New details on Prescott 0.09m P4

<A HREF="http://www.xbitlabs.com/news/story.html?id=1019274490" target="_new">http://www.xbitlabs.com/news/story.html?id=1019274490</A>

1MB Cache? Possible Yamhill technology using "similar" 64-bit x86 but NOT x86-64? Sounds odd to me...

Comments...

--
Thunderbirds in wintertime, Northwoods in summertime!

Maybe w/ more L1 which the P4 has little to none might help of benchmarks. Uping that extremely fast Trace cache might be a key. 1mb L2 cache is always welcomed in my pc

You GeForce Ti4xxx is faster then my R8500 but my R8500 is the king of Aniso baby

BTW Meltdown is racist to Canadians.

--
Thunderbirds in wintertime, Northwoods in summertime!

lol

If you can find it on the internet, it HAS to be true!

That's because Intel has a killer processor right now. The Northwood is an excellent extreme overclocker that provides excellent performance with the 512KB.

<b><i>"A penny saved is a penny earned!"</i></b>

Come on, IPC doesn't matter as long as there is MHz to compensate.

<b><i>"A penny saved is a penny earned!"</i></b>

I have read the source link chiptech
Nice by the way lot on information

This not not a inside info.
This look like they look at the patent and early prototype.So that what intel can put under the hood of presscott.If that all true Prescott will as big as Willimete


16 K L1 data
More X-trance-cache
New HT instruction
twin pump ALU twin pump FPU twin registery twin pump L1
body bias forwarding transitor if i remember well.
bigger faster L2
Not sure but also on moving ALU to 32 bit wide.

cheap, cheap. Think cheap, and you'll always be cheap.AMD version of semi conducteur industrie

http://www.chip-architect.com/

cheap, cheap. Think cheap, and you'll always be cheap.AMD version of semi conducteur industrie

The problem is, the Northwood has the lead right now, so that justfies my "excellent performance" claim. However, the 2.4GHz P4 costs a fortune, but you can always get a 1.6A or 1.8A P4 an overclock it beyond the 2.4GHz P4 in performance. That's where I get the excellent performance from.

What annoys me is that there seem to be only two distinct groups here:

* those who bash AMD, and defend Intel
* those who defend AMD (the only person I've ever seen bash Intel is Willamette_sucks)

The problem is, as much as I think the Northwood is a great processor, at times I find that Fugger or AMDMeltdown's claims about AMD inferiority unjustified. This summer, I'm getting a new processor, whether it's AMD or Intel, it doesn't matter. I'm just going to the get the best available at the time. I'm getting tired of defending AMD all the time and being called a troll by Fugger. Does being neutral (in terms of preference), deny me the right to debate in favour of Intel or AMD? Just because I see something in a certain point of view, doesn't me I've lost my stance.

Ok, I'm blabbering, that's enough.

<b><i>"A penny saved is a penny earned!"</i></b>

Hello Melty, See you after work then?

Can you pick me up as usual?
Ta!

I am the MAN, the Man_Man!

That was a joke, right?

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft

Good point, I wasn't thinking about it from a power consumption point of view.
But don't forget the P4 has SSE2, better hardware data prefetch, thermal protection, etc.
Those are the things that make me say it's a more advanced core. I'm not talking about speed ramping at all, that has nothing to do with being "advanced" IMO (conversely, neither does IPC).

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft

To an engineer... is rather cheap engineering? Eden, it seems that you are purposely making bad comments about Intel. First off, the Northwood is SMALLER than the XP(although it's made using 0.13), and it's not the worst heat consuimng thing ever. You must have mistaken that for the Athlon XP; Northwood will stay cool as long as you don't overclock it too much. As Fatburger said, the Northwood has alot of advanced stuff which Athlon doesn't have (SSE2, hardware prefetch, etc). And don't forget, right now, 30% of the northwood core is not being used (for whatever reasons). I guess Intel "reserved" that space for Hyperthreading when they first released it. Intel did stated a few months ago that Hyperthreading will use existing P4 circuitry to work, which I assume is that 30% of "unused" circuitry.

well, interesting link.






