NVIDIA48 GB VRAM864 GB/s$0.7/hr

GPT-OSS 120B (MoE) on L40S 48GB: Does It Fit, How Fast, What It Costs

GPT-OSS 120B (MoE) at AWQ 4-bit needs 64.3 GB for its weights and 66.7 GB in total at 8K context with 8 concurrent users. That means it needs 2× L40S 48GB. Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 2×
66.7 GB needed / 96 GB available
Decode speed
299.5
tok/s per stream
Aggregate
7188
tok/s @ 8 concurrent
Cost / 1M tokens
$0.0541
2× L40S 48GB

L40S 48GB has no NVLink, so 2-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 GB66.7 GB0.0000.000——✗
2× L40S96 GB66.7 GB299.57188945.5 ms$0.0541✓
4× L40S192 GB66.7 GB599.028754487.8 ms$0.0270✓
8× L40S384 GB66.7 GB1198115016258.9 ms$0.0135✓
Concurrency vs context length
On 2× 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?
4K203309.2 MB✓ fits
8K103618.5 MB✓ fits
33K262.42 GB✓ fits
66K134.83 GB✓ fits
131K69.66 GB✓ fits
Other GPUs that run GPT-OSS 120B (MoE)
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 5090 32GB4×59638$0.009316
AMD MI300X 192GB1×12969$0.0535
AMD MI325X 288GB1×14682$0.0662
A100 80GB SXM41×4894$0.0681
Intel Gaudi 3 128GB1×8075$0.0688
H100 SXM5 80GB1×8198$0.0847
Frequently asked questions

Can GPT-OSS 120B (MoE) run on L40S 48GB?

Yes, but it needs 2× L40S 48GB. At AWQ 4-bit the total footprint is 66.7 GB (64.3 GB of weights plus 2.5 GB of KV cache at 8K × 8 concurrent users), against 96 GB of available VRAM across 2 GPUs.

How fast is GPT-OSS 120B (MoE) on L40S 48GB?

299.5 tokens/sec per stream and 7188 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 945.5 ms, and inter-token latency is 3.3 ms.

What does GPT-OSS 120B (MoE) on L40S 48GB cost per million tokens?

$0.0541 per 1M output tokens at 2× 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 GPT-OSS 120B (MoE) need on L40S 48GB?

64.3 GB for the weights at AWQ 4-bit, plus 2.5 GB of KV cache for 8 concurrent requests at 8K context — 66.7 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is L40S 48GB the right choice for GPT-OSS 120B (MoE)?

It works. Compare $0.0541 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 8× RTX 3080 10GB at $0.004393, and the fastest is 2× Cerebras CS-3 (WSE-3) at 166400 tok/s.

Adjust the workload yourself

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

Open in calculator