Segmentation fault Error while runing riva_start.sh

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

Hello,

I just want to run the riva English ASR service
Able to run bash riva_init.sh
but I got error when running bash riva_start.sh
/opt/riva/bin/start-riva: line 4: 108 Segmentation fault (core dumped)
The details are shown in the attachment

Even if I disable all Riva services
I still get the same error

service_enabled_asr=false
service_enabled_nlp=false
service_enabled_tts=false
service_enabled_nmt=false

I also had tried to clean and re-init riva
but still have same issue.

How much VRAM do I need to run a single English ASR model ?
How do I run it on multiple GPUs? (I have 4 RTX2080 )
Or does it only run on a single GPU?

Hardware - GPU : RTX2080 * 4
Operating System : Ubuntu 20.04
Riva Version : riva_quickstart_v2.11.0

docker_log.txt (163.5 KB)
config.sh (13.3 KB)

2

Hi @arthurwu4work

Thanks for your interest in Riva

Thanks for sharing the config.sh and logs

We are not very sure about Multiple RTX 2080 support functioning for Riva,

In General, usage of multiple GPU’s is an experimental feature and may result in undefined behaviours.

Can you share with us the driver version used and also can you try modifying the config.sh

gpus_to_use ="device=0,1,2,3"

and provide us feedback

Thanks

3

Hi @rvinobha ,

thank you for quickly reply!

I had tried
gpus_to_use ="device=0,1,2,3"
but still get the same error

This morning(2023/06/28) I tried changing the GPU to a single RTX2080TI and upgrading the driver from 460.73.01 to 535.54.03.
It worked!

Is the main problem due to insufficient VRAM or driver version ?

So currently RIVA can only use a single GPU for inference at runtime, right?

thank you