Is my GPU compatible with Riva?

AI & Data Science Deep Learning (Training & Inference) Riva
1

Hi,

riva_start.sh doesn’t load every models, I wonder if this happen due to my GPU.

Hardware - GPU TU104 [GeForce RTX 2070 SUPER]
Hardware - CPU Intel® Core™ i9-9900 CPU @ 3.10GHz × 16
Operating System Ubuntu 20.04.5 LTS
Riva Version 2.6.0

log.txt (105.6 KB)

2

Hi @hadrien.goutas

Thanks for your interest in Riva

Yes we recommend having GPU with 16+ GB VRAM
Please find the support matrix for reference
https://docs.nvidia.com/deeplearning/riva/user-guide/docs/support-matrix.html

In config.sh

However we can try running only one model for example if we want to run ASR, we can set service flag for what we are interested to true and rest all to false

image
image985×648 25 KB

we can also further drill down to the service we are interested to run (lets say nlp) and further disable the models that we don’t need by adding # to start

image
image1228×279 13.9 KB

Thanks

1 Like
3

Thanks for your reply,
I tried with tts only but got the same output…

4

Hi @hadrien.goutas

Can we try running one nlp model

Thanks

5

Hi,

Same output, here is my config.sh
config.sh (9.6 KB)

6

Hi @hadrien.goutas

If the riva logs states CUDA out of memory or related errors, then it is something with GPU

Can you share the riva_init and riva_start logs

7

Hi,

Here are the logs:
init_logs.txt (11.8 KB)

8

I can’t put 2 files in a message, here is the start logs
start_logs.txt (103.5 KB)

9

Hi @hadrien.goutas

My Apologies, Logs indicate problem/issue with GPU Memory,
So we recommend trying a GPU suggested as per below link
https://docs.nvidia.com/deeplearning/riva/user-guide/docs/support-matrix.html#server-hardware

Also on another note, some other Applications/process might be using the GPU (apart from Riva), so we can check that using command nvidia-smiand try to close them and give a try once