main() {
log.Info("Starting lotus-bench")
build.SectorSizes = append(build.SectorSizes, 1024)
app := &cli.App{
Name: "lotus-bench",
Usage: "Benchmark performance of lotus on your hardware",

View File

@ -14,6 +14,9 @@ The setup below is a minimal example for sealing 32 GiB sectors on Lotus:
* 8 core CPU
* 128 GiB of RAM
Note that 1GB sectors don't require as high of specs, but are likely to be removed
as we improve the performance of 32GB sector sealing.
## TestNet discoveries
* If you only have 128GiB of ram, enabling 256GB of **NVMe** swap on an SSD will help you avoid out-of-memory issues while mining.

View File

@ -35,3 +35,19 @@ lotus-storage-miner pledge-sector
```
If you see this, that means `pledge-sector` wrote too much data to `$TMPDIR` which by default is the root partition (This is common for Linux setups). Usually your root partition does not get the largest partition of storage so you will need to change the environment variable to something else.
## GPU not being used
If you suspect that your GPU is not being used, first make sure it is properly configured as described in the [testing configuration page](hardware-mining.md). Once you've done that (and set the `BELLMAN_CUSTOM_GPU` as appropriate if necessary) you can verify your GPU is being used by running a quick lotus-bench benchmark.
First, to watch GPU utilization run `nvtop` in one terminal, then in a separate
terminal, run:
```
lotus-bench --sector-size=1024
```
This process uses a fair amount of GPU, and generally takes ~4 minutes to
complete. If you do not see any activity in nvtop from lotus during the entire
process, it is likely something is misconfigured with your GPU.