BREAKING
Google Building 'Frozen v2' AI Chip
0
x
low estimate
0
x
high estimate
0
target year
How Frozen v2 Works
1
Fix Gemini logic in silicon
↓
2
Reduce data movement
↓
3
Faster, cheaper inference
TPU vs Frozen v2
TPU
General
●
Handles training
●
Flexible workloads
●
Stays the workhorse
Frozen v2
Specialized
●
Inference focused
●
Model baked in
●
Less flexible
The Rigidity Trade-Off
Experiment, Not Product Yet
AI NEWS BLITZ
Google is reportedly developing a chip that bakes Gemini logic straight into silicon.