DEPLOYDatabase

Chip comparison

Google TPU v6 (Trillium) vs NVIDIA H100 Tensor Core GPU

Side-by-side specs

Straight from each vendor's datasheet. Winner in bold: higher throughput, memory, or transistor count; lower power draw or price. Sparse (2:4) throughput roughly doubles the FP8/INT8 numbers shown.

FieldGoogle TPU v6 (Trillium)NVIDIA H100 Tensor Core GPU
Process nodeTSMC 4N
Transistors (B)80
Die size814 mm²
TDP200 W700 W
Memory32 GB HBM380 GB HBM3
Memory bandwidth1,600 GB/s3,350 GB/s
FP16 (dense)989 TFLOPS
BF16 (dense)918 TFLOPS989 TFLOPS
FP8 (dense)1,836 TFLOPS1,979 TFLOPS
INT8 (dense)1,836 TOPS1,979 TOPS
Launch price (list)$30,000
Form factorOAM (Trillium pod, up to 256 chips)SXM5
Announced2024-05-142022-03-22
Released2024-12-012022-10-13

Robots running Google TPU v6 (Trillium)

No robots publicly running it yet.

Robots running NVIDIA H100 Tensor Core GPU

No robots publicly running it yet.

Data centers with Google TPU v6 (Trillium)

Common questions

Google TPU v6 (Trillium) vs NVIDIA H100 Tensor Core GPU: which is faster for training?

NVIDIA H100 Tensor Core GPU has 1.08x the dense FP8 throughput of the other (Google TPU v6 (Trillium): 1,836 TFLOPS; NVIDIA H100 Tensor Core GPU: 1,979 TFLOPS). Real-world training speed also depends on memory bandwidth, interconnect, and software stack: see the spec diff for those inputs.

Google TPU v6 (Trillium) vs NVIDIA H100 Tensor Core GPU: which has more memory?

NVIDIA H100 Tensor Core GPU carries more HBM (Google TPU v6 (Trillium): 32 GB HBM3; NVIDIA H100 Tensor Core GPU: 80 GB HBM3). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.

Google TPU v6 (Trillium) vs NVIDIA H100 Tensor Core GPU: which is more power-efficient?

Dense FP8 TFLOPS per watt: Google TPU v6 (Trillium) 9.18 (1,836 ÷ 200 W); NVIDIA H100 Tensor Core GPU 2.83 (1,979 ÷ 700 W). Google TPU v6 (Trillium) wins at the die level; system-level efficiency also depends on cooling and interconnect.

Google TPU v6 (Trillium) vs NVIDIA H100 Tensor Core GPU: which is more widely used?

Google TPU v6 (Trillium): 0 robots and 1 data center publicly running it. NVIDIA H100 Tensor Core GPU: 0 robots and 7 data centers. Empty here means no operator has said publicly, not zero adoption in the market.

See also: every chip comparison · Google TPU v6 (Trillium) full page · NVIDIA H100 Tensor Core GPU full page.