NVIDIA48 GB VRAM864 GB/s$0.7/hr

Llama 3.1 405B on L40S 48GB: Does It Fit, How Fast, What It Costs

Llama 3.1 405B at AWQ 4-bit needs 222.8 GB for its weights and 256.6 GB in total at 8K context with 8 concurrent users. That means it needs 8× L40S 48GB. Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 8×
256.6 GB needed / 384 GB available
Decode speed
15.09
tok/s per stream
Aggregate
1448
tok/s @ 8 concurrent
Cost / 1M tokens
$1.07
8× L40S 48GB

L40S 48GB has no NVLink, so 8-way tensor parallelism runs over PCIe. tokcalc applies a flat 0.85× multi-GPU efficiency factor for this, which is realistic for NVLink-equipped datacenter cards but optimistic for PCIe scale-out — treat the throughput and cost figures for this configuration as an upper bound, and measure before committing.

What changes as you add L40S 48GBs
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× L40S48 GB256.6 GB0.0000.000——✗
2× L40S96 GB256.6 GB0.0000.000——✗
4× L40S192 GB256.6 GB0.0000.000——✗
8× L40S384 GB256.6 GB15.09144818.21 s$1.07✓
Concurrency vs context length
On 8× L40S 48GB 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?
4K762.11 GB✓ fits
8K384.23 GB✓ fits
33K916.91 GB✓ fits
66K433.82 GB✓ fits
131K267.65 GB✓ fits
Other GPUs that run Llama 3.1 405B
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
A100 80GB SXM44×838.2$1.59
H100 SXM5 80GB4×1404$1.98
B200 SXM 192GB2×838.2$3.98
H200 SXM5 141GB2×502.9$4.42
AMD MI325X 288GB1×184.9$5.26
B300 SXM 288GB1×246.5$9.01
Frequently asked questions

Can Llama 3.1 405B run on L40S 48GB?

Yes, but it needs 8× L40S 48GB. At AWQ 4-bit the total footprint is 256.6 GB (222.8 GB of weights plus 33.8 GB of KV cache at 8K × 8 concurrent users), against 384 GB of available VRAM across 8 GPUs.

How fast is Llama 3.1 405B on L40S 48GB?

15.09 tokens/sec per stream and 1448 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 18.21 s, and inter-token latency is 66.3 ms.

What does Llama 3.1 405B on L40S 48GB cost per million tokens?

$1.07 per 1M output tokens at 8× L40S 48GB ($0.7000/hr). That is the marginal hardware cost — add idle capacity, ops, and egress for the real figure.

How much VRAM does Llama 3.1 405B need on L40S 48GB?

222.8 GB for the weights at AWQ 4-bit, plus 33.8 GB of KV cache for 8 concurrent requests at 8K context — 256.6 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is L40S 48GB the right choice for Llama 3.1 405B?

It works. Compare $1.07 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 8× A100 40GB SXM4 at $0.7697, and the fastest is 8× Cerebras CS-3 (WSE-3) at 33527 tok/s.

Adjust the workload yourself

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

Open in calculator