Sign in with
Sign up | Sign in
Your question

3.73Ghz P4 2MB L2 Cache in Q3: 1066Mhz FSB

Last response: in CPUs
Share
May 11, 2004 2:08:05 AM

Very strange news by <A HREF="http://www.anandtech.com/guides/showdoc.html?i=2048" target="_new">Anandtech</A>: Intel will be coming out with a 3.73Ghz P4 by Q3. It will feature a full 2MB L<b>2</b> cache, like Dothan, (not P4EE, that's 2MB L3), 1066Mhz FSB, and a 720 name. Highly puzzling: this CPU is the only 7xxish pentium that is not a Dothan core; it <i>should</i> be a Prescott.

Upping the FSB and the L2 cache does sound as if Intel's doing whatever is possible to increase performance...

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>
May 11, 2004 2:18:12 AM

Post deleted by juin
Related resources
Can't find your answer ? Ask !
May 11, 2004 3:03:56 AM

The real question is.....how fast is that cache? Size is nice and all but in the end, it's about speed.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
May 11, 2004 3:19:13 AM

Nice;but I don't trust any intel roadmap-they've got to show me the CPU...still waiting for tejas and jayhawk and have not forgotten they claimed to have a 64 bit cpu in the can ready for release whenever AMD released theirs. :mad: 
May 11, 2004 3:24:16 AM

Personnaly i still waiting for some answer from intel on prescott desigh.

Not only the core is slower but the cache systemes seen to have take a very large hit i wonder if the 64 bit path to the ALU have to do something with this.

i need to change useur name.
May 11, 2004 3:26:27 AM

Last news if reliable say to intel have postpone 64 bit to 2H 0f 2005 if a bit before nephalem.Anyway as long that intel can put on confusion on 64 bit support that good for them.

i need to change useur name.
May 11, 2004 8:12:31 AM

I have just read Anandtech's article. IMO the number system is a mess. At least that's the first idea that comes to my mind. People is going to get very confused, because the number doesn't indicate performance nor increases in performance between CPUs.

Exemple:

a) Pentium M 1.60GHz 90nm 400MHz 2MB 725
b) Pentium 4 3.73GHz 90nm 1066MHz 2MB 720

Can anyone explain me the relationship? And among all CPUs/numbers?

I understand some generic rules for 5xx roadmap ( -2 if FSB is 533 or +2 if it's slightly above rounded Mhz), but all in all only creates confusion.

More confusion = better for Intel?


Still looking for a <b>good online retailer</b> in Spain :frown:
May 11, 2004 8:33:52 AM

Model number between mobile and desktop part should not be directly compare.

i need to change useur name.
May 11, 2004 8:59:05 AM

>The real question is.....how fast is that cache? Size is
>nice and all but in the end, it's about speed.

If I trust intel on anything, it is designing fast, large and (space/power) efficient caches.

No, the real question is: is this a Prescott or Northwood based design ? If its prescott, its most likely gonna be HOT, if its northwood (on 90nm ?) its gonna be extremely telling :) 

Either way, it should be a fast chip, surely keeping them in the performance race if it isnt yet another paper launch.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
May 11, 2004 9:03:46 AM

> IMO the number system is a mess

Its worse than a mess. Numbers seem completely meaningless, they don't even offer a clue as to remember which processor is called what. There seems to be no logic whatsoever, besides "if its a different number, than something (or everything) is different". Big help. I begin to agree with those that claim that the only purpose of this numbering scheme is to confuse

= The views stated herein are my personal views, and not necessarily the views of my wife. =
May 11, 2004 9:03:46 AM

> IMO the number system is a mess

Its worse than a mess. Numbers seem completely meaningless, they don't even offer a clue as to remember which processor is called what. There seems to be no logic whatsoever, besides "if its a different number, than something (or everything) is different". Big help. I begin to agree with those that claim that the only purpose of this numbering scheme is to confuse

= The views stated herein are my personal views, and not necessarily the views of my wife. =
May 11, 2004 9:43:39 AM

That's right, but both are part of 7xx series. The P4 should be in another league SO it should start with 5. But that's only one exemple. Take also this two:

a) Pentium 4 4.0GHz 90nm 800MHz 1MB 580
b) Pentium 4 2.8GHz 90nm 800MHz 1MB 520

So the average Joe will say: "hey, number 5 only indicates desktop, so performance for (5)80s are FOUR TIMES (5)20s"

At least AMD XP+ rating gives you a more realistic relation (yes, sometimes is not accurate, as 2500+ for exemple), but this system does not give you any clue!


Still looking for a <b>good online retailer</b> in Spain :frown:
May 11, 2004 1:51:45 PM

Well, at least you can distinguish between them: let us not forget that the numbers are to be accompanied by processor name, so it's P<b>4</b> 720 and P<b>M</b> 725 for those two processors... At the very least...

As for the numbers being meaningless if compared to XP's ratings...

Erm... Opteron's numbers don't either.
FX's numbers don't make any sense.
nVidia's numbers don't either.
ATi's Radeons all have meaningless numbers.

