NVIDIA
Designs the GPUs, superchips, rack systems and networking that most AI training and inference runs on today.
Accelerators · Chip Design · Systems · Platforms · Networking
Santa Clara, United States · Founded 1993
GPUs and accelerators
68
Systems
24
Desktops
5
Architectures
4
NVIDIA hardware on Flopper
The highest figure we hold for each spec, and the part it belongs to.
Highest memory
576 GB
Vera Rubin Superchip
Vendor claimedHighest memory bandwidth
44,000 GB/s
Vera Rubin Superchip
Vendor claimedHighest TDP
1,400 W
B300 SXM 288GB
Official specHighest FP8
35,000 TFLOPS
Vera Rubin Superchip
Vendor claimedHighest BF16
8,000 TFLOPS
Vera Rubin Superchip
Vendor claimedLargest system
2,304 GPUs
DGX SuperPOD GB200
Official specHighest system power
3,840 kW
DGX SuperPOD GB200
Official specGPUs and accelerators
68 parts from NVIDIA, newest first.
Systems
24 systems from NVIDIA, newest first.
| System | GPUs |
|---|---|
| DGX Rubin NVL8 | 8 |
| Groq 3 LPX | 256 |
| HGX Rubin NVL8 | 8 |
| Vera Rubin NVL72 | 72 |
| DGX B300 | 8 |
| DGX GB300 NVL72 | 72 |
| DGX SuperPOD GB200 | 2304 |
| GB200 NVL32 | 32 |
| GB200 NVL36 | 36 |
| GB200 NVL576 | 576 |
| GB200 NVL72 | 72 |
| HGX B200 8-GPU | 8 |
| HGX B300 8-GPU | 8 |
| DGX B200 | 8 |
| DGX H200 | 8 |
| HGX B100 8-GPU | 8 |
| HGX H200 | 8 |
| DGX H100 | 8 |
| HGX H100 4-GPU | 4 |
| HGX H100 8-GPU | 8 |
| DGX A100 320GB | 8 |
| DGX A100 640GB | 8 |
| HGX A100 4-GPU | 4 |
| HGX A100 8-GPU | 8 |
Desktops and workstations
Architectures
Rubin 2026Blackwell Ultra 2025Blackwell 2024Hopper 2022