ComputeAssay

Model guide · prices from 2026-09-10 · API prices daily, full survey weekly

Cheapest way to run Llama 3.1 405B

Llama 3.1 405B needs ≈243 GB of VRAM at 4-bit — this week that starts at $3.21/hour (8× 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.

FP16 — ≈972 GB needed

GPUVRAMGPUs neededCheapest trackedVerifiable A/B
A100-SXM80 GB13 (multi-node)$11.79/hr$29.90/hr
RTX509032 GB31 (multi-node)$12.43/hr
RTX409024 GB41 (multi-node)$15.33/hr
A100-PCIe80 GB13 (multi-node)$17.55/hr
MI300X192 GB6$20.70/hr
H100-PCIe80 GB13 (multi-node)$25.87/hr
H200-SXM141 GB7$27.93/hr$30.03/hr
L40S48 GB21 (multi-node)$28.77/hr

FP8/INT8 — ≈486 GB needed

GPUVRAMGPUs neededCheapest trackedVerifiable A/B
A100-SXM80 GB7$6.35/hr$16.10/hr
RTX509032 GB16 (multi-node)$6.42/hr
RTX409024 GB21 (multi-node)$7.85/hr
A100-PCIe80 GB7$9.45/hr
MI300X192 GB3$10.35/hr
H100-PCIe80 GB7$13.93/hr
B300288 GB2$15.00/hr$35.60/hr
L40S48 GB11 (multi-node)$15.07/hr

4-bit — ≈243 GB needed

GPUVRAMGPUs neededCheapest trackedVerifiable A/B
RTX509032 GB8$3.21/hr
A100-SXM80 GB4$3.63/hr$9.20/hr
RTX409024 GB11 (multi-node)$4.11/hr
A100-PCIe80 GB4$5.40/hr
MI300X192 GB2$6.90/hr
B300288 GB1$7.50/hr$17.80/hr
H100-PCIe80 GB4$7.96/hr
H200-SXM141 GB2$7.98/hr$8.58/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