NVIDIA Google TPU
TPU v2-8
TPU v2 pod slices provide eight TPU cores for TensorFlow-intensive training and inference workloads.
- Launch year
- 2018
- Memory
- 64 GB HBM
- Memory bandwidth
- 600 GB/s
- Peak FP16 / FP32
- 90 TFLOPS · 0 TFLOPS
Market snapshot
$1.26 /hr
Range $1.26 – $5.22
Catalog coverage
10 live offerings
Across 1 providers · 5 regions
- Systolic array compute
- Integrated with Google Cloud
- Great for TensorFlow training
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×0.60
- TensorFlow Inference×0.55
- Power Efficiency×0.70
Provider availability
Price bands per provider pulled from the live catalog.
Google Cloud10 offers$1.26 – $5.22 /hr
Popular regions
- europe-west4-a2 offers
- us-central1-b2 offers
- us-central1-c2 offers
- us-central1-f2 offers
- asia-east1-c2 offers
Weighted average price $3.06 /hr · median $4.50
- Google Cloud$1.26/hr
- Google Cloud$1.35/hr
- Google Cloud$1.35/hr
- Google Cloud$1.35/hr
- Google Cloud$1.57/hr
- Google Cloud$4.50/hr
- Google Cloud$4.50/hr
- Google Cloud$4.50/hr
- Google Cloud$4.95/hr
- Google Cloud$5.22/hr
| Google Cloud | v2-8 | europe-west4-a | $1.26/hr |
| Google Cloud | v2-8 | us-central1-b | $1.35/hr |
| Google Cloud | v2-8 | us-central1-c | $1.35/hr |
| Google Cloud | v2-8 | us-central1-f | $1.35/hr |
| Google Cloud | v2-8 | asia-east1-c | $1.57/hr |
| Google Cloud | v2-8 | us-central1-b | $4.50/hr |
| Google Cloud | v2-8 | us-central1-c | $4.50/hr |
| Google Cloud | v2-8 | us-central1-f | $4.50/hr |
| Google Cloud | v2-8 | europe-west4-a | $4.95/hr |
| Google Cloud | v2-8 | asia-east1-c | $5.22/hr |