LLM Roofline
NVIDIA Cloud · SXM · Backend: vllm

Cloud NVIDIA H200 SXM (141 GB) — LLM Sizing & Speed Matrix

4.8 TB/s HBM3e single-GPU cloud node; effortlessly serves 70B FP16 or 235B Q4_K_M.

Reference Pricing
~$3.49/hr Cloud
Usable Memory
140.4 GiB
Memory Bandwidth
4800 GB/s (84% eff.)
FP16 Compute
989 TFLOPS
Roofline Ridge
176.6 FLOP/B
Power (Load / Idle)
700W / 0W

Buy or Rent Cloud NVIDIA H200 SXM (141 GB)

Vendor & Cloud Links
Deploy on DigitalOcean GPU Droplets
DigitalOcean · On-Demand / Spot GPU
↗
Launch Pod on RunPod
RunPod · On-Demand / Spot GPU
↗
Deploy on Vast.ai Spot GPU
Vast.ai · On-Demand / Spot GPU
↗
Run Serverless / GPU on Novita AI
Novita AI · On-Demand / Spot GPU
↗

All 32 Open-Weight Models on Cloud NVIDIA H200 SXM (141 GB) (Q4_K_M @ 32k)

Model Status Weights Total @ 32k Decode TTFT (4k) Max Safe Ctx
Mixtral 8x22B Instruct COMFORTABLE 81.2 GiB 89.8 GiB 112.2 tok/s 0.50s 64k
Command A 111B COMFORTABLE 63.9 GiB 69.1 GiB 56.3 tok/s 1.35s 256k
Llama 4 Scout 109B-A17B COMFORTABLE 62.8 GiB 70.2 GiB 209.3 tok/s 0.24s 406k
Qwen2.5 72B Instruct COMFORTABLE 41.9 GiB 53.7 GiB 72.4 tok/s 0.90s 32k
DeepSeek-R1 Distill Llama 70B COMFORTABLE 40.7 GiB 52.5 GiB 74.1 tok/s 0.88s 128k
Llama 3.3 70B Instruct COMFORTABLE 40.7 GiB 52.5 GiB 74.1 tok/s 0.88s 128k
Llama 3.1 70B Instruct COMFORTABLE 40.7 GiB 52.5 GiB 74.1 tok/s 0.88s 128k
Llama-3.3 Nemotron Super 49B COMFORTABLE 28.7 GiB 36.7 GiB 107.7 tok/s 0.63s 128k
Mixtral 8x7B Instruct COMFORTABLE 26.9 GiB 32.2 GiB 289.1 tok/s 0.19s 32k
Qwen3 32B COMFORTABLE 18.9 GiB 28.3 GiB 139.7 tok/s 0.44s 40k
Qwen2.5 Coder 32B COMFORTABLE 18.9 GiB 28.3 GiB 139.7 tok/s 0.43s 32k
DeepSeek-R1 Distill Qwen 32B COMFORTABLE 18.9 GiB 28.3 GiB 139.7 tok/s 0.43s 128k
QwQ 32B Reasoning COMFORTABLE 18.9 GiB 28.3 GiB 139.7 tok/s 0.43s 40k
GLM-4 32B 0414 COMFORTABLE 18.6 GiB 22.1 GiB 183.1 tok/s 0.42s 32k
Qwen3 30B-A3B COMFORTABLE 17.6 GiB 21.6 GiB 674.3 tok/s 0.08s 40k
Gemma 3 27B IT COMFORTABLE 15.8 GiB 20.2 GiB 201.0 tok/s 0.36s 128k
Mistral Small 3.1 24B COMFORTABLE 13.8 GiB 20.3 GiB 199.5 tok/s 0.32s 128k
Codestral 22B COMFORTABLE 12.8 GiB 21.4 GiB 189.8 tok/s 0.31s 32k
Qwen3 14B COMFORTABLE 8.5 GiB 15.0 GiB 277.7 tok/s 0.21s 40k
Qwen2.5 14B Instruct COMFORTABLE 8.5 GiB 16.0 GiB 258.6 tok/s 0.22s 32k
DeepSeek-R1 Distill Qwen 14B COMFORTABLE 8.5 GiB 16.0 GiB 258.6 tok/s 0.22s 128k
Phi-4 14B COMFORTABLE 8.5 GiB 13.0 GiB 324.0 tok/s 0.21s 16k
Gemma 3 12B IT COMFORTABLE 7.0 GiB 10.6 GiB 402.1 tok/s 0.18s 128k
Qwen3 8B COMFORTABLE 4.7 GiB 10.5 GiB 407.2 tok/s 0.14s 40k
Llama 3.1 8B Instruct COMFORTABLE 4.6 GiB 9.9 GiB 435.4 tok/s 0.13s 128k
Qwen2.5 7B Instruct COMFORTABLE 4.4 GiB 7.4 GiB 612.9 tok/s 0.13s 32k
Gemma 3 4B IT COMFORTABLE 2.5 GiB 4.4 GiB 1125.7 tok/s 0.09s 128k
DeepSeek-R1 671B OOM 386.4 GiB 390.3 GiB OOM — OOM
DeepSeek-V3 671B OOM 386.4 GiB 390.3 GiB OOM — OOM
Llama 3.1 405B Instruct OOM 233.8 GiB 252.5 GiB OOM — OOM
Llama 4 Maverick 400B-A17B OOM 231.5 GiB 239.0 GiB OOM — OOM
Qwen3 235B-A22B OOM 135.4 GiB 142.6 GiB OOM — 20k