It's like, "ooo, we need a nice name"... "how about X600?" "no, X800 is better" "No, wait!!! X800-Platinum!!! Make them think of something shiny and expensive!" "oooh, great, that's what we'll be doing"....

Completely at random.

And Intel's numbering system has just now been introduced, which suggests that the clearing up will still happen. Granted, it looks like an unbearable mess! ***holy crap!***

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>
May 11, 2004 8:44:43 PM

Nope, the numbers don't all make a whole lot of sense, but some are better than others. Opteron lets you know how multiprocessor it is and supposedly which is better in the series. I would also argue that unlike the pentiums or "PRed" A64s, these cpus aren't targeted at the average consumer. nVidia's numbers make sense to me at least. The first number lets you know if its the same generation and the latter numbers supposedly ranked performance within the generation. 4200<4400<4600; 5200<5600<5900 etc. It may lead to some people assuming that lower end new generation is better than high end old, but overall it works for people who bother to spend five minutes and pay attention. I think ATI's works about as well as nVidia's, though mixing in a roman numeral so they didn't end up with 10800 seems a little silly to me. Of course, when you start throwing "xt," ultra" etc into the mix graphics cards aren't so nice.
May 11, 2004 10:15:09 PM

>Erm... Opteron's numbers don't either.

They are perfectly logical. I don't think I need to expain them to you ? First digit number of cpu's supported, next two numbers relative performance, comparable across series (a 246 is just as fast as a 846), and higher is always faster. In fact, each 2 "points" each time correspond to a very similar performance jump (200 Mhz). I can see that logic.

>FX's numbers don't make any sense.

There is only 2 numbers to remember, which is quite a bit easier than intels 157 different product numbers :)  and even there you could argue the same logic as used as for opteron. Since there are no MP versions, the first digit is "blank", next two give indication of relative performance. Each 2 points is ~200 Mhz again.

Too bad they didnt use the Opteron scheme though, so FX51 would be called FX48. Then it would be completely logical.

By comparison the intel numbering doesnt tell you a damn thing. I can predict the next FX will be a FX55 and will most likely clock around 2.6GHz, the next opterons will be 152, 252, and 852, and will clock like a FX55. Can you predict the code of the next P4, P4EE or PM ? it could be *any* number afaict.

>nVidia's numbers don't either.
>ATi's Radeons all have meaningless numbers.

The numbers are still more or less logical, its the SE/XT/Pro/.. suffixes that totally mess it up.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
May 11, 2004 11:08:01 PM

I agree.
The best case is if model numbers just replace mhz (only 1 variable).
I think Opteron/FX have it best so far. Very simple to understand.

But when ever other variables change, the model numbers have to reflect that.
If they made a series for eache variable that would be too much and just add to the confusion. Adding A/B/C is a good way to denote there is a difference but doesn't say what it is. Once you know what level SE,Pro and Ultra represent that is a decent way to show the difference from the base model.

But I think Intel's system is confusing because we are going from using all specific system specs to only core type and model number advertised. Plus that one number stands for that model but hides the differences. Class,Mhz,Cache,FSB etc. That is not the best way to do it. At least Athlon's model numbers change based on other factors (performance dictates the model number).<P ID="edit"><FONT SIZE=-1><EM>Edited by piccoro on 05/11/04 07:09 PM.</EM></FONT></P>
May 12, 2004 12:59:02 AM

Intel's numbering system works fine for me. If I can afford it, it's a celeron.
May 12, 2004 1:06:29 AM

i sitll dont see why they couldnt have made two set of numbers, one for mobile and one for desktop chips, like the 600 series be mobile and 700 sereis be desktop or whatever, i mean arent the desktop chips suppose to be pushed to be better then the mobile ones?

I dont see how the average user will see the difference between p4720 and pm725, they will still think the pm is better lol. its still very ocnfusing, hope they straighten things out
a b à CPUs
May 12, 2004 5:18:17 AM

You could argue that Intel is trying to provide their numbering system based on price, that is they seriously want the Mobiles to sell for more money than the Desktop processors.

<font color=blue>Only a place as big as the internet could be home to a hero as big as Crashman!</font color=blue>
<font color=red>Only a place as big as the internet could be home to an ego as large as Crashman's!</font color=red>
May 12, 2004 7:06:38 AM

Quote:
If I trust intel on anything, it is designing fast, large and (space/power) efficient caches.


Not really. Historically, Intel has always had to deal with the issues of size vs speed. Look at the transition to 256KB of on-die cache vs 512KB of off-die. Or Netburst's small but extremely fast L1 cache. They made a mistake with Prescott (among many) and that was to sacrifice cache speed on an architecture that relied heavily on a strong and fast memory subsystem.

They should go back to Northwood and add some of Prescott's features (the good ones).

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
May 12, 2004 7:18:44 AM

if thats the case then im speechless lol
May 12, 2004 9:29:38 AM

It not even a size vs lantency issue with prescott.Never a increase in size result in a linear increase in latency.

