Difference between revisions of "New Nvidia RTX3090 GPUs"

From ScientificComputing
Jump to: navigation, search
(Introduction)
(Introduction)
 
Line 1: Line 1:
 
==Introduction==
 
==Introduction==
We added 10 GPU nodes with 8 Nvidia RTX3090 GPUs each to the Euler cluster. For the initial phase, we restrict the maximal run time of jobs for those nodes to 24 hours.
+
We added 10 GPU nodes, each with 8 Nvidia RTX3090 GPUs, to the Euler cluster. For the initial phase, we restrict the maximal run time of jobs for those nodes to 24 hours.
  
 
{| class="wikitable"
 
{| class="wikitable"

Latest revision as of 13:19, 27 July 2022

Introduction

We added 10 GPU nodes, each with 8 Nvidia RTX3090 GPUs, to the Euler cluster. For the initial phase, we restrict the maximal run time of jobs for those nodes to 24 hours.

Specs
Architecture Ampere
CUDA cores 10496
Base clock 1.4 GHz
Boost clock 1.7 GHz
Memory 24 GB GDDR6X

Compatibility with CUDA

The support for the Nvidia RTX30XX series of GPUs was added in CUDA 11.1, therefore you cannot use the RTX3090 GPUs with CUDA<11.1.

Access to the new GPUs

Please note that access to the new GPUs is restricted to shareholder groups that invested into GPU nodes in Euler.