GPU Tray
CS-Storm 500NX 1U server GPU tray components and numbering.
GPU Numbering
Numbering of the GPUs varies depending on the interface (bus number). GPU numbers are silk screened on the SXM board, displayed when using Intelligent Platform Management Interface (IPMI) commands, and referenced by the operating system. The different numbering systems are listed below.
| Bus Number | 09:00.0 | 0A:00.0 | 04:00.0 | 05:00.0 | 89:00.0 | 8A:00.0 | 85:00.0 | 86:00.0 |
| SXM2 Board | GPU 0 | GPU 1 | GPU 2 | GPU 3 | GPU 4 | GPU 5 | GPU 6 | GPU 7 |
| IPMI | GPU 1 | GPU 2 | GPU 3 | GPU 4 | GPU 5 | GPU 6 | GPU 7 | GPU 8 |
| OS | GPU 2 | GPU 3 | GPU 0 | GPU 1 | GPU 6 | GPU 7 | GPU 4 | GPU 5 |
NVIDIA Tesla GPUs
NVIDIA Tesla GPUs use the NVLink™ high-speed GPU interconnect to provide data transfers between GPUs. The NVLink connection provides GPU-to-GPU data transfers at up to 160 Gb/s of bidirectional bandwidth—5x the bandwidth of PCIe Gen 3 x16. The GPUs still communicate using the PCI-e protocol when transferring data to and from the CPUs, memory, and expansion slots.
NVIDIA Pascal GP100 Specifications
| GPU Architecture: | NVIDIA Pascal | |
| Performance: |
| |
| NVIDIA CUDA® Cores: | 3584 | |
| GPU Memory: | 16 GB CoWoS HBM2 | |
| Memory Bandwidth: | 732 GB/s | |
| Interconnect: | NVIDIA NVLink | |
| Max Power Consumption: | 300 W | |
| Thermal Solution: | Passive heatsink | |
| Form Factor: | SXM2 |
NVIDIA Volta GV100 Specifications
| GPU Architecture: | NVIDIA Volta | |
| Performance: |
| |
| NVIDIA CUDA® Cores: | 5376 | |
| GPU Memory: | 16 GB CoWoS HBM2 | |
| Memory Bandwidth: | 900 GB/s | |
| Interconnect: | NVIDIA NVLink | |
| Max Power Consumption: | 400 W | |
| Thermal Solution: | Passive heatsink | |
| Form Factor: | SXM2 |