Intel roadmaps pointing to new Netburst era?

[amdnVuser]

First of all, let me point out that I'm by no means trying to flame Intel. I usually go with whatever camp gives the best performance per $ (but nowadays $ and watt, which, on the desktop front has been AMD for quite a while now). So please do not misread my intentions here...

I've recently been perusing Intel mobile roadmaps, and it is clear that Intel is going to rather quickly start ramping FSB on the PM (i.e. Merom). While it is certainly impressive that they're able to keep the (projected) TDP down so low (< 50W on the 800MHz FSB "performance" Merom's) while ramping up to 667/800MHz FSB, is this not indicative of their missteps in running out of FSB and frequency scaling headroom with the Netburst architecture? I'm by no means a chip architect, but it seems like unless they have some sort of really innovative engineering in the works, the situation is going to get worse with multi-core (> 2 cores) solutions (isn't the PM architecture eventually going to be the basis for their desktop/server products as well?).

I haven't seen much information on what dual core Turions are supposed to be like in terms of TDP, FSB, etc., but aren't they also moving to DDR2 and a new socket? I know on the desktop front, the move to DDR2 with socket M2 is really supposed to pay off when we start seeing 800MHz DDR2 become mainstream. However, if this (along with the rumored on-chip PCIe controller) happens on the mobile front as well in addition to FSB increases, won't AMD be heading down this same avenue as what are now considered mainstream/performance TDP levels eventually become ULV levels?

I know there are upcoming battery technologies that are supposed to greatly increase battery life, but if chip companies keep slapping more power consuming components into notebooks, common sense would dictate that we're eventually going to be no better off than we are today. I'm all for mobile performance, but if it means carrying around huge, heavy batteries and AC adapters, or not even making it through an entire DVD (and possibly half an HD DVD/ Blu Ray DVD in the future) when I'm away from a power outlet, then I think this is all for nothing.

Anyway, sorry for the rant. Opinions?

 

[stevty2889]

While mostly pentium-m based from the limited info out there, Merom isn't a really a next generation pentium-M. It's a differant architecture, and supposedly 4 IPC wide. Conroe is the desktop version, and Woodcrest the workstation/server version. Rumor is that Conroe will have a 1333mhz FSB. Can't really compare it to netburst, since netburst chips tend to be bandwidth starved, but thats not the case for the pentium-M, and shouldn't be the case for Merom/Conroe either. They are supposed to have a pipeline thats slightly longer than that of current Pentium-M chips, but no where near as long as netburst. And supposedly the next generation after that is supposed to have the point-to-point bus, like Hypertransport, so the FSB won't really be an issue anymore.

 

[amdnVuser]

Damn, forgot about the pipeline stages and Intel's upcoming version of HTT. Well, that sounds promising, I guess.

 

[BrownTown]

Intels version of HTT is still a sopule years off for desktops, so we will still be seeing a FSB bottleneck and higher memmory latentcies for the forseeable future. Giant caches on Intel chips help to allieviate this a little, but still it kinda sucks that the FSB is to slow to even keep up with dual channel memmory bandwidth.

 

[DarkKnight69]

You also need to take not that a large majority of the P-M's that the people are putting ont he desktops are able to run 200FSB on stock voltage. That said, 166 will be very easy to attain with very basic cooling!

Will not be too hard for the units to be moved to 800 mhz fsb!

 

[BrownTown]

the pentium M runs at a lower FSB becasue that uses less power. Intel could have chosen 800mhz, but chose not to becasue the benefits would be outwayed by the increased power consumption. For conroe power consumption is less of an issue since it is a desktop processor, so they are moving to 1066mhz FSB (1333 for XE and server processors based on current rumors)

 

[dmens]

Intel could have had an on-die memory controller right now if it wanted, but is holding off because it directly conflicts with the idea of using the same core for all markets. It would require massive changes to the controller for different power levels.

As for starvation, it won't be a problem with woodcrest since it's at 1333 using FB-DIMM, which is good enough for a couple more years until CSI is out.

 

[BrownTown]

Well, Intel does make good money off its chipset buisness, so it doesn't wanna kill that by integrating it into the chip. But not having an on die memmory controller really hurts performance. As we all know, most CPU clockcycles are spent waiting for data from main memmory or the cache. Teh onboard memmory controller greatly reduced the wait for data to load from main memmory. Intel can try to make up for it by making huge caches, but the CPU woudl be alot better off with a smaller and faster L2 chache, and lower latentcy memmory accesses.

 

[dmens]

It's not really that bad when you have smart prefetchers and a balanced cache system. Not that many ultralarge workloads have random access patterns.

As for memory controllers, it is a requirement only for high end systems, but again, if you design a controller that is modular, problems solved.

 

[BrownTown]

Memmory accesses waste so many clockcycles that even if they happen very rarely they can still have a signifigant impact on performance.