GPU Selector
Recommend GPU configurations based on model size and concurrency requirements
Requirements
GPU Reference (estimated street prices, USD)
| GPU | VRAM | Price per card | TDP | Class |
|---|---|---|---|---|
| NVIDIA RTX 3090 | 24GB | $900 | 350W | Consumer |
| NVIDIA RTX 4090 | 24GB | $1,600 | 450W | Consumer |
| NVIDIA RTX 5090 | 32GB | $2,000 | 575W | Consumer |
| NVIDIA L4 | 24GB | $2,500 | 72W | Datacenter |
| NVIDIA RTX 6000 Ada | 48GB | $6,800 | 300W | Workstation |
| NVIDIA L40S | 48GB | $8,000 | 350W | Datacenter |
| NVIDIA A100 40GB | 40GB | $9,000 | 250W | Datacenter |
| NVIDIA A100 80GB | 80GB | $14,000 | 300W | Datacenter |
| AMD Instinct MI300X | 192GB | $20,000 | 750W | Datacenter |
| NVIDIA H100 80GB | 80GB | $28,000 | 700W | Datacenter |
| NVIDIA H200 141GB | 141GB | $32,000 | 700W | Datacenter |
| NVIDIA B200 192GB | 192GB | $42,000 | 1000W | Datacenter |
Prices are estimates for reference only — check vendor quotes for actual pricing. VRAM estimates assume a 4096-token context with FP16 KV cache; longer contexts increase KV cache usage significantly. All computation happens locally in your browser.