My AMD Bulldozer 2 (Second Generation) Wish list (2012 ~2013):
Front-end:
- 5 decoder, instead of 4 < Guys it’s Dual Core Module 4 is so little to feed dual core
Engine:
- 2 x 256 SMID FB units
- Bigger L1 Cache - 32KB Instructions and 32KB Data
- Mico-ops Cashe if it’s important (After the decoding) - like Sandy Bridge
- Faster ALU units - Improve the throughput and the latency - AMD you need to work on the single-thread performance more
- Fewer Pipelines Stages, like Phenom II
- Improve the Pipelines stages gate delay
Cache:
- Sub 35ns L3 latency, and fix the writing performance problems.
- Segment the L3 Cache into slice per core, each slice have cache pipelines for better writing performance.
- L3 working in same CPU clock
Memory:
- 3 channels DDR 3 for 4 cores and less
- 4 channels DDR 3 for 6, 8, and 10 cores
Bulldozer 3 that maybe will come 2014 ~ 2015:
- ATi Stream Engine that take care of all the FB workload