8KB to 16 KB 2 cycle to 4 cycle.
512 KB to 1MB 12 cycle to 25 cycle.

Almost a perfect linear increase in size and lantency.I know longer stage result in more cache latency.Larger path to the ALU FPU may have also decrease is performance.

I think that a deep probleme in the circuit on all of those reason.

i need to change useur name.
May 12, 2004 3:35:49 PM

Yup The new numbering system is a royal pain in the butt. How is Joe Q public to know what the heck he is buying. :frown:
May 12, 2004 4:47:38 PM

yeah, cuase if its not based on performance, then its pretty useless to consumers since they want to know which chips are better.
May 12, 2004 5:38:34 PM

As an extra: The inquirer now reports a 3.6Ghz Prescott won't be around until 22nd August. This is, for the unsuspecting, Q3.

I think we should stop speculating when Intel will get what out. I get the impression they're on an ASAP basis anyway... So they'll get something out by the time they can get this something to work!

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>
May 13, 2004 12:59:24 AM

The Register is saying that will be a Dothan part. That would be some fast. Watch out Amd, Intel gona eat you.
May 13, 2004 1:46:32 AM

My god, a 3.73Ghz Dothan on 1066Mhz FSB? In 2004? No, I seriously doubt that that would be possible. That would be the holy grail of processors in 2004. It would be a royal kick in AMD's a$$. Hard to believe...

If this baby is Dothan-based, it is more likely a fully-featured dual-core dothan @ 1.86Ghz and higher and improved FSB. This might... <i>might</i> be remotely possible in 2004, <i>if</i> it were a chip <i>that had already been considered</i> and therefore wasn't so far from completion.

Still sounds an awful lot like crazy talk to me, though. I'd love to believe it. We'll see, I guess.

What does seem to be the case is that Prescott is truly having problems. And I agree, it sounds suspicious that Intel suddenly has a doubled L2 cache in the wings for prescott for Q3, with increased FSB, while the standard 3.6Ghz, which had been promised for Q2, is now slated for a 22nd August release.

Something just doesn't feel right. We're missing part of the story...

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>
May 13, 2004 2:20:26 AM

More fuel for the fire. To keep everyone quessing as to whats coming next.
May 13, 2004 2:54:12 AM

WOW!! if that is true, that would be a huge break through for intel. even the anandtech mention 3.6giz pentium m in intels road map.

If true this is a break though for intel.

If I glanced at a spilt box of tooth picks on the floor, could I tell you how many are in the pile. Not a chance, But then again I don't have to buy my underware at Kmart.
May 13, 2004 3:10:11 AM

Quote:
It not even a size vs lantency issue with prescott.Never a increase in size result in a linear increase in latency.

8KB to 16 KB 2 cycle to 4 cycle.
512 KB to 1MB 12 cycle to 25 cycle.

Almost a perfect linear increase in size and lantency.I know longer stage result in more cache latency.Larger path to the ALU FPU may have also decrease is performance.

I think that a deep probleme in the circuit on all of those reason.


Pipeline length would not affect the cache latency as it has nothing to do with instruction latency (from the point after you fetch it). I'm guessing Intel relaxed on the cache latency intentionally in order to make the processor scale higher (only to hit a brick wall with thermal problems).

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
May 13, 2004 11:06:26 AM

Quote:

They say single core, and 3.73Ghz.

They say <b>they think</b> it's single core Dothan. Quote from the article itself:
Quote:

The odd thing is that the regular 3.73MHz part not only contains 2MB of L2 cache but has been described as carrying the model number 720. Both factors suggest that the part is based on 'Dothan'

It's all speculation at this point and the article is just speculating as the discussion here.


BigMac

<A HREF="http://www.p3int.com/product_center_NWO_The_Story.asp" target="_new">New World Order</A>
May 13, 2004 1:24:27 PM

Pipeline length would not affect the cache latency as it has nothing to do with instruction latency (from the point after you fetch it). I'm guessing Intel relaxed on the cache latency intentionally in order to make the processor scale higher (only to hit a brick wall with thermal problems).


Delay in the back end and front end have to be about equal.Like you say if the cache pipeligne are too large the overall delay of the front end will be higher and as usual the slowest pipeligne will give the frequency.I thing that come to me was P4 was using speculation load for the ALU and not for the FPU.I wonder if prescott use standart fetching for the ALU and FPU.

Personnaly i think that a bad choice.

i need to change useur name.
May 13, 2004 2:11:41 PM

The inquirer posted an e-mail clarifying this a little bit.

Apparently, the Tejas development team had come up with several new features/layouts, including the 1066Mhz FSB and 2MB L2 cache. Now that Tejas has been definitely canned, this means that some of Tejas' features might show up in prescott much earlier, so they didn't waste money developing these new features.

In any case, Prescott will get a boost, which can only be good. Hopefully, it will mature into a respectable product <i>with 64-bit support.</i>

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>
!