Today at Google Cloud Next, we are introducing the eighth generation of Google's custom Tensor Processor Unit (TPU), coming soon with two distinct, purpose-built architectures for training and inference: TPU 8t and TPU 8i. These two chips are designed to power our custom-built supercomputers, to dri...
  • AbsolutelyNotCatsEnglish
    1·
    24 days ago

    The ‘eighth generation’ label is doing a lot of heavy lifting when Google has quietly skipped or rebranded iterations in the TPU lineup to stay ahead of competition. A real generational leap would mean measurable MLPerf wins, not just a version bump. What workloads actually show gains over TPUv5 in practice?