Quote:
Originally Posted by Brutuz 
First off, IPC has nothing to do with single-core speed, you can have a low-IPC but highly clocked fast single core and a high-IPC and lower clocked single core and depending on the techs used internally they'd likely be as fast as each other, IPC can be measured either per-thread or for as many cores as you want as well. (eg. AMDs IPC is much closer to Intels when you're comparing tasks that use all 8 cores, as an FX-8350 gets a lot closer to an i7 3770k at even the same clock speeds than it does in gaming)
Secondly, Intel certainly does which is why they're improving their IGP as much as they possibly can right now, but AMDs HSA implementation will be better afaik. (So even if Intel has a much faster iGPU in future, AMDs will be that much easier to code for it won't matter so much)
And AMD already has a number of partners for HSA, if the tools are as easy as marketed (I doubt it, but reckon they'll still be easier than current GPGPU implementations) then it'll take off on its own accord.
You must be blind to not see the leaked Haswell benches showing a 10% improvement over Ivy Bridge...Steamroller is going to be showing at least that much, if not more and will continue bringing AMD slowly closer to Intel.

First off, IPC has nothing to do with single-core speed, you can have a low-IPC but highly clocked fast single core and a high-IPC and lower clocked single core and depending on the techs used internally they'd likely be as fast as each other, IPC can be measured either per-thread or for as many cores as you want as well. (eg. AMDs IPC is much closer to Intels when you're comparing tasks that use all 8 cores, as an FX-8350 gets a lot closer to an i7 3770k at even the same clock speeds than it does in gaming)
Secondly, Intel certainly does which is why they're improving their IGP as much as they possibly can right now, but AMDs HSA implementation will be better afaik. (So even if Intel has a much faster iGPU in future, AMDs will be that much easier to code for it won't matter so much)
And AMD already has a number of partners for HSA, if the tools are as easy as marketed (I doubt it, but reckon they'll still be easier than current GPGPU implementations) then it'll take off on its own accord.
You must be blind to not see the leaked Haswell benches showing a 10% improvement over Ivy Bridge...Steamroller is going to be showing at least that much, if not more and will continue bringing AMD slowly closer to Intel.
That's why I said Instructions Per Second to account for low-IPC high clock and high-IPC low clock CPUs.
As for the multi-threaded tasks, I wouldn't be surprised if Intel respond by throwing an additional pair of cores at the desktop i7s and HT for the i5s and i3s. They got an IPS advantage per core, and they could keep scaling up the cores if necessary.
My suspicion of why Intel has little interest in GPU computing is because they know that GPUs are their weakness, but CPUs is obviously their strength. So it's reasonable to expect Intel to delay GPU computing, such as with their x86 based thousand core cards that they released more than a year ago.
I don't hate AMD, but Intel is far ahead in everything (finances, engineers, fabs, etc) but GPUs.









