NVIDIA80 GB VRAM3,350 GB/s$2.5/hr

GLM-5.3 (MoE) on H100 SXM5 80GB: Does It Fit, How Fast, What It Costs

GLM-5.3 (MoE) at AWQ 4-bit needs 414.3 GB for its weights and 420.2 GB in total at 8K context with 8 concurrent users. That means it needs 8× H100 SXM5 80GB. Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 8×
420.2 GB needed / 640 GB available
Decode speed
1239
tok/s per stream
Aggregate
118970
tok/s @ 8 concurrent
Cost / 1M tokens
$0.0467
8× H100 SXM5 80GB
What changes as you add H100 SXM5 80GBs
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× H10080 GB420.2 GB0.0000.000——✗
2× H100160 GB420.2 GB0.0000.000——✗
4× H100320 GB420.2 GB0.0000.000——✗
8× H100640 GB420.2 GB1239118970109.9 ms$0.0467✓
Concurrency vs context length
On 8× H100 SXM5 80GB 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?
4K61315.70 GB✓ fits
8K30631.41 GB✓ fits
33K76125.63 GB✓ fits
66K38251.26 GB✓ fits
131K19502.51 GB✓ fits
Other GPUs that run GLM-5.3 (MoE)
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
A100 80GB SXM48×71027$0.0375
B200 SXM 192GB4×71027$0.0939
H200 SXM5 141GB4×42616$0.1043
Frequently asked questions

Can GLM-5.3 (MoE) run on H100 SXM5 80GB?

Yes, but it needs 8× H100 SXM5 80GB. At AWQ 4-bit the total footprint is 420.2 GB (414.3 GB of weights plus 5.9 GB of KV cache at 8K × 8 concurrent users), against 640 GB of available VRAM across 8 GPUs.

How fast is GLM-5.3 (MoE) on H100 SXM5 80GB?

1239 tokens/sec per stream and 118970 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 109.9 ms, and inter-token latency is 806.9 µs.

What does GLM-5.3 (MoE) on H100 SXM5 80GB cost per million tokens?

$0.0467 per 1M output tokens at 8× H100 SXM5 80GB ($2.50/hr). That is the marginal hardware cost — add idle capacity, ops, and egress for the real figure.

How much VRAM does GLM-5.3 (MoE) need on H100 SXM5 80GB?

414.3 GB for the weights at AWQ 4-bit, plus 5.9 GB of KV cache for 8 concurrent requests at 8K context — 420.2 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is H100 SXM5 80GB the right choice for GLM-5.3 (MoE)?

It works. Compare $0.0467 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 8× A100 80GB SXM4 at $0.0375, and the fastest is 8× H100 SXM5 80GB at 118970 tok/s.

Adjust the workload yourself

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

Open in calculator