Google95 GB VRAM819 GB/s$4.2/hr

Phi-4 14B on TPU v5p (per-chip, 95GB): Does It Fit, How Fast, What It Costs

Phi-4 14B at FP16 needs 28.0 GB for its weights and 38.7 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
38.7 GB needed / 95 GB available
Decode speed
19.01
tok/s per stream
Aggregate
228.1
tok/s @ 8 concurrent
Cost / 1M tokens
$5.11
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 GB38.7 GB19.01228.11.08 s$5.11✓
2× TPU190 GB38.7 GB32.32775.7560.7 ms$3.01✓
4× TPU380 GB38.7 GB64.643103295.3 ms$1.50✓
8× TPU760 GB38.7 GB129.312411162.7 ms$0.7520✓
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?
4K99687.2 MB✓ fits
8K491.34 GB✓ fits
33K125.37 GB✓ fits
66K610.74 GB✓ fits
131K321.47 GB✓ fits
Other GPUs that run Phi-4 14B
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 5090 32GB2×1697$0.1637
AMD MI300X 192GB1×1476$0.4704
A100 40GB SXM41×431.8$0.5790
AMD MI325X 288GB1×1671$0.5817
A100 80GB SXM41×557.1$0.5983
Intel Gaudi 3 128GB1×919.3$0.6043
Frequently asked questions

Can Phi-4 14B run on TPU v5p (per-chip, 95GB)?

Yes. Phi-4 14B at FP16 needs 38.7 GB including the KV cache, and TPU v5p (per-chip, 95GB) has 95 GB — so it fits on a single GPU.

How fast is Phi-4 14B on TPU v5p (per-chip, 95GB)?

19.01 tokens/sec per stream and 228.1 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 1.08 s, and inter-token latency is 52.6 ms.

What does Phi-4 14B on TPU v5p (per-chip, 95GB) cost per million tokens?

$5.11 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 Phi-4 14B need on TPU v5p (per-chip, 95GB)?

28.0 GB for the weights at FP16, plus 10.7 GB of KV cache for 8 concurrent requests at 8K context — 38.7 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is TPU v5p (per-chip, 95GB) the right choice for Phi-4 14B?

It works. Compare $5.11 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 4× RTX 3080 10GB at $0.0772, and the fastest is 1× Cerebras CS-3 (WSE-3) at 5571 tok/s.

Adjust the workload yourself

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

Open in calculator