 


<pre>version of semi conductor industry</pre><p>

so you think Intel is pulling an nvidia and one day all of a sudden peoples pentium 4s will perform considerably better? with just a new piece of hardware or software (bios upgrade)?

If my response is brief and vague its because the info you provided is too!

yeah - but dont forget the Athlon has the 9 issue execution core and the 3 way instruction decoder (who can decode 3 instruction at a time). it also has the best FPU unit which is super-scalr due to its abilty to "understand" RISC instruction (the decoder translte x87 into a RISC form instruction).

there are obvious advanced basis to build on in the Athlon core (with the addition of SSE2). thats why the hammer will use the same execution core as the athlons only improveing the front end of the processor. where, clearly, the P4 holds its strong points (Data prefetch, L2 cash. 400Mhz FSB etc.).

can you really say which processor is "better"? I think thats rather tricky - the Athlon XP preformed better then P4 on 180 nanometter process.
but with such a slim margin in preformance minor architecure changes (NW) or process changes or just minor core teaweking can change the leader.

This post is best viewed with common sense enabled

Missed this post the first time around.

<A HREF="http://www.tech-report.com/reviews/2002q1/northwood-vs-2000/index.x?pg=1" target="_new">http://www.tech-report.com/reviews/2002q1/northwood-vs-2000/index.x?pg=1</A>
"This so-called die shrink does several things for the Pentium 4. Northwood is smaller, runs cooler, and requires less power than Willamette. The Pentium 4's die size shrinks from 217 square millimeters to 145 square millimeters. Because Intel can fit more chips on a wafer, Northwood should be cheaper to manufacture. The process shrink should also enable Northwood to run at even higher clock frequencies with ease. "


"The Athlon XP hasn't yet undergone the die shrink to 0.13 microns. Like Northwood and unlike Willamette, however, that Athlon XP is made with copper interconnects, which AMD has been using on Athlon chips for quite some time now. AMD has plans to take the Athlon line to 0.13 microns this quarter; that chip is code-named Thoroughbred. However, even without the die shrink, the Athlon XP is only 128 square millimeters. Because Athlon XPs are made up of only 37.5 million transistors, they're much smaller than the Pentium 4—even smaller than die-shrunk Northwood. All other things being equal, Athlon XPs ought to be cheaper to make, as well. "


P4 northwood 145mm2
Axp 128mm2

The northwood EVEN ON .13 is larger than the .18 axp, imagine the tbred.


Strike 1 for dark archonis.

On a similar technology the axp usually is cooler than the northwood, so while this is objective(so I wont give it a strike) you are being misleading.

Fatburger is well respected, dont use him to promote incorrect information, while the axp does not have sse2, it DOES HAVE HARDWARE PREFETCH.

Strike 2 for you dark.

The current p4 implimentation of HT is flawwed, and sometimes results in a loss of performance, they will correct this flaw in prescott(so they claim) thus the 30% of the p4 die is in fact WASTED, never to be recovered.

Strike 3, your out.


(for future reference please dont attack forum members with incorrect information, its one thing to think eden is being hard on intel, its another thing to make [-peep-] up to make him look wrong).


Have a nice day :-)





The Cash Left In My Pocket,The BEST Benchmark

the current HT implemntation takes 5% of the Pentium 4 die.
and i dont think its flawd - the preformance hit of activating HT of the box is in the >10% range while the preformance boost (of the box) is in the >30% range. - thats not bad at all for only 5% of the die space.

This post is best viewed with common sense enabled

thats why HT is only avalible for Xeons right now.
its not just a few "selected" applications - its a major group of applications like Data-bases, masive content creation and other <b>commanly used</b> server/web related applications. so its excelent for its current target market (Xeon bassed servers).

for only 5% of the die - I would keep it there.

By the time HT is implimented on desktops it will have a new core with greater sofware support as-well.


