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

GPT-NeoX 20B on H100 SXM5 80GB: Does It Fit, How Fast, What It Costs

GPT-NeoX 20B at FP8 (E4M3) needs 20.0 GB for its weights and 90.9 GB in total at 8K context with 8 concurrent users. That means it needs 2× H100 SXM5 80GB. Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 2×
90.9 GB needed / 160 GB available
Decode speed
277.6
tok/s per stream
Aggregate
6663
tok/s @ 8 concurrent
Cost / 1M tokens
$0.2084
2× 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 GB90.9 GB0.0000.000——✗
2× H100160 GB90.9 GB277.66663364.6 ms$0.2084✓
4× H100320 GB90.9 GB555.326653197.3 ms$0.1042✓
8× H100640 GB90.9 GB1111106610113.6 ms$0.0521✓
Concurrency vs context length
On 2× H100 SXM5 80GB at FP8 (E4M3). KV cache scales linearly with context, so this is the number of users your GPUs can actually serve.
ContextConcurrent usersKV per requestFits?
4K314.43 GB✓ fits
8K158.86 GB✓ fits
33K335.43 GB✓ fits
66K170.87 GB✓ fits
131K0141.73 GB✗ does not fit
Other GPUs that run GPT-NeoX 20B
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 5090 32GB4×14257$0.0390
A100 80GB SXM42×3978$0.1676
AMD MI300X 192GB1×3101$0.2240
L40S 48GB2×1718$0.2263
AMD MI325X 288GB1×3510$0.2770
Intel Gaudi 3 128GB1×1931$0.2878
Frequently asked questions

Can GPT-NeoX 20B run on H100 SXM5 80GB?

Yes, but it needs 2× H100 SXM5 80GB. At FP8 (E4M3) the total footprint is 90.9 GB (20.0 GB of weights plus 70.9 GB of KV cache at 8K × 8 concurrent users), against 160 GB of available VRAM across 2 GPUs.

How fast is GPT-NeoX 20B on H100 SXM5 80GB?

277.6 tokens/sec per stream and 6663 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 364.6 ms, and inter-token latency is 3.6 ms.

What does GPT-NeoX 20B on H100 SXM5 80GB cost per million tokens?

$0.2084 per 1M output tokens at 2× 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 GPT-NeoX 20B need on H100 SXM5 80GB?

20.0 GB for the weights at FP8 (E4M3), plus 70.9 GB of KV cache for 8 concurrent requests at 8K context — 90.9 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is H100 SXM5 80GB the right choice for GPT-NeoX 20B?

It works. Compare $0.2084 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 8× RTX 4080 16GB at $0.0292, and the fastest is 4× Cerebras CS-3 (WSE-3) at 159120 tok/s.

Adjust the workload yourself

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

Open in calculator