The memory controller got a major overhaul though. AMD claims up to 40% bandwidth improvement from:
- Larger buffers (~2-4x) - Write bursting - Optimized paging - New and improved prefetching
I suspect the typical improvement seen by apps will be substantially less than 40%. If it is even 10% then the peak FLOP/s to bandwidth "machine balance" ratio of Barcelona will still worsen by 72% vs K8 at the same CPU frequency. Ironically for AMD the much lower CPU clock frequencies of Barcelona will cushion growth in application inefficiency from worsened machine balance. ;-)