Google95 GB VRAM819 GB/s$4.2/hr

Falcon 3 10B Instruct on TPU v5p (per-chip, 95GB): Does It Fit, How Fast, What It Costs

Falcon 3 10B Instruct at FP16 needs 20.0 GB for its weights and 30.7 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
30.7 GB needed / 95 GB available
Decode speed
26.62
tok/s per stream
Aggregate
319.4
tok/s @ 8 concurrent
Cost / 1M tokens
$3.65
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 GB30.7 GB26.62319.4781.5 ms$3.65✓
2× TPU190 GB30.7 GB45.251086409.0 ms$2.15✓
4× TPU380 GB30.7 GB90.504344219.5 ms$1.07✓
8× TPU760 GB30.7 GB181.017376124.8 ms$0.5371✓
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?
4K111687.2 MB✓ fits
8K551.34 GB✓ fits
33K135.37 GB✓ fits
66K610.74 GB✓ fits
131K321.47 GB✓ fits
Other GPUs that run Falcon 3 10B Instruct
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 5090 32GB1×698.9$0.1987
AMD MI300X 192GB1×2067$0.3360
V100 SXM2 32GB1×351.0$0.3957
A100 40GB SXM41×604.5$0.4136
AMD MI325X 288GB1×2340$0.4155
A100 80GB SXM41×780.0$0.4274
Frequently asked questions

Can Falcon 3 10B Instruct run on TPU v5p (per-chip, 95GB)?

Yes. Falcon 3 10B Instruct at FP16 needs 30.7 GB including the KV cache, and TPU v5p (per-chip, 95GB) has 95 GB — so it fits on a single GPU.

How fast is Falcon 3 10B Instruct on TPU v5p (per-chip, 95GB)?

26.62 tokens/sec per stream and 319.4 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 781.5 ms, and inter-token latency is 37.6 ms.

What does Falcon 3 10B Instruct on TPU v5p (per-chip, 95GB) cost per million tokens?

$3.65 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 Falcon 3 10B Instruct need on TPU v5p (per-chip, 95GB)?

20.0 GB for the weights at FP16, plus 10.7 GB of KV cache for 8 concurrent requests at 8K context — 30.7 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is TPU v5p (per-chip, 95GB) the right choice for Falcon 3 10B Instruct?

It works. Compare $3.65 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 4× RTX 3080 10GB at $0.0551, and the fastest is 1× Cerebras CS-3 (WSE-3) at 7800 tok/s.

Adjust the workload yourself

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

Open in calculator