backtop


Print 76 comment(s) - last by clnee55.. on May 30 at 2:24 PM

Intel promises significant price cuts across the board in preperation for dual-core ramp

Previously, DailyTech revealed a number of changes that will be taking place with Intel's family of desktop processors: Pentium 4 processors will drop down to roughly 30% of Intel's overall desktop processor shipments and Pentium D processors falling to 45% of desktop sales.

Core 2 Duo processors will be introduced on July 23 of this year and, according to Intel's estimates, and will be accompanied by the new 96x "Broadwater" chipset. According to Intel's updated roadmap, Core 2 processors will expand to roughly 35% of Intel's shipments in Q1'07. Conroe, as the desktop version of Core 2 Duo is dubbed, will be the first Intel processor with the new letter/number naming schedule, as noted below.

Intel's flagship, the Core 2 Extreme processor, is also expected to launch on July 23 with the X6800 name. 

Intel Desktop Processor Roadmap for Dual Core
Processor
Brand
Processor
Number

Clock Speed
FSB
Cache Launch
Date
Price @
Launch
C2E
X6800
2.93GHz / 1066MHz 4MB 23-Jul $999
PPXE 955
3.46GHz / 1066MHz 2z2MB Now $999
PPXE 840
3.20GHz / 1066MHz 2x1MB Now $999
C2D
E6700
2.67GHz / 1066MHz 4MB 23-Jul $530
C2D E6600
2.40GHz / 1066MHz 4MB 23-Jul $316
C2D E6400
2.13GHz / 1066MHz 2MB 23-Jul $224
C2D E6300
1.86GHz / 1066MHz 2MB 23-Jul $183
P4D
960
3.6GHz / 800MHz 2x2MB Now
$530
P4D 950
3.4GHz / 800MHz 2x2MB Now $224 (23-Jul)
P4D 940
3.2GHz / 800MHz 2x2MB Now $183 (23-Jul)
P4D
930
3.0GHz / 800MHz 2x2MB Now
$178 (4-Jun)
P4D 925 (no VT)

3.0GHz / 800MHz 2x2MB Q4'06 $133
P4D 920
2.8MHz / 800MHz 2x2MB Now $178 (4-Jun)

Intel has dropped desktop processor prices throughout the year. Many of the Pentium D processors have seen anywhere from $30 cuts to more than 50% price cuts. All entry level Celeron 300-series desktop processors from Intel will  be priced well under $80 by the time Q4'06 comes around.

Intel Desktop Processor Roadmap for Dual Core Without VT
Processor
Brand
Processor
Number

Clock Speed
FSB
Cache   
Price
   Now   
 Price @ Launch
PDP
945
3.4GHz / 800MHz 2x2MB N/A $163 (23-Jul)
PDP 925
3.0GHz / 800MHz 2z2MB N/A $133 (Q4'06)
PDP 915
2.8GHz / 800MHz 2x2MB N/A $133 (23-Jul)

Intel will also be launching VT-disabled Preslers laster on this year with the 945, 925 and 915 series. VT, or Virtualization Technology allows virtual operating systems running on VMs such as VMWare or Parallels Desktop to access processor ops directly. The mainstream Pentium 4 processors with 2MB and 1MB caches will also see significant price drops. For example, a Pentium 4 661 (LGA775) processors running at 3.6GHz with 2MB of L2 cache will be priced at $183 on the 23rd of July. Right now, the same processor sells for $401.

Intel Desktop Processor Roadmap for Single Core
Processor
Brand
Processor
Number

Clock Speed
FSB
Cache    Price
   Now
   Price on
  23-Jul
P4P
661
3.6GHz / 800MHz 2MB $401 $183
P4P 651
3.4GHz / 800MHz 2MB $273 $163
P4P 641
3.2GHz / 800MHz 2MB $218 $163
P4P 631
3.0GHz / 800MHz 2MB $178 $163
P4P 541
3.2GHz / 800MHz 1MB $218 $84
P4P 531
3.0GHz / 800MHz 1MB $178 $74
P4P 524
3.06GHz / 533MHz 1MB $143 $69

All of Intel's single core processors see significant price drops across the board. According to Intel's roadmaps, dual core ramp will accelerate after July 23rd. Core 2 Duo will become Intel's flagship processor while Woodcrest takes the helm for enterprise and server level performance in 2007.

