Not sure if scvi is using my GPU or not

General scRNA-seq
,
1

I got this message and I am not sure if scvi is using my GPU or not
Another thing that I have in mind,
Is there a way to get normalised gene expression from model to my Seurat object?

SCVI Model with the following params: 
n_hidden: 128, n_latent: 20, n_layers: 1, dropout_rate: 0.1, dispersion: gene, 
gene_likelihood: zinb, latent_distribution: normal
Training status: Not Trained
Model's adata is minified?: False
> model$train()
GPU available: True (cuda), used: True
TPU available: False, using: 0 TPU cores
IPU available: False, using: 0 IPUs
HPU available: False, using: 0 HPUs
LOCAL_RANK: 0 - CUDA_VISIBLE_DEVICES: [0]
2

Hi, you can check that the model training is using your GPU on the following line:

GPU available: True (cuda), used: True

Let me get back to you regarding getting normalized expression values into Seurat objects.

3

For anyone wondering, you can also run:

nvidia-smi

in the command line which will show you if GPU is running anything.

1 Like
4

You can also run: gpustat

1 Like
5

Thanks alot for you help

6

So with model training not running

Mon May  8 13:36:28 2023       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 520.61.05    Driver Version: 520.61.05    CUDA Version: 11.8     |
|-------------------------------+----------------------+----------------------+
| 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 P4000        On   | 00000000:2D:00.0  On |                  N/A |
| 53%   58C    P8    12W / 105W |   1319MiB /  8192MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A      1150      G   /usr/lib/xorg/Xorg                334MiB |
|    0   N/A  N/A      1387      G   /usr/bin/gnome-shell               79MiB |
|    0   N/A  N/A     72614      G   ...RendererForSitePerProcess       71MiB |
|    0   N/A  N/A    135703      G   ...mviewer/tv_bin/TeamViewer        5MiB |
|    0   N/A  N/A    181236      C   ...esources/app/bin/rsession      730MiB |
|    0   N/A  N/A    227985      G   ...715383201972569677,131072       91MiB

with model training running

Mon May  8 13:38:43 2023       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 520.61.05    Driver Version: 520.61.05    CUDA Version: 11.8     |
|-------------------------------+----------------------+----------------------+
| 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 P4000        On   | 00000000:2D:00.0  On |                  N/A |
| 53%   55C    P0    34W / 105W |   1357MiB /  8192MiB |     18%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|    0   N/A  N/A      1150      G   /usr/lib/xorg/Xorg                348MiB |
|    0   N/A  N/A      1387      G   /usr/bin/gnome-shell              100MiB |
|    0   N/A  N/A     72614      G   ...RendererForSitePerProcess       85MiB |
|    0   N/A  N/A    135703      G   ...mviewer/tv_bin/TeamViewer        5MiB |
|    0   N/A  N/A    181236      C   ...esources/app/bin/rsession      728MiB |
|    0   N/A  N/A    227985      G   ...715383201972569677,131072       83MiB |
+-----------------------------------------------------------------------------+

So apparently the GPU is not used, is there any fix for that?? I mean more power is running thru the GPU but RAM usage is not going up…
When using Python and scanpy it runs faster like 3x faster

7

scvi does not use much gpu memory. It only uses the memory of the model (~15 mb) and the memory of 128 cells (also should be in low MB range).

how fast is the model training? How fast does it train if you add (use_gpu=False) to .train(use_gpu=False)

8

I tried it without GPU and it took so much longer,
Thanks alot I think it using the GPU after all

and for normalized gene expression is used this

Nor <- model$get_normalized_expression(adata)
Nor1  <- py_to_r(Nor)
Nor1 <- t(Nor1)
GEXscvi[["Nor"]] <- CreateAssayObject(counts = Nor1[, colnames(GEXscvi)])
DefaultAssay(GEXscvi) <- "Nor"