Intel's 65nm Gameplan: Presler and Cedar Mill Updates
by Kristopher Kubicki on September 9, 2005 6:42 AM EST- Posted in
- CPUs
65nm and Dual Core for Everyone
Intel's WW36 and WW31 roadmaps came and went with relatively little fanfare. Even though we just had an extensive Fall IDF, there were several things mentioned in the roadmaps that were not at the show; like Intel's 975X chipset. We have some in depth articles concerning Broadwater, 975X and other finer points of the roadmap, but in the meantime it's important that we list the upcoming processor codenames and cores.
Intel's Extreme Edition series is nothing more than a series of low volume chips designed to showcase Intel's best of breed, but then again many argue the same for AMD's FX line. We don't expect any of the chips listed below to work their way into your family PC anytime soon, but they do give us a frame of reference for the desktop processor direction.
Intel Extreme Edition |
||||
| CPU | Core | Clock | FSB | Mass Production |
| EE 955 | Presler | 3.46GHz | 1066MHz | Q1'06 |
| EE 840 | Smithfield | 3.2GHz | 800MHz | Now |
| EE 3.73 | Prescott | 3.73GHz | 1066MHz | Now |
The Intel 955 Extreme Edition will be the first 65nm dual core Presler introduced. Presler, as you may recall, differs slightly from Smithfield in the fact that each individual Cedar Mill core comes from a separate die on the processor packaging. This is partially due to the fact that the smaller 65nm process changes the optimal configuration of the chip layout, but mostly due to the increased yield -- it's easier to throw out one core at a time than to scrap both cores even if only one fails QA. The new Presler Extreme Edition will feature two banks of 2MB L2 cache, and will feature Intel's VT virtualization technology. As with the 840EE, HyperThreading will also come enabled on the 955EE.
The desktop dual core roadmap is significantly more relevant and interesting. Intel and AMD both predict the mainstream desktop to fully embrace dual core, so the majority of both companies' roadmaps are targeted at that sector.
Intel Dual Core Desktop |
||||
| CPU | Core | Clock | FSB | L2 Cache |
| ??? | Conroe | ??? | ??? | 4MB |
| ??? | Conroe | ??? | ??? | 2MB |
| 950 | Presler | 3.4GHz | 800MHz | 2x2MB |
| 940 | Presler | 3.2GHz | 800MHz | 2x2MB |
| 930 | Presler | 3.0GHz | 800MHz | 2x2MB |
| 920 | Presler | 2.8GHz | 800MHz | 2x2MB |
We've already talked a little bit about Presler in previous roadmaps and with the 955EE. All of the 65nm Presler processors listed are scheduled for a Q1'06 launch and are virtually identical to the existing Intel Pentium D 8xx series with a few new features:
- 65nm die process with two separate dies, but a single packaging still
- 2x2MB L2 cache
- Virtualization Technology (VT)
As with Smithfield, HT will not be enabled on non-Extreme parts. Barring a miracle, the new Presler chips would need to have some new feature not listed on the roadmap to really make them outshine the existing Smithfield and Athlon 64 X2 chips. The smaller process might allow them to run slightly cooler, and we might even see 4.0GHz chips at some point, but that's about the only changes.
It's clear the NetBurst architecture is nearing the end of the road, but we find it odd that the Conroe unveiling is less than a year away from Intel's 65nm NetBurst launch! The current roadmap claims an H2'06 launch for Conroe. Conroe features a completely new architecture for Intel, with new features like a 4-issue core and a shared L2 cache between cores. Although the roadmaps have virtually no information on clock speeds, we do know that there will be two versions of the chip, one with a 4MB shared L2 cache and another with a 2MB shared L2 cache. The chips will also feature EIST, VT and EM64T.
Intel still thinks there is life in Prescott, and even after the die shrink to 65nm (Cedar Mill), the Intel roadmap claims new single core Prescott-2M versions will be made available with VT. Perhaps the only interesting detail below is the fact that Intel canceled the VT versions of Cedar Mill. Why Intel will even bother to pursue VT on the 90nm core and not the 65nm core is a total mystery. Some of the product names have also changed a bit since the previous roadmap.
Intel Desktop Performance Roadmap |
||||
| Processor | Core Name | Clock Speed | Socket | Launch Date |
| Pentium 672 | Prescott 2M + VT | 3.8 2MB | LGA 775 | Q4'05 |
| Pentium 671 | Cedar Mill | 3.8 2MB | LGA 775 | 2H'06 |
| Pentium 662 | Prescott 2M + VT | 3.6 2MB | LGA 775 | Q1'06 |
| Pentium 661 | Cedar Mill | 3.6 2MB | LGA 775 | Q1'06 |
| Pentium 651 | Cedar Mill | 3.4 2MB | LGA 775 | Q1'06 |
| Pentium 641 | Cedar Mill | 3.2 2MB | LGA 775 | Q1'06 |
| Pentium 631 | Cedar Mill | 3.0 2MB | LGA 775 | Q2'06 |
| Pentium 670 | Prescott 2M | 3.8 2MB | LGA 775 | Now |
Virtually everything else on the CPU roadmap remains unchanged since our last revision. Intel has been good to ship processors and chipsets on its set launch dates over the last year, but the 65nm chip launch looks to be one of Intel's shortest product cycles yet with Conroe just around the corner in the same year. In all likelihood, Intel is probably just testing their 65nm yields with the initial Cedar Mill and Presler chips before moving onto bigger and better projects like Conroe. Intel's launch of Broadwater (i965X) also comes about the same time as the new CPU launch; and nothing sells chipsets quite like new CPUs. We have more details coming up on Intel's new 965X, 975X, Viiv and Yonah very soon, so stay tuned!
