Google95 GB VRAM819 GB/s$4.2/hr

Qwen2 7B on TPU v5p (per-chip, 95GB): Does It Fit, How Fast, What It Costs

Qwen2 7B at FP16 needs 15.2 GB for its weights and 19.0 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
19.0 GB needed / 95 GB available
Decode speed
34.93
tok/s per stream
Aggregate
419.2
tok/s @ 8 concurrent
Cost / 1M tokens
$2.78
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 GB19.0 GB34.93419.2602.6 ms$2.78✓
2× TPU190 GB19.0 GB59.381425318.8 ms$1.64✓
4× TPU380 GB19.0 GB118.85701174.4 ms$0.8186✓
8× TPU760 GB19.0 GB237.522803102.2 ms$0.4093✓
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?
4K339240.5 MB✓ fits
8K169481.0 MB✓ fits
33K421.88 GB✓ fits
66K213.76 GB✓ fits
131K107.52 GB✓ fits
Other GPUs that run Qwen2 7B
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 3090 24GB1×479.1$0.1450
RTX 5090 32GB1×917.2$0.1514
RTX 4090 24GB1×515.9$0.2154
AMD MI300X 192GB1×2713$0.2560
V100 SXM2 32GB1×460.6$0.3015
A100 40GB SXM41×793.3$0.3151
Frequently asked questions

Can Qwen2 7B run on TPU v5p (per-chip, 95GB)?

Yes. Qwen2 7B at FP16 needs 19.0 GB including the KV cache, and TPU v5p (per-chip, 95GB) has 95 GB — so it fits on a single GPU.

How fast is Qwen2 7B on TPU v5p (per-chip, 95GB)?

34.93 tokens/sec per stream and 419.2 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 602.6 ms, and inter-token latency is 28.6 ms.

What does Qwen2 7B on TPU v5p (per-chip, 95GB) cost per million tokens?

$2.78 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 Qwen2 7B need on TPU v5p (per-chip, 95GB)?

15.2 GB for the weights at FP16, plus 3.8 GB of KV cache for 8 concurrent requests at 8K context — 19.0 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is TPU v5p (per-chip, 95GB) the right choice for Qwen2 7B?

It works. Compare $2.78 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 2× RTX 3080 10GB at $0.0840, and the fastest is 1× Cerebras CS-3 (WSE-3) at 10236 tok/s.

Adjust the workload yourself

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

Open in calculator