Intel's roadmap also confirms the existence of Kentsfield, the first quad core desktop component for Intel.  Kentsfield will launch as an "Extreme" processor in Q1'07, but Intel isn't saying the official name yet.  AMD is expected to introduce quad-core processors in 2007 as well. Called the K8L, AMD is expected to introduce new HyperTransport protocols as well as support for third party co-processors.


Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: 940 vs E6300
By slashbinslashbash on 5/28/2006 1:46:30 AM , Rating: 4
Dude, what's your deal? Do you forget who we're talking about here? INTEL. If it turns out that they've been lying or cheating somehow with the Conroe preview chips, the backlash will come hard and fast. Newspaper headlines will read "Intel stock drops 23% upon disappointing performance of latest processor" and hardware sites will have headlines like "Intel full of bullsh*t with Conroe."

I don't see what Intel could possibly be doing to cheat anyway. If Intel gives somebody a preview chip to benchmark, and that chip performs very well, then that's proof that Intel has the capability to produce chips capable of such performance. Plain and simple: "We can do it and we are doing it now." If a chip *exists* with that kind of performance, it's hard to imagine what they could do to actually slow themselves down between now and launch time. Ok, so maybe yields are terrible; it often happens, and prices rise on the end-user market. But you can STILL buy a chip that fast if you want to pay $1000 for it, even if they're super rare. And it's a guarantee that the bugs will be worked out fairly quickly and yields will improve.

What I *really* don't understand is what you think they could possibly be doing to these chips in order to cheat. So it's unbinned. So what? It's still a chip of Conroe architecture running at X GHz, and therefore, its performance is representative of the performance of future chips of Conroe architecture running at X GHz. I'm wondering if you fully understand the process of lithography and processor design and manufacture.

It can ONLY get better from there. So say that some super-genius Intel engineer says "Hey guys! Here's something we can do to speed up the pre-production samples of Conroe!" And they put this improvement into the pre-production samples. What possible reason could they have for not carrying over the same improvement into the production versions? We saw how quickly improvements could be made with the recent AM2 processors. In a matter of a couple of months, AMD brought performance from "OMG WTF?" to "Ok, that's what we expected." Going the other direction is, I assert, impossible. Manufacturing processes (masks, layouts, etc.) can only get better over time. It's a monotonically increasing function. It's called "learning." Companies don't get stupider, not when it comes to things like this. And given recent market history with binning (both Intel and AMD), it is likely that virtually all processors of the same architecture and revision will be able to overclock to high-end levels or even higher.

Just about the ONLY thing that I can think of that would be practical to "cheat" with these pre-production processors is to use some specially doped silicon that's too expensive to use in regular production. But the ONLY effect that would have is to lower the binning speeds of production processors; in other words, they might not be able to hit X GHz with production processors, while they provided an X GHz pre-production sample. But as long as the benchmarks line up where a hypothetical production processor at X GHz (or maybe a production processor overclocked to X GHz) hits the same benchmark numbers, it's all cool. Of course it is HIGHLY unlikely that Intel would provide a pre-production sample clocked any higher than at least the high-end production CPU, and they may even aim it more towards the middle of the line.

So, I ask: What else could they possibly be doing? Sneaking more cache in there? Even the most basic benchmark programs would give that away immediately. Crank up the clockspeed? I think I've already covered that. Uh... hmm. Maybe speed up the cache, independently of the rest of the CPU? Ok, but again, why would they do it on pre-production samples but not on production units? Your whole argument seems to reduce to "THESE ARE PRE-PRODUCTION!" as if that stands on its own.

It's not as if this is a car which has thousands of independent parts which can be tweaked and replaced individually, so an inscrupulous manufacturer could provide a pre-production sample with a hot-rodded engine. Processor cores are monolithic and created "all at once." You can't swap out the engine in a processor... nor the tires, nor the suspension, nor the exhaust. You can't even polish the intake manifold for an extra 3 horsepower. It's an all or nothing deal. (Yes, I know that CPU manufacturers can perform "surgery" to repair individual parts of individual chips during pre-production, but that happens very early in the design and layout process, and I doubt that such "healed" chips could even be put into functioning systems with real power supplies and heatsinks and RAM.)