Facebook
Twitter
RSS
27 Comments
View All Comments
Shintai - Friday, September 9, 2005 - link
VT is basicly virtualization of the hardware. Think of Virtual PC or VMware at a hardware level. 2 Windows XP or Windows XP and Linux running on the same machine without one being dependant of the other. Dad is doing some work on the same PC the son plays on.Both AMD and Intel won´t ramp up speeds with 65nm. They both look for cheaper dualcores, lower power and to some extent multicores. What people will have to learn is ramping of speed in terms of Ghz ain´t really that easy. Laws of physics apply here, so even tho you can sell a 6GHz P4 or a 4Ghz AMD64 it doesn´t help if it stops working or random crashes after 2 years due to degraded materials inside the core.
And what would such a CPU use of power! 150W? Add another 150 or even 300W for GFX if you got SLI/Crossfire. Then we need a PhysX chip etc. Can you say central heatingsystem for your home :x
But again..I´m a fan of low power and ecological aware production ;)
Windows Vista is currently scheduled for december 7 2006. And thats RTM only. However this can easily slip 3-6months, aswell as adding 1-2 months before its avalible in retail and OEM. But remember Windows Vista ain´t 64bit only versions. There is 32bit versions aswell. Servers is also avalible in both flavours, except for the small business server.
But again..corporations don´t care much. Alot will still use windows 2000 when Vista comes. And only about 10-20% will even move to Vista in 2007.
Conroe and Meron is scheduled for Q3 2006. Woodcrest is scheduled for Q4. AMD should also have 4core Opterons out in Q4. And maybe X4? too :D
Furen - Friday, September 9, 2005 - link
I wouldnt say VT is virtualization on hardware, per se, but rather it provides tools for virtualization software to work with less of a CPU overhead. The main benefit is, of course, ring -1 execution. This is because the virtualization software will not have to intercept (and process) every OS instruction to keep it from messing up the system, but rather will have another layer from which it can create a ring 0 of sorts for each OS.Shintai - Friday, September 9, 2005 - link
I can forecast alot of exploits on all platforms, VT will be a mess like Wifi was/is.Average Joe doesn´t use VT, but mr. Evil viruswriter does and now Mr. Average Joe got 2 OS running. Tho he only knows that 1 is running.
I wonder what the next step is..firewall and antivirus wont fix it.
Furen - Friday, September 9, 2005 - link
I was wondering what the minimum multiplier will be on the 65nm netburst chips. It seems Intel did away with the 14x multiplier limitation on the Prescott so I'd expect EIST to be much more useful this time around. Sucks that Intel (and AMD, to some extent) doesnt plan on ramping clockspeeds at all (Yonah will be released at 2.16GHz max)... talk about reversals.Shintai - Friday, September 9, 2005 - link
Yonah will ramp to 2.33Ghz tho. But remember it´s a mobile chip.You basicly get dualcores and alittle more for the same power you had single core with Pentium M. Sure a 3.5Ghz Yonah would WTFPWN, but it would do that to the battery aswell.
Furen - Friday, September 9, 2005 - link
Yeah, what I meant is that they are completely neglecting the single-core performance. Even these past two years have brought a steady (albeit slow) increase in clock speed. The next two years look rather gloomy in that regard. The Pentium 4 scaled back to 3.2 when it went dual-core and it, most likely, wont go over 3.46 before being killed off. Yonah will basically give us one speed bump before being killed off as well. And who knows how well Conroe performs in single-thread apps and at what clocks it comes out. AMD will probably clock its K9 (dual-core k8) up to 3GHz if its 65nm shrink works out ok, but I dont think we can expect this speed grade to happen any time soon.Furen - Friday, September 9, 2005 - link
Bah, I meant single-thread performance, not single core.Shintai - Friday, September 9, 2005 - link
We do know, since intel have shown working conroe that it will be about 25% faster than Pentium-M at the same clock. it will also contain a 16 stage pipeline, Dothan got 14 or 15, Prescoot got 31.So if we mix it up with intels new lowpower tactics. I say something like a 2.6-3.0Ghz Conroe (desktop) that performs like a 4.5Ghz P4 while using a max of 45W. So Conroe/Meron/Woodcrest should be something equal to a 3-3.2Ghz AMD64 that should be easy to reach in Q3/06
But it´s very clear..it´s more cores before speedsteps right now and the next 1-2 years and maybe more.
IntelUser2000 - Friday, September 9, 2005 - link
Umm did you see the article from Anandtech about IDF AT ALL?!? Merom/Conroe/Woodcrest is 14 stage pipeline, for Pentium M, in reality we DON'T KNOW. Some people say 12, some say 13, some say up to 16-17. Some say P6 cores are 12 stage pipeline, but according to Anandtech's Pentium 4 Willamette article, there is a pic from Intel that says P6 is 10 stages.
2.6-3.0GHz Conroe that performs like 4.5GHz P4?? Pentium M already performs like 4.5Ghz P4 at 3.0GHz.
Furen - Friday, September 9, 2005 - link
Conroe could theoretically be 25% faster than the Pentium M but good luck keeping the 4 issue execution core fed. AMD had to integrate the memory controller in order to keep its 3 issue cpu fed... not to mention that the instruction level parallelism of x86 code leave a bit to be desired. I dont doubt that Intel's cpu could be fantastic but they have been downplaying single-thread performance a bit too much for my taste, which makes me wonder whether or not they can actually deliver the 25% improvement they claim. Just adding cores is not necessarily going to lead to performance benefits since, even when multithreaded applications go mainstream, not all operations are inherently parallelable (most sounding word ever, huh?).