Intel's Netbust

[Venom]

Intel's netburst is turning into netbust.

Roadmaps show Prescott will intro @3.4 GHz around the end of this year / early next year and scale to 3.8 GHz by the end of next year. 4 GHz in december is a remote possibility. That is 18% in a year from here.

Even worse, if you compare it to the 3.06 GHz P4 launched in november 2002, reaching 3.8 GHz by november 2004 gives us a 12% clock increase per year over 2 years ! I don't think intel executed that poorly ever. TWELVE PERCENT ! Thats like going from 60 to 66 MHz.

Northwood/Willamette on the other hand, scaled from 1.5 GHz to 3 GHz in exactly 2 years. A 50% clock increase per year over 2 years. To match that again, intel would need a 5 GHz chip somewhere by the end of 2004 and a ~7 GHz by 2005. Just doesnt seem very likely.

AMD has had a tough year scaling Tbred and Barton, but K8 is looking healthy. a 2.4 GHz FX is expected this year, that is a 33% increase over the initial 1.8 GHz K8 (opteron) launched in may, roughly a 40% increase per year.

Also, let's assume a 2 GHz A64 to roughly equal a 3.2C and ignore better clock scaling of the K8 and better IPC for Prescott; from 3.2 -> 4 GHz is 25%. Does anyone seriously doubt AMD will have trouble releasing a ~2.5 GHz chip by the <b>end of next year </b>? They may not even need a die shrink for that. 2.4 GHz parts (FXs) are expected this year. Leaves them a full year and a die shrink to find 100 extra MHz and move those bin splits from low volume FX's to medium volume A64's.

Looks like intel's netburst effect is wearing out, and they are hitting a brick wall, and AMD will have an easy year. They can screw up their .09 migration and still conformtably lead.

 

[twofruitz]

yeah your right. intel is great at the moment but there isnt much more petrol in their tank. AMD is the underdog and has along way to go. its a tug of rope and both companies feed each other. just means competitive prices for me

<A HREF="http://service.futuremark.com/compare?2k1=6973142" target="_new">3dmark2001</A>

2.4C @ 3.0G
Asus P4C800
Corsair 2x512 3200
Sapphire 9600pro @ 500/700
80G Seagate S-ATA150

 

[Venom]

I agree on the last statement; we all benefit.

However, AMD has moved up from an underdog making compatible but slower intel clones to an innovator and performance leader. They had the performance crown through most of the Athlon's life (from 500 to ~AXP 2000+), and they have it now again, and probably for the next years.

They are also being more innovative than intel, at least on the desktop. All we are seeing from intel is higher clockspeeds, higher FSB's and huge caches. Hyperthreading was one nice idea, but doesnt bring much. AMD on the other hand has brought us innovative process tech (SOI), a real and usefull 64 bit ISA extention, integrated memory controllers, revolutionary hypertransport interconnect mesh, variable clockspeed/voltage for the desktop (cool & quiet), etc, etc. By comparison, intel has brought pretty much nothing to the desktop, and in the mobile world, they are hyping 5 year old slow and outdated wireless technology with $300M marketing campaigns, as if they invented it (ever heard of Apple's Airport half a decade ago ?), a mobile cpu, which is nice, but based on a EIGHT year old core.. where is the innovation ?

Intel is quickly becoming the technology and performance underdog, a dinosaur, all they have left is fab capacity to make HUGE face saving caches, historical market share and $$$. That won't last too long however.

 

[GIPNOR]

What about Intel's power saving Centrino? HT, IMO is a more important advancement than higher clock speeds too. My point is that it's not ALL about clock speeds. That's probably not the only measurement you should use.

AMD's 64bit processors are impressive... too bad they're useless if you want to run Windows. IMO, the new AMD processors aren't worth the price/performance difference you get with a P4 2.8c.

I run both an Intel and AMD rig as well as an Intel laptop. If I were building a new box from scratch right now, I'd definitely go with an Intel processor. If I'm upgrading one of my current machines, I'd slap an Athlon XP with a 333mhz FSB in my nforce2 board and upgrade the 2100 I'm currently running.

Innovation and competition is good for us (the consumers). I hope AMD and Intel continue to duke it out for years to come!

<P ID="edit"><FONT SIZE=-1><EM>Edited by gipnor on 11/09/03 07:37 AM.</EM></FONT></P>

 

[Zoron]

However, AMD has moved up from an underdog making compatible but slower intel clones to an innovator and performance leader.

It's going to take a lot more than being the performance leader to establish themselves. If you listen to g5_inside, Apple is the performance 'leader' and yet look at their market share.

Don't get me wrong. I love AMD. Their new A64 line is great. However, they now have to turn a great chip into great profit. It's not enough to make a great CPU, you actually have to sell it.

<font color=red> If you design software that is fool-proof, only a fool will want to use it. </font color=red>

 

[Guest]

>What about Intel's power saving Centrino?

