Table of Contents
Advertisement
Table 4-22 Comparison of NVIDIA GPUs
Use Case
Technical
Tesla K40
Specifications
Peak
1.43
double-precision
Tflops
FP performance
(board)
Peak
4.29
single-precision FP
Tflops
performance
(board)
Number of physical
1x GK110B
GPUs
Number of CUDA
2,880
cores
Memory size per
12 GB
board
Memory bandwidth
288 Gb/s
per board
Memory I/O
Max Power
235W
Aux Power
PCI Interface
Gen 3 x16
Physical display
connectors
Cooling
Memory Clock
3.0 GHz
NVIDIA GPU Boost
Yes
Base Core Clock
745 MHz
Boost Clocks
810 MHz
875 MHz
Memory Error
Protection
High Performance Computing
Tesla K20X
Tesla K20
1.31
Tflops
Tflops
3.95
Tflops
Tflops
1x GK110
1x GK110
2,688
2,496
6 GB
5 GB
250 Gb/s
208 Gb/s
384-bit GDDR5
320-bit GDDR5
1 x 8-pin and 1 x 6-pin
Gen 2 x16
Passive, Active
2.6 GHz
732MHz
706MHz
Yes Enabled (External & Internal)
Chapter 4. IBM NeXtScale nx360 M4 compute node
Tesla K10
1.17
0.19
Tflops
3.52
4.58
4.30 Tflops
Tflops
2x GK104
3,072 (2 x 1,536)
8 GB (2 x 4 GB)
320 Gb/s
256-bit GDDR5
225W
None
Passive
Passive, Active
2.5 GHz
No
745 MHz
Not applicable
Yes (External Only)
Virtual Graphics
Grid K2
Grid K1
N/A
N/A
N/A
(2 x 2.15
Tflops)
4x GK107
768
(4 x 192)
16 GB
(4 x 4 GB)
116 Gb/s
128-bit DDR3
130W
1 x 6-pin
Gen3 x16
Passive
891 MHz
850 MHz
No
95
Advertisement
Table of Contents
