🔌
New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
// RUN ON DATACENTER GPU
Run Tri-21B on Datacenter GPU hardware
13 options ranked by estimated speed at Q4 / 8k context.
| Hardware | Est tok/s | Fits 8k? | Max ctx | ~Price |
|---|---|---|---|---|
| NVIDIA B200 → | ~599.5 | yes | 624k | — |
| AMD Instinct MI325X → | ~398 | yes | 848k | — |
| NVIDIA H200 → | ~359.7 | yes | 445k | — |
| AMD Instinct MI300X 192GB → | ~351.5 | yes | 624k | ~$15,000 |
| NVIDIA H100 SXM5 80GB → | ~251.1 | yes | 231k | ~$30,000 |
| Intel Gaudi 3 → | ~229.5 | yes | 399k | — |
| NVIDIA A100 SXM4 80GB → | ~152.8 | yes | 231k | ~$15,000 |
| NVIDIA H100 PCIe 80GB → | ~149.9 | yes | 231k | ~$25,000 |
| NVIDIA A100 PCIe 80GB → | ~145 | yes | 231k | ~$12,000 |
| NVIDIA RTX PRO 6000 Blackwell → | ~134.3 | yes | 287k | ~$8,500 |
| NVIDIA RTX 6000 Ada 48GB → | ~71.9 | yes | 118k | ~$6,800 |
| NVIDIA L40S 48GB → | ~64.8 | yes | 118k | ~$10,000 |
| AMD Radeon Pro W7900 48GB → | ~57.3 | yes | 118k | ~$3,999 |
Estimate at Q4 (memory-bound). Click any option for its full per-hardware estimate and measured numbers.
FAQ
What Datacenter GPU hardware runs Tri-21B?▶
On GenAIList, 13 Datacenter GPU options can run Tri-21B at Q4 — the fastest is the NVIDIA B200 at ~599.5 tok/s.
How fast is Tri-21B on Datacenter GPU hardware?▶
Estimated generation speed ranges from ~57.3 to ~599.5 tok/s at Q4 (memory-bandwidth bound). Click any option for its full estimate and any measured numbers.