Its a good cpu, a very good one; probably the best one to come out of intel's factories in a long time; but I have to agree with Venom (despite his fanboy-ish language); its hardly innovating. Its just taking agood, old core (P6 core), slapping on a larger cache, the P4 FSB bus and tweaking the process, and to some extent the core to lower power consumption. Compared to P4 that makes it look fabulous, but when you'd compare it with 2 year old tualatin mobile P3's its a lot less impressive, evolutionary at best.

"Old" certainly doesnt mean bad in this case, but its really nowhere near as revolutionary as, say the Crusoe or even the A64. Now the P-M is without doubt the most interesting part of "centrino", the rest is really pathetic; completely obsolete WiFi chip (802.11b is actually the same standard I've had 5 years ago, agonizing slow 11 Mb/s while 5x faster 802.11a and 802.11g are defacto standards), and the chipset that comes with it is nothing but a rebadged i845 chip with a bus disconnect feature that even the venerable athlon supported on the desktop (and maybe so does the P4, not sure). Centrino definately NOT a good example of innovation, its more a case of extreme marketing.

>HT, IMO is a more important advancement than higher clock
>speeds too.

HT is technically definately interesting, but so far, on the majority of benchmarks, it doesnt bring a lot to the table. Overall, certainly no more than maybe 10%. On server apps this can be substantially more though

>My point is that it's not ALL about clock
>speeds.

Well, I have to agree with Venom once more; with intel, beside HT, these last 4 years its only been about clockspeed and caches with nothing really innovating on the horizon either. Prescot with a bigger cache, higher clockspeed, and yet another few SIMD instructions. No bigger deal than P2->P3. They may need a more refreshing approach to keep up with AMD in the performance race or at least have the balls to copy some good idea's from AMD like ondie MC and HTT for their server chips.


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

 

[Guest]

I agree with your point, even though the Apple comparison doesnt hold. Unlike Apple, AMD is not dependant on a more or less static customer base that wants to use their propriarity (sp?) OS and apps. Pretty mych every potential intel customer is also a potential AMD customer, which can not be said of Apple.

That being said, and while stil agreeing with you, I think it must be said todays AMD is not the same as the one that launched the Athlon. Athlon was the performance leader for a while, but it came out of nowhere, from a company with no name, no serious trackrecord and that produced a chip that didnt have any significant advantages over intel's alternatives beside temporarely, a relatively better performance (or and/or price).

Today's AMD is the company that is being backed by Sun, IBM, Cray, for which MS is writing a tailored 64 bit OS. And the opteron/A64 chips have a few very compelling advantages over intel's products beside just speed. They run cooler, run 32 and 64 bit apps simultaneously, opteron makes it *much* easier to build and scale MP systems, ..etc. Also, unlike 3 years ago with the K7 , the K8 has already a very decent infrastructure behind it, with good chipsets, plenty of OS and ISV support, and OEM support that is beginning to become sizeable.

WHile I am not the one predicting the end of Xeon because of Opteron, I do expect AMD to make some very serious inroads into the high end markets the next few years.

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

 

[pIII_Man]

Its just taking agood, old core (P6 core), slapping on a larger cache, the P4 FSB bus and tweaking the process.

Actually the pentium M still uses the same old ATGL+ bus as the tualatin core piii's just they are running it at a very high clock speed (400mhz).


If it isn't a P6 then it isn't a procesor
110% BX fanboy

 

[Guest]

The P4 also uses a AGTL+ bus, as does Itanium !

Make no mistake about it though, the P-M should work in a bog standard P4 motherboard if it weren't for two pins they swapped to make this impossible. Also, the i855 (or whatever its name) chipset that is part of the "Centrino" package is nothing more or less than a rebadged i845 chipset, maybe with added or enabled bus disconnect. The PM uses the P4's quad pumped databus, not the P3 bus, its electrically 100% compatible with the P4.

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

 

[eden]

The P4 is also AGTL+ like the P3. They're both pretty similar, so you CAN equate the Pentium M's bus to the P4's.

Also, it IS a QDR bus like the P4: <A HREF="http://www6.tomshardware.com/mobile/20030205/centrino-06.html" target="_new">http://www6.tomshardware.com/mobile/20030205/centrino-06.html</A>
Damn the info was hard to find... Little is revealed on Banias.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>This just in, over 56 no-lifers have their pics up on THGC's Photo Album! </b></font color=blue></A> :lol:

 

[simwiz2]

Your slow clockspeed scaling can be applied to AMD as well. AMD had a 2.25 GHz T-Bred B at the beginning of 2003, and at the end of 2003 they are expected to have a 2.4 GHz A64, for a 7% increase. SEVEN PERCENT !

Clock speed does not equal performance. Intel made up for the low clock increases with the P4C's, AMD with the Barton and now the A64 to increase IPC.

 

[pIII_Man]

Heh, thats new to me, i figured with the intro of QDR intel would have ditched Gunning logic er whatever it stands for.

I guess thats pretty interesting that intel has ben using pretty much the same bus logic since the pII's and probably since the pentium MMX.


If it isn't a P6 then it isn't a procesor
110% BX fanboy

 

[Guest]

Comparing different architectures makes little sense. I could just as well claim intel's fastest cpu is currently running at a 50% LOWER clockspeed as what they had 1 year ago (ie, itanium). What is somewhat more relevant, is how much either company has been able to scale the same core. K7<>K8.

