Shop by GPU
Every NVIDIA data center and professional GPU we supply systems around, with the figures that decide between them. Pick the GPU and we show the server, workstation and cluster classes that carry it.
Compare GPUs
| GPU | Memory | Bandwidth | FP4 | NVLink | Form | Best for | Availability | Systems that carry it |
|---|---|---|---|---|---|---|---|---|
| HGX B300 | 288 GB HBM3e | 8 TB/s | Yes | Yes, 1.8 TB/s | HGX SXM, 8 GPU | Training and inference at scale | Allocation required | 5 systems |
| HGX B200 | 180 GB HBM3e | 8 TB/s | Yes | Yes, 1.8 TB/s | HGX SXM, 8 GPU | Training and inference | Build to order | 2 systems |
| HGX H200 | 141 GB HBM3e | 4.8 TB/s | No, FP8 | Yes, 900 GB/s | HGX SXM, 8 GPU | Hopper cluster expansion | Available on request | 2 systems |
| H200 NVL | 141 GB HBM3e | 4.8 TB/s | No, FP8 | Yes, bridge, 2 or 4 GPUs | PCIe, dual slot | PCIe inference and fine tuning | Available on request | 6 systems |
| RTX PRO 6000 Blackwell Server Edition | 96 GB GDDR7 | 1.6 TB/s | Yes | No | PCIe, dual slot, passive | Dense PCIe inference and rendering | Build to order | 7 systems |
| RTX PRO 6000 Blackwell | 96 GB GDDR7 | 1.8 TB/s | Yes | No | PCIe, dual slot, active | Workstations, 1 to 4 GPUs | Build to order | 5 systems |
| RTX PRO 4500 Blackwell | 32 GB GDDR7 | 896 GB/s | Yes | No | PCIe, dual slot, active | Compact and edge systems | Build to order | 4 systems |
| L40S | 48 GB GDDR6 | 864 GB/s | No, FP8 | No | PCIe, dual slot, passive | Inference, graphics and video | Build to order | 6 systems |
| GeForce RTX 5090 | 32 GB GDDR7 | 1.8 TB/s | Yes | No | PCIe, 2 to 3 slot, active | Low cost inference, media generation, content creation | Build to order | 4 systems |
| GB300 NVL72 | 20 TB HBM3e per rack | 576 TB/s per rack | Yes | Yes, 72 GPU domain | Liquid cooled rack | Rack scale inference and training | Allocation required | 2 systems |
| Vera Rubin NVL72 | 20.7 TB HBM4 per rack | 1,400 TB/s per rack | Yes | Yes, 72 GPU domain | Liquid cooled rack | Next generation AI factories | Coming soon | 2 systems |
| HGX Rubin NVL8 | 288 GB HBM4 | 22 TB/s | Yes | Yes, 3.6 TB/s | HGX SXM, 8 GPU | Next generation 8 GPU servers | Coming soon | 2 systems |
Figures are NVIDIA's published values for the GPU. Memory and bandwidth for NVL72 racks are per rack.
Not sure which GPU fits?
Tell us the workload and we propose the GPU and the chassis, with the figures that decided it.
- (858) 613-1818
- sales@computermodules.com
- Monday to Friday, 8am to 5pm Pacific