Intel Northwood Potential

[IInuyasha74]

Hello,

Anyone ever wonder if Intel might of actually had a reasonably good CPU with the Pentium 4 Northwood had they put all the advancements of Prescott into it but maintained the pipeline or possibly shrank it a little? Prescott had a lot of features to improve performance, but performed worse since it extended the pipeline to 31 stages. All those advancements just barely made it able to compete with Northwood. I can't help but wonder about the potential for a good CPU that could of existed had Intel added all the changes to Northwood, performed a die shrink, and made it dual-core.

 

[RaDiKaL_]

That sounds like Conroe 😛

 

[IInuyasha74]

A little yes and a little no.
http://en.wikipedia.org/wiki/Intel_Core_(microarchitecture)

Conroe is based on the Pentium M, which was based on the Pentium III and the Intel P6 micro-architecture. The Pentium III only had 10 pipelines and Pentium M, Conroe, Core 2 Duo, and even Nahelm all have 12-14 pipelinse. Also the architecture inside the ALUs, cache, FPU, pipeline itself, and several other parts are all drastically different from the Netburst P68 architecture. So it would be a little like Conroe, since they basically took the bus interface, RAM interface, instruction set advancements for Pentium D and worked it into Conroe which would be part of it, but everything past that is a completely different thing entirely.

Where as Prescott and Northwood have a lot more in common with each other.

 

[RaDiKaL_]

Hehe yeah I know that Pentium-M was practically the basis for Conroe, Intel eventually realized that the Northwood architecture was a no go and went back to the more efficient Pentium-M, thankfully.

 

[IInuyasha74]

Yea i think the Pentium-M architecture was best in the long run, it just seems like they could of done better with netburst. Prescott was like the worst CPU ever cause it had all these advancements and still couldn't outperform Northwood. Seems to think had they not extended the pipeline and kept with Northwood's 20 pipelines they might of been able to outperform AMD Athlon 64.

 

[RaDiKaL_]

Possible, though if I remember correctly they said that there was simply no path to follow other than extend the pipelines which lead to the inefficient Prescott, perhaps they couldn't see another possibility for Northwood or maybe it was really the apex of that architecture, what I wonder now is with the present situation, if AMD doesn't make it with it's next gen architecture Intel won't make any significant improvements either (we've seen that already from Sandy to Haswell, more efficient indeed but performance wise it's not really that promising if such tendency continues).

 

[IInuyasha74]

See that is whats weird cause that might of been the only potential path for increasing clock speeds but not performance. Had it worked might of been good, but since it was locked at the same clock speed because of heat and I don't think they saw that coming. Maybe it was too late to back track and meet the launch date they wanted but I think they would of had a better chance at the time had they.

Yea AMD is kinda locked into that at the moment but I think they have real potential to fix it. Hopefully they are able to, but I was actually thinking about that the other day too. Their trouble seems to stem from the Modules they use instead of just one core. I know that there is a lot of other stuff in there but it seems they could improve things right off by abandoning module design, concentrating the work resources down to a single core, and enabling SMT on the Integer side as well. A lot of the older designs like Athlon 64 for example had multiple ALUs which is one of the biggest parts of the Integer unit, so if they were to form it all into one core instead of partitioning it into a dual-core module, the performance would increase on single-threaded performance, and die size and power consumption would decrease as a result of removing unnecessary overlapping hardware which is under utilized as a result of the dual-core module design design (such as the pipelines in the 2nd core would be completely able to removed as the working resources could share the pipeline connection of a single core). Since they already have SMT technology in use in the FPU it wouldn't be too hard to implement it on a core wide level though might be time consuming. So I think they should be able to pull something out, I just hope they will actually do it instead of pushing on with this module nonsense.

You are right if they don't do something, Intel is going to be sitting pretty still. The GPU will keep improving and power consumption too, but performance will probably sit still until IBM, Via, ARM, Oracle, or Imagination manages to make their CPUs competitive in the desktop market again.