Google TPU v3 32GB vs Google TPU v6e 32GB

The TPU v6e delivers 7.5x the BF16 throughput of the TPU v3 (918 vs 123 TFLOPS dense).

TPU v3: TPU, 2018 TPU v6e: TPU, 2024
BF16 dense lead
7.5x
TPU v6e: 918 vs 123 TFLOPS
VRAM
32 vs 32 GB
Same capacity
Bandwidth
900 GB/s vs 1.6 TB/s
TPU v6e moves data faster

Specifications

TPU v3
TPU v6e
Architecture
TPUTPU
Launch Year
20182024
Form Factor
VRAM
32 GB32 GB
Memory Bandwidth
900 GB/s1.6 TB/s
TDP
Process Node

Performance (TFLOPS)

TPU v3
TPU v6e
BF16
123.0 TFlops 918.0 TFlops
INT8
1836.0 TFlops

FLOPS by Precision

What actually differs

The TPU v6e is the newer part: TPU, launched in 2024, against the TPU v3's TPU from 2018. Newer architectures typically add lower-precision formats and better throughput per watt, so check the precision rows your workload actually uses.

Memory is 32 GB against 32 GB, fed at 900 GB/s versus 1.6 TB/s. More memory per GPU means larger models fit before you have to shard across cards, which often matters more than raw TFLOPS for inference.

For LLM training and inference, weight the FP8 and FP16 rows and memory capacity most heavily. For scientific and HPC workloads, the FP64 row is the one to read.

Frequently asked questions

Is the TPU v3 faster than the TPU v6e?

At BF16 precision the TPU v6e reaches 918 TFLOPS dense against 123 TFLOPS for the TPU v3. The performance table on this page lists every published precision for both GPUs.

Which has more memory, the TPU v3 or the TPU v6e?

Both GPUs carry 32 GB of VRAM. Memory bandwidth is 900 GB/s for the TPU v3 and 1.6 TB/s for the TPU v6e.

Get Comparison Updates

New GPUs added weekly. Be the first to see how they compare.