Nintendo's (IBM's) Broadway coming to ArsTechnica soon

[Gubbi]

well, the problem is that 750x's oooe does not need to be earthshakingly efficient to give an x4-clock in-order sibling a run for its money, to begin with : )

Huh?

Decent OOO (40-130 instruction scheduling window) normally gives about 30-50% performance gain for the same issue width, same clock and same architecture.

And as Faf said, the 750's OOOE implementation isn't decent. It's OOOE can only absorb pipeline execution latencies as well as D$ latencies and are not even enough to deal with the latency of the on die L2 cache.

Cheers
Gubbi

 

[V3]

It seems that the engineer of the DEC Alpha is making a PowerPC 970MP at 2Ghz that only has a power consumption of 13-15W.

http://www.eweek.com/article2/0,1895,1875302,00.asp

That article only said it will have lower power consumption compare to 970MP, they aren't making 970MP though.

P.A. Semi's first processor, a chip dubbed PA6T-1682M, will include two 2GHz processor cores, a pair of DDR2 memory controllers, 2MB of Level 2 cache, and an I/O subsystem consisting of eight PCI Express controllers, two 10-gigabit Ethernet controllers, and four gigabit Ethernet controllers, the company said.

 

[darkblu]

Decent OOO (40-130 instruction scheduling window) normally gives about 30-50% performance gain for the same issue width, same clock and same architecture.

"normally gives about 30-50%" does not concern the worst/best cases. as i already mentioned, to substitute an ooo for an in-order while meeting your timings you have to make sure the latter handles the most extreme ooo-favorable case adequaltely. now, if somebody provides a worst-case scenario of gekko code running on ppe and we examine that - that would give us some estimate. average performance differences dont help us, unfortunately.

And as Faf said, the 750's OOOE implementation isn't decent. It's OOOE can only absorb pipeline execution latencies as well as D$ latencies and are not even enough to deal with the latency of the on die L2 cache.

i take Faf's word that gekko's oooe is not decent - i'm not questioning it (that would be foolish - my experience in that direction doesn't get any futher than the 7450). i'm also sure your theoretical knowlege of gekko is way greater than mine, Gubbi . and yet, excuse me for repeating myself, it's not a matter of average performance difference, rather than of worst such. and for the worst case a 4x clock difference may or may not cut it - and that's what i'm curious about.