Surface is an unclassified resource for visualization and data analysis work.  It is limited access.

Here is a breakdown of the "mixed" composition of CPU/GPU architectures on Surface:

Architecture Clock Speed (GHz) CPUs

GPUs
Cores/CPU Node

Cores/GPU Node
Cores Total Memory/Node (GB) Memory Total (GB) TFLOPS peak
Intel 8-core Xeon E5-2670

NVIDIA Tesla K40 GPU
2.6

745 MHz
162

158*2
16

2880
2,592

910,080
256

12*2
41,472

3,792
53.9

451.9

The chart, below, is necessarily simplified to the limitations of the database and consistency with other systems.

Zone
CZ
Vendor
Appro
User-Available Nodes
Login Nodes*
2
Batch Nodes
158
Total Nodes
162
CPUs
CPU Architecture
Intel Xeon E5-2670
Cores/Node
16
Total Cores
2,592
GPUs
GPU Architecture
NVIDIA Tesla K40
Total GPUs
316
GPUs per compute node
2
GPU peak performance (TFLOP/s double precision)
452.00
GPU global memory (GB)
3792.00
Memory Total (GB)
41,500
CPU Memory/Node (GB)
256
Peak Performance
Peak TFLOPS (CPUs)
53.9
Peak TFLOPs (GPUs)
451.9
Peak TFLOPS (CPUs+GPUs)
505.80
Clock Speed (GHz)
2.6
Peak single CPU memory bandwidth (GB/s)
51
OS
TOSS 3
Interconnect
IB QDR
Parallel job type
multiple nodes per job
Run Command
srun
Recommended location for parallel file space
Program
ASC, M&IC
Password Authentication
OTP, Kerberos
Compilers
Documentation