Torch Spec Sheet

The Torch cluster has 518 Intel "Xeon Platinum 8592+ 64C" CPUs, 29 NVIDIA H200 GPUs & 68 NVIDIA L40S GPUs connected together via Infiniband NDR400 interconnect. Further details on each kind of node is provided in the table below.

Torch System Resources

Type	Nodes	CPU Cores	GPUs	Memory (GB)	CPUs per Node	GPUs per Node	Memory per Node (GB)
Standard Memory	186	23,808	N/A	95,232	128	N/A	512
Large memory	7	896	N/A	21,504	128	N/A	3,072
H200 GPU	29	3,712	232	59,392	128	8	2,048
L40S GPU	68	8,704	272	34,816	128	4	512
RTX Pro 6000 GPU	2	256	16	3,072	128	8	1,536
H100 GPU	15	1,440	60	23,040	96	4	1,536
A100 GPU	9	576	36	4,608	64	4	512
A100 GPU	34	2720	136	34,816	80	4	1,024
Login	4	512	N/A	1024	128	N/A	256
Data Transfer	2	64	N/A	512	32	N/A	256
Provisioning	4	320	N/A	1024	80	N/A	256
Scheduler	2	64	N/A	1024	32	N/A	512
Total	N/A	43,072	752	280.1(TB)	NA	NA	NA

Torch was tested in June 2025 using the LINPACK benchmark system, which is the basis for all HPC systems ranked on the Top500 list. It had a theoretical maximum performance of 12.25 PF/s thanks to its powerful GPU resources, of which LINPACK was able to use 10.79 PF/s, thus placing it at #133 on the listed.

Torch was recently ranked #40 on the Green 500 list, a global list of the most energy efficient supercomputers in the world thanks to its advanced liquid cooling system.

Torch Stakeholder Resources

Many GPUs on Torch are owned by stakeholder groups who have prioirty access to the resources. The breakdown of these GPUs into stakeholder groups is as follows:

Stakeholders	H200	L40S	H100	A100	RTX Pro 6000	Total
Tandon SoE	112		60	60		232
Courant (CS)	48	52				100
CDS	24			24		48
Chemistry				56		56
CILVR				32		32
Individual PIs	24	12			16	52
IT Public	24	208				232
Total	232	272	60	172	16	752