Model guide · prices from 2026-09-10 · API prices daily, full survey weekly
Cheapest way to run DeepSeek-R1 (671B MoE)
DeepSeek-R1 (671B MoE) needs ≈403 GB of VRAM at 4-bit — this week that starts at $5.21/hour (13× RTX5090, cheapest tracked walk-up price). FP8 and FP16 below. VRAM calculators stop at the memory number; this connects it to live, graded prices across 20 providers.
MoE note: DeepSeek-R1 (671B MoE) activates ~37B active per token, but the full 671B weights must fit in VRAM — memory scales with total parameters, speed with active ones.
FP16 — ≈1610 GB needed
| GPU | VRAM | GPUs needed | Cheapest tracked | Verifiable A/B |
|---|---|---|---|---|
| A100-SXM | 80 GB | 21 (multi-node) | $19.05/hr | $48.30/hr |
| RTX5090 | 32 GB | 51 (multi-node) | $20.45/hr | — |
| RTX4090 | 24 GB | 68 (multi-node) | $25.43/hr | — |
| A100-PCIe | 80 GB | 21 (multi-node) | $28.35/hr | — |
| MI300X | 192 GB | 9 (multi-node) | $31.05/hr | — |
| H100-PCIe | 80 GB | 21 (multi-node) | $41.79/hr | — |
| B300 | 288 GB | 6 | $45.00/hr | $106.81/hr |
| L40S | 48 GB | 34 (multi-node) | $46.58/hr | — |
FP8/INT8 — ≈805 GB needed
| GPU | VRAM | GPUs needed | Cheapest tracked | Verifiable A/B |
|---|---|---|---|---|
| A100-SXM | 80 GB | 11 (multi-node) | $9.98/hr | $25.30/hr |
| RTX5090 | 32 GB | 26 (multi-node) | $10.43/hr | — |
| RTX4090 | 24 GB | 34 (multi-node) | $12.72/hr | — |
| A100-PCIe | 80 GB | 11 (multi-node) | $14.85/hr | — |
| MI300X | 192 GB | 5 | $17.25/hr | — |
| H100-PCIe | 80 GB | 11 (multi-node) | $21.89/hr | — |
| B300 | 288 GB | 3 | $22.50/hr | $53.41/hr |
| L40S | 48 GB | 17 (multi-node) | $23.29/hr | — |
4-bit — ≈403 GB needed
| GPU | VRAM | GPUs needed | Cheapest tracked | Verifiable A/B |
|---|---|---|---|---|
| RTX5090 | 32 GB | 13 (multi-node) | $5.21/hr | — |
| A100-SXM | 80 GB | 6 | $5.44/hr | $13.80/hr |
| RTX4090 | 24 GB | 17 (multi-node) | $6.36/hr | — |
| A100-PCIe | 80 GB | 6 | $8.10/hr | — |
| MI300X | 192 GB | 3 | $10.35/hr | — |
| H100-PCIe | 80 GB | 6 | $11.94/hr | — |
| H200-SXM | 141 GB | 3 | $11.97/hr | $12.87/hr |
| L40S | 48 GB | 9 (multi-node) | $12.33/hr | — |
Method: VRAM ≈ parameters × bytes/parameter × 1.2 (20% headroom for KV cache and activations at ~8K context). Longer context or high concurrency needs more; training needs far more (optimizer states: ~4–8× weights). GPU counts assume even sharding. Above one 8-GPU node, the interconnect becomes the constraint — restrict to Grade A/B listings. Grades measure what sellers publish — deliverability of the written claim, not measured performance. No fleet holds a measured (Verified) badge yet; that tier arrives with the verification harness. Prices are this week's cheapest tracked walk-up listings (teaser prices excluded); every listing with grades and sources on the GPU pages. Interactive version for any model size: Model Fit.
These prices move weekly
The Delivered Compute Report tracks them — free, every row sourced.
Subscribe free