More Performance Per Core

Seite also confirmed that high frequencies were a key design goal when creating Bulldozer. He summarised the two basic approaches to CPU design brilliantly: you can either try to process as many instructions per clock as possible, or have as many clocks per second as possible. Whether you choose IPC or a high frequency, the end result is speed. Improving the IPC of an execution core is really tricky, so AMD has taken the other option.

This is entirely sensible, and perfectly understandable considering that AMD has already designed a completely new processor that beats Intel’s much-marketed Atom and a new processor for low-end to mid-range desktop and laptop PCs this year. AMD has obviously been furiously busy for the past couple of years and clearly needs to work as efficiently as possible if it’s meet its own ambition. More IPC? Too hard, let’s just go for frequency instead.

To this end, the execution core of Bulldozer features a longer pipeline than the almost absurdly short pipeline of the K10 generation. We couldn’t get a precise number from Seite, but the K10 pipeline is only 12 stages and the mention of Prescott’s stupidly long 31-stage pipeline seemed way off the mark too, judging by the reaction.

Bulldozer Super Street Fighter Turbo

AMD has also updated its Turbo Core feature, which aims to overclock the CPU based on the load its under. A Bulldozer CPU can now overclock itself even when all eight cores are at 100 per cent load as long as there’s TDP headroom to do so, which is a new addition. The CPU can also overclock itself by two multipliers if half the CPU is shut down, a feature called Max Core.


We tested this feature in a range of multi- and lightly and single-threaded applications and saw a benefit in each scenario. Cinebench ran 2 per cent faster despite using every last bit of processing resource, while Arma II (which has a main thread with a few minor threads) saw gains of 11-13 per cent