Google TPU v6e Pod

pod
256× TPU v6e
2D torus ICI, 800 GBps bidirectional per chip
2024
BF16
235.0
PFLOPS
INT8
470.0
PFLOPS
Power
kW Total
Memory
8389
GB Total

A full TPU v6e (Trillium) pod is 256 chips in a 2D torus, the largest slice being a 16x16 topology across 64 VMs. Per chip Google publishes 918 TFLOPS bf16 and 1,836 TOPS int8, giving 235.0 PFLOPS and 470.0 POPS across the pod: a 4.7x per-chip gain over v5e at the same pod size. Memory is 256 x 32 GB HBM at 1,638 GBps per chip, with 800 GBps of bidirectional inter-chip interconnect per chip. v6e is the first generation to widen the matrix unit to 256x256 multiply-accumulators, from 128x128 on every prior TPU. Google publishes no pod-level compute or power figure; the compute above is chip count times published per-chip peak. No sparse figures are published, so all figures are dense.

We will point you at suppliers who have it. Free, and no signup.

BF16
235.01
PFLOPS
INT8
470.02
PFLOPS

System Details

GPU Configuration

GPU Count: 256 GPUs
Architecture: TPU
Interconnect: 2D torus ICI, 800 GBps bidirectional per chip

System Specifications

Form Factor: pod
Total Power:
Total Memory: 8389 GB
Memory Bandwidth: 419328 GB/s

Precision Performance Breakdown

PrecisionSystem PerformancePer GPUEfficiency
235.008 PFLOPS 918.0 TFLOPS
470.016 PFLOPS 1836.0 TFLOPS

All figures are dense. Vendors commonly headline the number, which is twice the dense one.

Powered by Google TPU v6e

This system utilizes 256 × Google TPU v6e 32GB GPUs, each delivering exceptional performance for AI and HPC workloads.

Per GPU TDP

Per GPU Memory

32 GB

Process Node

Architecture

TPU

Documentation & Resources

Official Datasheet

Google TPU v6e Pod technical specifications

Download PDF

Cloud TPU v5p

Google • 2023-12-07

View Document ↗

Typical Use Cases

AI/ML Training
High-Performance Computing
Data Analytics

The Google TPU v6e Pod runs 256× TPU v6e GPUs over 2D torus ICI, 800 GBps bidirectional per chip, delivering 235 PFLOPS BF16 dense.