Pemalite said:
fatslob-:O said:
Pemalite said:
fatslob-:O said:
I'm sure an FX-8350 will perform fine on integer workloads but it falls way behind when attempting to go against applications that need the floating point performance and even more behind when depending on single threaded peformance.
|
That goes without saying because of how AMD designed the chips where they shared the floating point unit across 2 "cores".
But that was for a reason, which won't come to fruition for a few years yet. (Having the IGP take over some of that task.)
|
That was a bad idea in general. What AMD still doesn't understand is that there is a need for serialized floating point and integer workloads. AMD fusion was a decent idea if not for all the delays but AMD bulldozer was a total disaster, period. It mostly only shows incompetence on their CPU division. I think AMD is better off copying intel's core architecture.
|
Well. Bulldozer was a disaster, I don't think anyone could argue with that.
But what it did accomplish is mid-range performance for a good price, granted only if you have cheap energy prices. :P
AMD should go back to it's Stars architecture and tweak it, it's a better starting point than the FX.
|
Definitely agreed that they should go back to the K10 but the K10 was just a slight modification of the K8. How should AMD go about competing with intel in the CPU space when they have loads of engineers that are better and more experienced than AMD's ?
The first order of business that I want AMD to do is increase the amount of stages in their integer and FP pipeline. That will essentially allow AMD processors to overclock very comfortably as this is important for increasing single threaded performance. The next thing AMD should do is improve the crap out of the branch predictor unit in the CPU so that they don't end up having alot of stalls in the pipelines. The other thing AMD has being lagging behind is their memory contollers. If they want fusion to come to fruition then they need to figure out a way to get quad channel memory configurations to their APUs. It wouldn't also hurt AMD to make the fetch and decode unit bigger as well as giving the whole processor more cache.
The important thing AMD should do is limit the amount of CPU cores to 4.
