NVIDIA141 GB VRAM4,800 GB/s$4/hr

Llama 4 Maverick 17B-128E (MoE, VLM) on H200 SXM5 141GB: Does It Fit, How Fast, What It Costs

Llama 4 Maverick 17B-128E (MoE, VLM) at AWQ 4-bit needs 220.0 GB for its weights and 232.9 GB in total at 8K context with 8 concurrent users. That means it needs 2× H200 SXM5 141GB. Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 2×
232.9 GB needed / 282 GB available
Decode speed
499.2
tok/s per stream
Aggregate
11981
tok/s @ 8 concurrent
Cost / 1M tokens
$0.1855
2× H200 SXM5 141GB
What changes as you add H200 SXM5 141GBs
Tensor-parallel degree vs VRAM, throughput, and cost. Cost scales linearly with GPU count; decode throughput scales near-linearly too, so cost per token stays roughly flat.
GPUsTotal VRAMVRAM neededDecode tok/sAggregate tok/sTTFTCost / 1M tokFits?
1× H200141 GB232.9 GB0.0000.000——✗
2× H200282 GB232.9 GB499.211981313.3 ms$0.1855✓
4× H200564 GB232.9 GB998.447923171.7 ms$0.0927✓
8× H2001128 GB232.9 GB1997191693100.8 ms$0.0464✓
Concurrency vs context length
On 2× H200 SXM5 141GB at AWQ 4-bit. KV cache scales linearly with context, so this is the number of users your GPUs can actually serve.
ContextConcurrent usersKV per requestFits?
4K76824.6 MB✓ fits
8K381.61 GB✓ fits
33K96.44 GB✓ fits
66K412.88 GB✓ fits
131K225.77 GB✓ fits
Other GPUs that run Llama 4 Maverick 17B-128E (MoE, VLM)
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 5090 32GB8×71565$0.0155
L40S 48GB8×34505$0.0451
A100 80GB SXM44×19968$0.0668
H100 SXM5 80GB4×33446$0.0831
B200 SXM 192GB2×19968$0.1669
AMD MI325X 288GB1×4405$0.2207
Frequently asked questions

Can Llama 4 Maverick 17B-128E (MoE, VLM) run on H200 SXM5 141GB?

Yes, but it needs 2× H200 SXM5 141GB. At AWQ 4-bit the total footprint is 232.9 GB (220.0 GB of weights plus 12.9 GB of KV cache at 8K × 8 concurrent users), against 282 GB of available VRAM across 2 GPUs.

How fast is Llama 4 Maverick 17B-128E (MoE, VLM) on H200 SXM5 141GB?

499.2 tokens/sec per stream and 11981 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 313.3 ms, and inter-token latency is 2.0 ms.

What does Llama 4 Maverick 17B-128E (MoE, VLM) on H200 SXM5 141GB cost per million tokens?

$0.1855 per 1M output tokens at 2× H200 SXM5 141GB ($4.00/hr). That is the marginal hardware cost — add idle capacity, ops, and egress for the real figure.

How much VRAM does Llama 4 Maverick 17B-128E (MoE, VLM) need on H200 SXM5 141GB?

220.0 GB for the weights at AWQ 4-bit, plus 12.9 GB of KV cache for 8 concurrent requests at 8K context — 232.9 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is H200 SXM5 141GB the right choice for Llama 4 Maverick 17B-128E (MoE, VLM)?

It works. Compare $0.1855 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 8× RTX 5090 32GB at $0.0155, and the fastest is 8× Cerebras CS-3 (WSE-3) at 798720 tok/s.

Adjust the workload yourself

Same formulas, your context length, concurrency, and quantisation.

Open in calculator