Google32 GB VRAM819 GB/s$0.8/hr

Qwen2 7B on TPU v6e Trillium (32GB): Does It Fit, How Fast, What It Costs

Qwen2 7B at FP16 needs 15.2 GB for its weights and 19.0 GB in total at 8K context with 8 concurrent users. That means it fits on a single TPU v6e Trillium (32GB). Every figure is computed by tokcalc's open-source capacity model.

Fits?
Yes, 1 GPU
19.0 GB needed / 32 GB available
Decode speed
34.93
tok/s per stream
Aggregate
419.2
tok/s @ 8 concurrent
Cost / 1M tokens
$0.5301
1× TPU v6e Trillium (32GB)
What changes as you add TPU v6e Trillium (32GB)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× TPU32 GB19.0 GB34.93419.2330.6 ms$0.5301✓
2× TPU64 GB19.0 GB59.381425182.8 ms$0.3118✓
4× TPU128 GB19.0 GB118.85701106.4 ms$0.1559✓
8× TPU256 GB19.0 GB237.52280368.2 ms$0.0780✓
Concurrency vs context length
On 1× TPU v6e Trillium (32GB) at FP16. KV cache scales linearly with context, so this is the number of users your GPUs can actually serve.
ContextConcurrent usersKV per requestFits?
4K71240.5 MB✓ fits
8K35481.0 MB✓ fits
33K81.88 GB✓ fits
66K43.76 GB✓ fits
131K27.52 GB✓ fits
Other GPUs that run Qwen2 7B
Same workload, same quantization — ranked by value (tokens per dollar-hour).
GPUGPUsAggregate tok/sCost / 1M tok
RTX 3090 24GB1×479.1$0.1450
RTX 5090 32GB1×917.2$0.1514
RTX 4090 24GB1×515.9$0.2154
AMD MI300X 192GB1×2713$0.2560
V100 SXM2 32GB1×460.6$0.3015
A100 40GB SXM41×793.3$0.3151
Frequently asked questions

Can Qwen2 7B run on TPU v6e Trillium (32GB)?

Yes. Qwen2 7B at FP16 needs 19.0 GB including the KV cache, and TPU v6e Trillium (32GB) has 32 GB — so it fits on a single GPU.

How fast is Qwen2 7B on TPU v6e Trillium (32GB)?

34.93 tokens/sec per stream and 419.2 tok/s aggregate with continuous batching at 8 concurrent users. Time to first token is 330.6 ms, and inter-token latency is 28.6 ms.

What does Qwen2 7B on TPU v6e Trillium (32GB) cost per million tokens?

$0.5301 per 1M output tokens at 1× TPU v6e Trillium (32GB) ($0.8000/hr). That is the marginal hardware cost — add idle capacity, ops, and egress for the real figure.

How much VRAM does Qwen2 7B need on TPU v6e Trillium (32GB)?

15.2 GB for the weights at FP16, plus 3.8 GB of KV cache for 8 concurrent requests at 8K context — 19.0 GB total. Every extra concurrent user at long context costs more, so leave headroom.

Is TPU v6e Trillium (32GB) the right choice for Qwen2 7B?

It works. Compare $0.5301 per 1M tokens against the alternatives below — the cheapest option that fits the same workload is 2× RTX 3080 10GB at $0.0840, and the fastest is 1× Cerebras CS-3 (WSE-3) at 10236 tok/s.

Adjust the workload yourself

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

Open in calculator