H100 SXM: 80 GB, what it runs and what it costs
Specs checked · Rental prices from 50m ago · Aliteq Team · How we compute this
The short answer
The NVIDIA H100 SXM has 80 GB of VRAM and 3350 GB/s of memory bandwidth (2022). At Q4_K_M quality or better with an 8K context it runs 40 of the open models we track; the largest that fits with room to spare is Qwen3.5 122B-A10B (MoE). The cheapest rental right now is $0.667/hr (Vast.ai, interruptible).
- VRAM
- 80 GB
- Bandwidth
- 3350 GB/s
- Board power
- 700 W
- Launched
- 2022
Source: NVIDIA specification · bandwidth not yet confirmed by a second source, so we don't estimate speed from it
Buy it or rent it
| Provider | Type | From, per hour | Rent |
|---|---|---|---|
| Vast.ai | Spot (can be interrupted) | $0.667/hr | Vast.ai |
| Vast.ai | On demand | $1.336/hr | Vast.ai |
| Lium | Spot (can be interrupted) | $1.990/hr | Lium |
| Lium | On demand | $2.090/hr | Lium |
| Runpod | On demand | $3.990/hr | Runpod |
Lowest price per provider in our hourly snapshot (50m ago). Rent buttons are referral links. Every card side by side: cloud GPU prices.
Which AI models run on the H100 SXM
Computed by our VRAM engine for an 8K context: the best quality each model fits at in 80 GB. Speeds are upper-bound estimates from memory bandwidth, not benchmarks.
| Model | Best fit | VRAM used | ~tok/s |
|---|---|---|---|
| Qwen3.5 122B-A10B (MoE) | Q4_K_M | 72 GB89% | — |
| NVIDIA Nemotron 3 Super 120B-A12B (MoE) | Q4_K_M | 71 GB88% | — |
| gpt-oss 120B (MoE) | Q5_K_M | 79 GB98% · tight | — |
| Llama 4 Scout 17B-16E (MoE) | Q5_K_M | 73 GB91% · tight | — |
| GLM-4.5-Air (MoE) | Q5_K_M | 72 GB90% · tight | — |
| Qwen3-Next 80B-A3B (MoE) | Q6_K | 62 GB78% | — |
| Qwen3-Coder-Next 80B-A3B (MoE) | Q8_0 | 80 GB100% · tight | — |
| DeepSeek R1 Distill 70B | Q8_0 | 73 GB91% · tight | — |
| Mixtral 8x7B (MoE) | Q8_0 | 48 GB60% | — |
| Qwen3.6 35B-A3B (MoE) | F16 | 68 GB85% | — |
| Ornith-1.5 35B-A3B (MoE) | F16 | 68 GB85% | — |
| Qwen3.5 35B-A3B (MoE) | F16 | 68 GB85% | — |
Fits only below Q4 quality: Qwen3.8-Flash-Next (MoE) (Q2_K).
Too big for 80 GB (rent a bigger card instead): what it costs to run MiniMax-M2.5 (MoE), what it costs to run Qwen3 235B-A22B (MoE), what it costs to run MiniMax-M3 (MoE), what it costs to run GLM-5.3 (MoE).
Check a model and context of your own in the VRAM calculator.
AI video on the H100 SXM
Alibaba's Wan 2.2 is an open video model you can run yourself in ComfyUI. With 80 GB the H100 SXM fits:
- Wan 2.2 14B at 720p without offloading (80 GB and up, per Runpod's Wan 2.2 guide)
- Wan 2.2 14B in fp16, or in fp8 at 720p (48 GB and up, per Runpod's Wan 2.2 guide)
- Wan 2.2 14B at 640x640 with fp8 weights (24 GB and up, per Runpod's Wan 2.2 guide)
- Wan 2.2 5B, with ComfyUI's native offloading (8 GB and up, per ComfyUI's Wan 2.2 docs)
What each setup needs, the cheapest card to rent and what one clip costs: cloud GPUs for ComfyUI with Flux and Wan 2.2. Against an app or an API: AI video cost per clip, app vs API vs rented GPU. Rather not run a model yourself? Hosted video models such as Seedance, Veo and Hailuo are compared in Kling AI alternatives, priced per clip.
Our guides on the H100 SXM
- CompareIs a B200 worth its rental premium over an H100? Only if it runs your job 2× faster
- CompareWhich GPU cloud is cheapest for AI: Runpod, Vast.ai or Lambda?
- the cheapest way to run Llama 4 in the cloud: Scout on one H100, Maverick only if you must
- Where can you rent an H100 cheapest? From $1.79 an hour, and the catch in each price
- Which big GPU cloud is cheapest for H100 and B200? CoreWeave vs Nebius vs Lambda vs Crusoe
- How much does it cost to rent an H200? From $3.76 an hour, and when 141GB is worth it
- How much does Runpod Serverless cost compared with a pod? The per-hour gap and the break-even
Cards to compare it with
- A100 SXM · 80 GB
- H100 PCIe · 80 GB
- A100 PCIe · 80 GB
- RTX PRO 6000 Blackwell Server Edition · 96 GB
- L40S · 48 GB
- A40 · 48 GB
Every card by VRAM tier: all GPUs for gaming and local AI.