This post is best viewed with common sense enabled

>HT is not only unstable in many apps

? Got any links to back that up ? HT is imho a very clever technology, and Im a great fan of it. true enough its not the holy grail, and doesnt deliver on every possible app, but I've seen improvements of up to 50% with HT enabled. For only 5% die space, thats nothing to sneeze at. As for the performance decrease, I seem to recall that had to do with win2k not knowing the difference between physical and logical CPU's, and because of the CPU numbering scheme, threads would be allocated to the "wrong" cpu. i thought win XP fixed that (as might a win2K service pack, not sure). Really, I cant wait for the first cpu with on-die (dye ? how do you spell that anyway ?) memory controller AND HyperThreading.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

Lets see, where to start. The fact about the Northwood dies being bigger than the AXP, i didn't know. I though it was always smaller, according to these pictures:
<A HREF="http://www6.tomshardware.com/cpu/02q2/020402/index.html" target="_new">http://www6.tomshardware.com/cpu/02q2/020402/index.html</A>
Guess they are a little misleading, and i didn't thoroughly read up on it. My bad.

What similar technology? Care to be specific? According to several real life experiences, a 2.4GHz Northwood will run cooler under the exact same conditions as an Athlon XP 2100+. In other words, standard stock HSF, same room temperature, and same amount of thermal gel (or none). I'm not really bashing you for this, it's simply to say that I've never seen or heard of a Northwood running hotter than an Athlon XP of comparable level under the same conditions. And if you really want to get picky, clock for clock, the Northwood definately runs cooler than the AXP. Northwood is reliable, Intel is announced in a few weeks, they will demo an air-cooled Northwood running at 5GHz!

You probably read this in the wrong way. Hardware prefetch i just threw in, but in general, I meant that P4 has things athlon doesn't, eg. a more advanced (although quite smaller) L1 cache.

I read on several sites it's 30% altogether of wasted circuitry in the P4, which includes the disabled HT. Right now, HT actually already exists in the Willy & Northwood, but it's just disabled. I admit, it's flawed a bit, but not that much. I reckon it was a last minute feature they threw into the P4, but then decided to disable it because it's not practical for desktops. For certain applications, it gives a huge preformance boost, like 3d Studio Max, or Maya. But the fact still remains, unfortunately, that right now, the Northwood does not utilize all of it's circuitry.
And yes, it's rumoured that Prescott is supposed to have a new, revamped version of HT. Some of the negative effects of HT are in fact, a result of Windows not being able to utilize it properly (98 anyways).

Eden, could you care to tell me what that solution is?

Also, here is an article on HT:
<A HREF="http://www.anandtech.com/showdoc.html?i=1576" target="_new">http://www.anandtech.com/showdoc.html?i=1576</A>

It's a good read.

Back to the orginal post, here are the confirmed (by Intel)features of Prescott:

- starting frequency at about 4GHz
- 800mhz. FSB (200mhz quad-pumped)

Now here are the rumoured features which are likely to be on the Prescott:

- Tweaked L1 cache; size of L1 increased
- Tweaked L2 cache; size of L2 increased
- Revamped HT
- 32-bit ALU's

Now, finally, here are the distantly rumoured features of Prescott, which may or may not be true:

- 64-bit Yamhill technology
- SSE III (64-bit Yamhill extensions)
- added FPU
- native support for both 32 & 64 bit modes

Sorry, I didn't add on-die as in on-die mem controller, which does its job in performance boost no matter what the situation, compared to HT.

I'll have to disagree with Mat if he meant clock for clock, because yes the NW is cooler at stock speeds. I would however like to see soon what a 1.8GHZ Tbred vs a 1.8GHZA P4 would yeild.

Personally the idea of a 5GHZ A NW demoed, signals one of these:
-temporary WCPUID purpose
-really unstable but put in with one program open, which could be WCPUID
-case is hidden behind curtains, with really expensive aircooling, but is designed to show that it does reach it

Personally yes it is amazing that it has reached this much, but it's nothing to fuss about, mem bandwidth is probably limited at such speeds, you'd need PC1200 RDRAM or higher to compensate, many apps will not work fully stable, or are bottlenecked by everything else.


