Google95 GB VRAM819 GB/s$4.2/hr

Llama 2 7B on TPU v5p (per-chip, 95GB): Does It Fit, How Fast, What It Costs

Llama 2 7B at FP16 needs 13.5 GB for its weights and 47.8 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
47.8 GB needed / 95 GB available
Decode speed
39.49
tok/s per stream
Aggregate
473.9
tok/s @ 8 concurrent
Cost / 1M tokens
$2.46
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 GB47.8 GB39.49473.9536.5 ms$2.46✓
2× TPU190 GB47.8 GB67.141611285.5 ms$1.45✓
4× TPU380 GB47.8 GB134.36445157.7 ms$0.7241✓
8× TPU760 GB47.8 GB268.52578093.9 ms$0.3620✓
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?
4K372.15 GB✓ fits
8K184.29 GB✓ fits
33K417.18 GB✓ fits
66K234.36 GB✓ fits
131K168.72 GB✓ fits
Other GPUs that run Llama 2 7B
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 5090 32GB2×3526$0.0788
AMD MI300X 192GB1×3067$0.2264
AMD MI325X 288GB1×3472$0.2800
A100 80GB SXM41×1157$0.2880
Intel Gaudi 3 128GB1×1909$0.2909
H100 SXM5 80GB1×1938$0.3583
Frequently asked questions

Can Llama 2 7B run on TPU v5p (per-chip, 95GB)?

Yes. Llama 2 7B at FP16 needs 47.8 GB including the KV cache, and TPU v5p (per-chip, 95GB) has 95 GB — so it fits on a single GPU.

How fast is Llama 2 7B on TPU v5p (per-chip, 95GB)?

39.49 tokens/sec per stream and 473.9 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 536.5 ms, and inter-token latency is 25.3 ms.

What does Llama 2 7B on TPU v5p (per-chip, 95GB) cost per million tokens?

$2.46 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 Llama 2 7B need on TPU v5p (per-chip, 95GB)?

13.5 GB for the weights at FP16, plus 34.4 GB of KV cache for 8 concurrent requests at 8K context — 47.8 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is TPU v5p (per-chip, 95GB) the right choice for Llama 2 7B?

It works. Compare $2.46 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 8× RTX 3080 10GB at $0.0186, and the fastest is 2× Cerebras CS-3 (WSE-3) at 39347 tok/s.

Adjust the workload yourself

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

Open in calculator