RE: 940 vs E6300
By Viditor on 5/28/2006 2:21:54 AM , Rating: 2
quote:
If it turns out that they've been lying or cheating somehow with the Conroe preview chips, the backlash will come hard and fast

And what if they just missed something (like they did with the Rambus based mobos)?
quote:
If Intel gives somebody a preview chip to benchmark, and that chip performs very well, then that's proof that Intel has the capability to produce chips capable of such performance

While they can produce them, it doesn't mean they can produce them in volume...the 2 processes are not the same.
quote:
So say that some super-genius Intel engineer says "Hey guys! Here's something we can do to speed up the pre-production samples of Conroe!" And they put this improvement into the pre-production samples. What possible reason could they have for not carrying over the same improvement into the production versions?

If it turns out that the process can't be made with an acceptable yield, then the cost would be prohibitive...
quote:
Processor cores are monolithic and created "all at once." You can't swap out the engine in a processor... nor the tires, nor the suspension, nor the exhaust. You can't even polish the intake manifold for an extra 3 horsepower

Actually it depends...there are any number of tweaks to a process that can be made. Let's say (hypothetically) that you discover that with doping level X, you can get chips that perform far more efficiently, but only 1 in 100 works properly.
Or that the new PMOS cap works fantastic, but it severely reduces yields...

I'm not saying that this is what they are doing (far from it), but it's certainly at least a small possibility.

As to production, both AMD and Intel change their production regularly...
AMD has far more granularity in these changes due to APM (they can change the doping level on any individual chip of a wafer at any point of the process), but Intel will also copy tweaks across as they find them (which is why some batches are better overclockers than others).


RE: 940 vs E6300
By slashbinslashbash on 5/28/2006 3:46:56 AM , Rating: 2
All of the problems you bring up (except for Rambus) were covered in my intial reply. It all has to do with yields. Again, if yields aren't there then they will simply be selling lower clocked processors at launch. And, to make a circular statement, we would expect these lower clocked processors to perform as expected (i.e. in a clock-for-clock proportional manner with the pre-production samples).

The only thing that is really going to change performance per clock is a significant change in architecture. The best way to change clock speed is through process updates -- process shrinks, doping, low-k, whatever. In other words, clock speed is related (inversely) to yield. Of course, performance per clock * clock speed = performance.

I am not debating that Intel may very well be doing something to increase clock speed on these pre-production samples. These samples may very well be fabbed on very low-yield processes that are cost prohibitive for mass production. However, barring significant changes to the architecture, the clock per clock performance will not change between pre-production samples and production units.

Anand previewed a Conroe at 2.66GHz. We now know that Intel is planning to release the 2.66GHz Conroe on July 23 at a price of $530. Maybe they won't be able to! Maybe they'll fail! Maybe at launch, yields will still not be high enough and they will only be able to come out with (say) 2.2GHz Conroe CPUs. IMO this is not a problem as far as Intel's credibility is concerned unless the 2.2GHz CPUs perform disproportionally slower than their clock speed would suggest, based on the samples that we've seen benchmarked so far. And there is no way for that to happen without significant changes to the architecture between the production units and the samples.

Lastly, your Rambus thing is a red herring. If there were some sort of flaw like in the Rambus motherboards, then that would not affect "performance" per se. That kind of problem affects basic functioning and would be covered by Intel in a recall, like they did with the Rambus boards and like they did with the original Pentiums before that. That's the kind of problem where most people would see the performance that we're expecting based on pre-production samples, but a small percentage would see big problems or even outright failures. That kind of flaw is obviously a very different situation than 100% of production CPU's performing slower than pre-production CPU's in some set of benchmarks. Of course, the pre-production CPU's haven't even been subjected to third-party reviewers for long-term stability etc. yet, so really that's an unrelated discussion anyway. We're talking about "how will the production CPU's benchmark compared to the sample CPU's that we've already seen benchmarked", not "will there be some flaw in the production CPU's that requires a recall."


RE: 940 vs E6300
By Viditor on 5/28/2006 4:51:20 AM , Rating: 2
Some very good points...I would agree that pure performance as a function of clockspeed for the CPU would remain a constant of the architecture.
However, what about the motherboard performance?
And more importantly, what about leakage? It's usually not a function of architecture and could also be vastly different in volume production...

Finally, I guess what I was trying to say (not very well) is that often it takes an architectual change to correct for a manufacturing difficulty (e.g. Prescott)...
If that does happen (doubtful but possible), then the old benchmarks get thrown out and new ones must be created.


“And I don't know why [Apple is] acting like it’s superior. I don't even get it. What are they trying to say?” -- Bill Gates on the Mac ads














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki