DEPLOYDatabase

Chip comparison

Google TPU v7 (Ironwood) vs NVIDIA L40S

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 v7 (Ironwood)NVIDIA L40S
Process nodeTSMC 4N
Transistors (B)76.3
TDP700 W350 W
Memory192 GB HBM3e48 GB GDDR6
Memory bandwidth7,400 GB/s864 GB/s
FP16 (dense)362 TFLOPS
BF16 (dense)4,614 TFLOPS362 TFLOPS
FP8 (dense)9,228 TFLOPS733 TFLOPS
INT8 (dense)733 TOPS
Form factorOAM (Ironwood pod, up to 9,216 chips)PCIe
Announced2025-04-092023-08-08
Released2025-12-01

Robots running Google TPU v7 (Ironwood)

No robots publicly running it yet.

Robots running NVIDIA L40S

No robots publicly running it yet.

Data centers with Google TPU v7 (Ironwood)

No data centers publicly running it yet.

Data centers with NVIDIA L40S

No data centers publicly running it yet.

Common questions

Google TPU v7 (Ironwood) vs NVIDIA L40S: which is faster for training?

Google TPU v7 (Ironwood) has 12.59x the dense FP8 throughput of the other (Google TPU v7 (Ironwood): 9,228 TFLOPS; NVIDIA L40S: 733 TFLOPS). Real-world training speed also depends on memory bandwidth, interconnect, and software stack: see the spec diff for those inputs.

Google TPU v7 (Ironwood) vs NVIDIA L40S: which has more memory?

Google TPU v7 (Ironwood) carries more HBM (Google TPU v7 (Ironwood): 192 GB HBM3e; NVIDIA L40S: 48 GB GDDR6). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.

Google TPU v7 (Ironwood) vs NVIDIA L40S: which is more power-efficient?

Dense FP8 TFLOPS per watt: Google TPU v7 (Ironwood) 13.18 (9,228 ÷ 700 W); NVIDIA L40S 2.09 (733 ÷ 350 W). Google TPU v7 (Ironwood) wins at the die level; system-level efficiency also depends on cooling and interconnect.

See also: every chip comparison · Google TPU v7 (Ironwood) full page · NVIDIA L40S full page.