Google95 GB VRAM819 GB/s$4.2/hr

Qwen 3 8B on TPU v5p (per-chip, 95GB): Does It Fit, How Fast, What It Costs

Qwen 3 8B at FP16 needs 16.0 GB for its weights and 25.7 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
25.7 GB needed / 95 GB available
Decode speed
33.27
tok/s per stream
Aggregate
399.3
tok/s @ 8 concurrent
Cost / 1M tokens
$2.92
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 GB25.7 GB33.27399.3631.2 ms$2.92✓
2× TPU190 GB25.7 GB56.561357333.2 ms$1.72✓
4× TPU380 GB25.7 GB113.15430181.6 ms$0.8594✓
8× TPU760 GB25.7 GB226.221720105.8 ms$0.4297✓
Concurrency vs context length
On 1× TPU v5p (per-chip, 95GB) at FP16. KV cache scales linearly with context, so this is the number of users your GPUs can actually serve.
ContextConcurrent usersKV per requestFits?
4K130618.5 MB✓ fits
8K651.21 GB✓ fits
33K164.83 GB✓ fits
66K89.66 GB✓ fits
131K419.33 GB✓ fits
Other GPUs that run Qwen 3 8B
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 5090 32GB1×873.6$0.1590
AMD MI300X 192GB1×2584$0.2688
V100 SXM2 32GB1×438.8$0.3166
A100 40GB SXM41×755.6$0.3309
AMD MI325X 288GB1×2925$0.3324
A100 80GB SXM41×975.0$0.3419
Frequently asked questions

Can Qwen 3 8B run on TPU v5p (per-chip, 95GB)?

Yes. Qwen 3 8B at FP16 needs 25.7 GB including the KV cache, and TPU v5p (per-chip, 95GB) has 95 GB — so it fits on a single GPU.

How fast is Qwen 3 8B on TPU v5p (per-chip, 95GB)?

33.27 tokens/sec per stream and 399.3 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 631.2 ms, and inter-token latency is 30.1 ms.

What does Qwen 3 8B on TPU v5p (per-chip, 95GB) cost per million tokens?

$2.92 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 Qwen 3 8B need on TPU v5p (per-chip, 95GB)?

16.0 GB for the weights at FP16, plus 9.7 GB of KV cache for 8 concurrent requests at 8K context — 25.7 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is TPU v5p (per-chip, 95GB) the right choice for Qwen 3 8B?

It works. Compare $2.92 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 4× RTX 3080 10GB at $0.0441, and the fastest is 1× Cerebras CS-3 (WSE-3) at 9750 tok/s.

Adjust the workload yourself

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

Open in calculator