~/talk

Confidential Computing within an AI Accelerator

Talkat USENIX ATC '23 · 2023-07-10..07-12· [security]
links:none on file
links:none on file
links:none on file
links:none on file
links:none on file
links:none on file
links:none on file
links:none on file
links:none on file
links:none on file
links:none on file
slides:none on file

Built an AI accelerator extension called IPU Trusted Extensions (ITX) that adds an on‑chip hardware root-of-trust and PCIe Gen4‑rate cryptographic engines, plus compiler/runtime changes to support multi‑party training without a CPU TEE. Implemented on Graphcore GC200 taped out at TSMC 7nm and tested on a dev board, it keeps models and data encrypted on the device, decrypts at chip boundary, and reports <5% overhead in typical DNN training and up to 17x speedups versus CPU SEV‑SNP solutions.

AI-written summary · about

  1. 0:00Confidential computing in AI
  2. 1:07Why confidential computing for AI cloud
  3. 3:15Low-TCB AI accelerator design
  4. 8:49Trust extensions for IPU
  5. 13:44Remote attestation and key provisioning
  6. 16:35Replay and reorder attack protection
  7. 17:19Performance overheads and benchmarks
  8. 19:09Future of confidential AI
~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.