B200: 192 GB, what it runs and what it costs
Specs checked · Rental prices from 18m ago · Aliteq Team · How we compute this
The short answer
The NVIDIA B200 has 192 GB of VRAM and 8000 GB/s of memory bandwidth (2025). At Q4_K_M quality or better with an 8K context it runs 43 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 $3.125/hr (Vast.ai, interruptible).
- VRAM
- 192 GB
- Bandwidth
- 8000 GB/s
- Board power
- 1000 W
- Launched
- 2025
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) | $3.125/hr | Vast.ai |
| Lium | On demand | $3.500/hr | Lium |
| Lium | Spot (can be interrupted) | $5.000/hr | Lium |
| Vast.ai | On demand | $5.875/hr | Vast.ai |
| Runpod | On demand | $7.990/hr | Runpod |
Lowest price per provider in our hourly snapshot (18m ago). Rent buttons are referral links. Every card side by side: cloud GPU prices.
Which AI models run on the B200
Computed by our VRAM engine for an 8K context: the best quality each model fits at in 192 GB. Speeds are upper-bound estimates from memory bandwidth, not benchmarks.
| Model | Best fit | VRAM used | ~tok/s |
|---|---|---|---|
| Qwen3 235B-A22B (MoE) | Q6_K | 182 GB95% · tight | — |
| MiniMax-M2.5 (MoE) | Q6_K | 177 GB92% · tight | — |
| Qwen3.8-Flash-Next (MoE) | Q8_0 | 179 GB93% · tight | — |
| Qwen3.5 122B-A10B (MoE) | Q8_0 | 125 GB65% | — |
| NVIDIA Nemotron 3 Super 120B-A12B (MoE) | Q8_0 | 123 GB64% | — |
| gpt-oss 120B (MoE) | Q8_0 | 117 GB61% | — |
| Llama 4 Scout 17B-16E (MoE) | Q8_0 | 109 GB57% | — |
| GLM-4.5-Air (MoE) | Q8_0 | 107 GB56% | — |
| Qwen3-Next 80B-A3B (MoE) | F16 | 150 GB78% | — |
| Qwen3-Coder-Next 80B-A3B (MoE) | F16 | 149 GB78% | — |
| DeepSeek R1 Distill 70B | F16 | 135 GB70% | — |
| Mixtral 8x7B (MoE) | F16 | 89 GB46% | — |
Fits only below Q4 quality: MiniMax-M3 (MoE) (Q2_K).
Too big for 192 GB (rent a bigger card instead): what it costs to run GLM-5.3 (MoE), what it costs to run DeepSeek V4.1 Flash (MoE), what it costs to run DeepSeek V4 Pro (MoE), what it costs to run Kimi K3 (MoE).
Check a model and context of your own in the VRAM calculator.
AI video on the B200
Alibaba's Wan 2.2 is an open video model you can run yourself in ComfyUI. With 192 GB the B200 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 B200
Cards to compare it with
- AMD Instinct MI300X · 192 GB
- H200 SXM · 141 GB
- RTX PRO 6000 Blackwell Server Edition · 96 GB
- H100 SXM · 80 GB
- A100 SXM · 80 GB
- H100 PCIe · 80 GB
Every card by VRAM tier: all GPUs for gaming and local AI.