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

 

[pIII_Man]

Thats true, although intel has not been increasing clock speed lately they have been implimenting HT higher fsbs, better chipsets (dual channel and pat) all of these things really help performance.


If it isn't a P6 then it isn't a procesor
110% BX fanboy

 

[eden]

Well, they had used better material for the AGTL+, especially for the P4 to ramp up to 200MHZ QDR, compared to the P3 who often couldn't do higher than 150MHZ SDR. The whole noise issue was serious with GTL+.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>This just in, over 56 no-lifers have their pics up on THGC's Photo Album! </b></font color=blue></A> :lol:

 

[simwiz2]

Comparing different architectures makes about as much sense as comparing P4B to P4C to Prescott. They all have/will have different performance per clock, yet Venom fails to mention that, just that Intel's rising clock speeds have slowed. Mentioning one without the other is very misleading and gives the impression that Intel has not raised performance by much recently, while in fact they have with the P4C's. With Prescott's IPC still largely unknown, how are assumptions based on clockspeed alone likely to be any more accurate than an assumption that AMD hasn't increased performance by much because they have not increased clockspeed much?

 

[pIII_Man]

I disagree, if you had good memory you could easily take a piii 800eb to 166mhz+ fsb, most of the times higher with a memory divider.


If it isn't a P6 then it isn't a procesor
110% BX fanboy

 

[Guest]

> Mentioning one without the other is very misleading and
>gives the impression that Intel has not raised performance
>by much recently, while in fact they have with the P4C's

I'm not sure I fully agree; FSB's speeds and cache sizes have been increasing steadily for like, forever; this is not something new or unique about the P4C or Prescott. For instance, going from 1.5 Willamette to 3.06 Northwood intel increased the FSB from 400 to 533, doubled the cache, added HT, created a new (dual channel ?) DDR platform while at the same time upping the cpu frequency by 50% per year. These last few years however, we've only seen another incremental FSB increase, and soon a L2 cache increase along with few tweaks, so not very different from what we've seen in the past, but this time without the fast 30-50% clock scaling we've seen before. Intel *is* slowing down or hitting a wall. If you think IPC increase from prescott compared to NW is bigger then willamette -> northwood -> P4C + dual channel, then you're in for a disappointment IMHO.

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

 

[eden]

Hmm, you do have more experience, I just based mine on ol' Crashman.

I do know it was possible, but I recall it wasn't always obvious, compared to the AGTL+ enhancements on the P4. Dunno when AGTL+ was added on P3. Could've sworn it was GTL+ basic!

Anyways, I was wondering also: Is EV6 really using its features on K7? I believe the main advantage was having dedicated resources not on the same bus all at once, but has it ever really yeilded advantages in performance in real life, like Hyper Transport does for Opteron?

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>This just in, over 56 no-lifers have their pics up on THGC's Photo Album! </b></font color=blue></A> :lol:

 

[pIII_Man]

the tualatin piii's got atgl+ however the only real diffrence between gtl+ and atgl+ was the voltage at which signals were transfered, i beleive Atgl+ required 1.25v while GTL+ used 1.85 (don't quote me on those numbers though). But the two buses were compatable, meaning that a tualatin with some pin mods (to trick the cpu into thinking it was runing in an ATGL+ supported chipset) the cpu could run fine in an older GTL+ mainboard.


If it isn't a P6 then it isn't a procesor
110% BX fanboy

 

[eden]

Dunno the exact workarounds for the AGTL+ specs, but I do remember the improved version includes better and reduced noise, which allowed them to go further.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=blue><b>This just in, over 56 no-lifers have their pics up on THGC's Photo Album! </b></font color=blue></A> :lol:

 

[pIII_Man]

I really doubt it as the tualatins used basically the same chipset as the Coppermines, the only diffrence being the voltage at which the bus operated and i beleive the movment of one of the reset pins, possibly they have done some fun stuff to the p4 to make it go a bit further, but in all honesty GTL+ could go pretty high, with isolated reports of 200+ fsb with people runnung extreme cooling on their EB processors, never have i heard of someone saying they were limited by the GTL+ bus.


If it isn't a P6 then it isn't a procesor
110% BX fanboy

 

[pIII_Man]

Thats possible but the tualatins used basically the same chipset as the Coppermines, the only diffrence being the voltage at which the bus operated and i beleive the movment of one of the reset pins, possibly they have done some fun stuff to the p4 to make it go a bit further, but in all honesty GTL+ could go pretty high, with isolated reports of 200+ fsb with people runnung extreme cooling on their EB processors, never have i heard of someone saying they were limited by the GTL+ bus. Whatever they have done GTL+ is compatable with ATGL+ it could be that intel has put stricter guidlines on the S:N on ATGL+ mobos?


If it isn't a P6 then it isn't a procesor
110% BX fanboy

 

[Unolocogringo]

I'm running a P3 1.26@1583 . A 167fsb with memory in sync.@3/2/2/7/9

I aint signing nothing!!!