GoogleTPU
Google TPU v3
TPU v3 architecture · 16GB memory · 123 FP16 TFLOPS
Cloud Pricing Today
About the Google TPU v3
The Google Cloud TPU v3 offers improved performance over v2 with liquid cooling and higher compute throughput. It supports larger models and is available in pod configurations for distributed training.
Memory (VRAM)
16 GB
FP16 Performance
123 TFLOPS
Architecture
TPU v3
Common Use Cases
Large-scale ML trainingJAX/TensorFlow workloadsDistributed training
Key Facts
- Manufacturer
- Architecture
- TPU v3
- Accelerator Type
- TPU
- Primary Use
- training
- Memory (VRAM)
- 16 GB
- FP16 Performance
- 123 TFLOPS
Related Accelerators
NVIDIA
NVIDIA GB300
Blackwell · 288GB · 2250 TFLOPS
training
NVIDIA
NVIDIA GB200
Blackwell · 192GB · 1800 TFLOPS
training
NVIDIA
NVIDIA B200
Blackwell · 192GB · 1800 TFLOPS
training
NVIDIA
NVIDIA B300
Blackwell · 288GB · 2250 TFLOPS
training
NVIDIA
NVIDIA H200
Hopper · 141GB · 990 TFLOPS
training
NVIDIA
NVIDIA H100
Hopper · 80GB · 990 TFLOPS
training
Track Google TPU v3 pricing over time
Get access to historical pricing data, regional analysis, and custom alerts.