Google95 GB VRAM819 GB/s$4.2/hr

SmolLM2 1.7B Instruct on TPU v5p (per-chip, 95GB): Does It Fit, How Fast, What It Costs

SmolLM2 1.7B Instruct at FP16 needs 3.4 GB for its weights and 16.3 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
16.3 GB needed / 95 GB available
Decode speed
156.6
tok/s per stream
Aggregate
1879
tok/s @ 8 concurrent
Cost / 1M tokens
$0.6209
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 GB16.3 GB156.61879157.7 ms$0.6209✓
2× TPU190 GB16.3 GB266.2638894.4 ms$0.3653✓
4× TPU380 GB16.3 GB532.42555362.2 ms$0.1826✓
8× TPU760 GB16.3 GB106510221146.1 ms$0.0913✓
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?
4K113824.6 MB✓ fits
8K561.61 GB✓ fits
33K146.44 GB✓ fits
66K712.88 GB✓ fits
131K325.77 GB✓ fits
Other GPUs that run SmolLM2 1.7B Instruct
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 3090 24GB1×2147$0.0323
RTX 5090 32GB1×4111$0.0338
RTX 4090 24GB1×2312$0.0480
AMD MI300X 192GB1×12159$0.0571
V100 SXM2 32GB1×2065$0.0673
A100 40GB SXM41×3556$0.0703
Frequently asked questions

Can SmolLM2 1.7B Instruct run on TPU v5p (per-chip, 95GB)?

Yes. SmolLM2 1.7B Instruct at FP16 needs 16.3 GB including the KV cache, and TPU v5p (per-chip, 95GB) has 95 GB — so it fits on a single GPU.

How fast is SmolLM2 1.7B Instruct on TPU v5p (per-chip, 95GB)?

156.6 tokens/sec per stream and 1879 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 157.7 ms, and inter-token latency is 6.4 ms.

What does SmolLM2 1.7B Instruct on TPU v5p (per-chip, 95GB) cost per million tokens?

$0.6209 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 SmolLM2 1.7B Instruct need on TPU v5p (per-chip, 95GB)?

3.4 GB for the weights at FP16, plus 12.9 GB of KV cache for 8 concurrent requests at 8K context — 16.3 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is TPU v5p (per-chip, 95GB) the right choice for SmolLM2 1.7B Instruct?

It works. Compare $0.6209 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 2× RTX 3080 10GB at $0.0187, and the fastest is 1× Cerebras CS-3 (WSE-3) at 45882 tok/s.

Adjust the workload yourself

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

Open in calculator