Skip to content

Commit 10f5360

Browse files
committed
Switch CUDA package & slight clarifications
1 parent f3682ed commit 10f5360

1 file changed

Lines changed: 20 additions & 11 deletions

File tree

content/hardware-failure.md

Lines changed: 20 additions & 11 deletions
Original file line numberDiff line numberDiff line change
@@ -123,39 +123,48 @@ Click the 'Run' button to begin the program.
123123

124124
### GPU Burn (for NVIDIA GPU's only)
125125

126-
We can also test the GPU by using GPU Burn; first, if we're on Ubuntu, we'll need to install git and CUDA with this command:
126+
We can also test the GPU by using GPU Burn. First, ensure git and CUDA are installed with this command:
127127

128128
```bash
129-
sudo apt install git system76-cuda-latest
129+
sudo apt install git nvidia-cuda-toolkit
130130
```
131131

132-
Next, we can clone the repository with this command:
132+
Next, clone the repository and enter its directory with these commands:
133133

134134
```bash
135135
git clone https://github.com/wilicc/gpu-burn.git
136+
cd gpu-burn
136137
```
137138

138-
Now that we have cloned it, we can move into that directory like so:
139+
Now we'll compile it:
139140

140141
```bash
141-
cd gpu-burn
142+
make
142143
```
143144

144-
Now we'll compile it:
145+
#### GCC Errors (Pop!_OS 22.04)
145146

146-
```bash
147-
make CUDAPATH=/usr/lib/cuda
147+
On some versions of Pop!_OS and Ubuntu (including 22.04), the default version of GCC may fail to compile gpu-burn. Install a compatible version with this command:
148+
149+
```
150+
sudo apt install g++-10
151+
```
152+
153+
Then, run `make` again, specifying which version of GCC to use:
154+
155+
```
156+
make NVCCFLAGS='-ccbin /usr/bin/g++-10'
148157
```
149158

150-
(If you receive an error about your version of GCC being too new, consider installing the appropriate version with e.g. `sudo apt install g++-10` and then passing `NVCCFLAGS='-ccbin /usr/bin/g++-10'` or equivalent as a Make argument.)
159+
#### Running GPU Burn Tests
151160

152-
For NVIDIA **RTX** GPUs we can use Tensor cores to run it like so (this example will run it for 60 minutes/1 hour):
161+
For NVIDIA **RTX** GPUs, run the test using Tensor cores like so (this example will run it for 3600 seconds/1 hour):
153162

154163
```bash
155164
./gpu_burn -tc 3600
156165
```
157166

158-
For NVIDIA **GTX** GPUs run it like so (this example will run it for 60 minutes/1 hour):
167+
For NVIDIA **GTX** GPUs, run the test using CUDA cores like so (this example will run it for 3600 seconds/1 hour):
159168

160169
```bash
161170
./gpu_burn 3600

0 commit comments

Comments
 (0)