I'll tell you one thing though, I started this thread because I found some real killer specs of Prescott, yet you show even more! I am really starting to fear Intel's next generation processor, it sounds more than it could possibly offer.
Assume the following approximate performance improvements:
-With improved-HT enabled(10-20% in any app)
-Higher improved L1 trace cache(2-5%)
-L2 Cache of 1MB (3-5% MAX, since the IPC is starting to go up, therefore adding cache like that starts to be less beneficial)
-800MHZ FSB, twice the current P4s (up to 20%, average 10%)
-32-bit ALUs, moving SSE to them, freeing the FPUs to work as they deem, as well as maybe making the ALUs double pumped EFFECTIVELY as said (double the speed).(5-10%)
-Starting speed of 4GHZ, roughly at the time the ClawHammer 4400 comes out in 0.09m technology.
All that will increase IPC fer sure, and most likely will finally put it above the P3 clock per clock, however I think adding back an FPU would definitly make it even more threatning, especially in 3d Studio Max, at the point where SSE 2 is also enabled once moved to the ALUs, thus the performance under that prog is at least up by 40%!

If all that happens as expected, the new P4, will have an increase clock per clock performance of at least 40% basic, ranging up to 50% more! That IMO can be more than just dangerous for AMD. However even with the move to 0.09m, these improvements are added on THAT process, thus it gets filled already, and with no doubt at all, will the P4 start to slow its incremental speed steps. At such point, the long pipeline no longer can suffer much from branch misprediction, as well as it being quickly fed with data thus chances are low of any slowdown. Couple that with the new speed of 4GHZ, you got yourself a powerhouse, most likely gonna be the record house-heater too, but definitly something to gaze at and drool. Just that is enough to put AMD in a huge competitive position, which I do hope the 0.09m Hammers won't be just a die shrink, I simply don't want AMD to play the MHZ race like Intel did for the past 2 years with P4!

So in conclusion, Hammer 4400 is maybe enough or close to compete Prescott if these specs are followed, however AMD HAS to get ready for that, otherwise Prescott will be deadly.

--
Thunderbirds in wintertime, Northwoods in summertime!

Yeah well here's hoping it does. But we're still far from BBUL, and the best thing they could do is take an example from AMD's engineers. From there they might know more of how to compact the chips down and even make better heat dissipation as well as high performance with low transistor count.

I agree with Mat, indeed you are quite informative and are open to debates, so yeah I find it very fun to debate with you! I think we're all gonna get along well on this one.

--
Thunderbirds in wintertime, Northwoods in summertime!

yeah, but all those roadmaps realistically say is that the chips will both come out SOMETIME in 03, although AMD'S is more likely to come out in 04 than Intel's is. Unless, of course, Intel is blowing smoke up everyone's ass, which is quite possible. Speaking of blowing smoke, when the hell IS the t-bred coming? It seems like it SHOULD be AMD's response that the fans have been anxiously awaiting for a LONG TIME, but now they're getting so hyped about the Opteron and all, is this thing EVER coming out?

<b>Studies have shown that most people prefer the taste of AMD to the taste of Intel</b>

U reall y disappointed me here... i think every programmers knows that there r synchronization problems.. and thats y parallelism is not easy to extract... rather it becomes harder... as a simple example many modern Operating System doesn;t support deadlock prevention cuz they think that the extra effort spent on it is not worth... So if due to some synchronization problem u get a little bit of lower performance is not intel's problem.. and claiming it to be flawed is grossly biase opinion... as everyone is aware that x86 ISA is too old and crappy and so there will be some bugs may be... but it is not flawed.. and with rite optimization u can get even more than 35% improvement... secondly HT will increase more in performance with increase cache... thats y i think u will see quite a good number with prescott cuz its going to have 1MB cache.. and thats y they originally developed P4 with three level cache..

running two seperate threads is just against one of the basic of OS.. and to set the new standard is always hard.. its really a wonderful and truly impressive design.. as a programmer i m really impressed....
definitely with the time this will mature.

lastly they only used 5-7% xtra die for HT..
it makes it even more impressive.. isn't it..?

just my opinion..
No flaming inteneded to anybody