Google TPU v3 32GB vs Google TPU v7 192GB

The TPU v7 delivers 18.8x the BF16 throughput of the TPU v3 (2,307 vs 123 TFLOPS dense). The TPU v7 also carries 160 GB more memory (192 GB vs 32 GB).

TPU v3: TPU, 2018 TPU v7: TPU, 2025
BF16 dense lead
18.8x
TPU v7: 2,307 vs 123 TFLOPS
VRAM
32 vs 192 GB
160 GB more for the TPU v7
Bandwidth
900 GB/s vs 7.4 TB/s
TPU v7 moves data faster

Specifications

TPU v3
TPU v7
Architecture
TPUTPU
Launch Year
20182025
Form Factor
VRAM
32 GB192 GB
Memory Bandwidth
900 GB/s7.4 TB/s
TDP
Process Node

Performance (TFLOPS)

TPU v3
TPU v7
BF16
123.0 TFlops 2307.0 TFlops
FP8
4614.0 TFlops

FLOPS by Precision

What actually differs

The TPU v7 is the newer part: TPU, launched in 2025, 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 192 GB, fed at 900 GB/s versus 7.4 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 v7?

At BF16 precision the TPU v7 reaches 2,307 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 v7?

The TPU v7 carries 192 GB of VRAM versus 32 GB for the TPU v3. Memory bandwidth is 900 GB/s for the TPU v3 and 7.4 TB/s for the TPU v7.

Get Comparison Updates

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