NVIDIA Google TPU
TPU v3-8
TPU v3 increases performance and memory for larger TensorFlow models with liquid cooling support.
- Launch year
- 2019
- Memory
- 128 GB HBM
- Memory bandwidth
- 900 GB/s
- Peak FP16 / FP32
- 180 TFLOPS · 0 TFLOPS
Market snapshot
$2.04 /hr
Range $2.04 – $8.80
Catalog coverage
9 live offerings
Across 1 providers · 5 regions
- Higher matrix throughput
- Liquid-cooled pods
- Integrated GCP access
Last refreshed Oct 20, 2025, 1:53 AM
Performance snapshot
Normalized versus NVIDIA A100 (=1.0). Values use public reference benchmarks for training and inference workloads.
- TensorFlow Training×0.80
- TensorFlow Inference×0.75
- Power Efficiency×0.70
Provider availability
Price bands per provider pulled from the live catalog.
Google Cloud9 offers$2.04 – $8.80 /hr
Popular regions
- us-central1-b2 offers
- us-central1-f2 offers
- us-central1-a2 offers
- europe-west4-a2 offers
- us-east1-d1 offers
Weighted average price $5.46 /hr · median $8.00
- Google Cloud$2.04/hr
- Google Cloud$2.04/hr
- Google Cloud$2.04/hr
- Google Cloud$2.24/hr
- Google Cloud$8.00/hr
- Google Cloud$8.00/hr
- Google Cloud$8.00/hr
- Google Cloud$8.00/hr
- Google Cloud$8.80/hr
| Google Cloud | v3-8 | us-central1-b | $2.04/hr |
| Google Cloud | v3-8 | us-central1-f | $2.04/hr |
| Google Cloud | v3-8 | us-central1-a | $2.04/hr |
| Google Cloud | v3-8 | europe-west4-a | $2.24/hr |
| Google Cloud | v3-8 | us-central1-b | $8.00/hr |
| Google Cloud | v3-8 | us-central1-f | $8.00/hr |
| Google Cloud | v3-8 | us-east1-d | $8.00/hr |
| Google Cloud | v3-8 | us-central1-a | $8.00/hr |
| Google Cloud | v3-8 | europe-west4-a | $8.80/hr |