Posted by: Anonymous Coward
on November 25, 2003 08:36 AM
ho-hum, so the overall conclusion so far is that you have FFT that's 10 times faster than<nobr> <wbr></nobr>... what?
you're comparing apples (sic!) to oranges here which makes you a bigger troll than the grandparent poster's author. SIMD processing unit can do things like FFT so much faster than the x86 offerings. Very wide architecture, 64bits, grabs 8 instructions per fetch cycle, for up to 215 instructions inflight. -- this is plain fanboy ranting. AltiVec and SSE2 are different beasts, they look/behave differently, but 'so much faster' is a plain lie no matter how you put it. All you can argue is that AltiVec is more clean/efficient, but ymmv. Keep on looking at Apple's benchmarks and forget that the gcc they used has almost no optimizations for pentium4 (see gcc.gnu.org for details, basically intel doesn't provide enough info for a complete description of the execution units). And don't even compare that to opterons (vanilla gcc barely started to incorporate k8 optimisations). That will make you happy. As to the proc. details, 'very wide architecture' -- ??? -- '64 bits' -- yay momma! -- are bulls***; 'many more registers', we're back to the ISA argument (the only valid point - Power's ISA is cleaner, easier to optimize for and has more on-cpu stores; but the difference is not that great); hypertransport is hardly unique, AMD introduced that first (and yes, it's for a 64bit cpu too).
finally, opteron uses SOI too - AMD's cpu research unit works with IBM and share manufacturing to some extent, go figure! (East Fishkill, rings any bells?) so how is power970 superior in that?
the rest of the post is indeed about growing up - but you had the person confused. Try looking in the mirror before throwing a fit next time over someone's preferences. he's going to stick with his cpu of chioce. you stick with yours (and accidently, it's Power970, not ppc970, as it's a power4 core inside). benchmarks are fine, as long as they are meaningful for what you will actually be using the machine for. I, for one, am yet to see a benchmark telling me I should have bought a mac for my number-crunching. while apple's offering is impressive, it's nowhere near compelling - and the price/performance ratio is also part of the deal.
Learn about Linux
Download Linux
Get Linux help
Re:alright super nerd!!!
Posted by: Anonymous Coward on November 25, 2003 08:36 AMyou're comparing apples (sic!) to oranges here which makes you a bigger troll than the grandparent poster's author. SIMD processing unit can do things like FFT so much faster than the x86 offerings. Very wide architecture, 64bits, grabs 8 instructions per fetch cycle, for up to 215 instructions inflight. -- this is plain fanboy ranting. AltiVec and SSE2 are different beasts, they look/behave differently, but 'so much faster' is a plain lie no matter how you put it. All you can argue is that AltiVec is more clean/efficient, but ymmv. Keep on looking at Apple's benchmarks and forget that the gcc they used has almost no optimizations for pentium4 (see gcc.gnu.org for details, basically intel doesn't provide enough info for a complete description of the execution units). And don't even compare that to opterons (vanilla gcc barely started to incorporate k8 optimisations). That will make you happy. As to the proc. details, 'very wide architecture' -- ??? -- '64 bits' -- yay momma! -- are bulls***; 'many more registers', we're back to the ISA argument (the only valid point - Power's ISA is cleaner, easier to optimize for and has more on-cpu stores; but the difference is not that great); hypertransport is hardly unique, AMD introduced that first (and yes, it's for a 64bit cpu too).
finally, opteron uses SOI too - AMD's cpu research unit works with IBM and share manufacturing to some extent, go figure! (East Fishkill, rings any bells?) so how is power970 superior in that?
the rest of the post is indeed about growing up - but you had the person confused. Try looking in the mirror before throwing a fit next time over someone's preferences. he's going to stick with his cpu of chioce. you stick with yours (and accidently, it's Power970, not ppc970, as it's a power4 core inside). benchmarks are fine, as long as they are meaningful for what you will actually be using the machine for. I, for one, am yet to see a benchmark telling me I should have bought a mac for my number-crunching. while apple's offering is impressive, it's nowhere near compelling - and the price/performance ratio is also part of the deal.
#