Google95 GB VRAM819 GB/s$4.2/hr

GPT-OSS 120B (MoE) on TPU v5p (per-chip, 95GB): 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 fits on a single TPU v5p (per-chip, 95GB). Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 1 GPU
66.7 GB needed / 95 GB available
Decode speed
167.0
tok/s per stream
Aggregate
2004
tok/s @ 8 concurrent
Cost / 1M tokens
$0.5821
1× TPU v5p (per-chip, 95GB)
What changes as you add TPU v5p (per-chip, 95GB)s
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× TPU95 GB66.7 GB167.02004400.1 ms$0.5821✓
2× TPU190 GB66.7 GB283.96814215.6 ms$0.3424✓
4× TPU380 GB66.7 GB567.827256122.8 ms$0.1712✓
8× TPU760 GB66.7 GB113610902576.4 ms$0.0856✓
Concurrency vs context length
On 1× TPU v5p (per-chip, 95GB) 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?
4K197309.2 MB✓ fits
8K100618.5 MB✓ fits
33K252.42 GB✓ fits
66K124.83 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
L40S 48GB2×7188$0.0541
AMD MI325X 288GB1×14682$0.0662
A100 80GB SXM41×4894$0.0681
Intel Gaudi 3 128GB1×8075$0.0688
Frequently asked questions

Can GPT-OSS 120B (MoE) run on TPU v5p (per-chip, 95GB)?

Yes. GPT-OSS 120B (MoE) at AWQ 4-bit needs 66.7 GB including the KV cache, and TPU v5p (per-chip, 95GB) has 95 GB — so it fits on a single GPU.

How fast is GPT-OSS 120B (MoE) on TPU v5p (per-chip, 95GB)?

167.0 tokens/sec per stream and 2004 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 400.1 ms, and inter-token latency is 6.0 ms.

What does GPT-OSS 120B (MoE) on TPU v5p (per-chip, 95GB) cost per million tokens?

$0.5821 per 1M output tokens at 1× TPU v5p (per-chip, 95GB) ($4.20/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 TPU v5p (per-chip, 95GB)?

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 TPU v5p (per-chip, 95GB) the right choice for GPT-OSS 120B (MoE)?

It works. Compare $0.5821 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