BREAKING
AWS AI Chips Gain Ground
0
chips
per UltraServer
0
x
compute vs prior
0
%
lower power
Inferentia vs Trainium3
Inferentia
Inference
●
Inf2 instances
●
Up to 4x throughput vs gen 1
●
Up to 70% lower cost
Trainium3
Train + Infer
●
UltraServer up to 144 chips
●
4x+ compute vs predecessor
●
30-50% ahead of Nvidia
Real-World Cost Savings
Leonardo.ai
80
Finch
80
Tomofun BLIP
83
CUDA Still Holds an Edge
A Practical Nvidia Alternative
AI NEWS BLITZ
Amazon's in-house AI chips are winning over cost-focused firms looking beyond Nvidia.