Layer 2 · head to head
Google TPU v6e (Trillium) vs Google TPU v7 (Ironwood)
One generation apart, and Google publishes far more detail about the newer one — which is itself a signal about how they want them compared.
| Google TPU v6e (Trillium) | Google TPU v7 (Ironwood) | |
|---|---|---|
| Can you get it | Cannot be bought. Exists only inside one cloud, so choosing it is choosing that cloud. | Cannot be bought. Exists only inside one cloud, so choosing it is choosing that cloud. |
| Memory | Not published in directly comparable terms | 192 GB HBM3E per chip |
| Bandwidth | Not published | 7.37 TB/s per chip |
| Compute | Not published in comparable FP8 terms | 4,614 FP8 TFLOPS per chip |
| Power | Not published | Not published |
| Interconnect | ICI | 9.6 Tb/s inter-chip (ICI) |
| Software | JAX, XLA, PyTorch/XLA | JAX, XLA, PyTorch/XLA |
| Workload | both | inference |
Google TPU v6e (Trillium)
For: The previous TPU generation, still widely provisioned on Google Cloud and cheaper per hour than Ironwood.
The catch: Same lock-in as any TPU: Google Cloud only. Google publishes far less per-chip detail for v6e than for v7, which is itself a signal about how they want it compared.
Economics: Priced as a cloud SKU. Compare against v7 on tokens delivered per dollar, not on any spec sheet.
Google TPU v7 (Ironwood)
For: Inference at scale on Google Cloud. Ships in 256-chip and 9,216-chip configurations.
The catch: You cannot buy this. It exists only inside Google Cloud, so choosing it is choosing a cloud, permanently. It is also inference-optimised — do not benchmark it as a training part.
Economics: Vertically integrated: the price you see is a cloud price, not a chip price, and it is not comparable to a $/GPU-hour rental line.
Before either — can you power it?
Not published against Not published. In 2026 that comparison usually matters more than the FLOPS one: the US interconnection queue exceeds 2,600 GW with waits approaching five years, and roughly 80% of projects withdraw before energising.
Want this whole layer as one table?
The accelerators on these pages as a single CSV — memory, bandwidth, compute, power, interconnect, software stack, availability, who each part is for and the catch — ready to sort, filter or load into a notebook.
Get the silicon table — free for research and evaluation
Four short lines and the download starts at once (CC-BY-NC-4.0, built from these pages the moment you ask). What you type is used only to answer the request and is never sold. Prefer Hugging Face? Request it there.
Using it inside a product, a client report or anything you sell? That needs a commercial licence.