Snowy Hardware

Nodes	CPUs	Cores	Memory	Scratch	GPUs	Name	Comment
122	2x Xeon E5-2660 2.2 GHz	16 (2 x 8)	128GB	3/4TB	N/A	s1-s12, s14-s40, s42-s120, s201-s204	.
49	2x Xeon E5-2660 2.2 GHz	16 (2 x 8)	128GB	3/4TB	Tesla T4	s151-s163, s164-s200	.
15	2x Xeon E5-2660 2.2 GHz	16 (2 x 8)	512GB	3/4TB	N/A	s121-s129, s131, s133-s137	-C mem512GB or -C fat
12	2x Xeon E5-2660 2.2 GHz	16 (2 x 8)	256GB	3/4TB	N/A	s139-s150	-C mem256GB or -C fat
1	2x Xeon E5-2660 2.2 GHz	80 (10 x 8)	4TB	3/4TB	N/A	s229	-C mem4TB -p veryfat
1	2x Xeon E5-2660 2.2 GHz	16 (2 x 8)	256GB	3/4TB	Tesla T4	s138	-C mem256GB or -C fat

• Job walltime : 30 days (720 hrs)
• interactive walltime : 12 hrs
• -p devel or -p devcore walltime : 1 hr
• No -t walltime : 1 hr
• Total jobs allowed (running and waiting) : 5000
• Path to node-local temp disk space (scratch folder) : $SNIC_TMP

GPU on Snowy: NVIDIA T4 (16GB)

Performance in TFLOPS and TOPS (10¹² operations per sec):

Data type	A100	T4
FP64	9.7 | 19.5*	0.25
FP32	19.5	8.1
TF32	156**	N/A
FP16	312**	65
BF16	312**	N/A
Int8	624**	130
Int4	1248**	260

*Performance on Tensor Core.
**Up to a factor of two faster with sparsity.

Submitting jobs : Slurm on snowy