GPUs available & how to use it

This version is outdated by a newer approved version.

This version (2024/06/06 15:40) was approved by goldenberg.The Previously approved version (2023/05/17 14:36) is available.

This is an old revision of the document!

Rank	Machine	Performance	Accelerators
1.	Frontier	1206 PFLOPs/s	AMD Instinct MI250X
2.	Aurora	1012 PFLOPs/s	Intel Data Center GPU Max
3.	Eagle	561 PFLOPs/s	NVIDIA H100
4.	Fugaku	442 PFLOPs/s
5.	LUMI	379 PFLOPs/s	AMD Instinct MI250X
6.	Alps	270 PFLOPs/s	NVIDIA GH200 Superchip
7.	Leonardo	241 PFLOPs/s	NVIDIA A100 SXM4
8.	MareNostrum 5 ACC	175 PFLOPs/s	NVIDIA H100
9.	Summit	148 PFLOPs/s	NVIDIA V100
10.	Eos NVIDIA DGX	121 PFLOPs/s	NVIDIA H100

Model	#cores	Clock Freq (GHz)	Memory (GB)	Bandwidth (GB/s)	TDP (Watt)	FP32/FP64 (GFLOPs/s)
19x GeForce RTX-2080Ti n375-[001-019] - only in a special project
	4352	1.35	11	616	250	13450/420
45×2 nVidia A40 n306[6,7,8]-[001-019,001-019,001-007]
	10752	1.305	48	696	300	37400/1169
62×2 nVidia A100-40GB n307[1-4]-[001-015]
	6912	0.765	40	1555	250	19500/9700

Interactive mode

1. VSC-5 >  salloc -N 1 -p zen2_0256_a40x2 --qos  zen2_0256_a40x2 --gres=gpu:2

2. VSC-5 >  squeue -u $USER

3. VSC-5 >  srun -n 1 hostname  (...while still on the login node !)

4. VSC-5 >  ssh n3066-012  (...or whatever else node had been assigned)

5. VSC-5 >  module load cuda/9.1.85    
            cd ~/examples/09_special_hardware/gpu_gtx1080/matrixMul
            nvcc ./matrixMul.cu  
            ./a.out 

            cd ~/examples/09_special_hardware/gpu_gtx1080/matrixMulCUBLAS
            nvcc matrixMulCUBLAS.cu -lcublas
            ./a.out

6. VSC-5 >  nvidia-smi

7. VSC-5 >  /opt/sw/x86_64/glibc-2.17/ivybridge-ep/cuda/9.1.85/NVIDIA_CUDA-9.1_Samples/1_Utilities/deviceQuery/deviceQuery

SLURM submission gpu_test.scrpt

#!/bin/bash
#
#  usage: sbatch ./gpu_test.scrpt          
#
#SBATCH -J A40     
#SBATCH -N 1                           #use -N only if you use both GPUs on the nodes, otherwise leave this line out
#SBATCH --partition zen2_0256_a40x2
#SBATCH --qos zen2_0256_a40x2
#SBATCH --gres=gpu:2                   #or --gres=gpu:1 if you only want to use half a node
 
module purge
module load cuda/9.1.85
 
nvidia-smi
/opt/sw/x86_64/glibc-2.17/ivybridge-ep/cuda/9.1.85/NVIDIA_CUDA-9.1_Samples/1_Utilities/deviceQuery/deviceQuery

Performance	Power Efficiency

GPUs available & how to use it

TOP500 List June 2024

Components on VSC-5

Working on GPU nodes Interactively

Working on GPU using SLURM

Real-World Example, AMBER-16