Interactive Cloud NVIDIA H200 SXM (141 GB) Simulator

141B (39B active) · GQA
4800 GB/s · 989 TFLOPS
3. Weight Quantization 4.85 bpw
4. KV Cache
32,768 tok
Compute Roofline
COMFORTABLE FIT Memory-Bandwidth Bound
Max Safe Ctx: 64k
Memory Allocation Footprint 89.80 / 140.40 GiB usable
Weights: 81.20 GiB KV Cache: 7.00 GiB Overhead: 1.59 GiB Headroom: 50.60 GiB
Decode Speed
112.2
tok/s (batch=1)
Prefill Speed
8811
prompt tok/s
TTFT Latency
0.50s
for 4,096 tok
Active Params
39.0B
of 141B total
vLLM Production Serve Command 
vllm serve mistralai/Mixtral-8x22B-Instruct-v0.1 --max-model-len 32768 --quantization fp8
Context KV Cache Total VRAM Decode Status
4k 0.88 GiB 83.25 GiB 141.6 tok/s COMFORTABLE
8k 1.75 GiB 84.55 GiB 136.5 tok/s COMFORTABLE
16k 3.50 GiB 86.30 GiB 127.3 tok/s COMFORTABLE
32k 7.00 GiB 89.80 GiB 112.2 tok/s COMFORTABLE
64k 14.00 GiB 96.80 GiB 90.6 tok/s COMFORTABLE

Buy-vs-Cloud-vs-API Monthly TCO Break-Even

RENT CLOUD GPU
Own Hardware/mo
$0.00
38.7h active + power
Cloud GPU/mo
$155.41
@$3.49/hr spot/pod
Serverless API/mo
$190.00
API cheaper

Recommended Hardware & Cloud Deployment

Contextual to fit status

Hardware FAQ — Cloud NVIDIA H200 SXM (141 GB)

How much usable VRAM / GPU memory does Cloud NVIDIA H200 SXM (141 GB) actually provide for LLMs?

Cloud NVIDIA H200 SXM (141 GB) provides 140.4 GiB usable VRAM after subtracting 0.6 GiB driver/display overhead.

What is the largest Q4_K_M model Cloud NVIDIA H200 SXM (141 GB) can run at 32k context?

With Q4_K_M weights and FP16 KV cache at 32k context, the largest fitting model in our verified catalog on Cloud NVIDIA H200 SXM (141 GB) is Mixtral 8x22B Instruct (141B total).