NVIDIA40 GB VRAM1,550 GB/s$0.9/hr

Llama 3 8B on A100 40GB SXM4: Does It Fit, How Fast, What It Costs

Llama 3 8B at FP16 needs 16.1 GB for its weights and 24.6 GB in total at 8K context with 8 concurrent users. That means it fits on a single A100 40GB SXM4. Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 1 GPU
24.6 GB needed / 40 GB available
Decode speed
62.73
tok/s per stream
Aggregate
752.8
tok/s @ 8 concurrent
Cost / 1M tokens
$0.3321
1× A100 40GB SXM4
What changes as you add A100 40GB SXM4s
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× A10040 GB24.6 GB62.73752.8889.3 ms$0.3321✓
2× A10080 GB24.6 GB106.62560461.1 ms$0.1953✓
4× A100160 GB24.6 GB213.310238245.5 ms$0.0977✓
8× A100320 GB24.6 GB426.640952137.8 ms$0.0488✓
Concurrency vs context length
On 1× A100 40GB SXM4 at FP16. KV cache scales linearly with context, so this is the number of users your GPUs can actually serve.
ContextConcurrent usersKV per requestFits?
4K44549.8 MB✓ fits
8K221.07 GB✓ fits
33K54.29 GB✓ fits
66K28.59 GB✓ fits
131K117.18 GB✓ fits
Other GPUs that run Llama 3 8B
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 5090 32GB1×870.3$0.1596
AMD MI300X 192GB1×2574$0.2698
V100 SXM2 32GB1×437.1$0.3177
AMD MI325X 288GB1×2914$0.3336
A100 80GB SXM41×971.4$0.3432
Intel Gaudi 3 128GB1×1603$0.3466
Frequently asked questions

Can Llama 3 8B run on A100 40GB SXM4?

Yes. Llama 3 8B at FP16 needs 24.6 GB including the KV cache, and A100 40GB SXM4 has 40 GB — so it fits on a single GPU.

How fast is Llama 3 8B on A100 40GB SXM4?

62.73 tokens/sec per stream and 752.8 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 889.3 ms, and inter-token latency is 15.9 ms.

What does Llama 3 8B on A100 40GB SXM4 cost per million tokens?

$0.3321 per 1M output tokens at 1× A100 40GB SXM4 ($0.9000/hr). That is the marginal hardware cost — add idle capacity, ops, and egress for the real figure.

How much VRAM does Llama 3 8B need on A100 40GB SXM4?

16.1 GB for the weights at FP16, plus 8.6 GB of KV cache for 8 concurrent requests at 8K context — 24.6 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is A100 40GB SXM4 the right choice for Llama 3 8B?

It works. Compare $0.3321 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 4× RTX 3080 10GB at $0.0443, and the fastest is 1× Cerebras CS-3 (WSE-3) at 9714 tok/s.

Adjust the workload yourself

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

Open in calculator