AMD's Modula cores are 'not fast'
To put it into context Bulldozer was 10% per core per clock slower than Thuban (AMD's own predecessor)
PileDriver is a little better but still 5% slower per core / clock than Thuban, it also has 10% better power efficiency than BulllDozer
AMD's 8 core CPU's could be very good with just a bit more per core / clock performance.
IMO they don't have to be as fast or faster than Intel per core / clock, they just have to be better than what PileDriver is.
SteamRoller slides say:
If that is right, and its a big IF, then we could be looking at 20% ~ per core / clock performance improvement.
Clocked at 4Ghz with 4.2Ghz Boost out of the box we would be looking at -15% Single threaded Haswell performance (@ stock) and a monster in fully threaded apps.
Sticking 8 cores in models was not necessarily a bad idea, a normal 8 core with each core in its own module would have been a much larger chip.
Its a bit like Porsche with the engine in the back behind the back wheels, early models did not work very well with that engine hanging out of the back of the car swinging about like a pendulum.
But they stuck with it and perfected it, now Porsche are arguably the best handling car there is, certainly enthusiasts say you cannot beat a 911 for handling and shear excitement, its a concept which works beautifully where on paper it shouldn't work at all, some how Porsche defied the laws of physics and made it work.
AMD had a radical idea and it looks like they are sticking with it, time will tell if they can perfect it.
To put it into context Bulldozer was 10% per core per clock slower than Thuban (AMD's own predecessor)
PileDriver is a little better but still 5% slower per core / clock than Thuban, it also has 10% better power efficiency than BulllDozer
AMD's 8 core CPU's could be very good with just a bit more per core / clock performance.
IMO they don't have to be as fast or faster than Intel per core / clock, they just have to be better than what PileDriver is.
SteamRoller slides say:
If that is right, and its a big IF, then we could be looking at 20% ~ per core / clock performance improvement.
Clocked at 4Ghz with 4.2Ghz Boost out of the box we would be looking at -15% Single threaded Haswell performance (@ stock) and a monster in fully threaded apps.
Sticking 8 cores in models was not necessarily a bad idea, a normal 8 core with each core in its own module would have been a much larger chip.
Its a bit like Porsche with the engine in the back behind the back wheels, early models did not work very well with that engine hanging out of the back of the car swinging about like a pendulum.
But they stuck with it and perfected it, now Porsche are arguably the best handling car there is, certainly enthusiasts say you cannot beat a 911 for handling and shear excitement, its a concept which works beautifully where on paper it shouldn't work at all, some how Porsche defied the laws of physics and made it work.
AMD had a radical idea and it looks like they are sticking with it, time will tell if they can perfect it.
Last edited: