DEPLOYDatabase

Chip comparison

Google TPU v6 (Trillium) vs Intel Gaudi 3

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)Intel Gaudi 3
Process nodeTSMC N5
TDP200 W900 W
Memory32 GB HBM3128 GB HBM2e
Memory bandwidth1,600 GB/s3,675 GB/s
FP16 (dense)1,835 TFLOPS
BF16 (dense)918 TFLOPS1,835 TFLOPS
FP8 (dense)1,836 TFLOPS1,835 TFLOPS
INT8 (dense)1,836 TOPS
Form factorOAM (Trillium pod, up to 256 chips)OAM
Announced2024-05-142024-04-09
Released2024-12-012024-09-01

Robots running Google TPU v6 (Trillium)

No robots publicly running it yet.

Robots running Intel Gaudi 3

No robots publicly running it yet.

Data centers with Google TPU v6 (Trillium)

Data centers with Intel Gaudi 3

No data centers publicly running it yet.

Common questions

Google TPU v6 (Trillium) vs Intel Gaudi 3: which is faster for training?

Google TPU v6 (Trillium) has 1.00x the dense FP8 throughput of the other (Google TPU v6 (Trillium): 1,836 TFLOPS; Intel Gaudi 3: 1,835 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 Intel Gaudi 3: which has more memory?

Intel Gaudi 3 carries more HBM (Google TPU v6 (Trillium): 32 GB HBM3; Intel Gaudi 3: 128 GB HBM2e). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.

Google TPU v6 (Trillium) vs Intel Gaudi 3: which is more power-efficient?

Dense FP8 TFLOPS per watt: Google TPU v6 (Trillium) 9.18 (1,836 ÷ 200 W); Intel Gaudi 3 2.04 (1,835 ÷ 900 W). Google TPU v6 (Trillium) wins at the die level; system-level efficiency also depends on cooling and interconnect.

Google TPU v6 (Trillium) vs Intel Gaudi 3: which is more widely used?

Google TPU v6 (Trillium): 0 robots and 1 data center publicly running it. Intel Gaudi 3: 0 robots and 0 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 · Intel Gaudi 3 full page.