Google95 GB VRAM819 GB/s$4.2/hr

Mistral 7B v0.3 on TPU v5p (per-chip, 95GB): Does It Fit, How Fast, What It Costs

Mistral 7B v0.3 at FP16 needs 14.5 GB for its weights and 23.1 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
23.1 GB needed / 95 GB available
Decode speed
36.76
tok/s per stream
Aggregate
441.2
tok/s @ 8 concurrent
Cost / 1M tokens
$2.64
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 GB23.1 GB36.76441.2574.1 ms$2.64✓
2× TPU190 GB23.1 GB62.501500304.4 ms$1.56✓
4× TPU380 GB23.1 GB125.06000167.2 ms$0.7778✓
8× TPU760 GB23.1 GB250.02400098.6 ms$0.3889✓
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?
4K149549.8 MB✓ fits
8K741.07 GB✓ fits
33K184.29 GB✓ fits
66K98.59 GB✓ fits
131K417.18 GB✓ fits
Other GPUs that run Mistral 7B v0.3
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 3090 24GB1×504.2$0.1377
RTX 5090 32GB1×965.3$0.1439
RTX 4090 24GB1×543.0$0.2046
AMD MI300X 192GB1×2855$0.2432
V100 SXM2 32GB1×484.8$0.2865
A100 40GB SXM41×834.9$0.2994
Frequently asked questions

Can Mistral 7B v0.3 run on TPU v5p (per-chip, 95GB)?

Yes. Mistral 7B v0.3 at FP16 needs 23.1 GB including the KV cache, and TPU v5p (per-chip, 95GB) has 95 GB — so it fits on a single GPU.

How fast is Mistral 7B v0.3 on TPU v5p (per-chip, 95GB)?

36.76 tokens/sec per stream and 441.2 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 574.1 ms, and inter-token latency is 27.2 ms.

What does Mistral 7B v0.3 on TPU v5p (per-chip, 95GB) cost per million tokens?

$2.64 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 Mistral 7B v0.3 need on TPU v5p (per-chip, 95GB)?

14.5 GB for the weights at FP16, plus 8.6 GB of KV cache for 8 concurrent requests at 8K context — 23.1 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is TPU v5p (per-chip, 95GB) the right choice for Mistral 7B v0.3?

It works. Compare $2.64 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 4× RTX 3080 10GB at $0.0399, and the fastest is 1× Cerebras CS-3 (WSE-3) at 10773 tok/s.

Adjust the workload yourself

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

Open in calculator