Skip to content
Updated minutes ago
google

TPU v5e

google · tpu · 16 GB HBM2e · 200W TDP

VRAM

16 GB

BF16 TFLOPS

200

Bandwidth

820 GB/s

From

$0.85/hr

Calculate ROI with this GPU →

Spec Sheet

VRAM16 GB HBM2e
Memory Bandwidth820 GB/s
BF16 TFLOPS200
FP16 TFLOPS200
FP8 TFLOPS400
INT8 TOPS400
TDP200W
InterconnectPCIE
Max per Node256
PCIe Gen4
Tensor CoresNo

Pricing by Provider

ProviderOn-DemandReservedSpotBadge
gcp$1.20/hr$0.85/hr-Cheapest

Compatible Models (281)

Training Capabilities

Estimated GPU count for full fine-tuning (AdamW, BF16) and QLoRA

Model SizeFull Fine-TuneQLoRA
7B model9 GPUs1 GPU
13B model16 GPUs1 GPU
70B model83 GPUs3 GPUs

Energy Efficiency

Estimated tokens/second per Watt for popular models

Mistral 7B
0.56 t/s/WFP8
Qwen 2.5 7B
0.54 t/s/WFP8
Llama 3.1 8B
0.51 t/s/WFP8
DeepSeek V3
0.11 t/s/WFP8
Llama 3.1 70B
0.06 t/s/WFP8
Qwen 2.5 72B
0.06 t/s/WFP8

Similar GPUs

GPUVRAMBF16 TFLOPSBW (GB/s)From
TPU v432 GB2751200$2.25/hr
TPU v6e (Trillium)32 GB4601640$1.75/hr
A400016 GB76448$0.17/hr
RTX 408016 GB97717$0.32/hr
T416 GB65300$0.12/hr

Methodology Note

Performance estimates for the TPU v5eare based on InferenceBench's roofline performance model with CUDA kernel-level optimization including FlashAttention v2 and PagedAttention. Memory calculations account for model weights (16 GB HBM2e available), KV-cache allocation, and activation memory. Throughput predictions use the TPU v5e's rated 820 GB/s memory bandwidth and 200 BF16 TFLOPS compute capacity as roofline ceilings, with empirical correction factors per GPU architecture (tpu). See our full methodology.

Frequently Asked Questions

How many AI models can run on TPU v5e?

The TPU v5e can run 281 AI models from our database within a single node. Compatible models range across various parameter sizes depending on the quantization precision (BF16, FP8, INT4). Smaller models fit on a single GPU while larger models may require multi-GPU setups up to 256x TPU v5e.

What is the TPU v5e inference throughput?

The TPU v5e delivers 200 BF16 TFLOPS and 400 FP8 TFLOPS with 820 GB/s memory bandwidth. Actual inference throughput (tokens/sec) depends on the model size, precision, and batch size. Use our calculator for model-specific throughput estimates.

How much does TPU v5e cost per hour?

The TPU v5e is available starting from $0.85/hour via gcp. Prices vary by provider and pricing tier (on-demand, reserved, spot). Compare pricing across all providers in the table above.