Can LFM2.5 8B A1B run on the H100 SXM?
Yes. At 8K context the best fit is F16, using about 32.3 GB of the 76.0 GB usable on the H100 SXM. Estimated speed ceiling: ~837 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|
| 2-bit (Q2_K) | 3.6 GB Fits | 3.7 GB Fits | 4.0 GB Fits | 8933 |
| 3-bit (Q3_K_M) | 4.5 GB Fits | 4.6 GB Fits | 4.9 GB Fits | 6854 |
| 4-bit (Q4_K_M) • | 5.5 GB Fits | 5.5 GB Fits | 5.8 GB Fits | 5502 |
| 5-bit (Q5_K_M) • | 6.3 GB Fits | 6.4 GB Fits | 6.6 GB Fits | 4704 |
| 6-bit (Q6_K) • | 7.2 GB Fits | 7.2 GB Fits | 7.5 GB Fits | 4076 |
| 8-bit (Q8_0) • | 9.1 GB Fits | 9.1 GB Fits | 9.4 GB Fits | 3148 |
| 16-bit (F16/BF16) • | 32.3 GB Fits | 32.3 GB Fits | 32.6 GB Fits | 837 |
• real GGUF file size from LiquidAI/LFM2.5-8B-A1B-GGUF. Speed is a bandwidth ceiling (3350 GB/s), not a benchmark. Methodology.
What fits instead
- Cheapest listed GPU that runs LFM2.5 8B A1B: Arc B580 (12 GB, launch MSRP $249) at Q8_0, fully in VRAM at 8K context.
- Largest popular model that fits the H100 SXM: gpt-oss 120B (117B, MoE) at F16, about 61.9 GB at 8K context.
"Runs" and "fits" mean the whole model in GPU memory at 8K context (4-bit or better for the model pick). Prices are US launch MSRPs, not today's street prices. "Popular" means the 60 most-downloaded models we track.
More
LFM2.5 8B A1B on every GPU and Mac · Every model that runs on the H100 SXM · Other models that fit 64 GB VRAM · Try other settings in the calculator
Guides: Best GPU for local LLMs by budget · Can I run gpt-oss, Llama 70B or Qwen3? · Mac vs PC for local LLMs
Data: Hugging Face Hub (LiquidAI/LFM2.5-8B-A1B), NVIDIA specs. Model data last checked 2026-11-17; refreshed hourly.