new AMD Overdrive video's on youtube

[Start]

Anyone who's OC'ed multi core chips knows the one core will be the weak link. Technically, the other cores might have higher OC potential, if it wasn't for their twin brother holding them back.

What AMD is doing is giving us a way around that by letting us OC to the highest level on each core. So say, one core might max out at 3.4Ghz, and the other 3.3Ghz. You'll get the benefit of 100Mhz that you wouldn't normally get on a chip that had only one multiplier for all 4 cores.

Question is, what anomaly's might appear when you do asynchronous over clocking? The result might be performance that seems uneven, like an egg rolling end over end. Or the result might be that the faster core will end up waiting on the slower core, so your effective speed will still be only as good as the slowest core, unless you manually set affinity. No one will know until this ends up in the hands of John Q Public.

Correcto. But what I'm talking about is in heavy multithreaded application which uses all the cores possible. The workload would be distributed to all the cores evenly, and if one of the cores is slower, it'll take longer to finish executing the instruction. Having this one instruction a bit slower may not seem like a big deal at first, but if the other instructions in the core are dependent on each other to finish a certain clock cycle, this could become problematic. My best guess of how AMD got around this was probably the shared L3 cache and using it as storage for all 4 instructions. Then fetching the finished executed result from there. The other solution is likely to be stalling for the slower core, which would reduce the purpose of indepedent core overclocking. Fixing this solution by far is not an easy task.