Google95 GB VRAM819 GB/s$4.2/hr

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

Llama 3 8B at FP16 needs 16.1 GB for its weights and 24.6 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
24.6 GB needed / 95 GB available
Decode speed
33.15
tok/s per stream
Aggregate
397.8
tok/s @ 8 concurrent
Cost / 1M tokens
$2.93
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 GB24.6 GB33.15397.8633.4 ms$2.93✓
2× TPU190 GB24.6 GB56.351352334.4 ms$1.73✓
4× TPU380 GB24.6 GB112.75410182.2 ms$0.8627✓
8× TPU760 GB24.6 GB225.421639106.1 ms$0.4313✓
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?
4K147549.8 MB✓ fits
8K731.07 GB✓ fits
33K184.29 GB✓ fits
66K98.59 GB✓ fits
131K417.18 GB✓ fits
Other GPUs that run Llama 3 8B
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 5090 32GB1×870.3$0.1596
AMD MI300X 192GB1×2574$0.2698
V100 SXM2 32GB1×437.1$0.3177
A100 40GB SXM41×752.8$0.3321
AMD MI325X 288GB1×2914$0.3336
A100 80GB SXM41×971.4$0.3432
Frequently asked questions

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

Yes. Llama 3 8B at FP16 needs 24.6 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 3 8B on TPU v5p (per-chip, 95GB)?

33.15 tokens/sec per stream and 397.8 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 633.4 ms, and inter-token latency is 30.2 ms.

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

$2.93 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 3 8B need on TPU v5p (per-chip, 95GB)?

16.1 GB for the weights at FP16, plus 8.6 GB of KV cache for 8 concurrent requests at 8K context — 24.6 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 3 8B?

It works. Compare $2.93 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 4× RTX 3080 10GB at $0.0443, and the fastest is 1× Cerebras CS-3 (WSE-3) at 9714 tok/s.

Adjust the workload yourself

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

Open in calculator