Can I Run Devstral-2 123B on NVIDIA H100 80GB (PCIe)?
Written by Jakub Rusinowski · Last updated July 21, 2026
Yes, but it's tight. Devstral-2 123B needs 77.9 GB at 8K context (74.3 GB weights + 3.6 GB KV cache/overhead) at Q4_K_M and the NVIDIA H100 80GB (PCIe) has 80 GB, leaving about 2.1 GB of headroom, at roughly 21 tok/s (estimated) with room for up to 8K of context.
See what else this hardware can run →
The numbers
| Model | Devstral-2 123B |
| Parameters | 123B |
| Quantization | Q4_K_M |
| VRAM needed (8K context) | 77.9 GB |
| NVIDIA H100 80GB (PCIe) VRAM | 80 GB |
| Verdict | Runs |
| Max context | 8K |
| Estimated speed | ~21 tok/s |
Devstral-2 123B on the NVIDIA H100 80GB (PCIe): VRAM by quantization
| Quant | VRAM needed | Fits 80 GB? | Max context | Whole PC |
|---|---|---|---|---|
| F16 | 249.7 GB | ✗ No | — | — |
| Q8_0 | 134.4 GB | ✗ No | — | — |
| Q6_K | 104.5 GB | ✗ No | — | — |
| Q5_K_M | 90.9 GB | ✗ No | — | — |
| Q4_K_M | 77.9 GB | ✓ Yes | 8K | — |
| Q3_K_M | 56.1 GB | ✓ Yes | 64K | — |
| Q2_K | 44.1 GB | ✓ Yes | 64K | — |
Assumes an 8K-token context with an f16 KV cache, on an estimated architecture — this model publishes no config we can read. A longer window needs more; a quantized KV cache needs less. “Max context” is the largest window that still fits in 80 GB. Figures are estimates from parameter count, quantization and memory bandwidth — the analyzer lets you tune KV-cache quant and context.
Context costs VRAM too. At Q4_K_M on the NVIDIA H100 80GB (PCIe): 8K 77.9 GB · 32K 86.6 GB ✗ · 128K 121.2 GB ✗. Past 32K it no longer fits 80 GB — a q8_0 KV cache buys roughly half of that back, which the calculator will price for you.
What to do next
Nothing to change — run it at Q4_K_M. If you want more capability, check whether the next size up also fits.
No compatible GPU? Devstral on 16 GB of system RAM, CPU only: runs, but it is tight.
Running Devstral-2 123B on this card:
At 2 hrs/day, buying (~$25,000) beats renting at $0.77/hr after about 45.1 years.
Affiliate links — we may earn a commission if you sign up, at no extra cost to you.
Cloud rates verified 2026-07 — estimates, and marketplace prices vary. Buying price is GPU MSRP only, not a full PC.
Other Devstral sizes on the NVIDIA H100 80GB (PCIe)
- Devstral Small 2505 24B — runs (16.6 GB)
- Devstral Small 2 24B — runs (16.6 GB)
Devstral-2 123B on nearby hardware
- Apple M3 Pro (36 GB) — does not fit
- NVIDIA RTX PRO 6000 Blackwell (96 GB) — runs
All Devstral sizes on the NVIDIA H100 80GB (PCIe) | Devstral-2 123B model page | NVIDIA H100 80GB (PCIe) GPU page | VRAM calculator | Check your hardware