Originally Posted by aeassa
I'm not sure decreasing cache latency is something that's obvious to actually fix...
I think a lot of it has to do with the crappy node they're on, actually. And keep in mind they'll have had longer than we think to fix up PD, I have a feeling they had a separate team on it well before BD launched. Likely when they delayed it from June.
Even if PD isn't a 15% IPC increase, what if the node has improved enough to let it clock up to 6Ghz? It'd compete fairly well with SB at the very least there. The entire point of a uarch like BD/Netburst is to clock it higher than the competing chips.
Originally Posted by xd_1771
Maybe we won't get higher IPC. But if we get higher clock speeds and higher clock speed capabilities at lower power envelope, it works just the same. Instructions per second. If Pentium 4 didn't consume exorbitant amounts of power consumption and heat at the higher ~3.8Ghz clock speed levels and could manage that clock speed with the same power envelope as the competitor at the same price offering the same performance, it would have been the most successful CPU, ever. People keep mentioning IPC; that is not the whole point of the argument.
If AMD released Piledriver and it required 6Ghz to match the performance provided at 4Ghz on an Intel 22nm Ivy Bridge processor, but did so at lower power consumption and would be able to overclock to frequencies at which it could provide performance equivalent to or higher than Intel's processor at the same or lower power envelope, they would surely have the win here, or are you going to continue using the IPC argument?
I don't think that Prescott is the direction AMD is going here. New technologies like resonant clock mesh (pretty much AMD's equal to Intel's 3D transistors) can and will allow for further clock speed improvements and if not further IPC, we will get further IPS at the same or lower power envelope and this is totally acceptable too. I'm sure AMD knows what BD's problems are/were; I believe they also have the capacity to pinpoint them and fix them.
I wouldn't say IPC is dead, but it's not the only thing you look at. If SB didn't clock so high, AMD would be in a much better position.
Lets say 4.5Ghz was the absolute max you'd ever get on a 2600k/2500k on air? A 5Ghz BD looks a bit more attractive.
Originally Posted by Dmac73
IPC is far from dead because Intel has been able to keep up clockspeed wise and has a SIGNIFICANT IPC advantage. In the years to come when CPU's are into 7+ ghz daily, only then will IPC start to not mean as much. As long as Intel keeps clocking as high or higher daily than AMD, IPC is a big factor. The biggest, actually.
Everything else, i agree with you though xD. Especially your last paragraph. I also believe AMD know's every little problem with BD and will fix what they can.
God knows what Intel did with their 32nm process, but it's an extremely good process. If it only let chips clock up as far as 45nm? AMD would probably be in a mcuh better position now. To be fair though, Bulldozer as an architecture has more long term clock scalability than P6.
It is, actually. Apart from the Atom and any Netburst chips, all of Intels architectures since the Pentium Pro are part of the P6 architecture, just dramatically tweaked and improved.
Think low how a Athlon64 was a tweaked Athlon XP.
Even your link says it: "Sandy Bridge’s CPU architecture is evolutionary from a high level viewpoint but far more revolutionary in terms of the number of transistors that have been changed since Nehalem/Westmere."
Intel tweaked and evolved Nehalem and got Sandy Bridge out of it, they'll shrink Sandy Bridge and tweak it to fit in the new node and fix a few bugs, etc to get Ivy Bridge. Then they'll tweak Ivy Bridge to get Haswell, see where I'm going?
And Bulldozer, like all entirely
new x86 CPUs, starts off bad. SB is based off the Pentium Pro, like I said before..It sold like crap. it's tweaked version (The Pentium 2) sold much better, same story with the K5 vs K6, 386 vs 486, 286 vs 386, P4 vs P3, etc.
Originally Posted by Nintendo Maniac 64
About the whole IPC thing, I think what is the real thing is IPW, that is, I
The current thinking ever since Prescott is to assume that IPC = IPW, but if it's somehow possible to get, say 6GHz with a 95w TDP that "only" has the current IPC Bulldozer has, then that would be quite the paradigm shift.
Originally Posted by Homeles
Unfortunately, Ivy Bridge has apparently been running extremely hot compared to Sandy Bridge... so the overclocking potential might not be there.
Not to mention, it might not have as much voltage tolerance as SB does. Even if you're able to get it to 30c load, if you can't run your chip above a certain voltage without the chip degrading fast...yeah.
Keep in mind, that's an ES though. It may be better at retail. BD was a bit better when you looked at retail chips.