
CA16010
The CA16010 adds 81,920 NVIDIA CUDA cores, 10,240 NVIDIA Tensor cores 262.4 TFlops of single-precision performance to one or more servers
PRODUCT HIGHLIGHTS
- 3U High
- Four removable canisters with four GPUs each
- Fully IPMI v2.0-compliant system monitoring capability
- Three 3000-watt power supplies
- Superior Cooling with eight temperature controlled fans
- Up to four PCIe x16 Gen3 cable connections to host server(s)
- Liquid cooling option available
|
Tesla M40
|
Tesla M60
|
Tesla P100 | Tesla V100 | Tesla V100S |
Peak Double Precision Performance
|
.2 TeraFLOPS
|
3.8 TeraFLOPS
|
4.7 TeraFLOPS | 7.8 TeraFLOPS | 8.2 TeraFLOPS |
Peak Single Precision Performance
|
7 TeraFLOPS
|
7.4 TeraFLOPS
|
9.3 TeraFLOPS | 15.7 TeraFLOPS | 16.4 TeraFLOPS |
Number of GPUs
|
1 Maxwell GM200
|
2 Maxwell GM204s
|
1 Pascal GP100 | 1 Volta GV100 | 1 Volta V100S |
Number of CUDA Cores
|
3072
|
4096
|
3584 | 5120 | 5120 |
Memory Capacity and Bandwidth
|
24GB GDDR5 at 288GB/s
|
16GB GDDR5 at 320GB/s
|
16GB CoWoS HBM2 at 720GB/s or 12GB CoWoS HBM2 at 540GB/s | 16GB HBM2 at 900GB/s | 32GB HBM2 at 1134GB/s |
Power Consumption | 250 W | 300 W | 300 W | 250 W | 250 W |