P4's future is looking GOOD to me.

[Degenerate]

Few points i have found:

1. Rambus will become cheaper with demand and Samsung's new design.

2. P4's Clock speed can be increased dramatically, outshadow the long pipline which slow the chip down at low clock speeds. as far as i can see, to reach 2ghz by the start of 2H.

3. 0.13 fab's to lower cost and possible improve performance as new feature will be added.

4. 300mm wafers to further lower costs

5. Intel will spend $7.6 billionQQ (thats way more than Amd's sales alone) on equipment for points 3 and 4 fo the 2nd half of this years, and introduce Tualatin- 0.13 version of p3.

5. SSE2 will be supported widely soon. even now the performace is very very good in multimedia compared to Amd. (point,2)

Few other points. Amd seem to run to an end with the Athlon just as the P3, both need reduction and redesign. (Amd seems to have an advantage with their core being about 200mhz faster.) And there for what it seems is that Intel will ramp the p4 with confidence as their processor is able to hit much faster speeds while Amd will need a comlete redesign. But who knows, it seems to me Intel has a chip that that Amd will soon have problems with.


Being a Taiwanees, my spelling is not so good but i cant be bother to check.

 

[Gunbuster]

It just has no future for people who have one now, since the socket will be changed.

 

[Czar]

you on something? Intel will not release the 2ghz p4 untill next christmas, Amd will have 1.5-2ghz by that time. The future of the p4 is good, but not when you compare it to what it has to go up against.

 

[Sunner]

It has a bright future indeed, but that doesnt mean AMD's future isnt bright as well, assuming AMD can keep up their execution.

As for Samsung's Rambus manufacturing tech, Paul DeMone wrote a bit on this, expressing some doubt as to if it would actually decrease costs by any significant margin.

 

[Dulanic]

<< 1. Rambus will become cheaper with demand and Samsung's new design. >>



As will DDR which is lower latency, now all we need to see is if Dual Channel DDR will work with Crush.... if so its the Rambus killer.



<< 2. P4's Clock speed can be increased dramatically, outshadow the long pipline which slow the chip down at low clock speeds. as far as i can see, to reach 2ghz by the start of 2H. >>



Not quite 2H buddy, but in AMDs press release we will see 1.5Ghz in 2Q 01, and 2Ghz sometime 2H, tho more likely Q4.



<< 3. 0.13 fab's to lower cost and possible improve performance as new feature will be added. >>



And imagine what AMD is gonna do with .13, don't think Intel is the only one with it or something.



<< 4. 300mm wafers to further lower costs >>



Again AMD is already working on this. No Intel advantage.



<< 5. Intel will spend $7.6 billionQQ (thats way more than Amd's sales alone) on equipment for points 3 and 4 fo the 2nd half of this years, and introduce Tualatin- 0.13 version of p3. >>



Thats all fine and stuff, they HAVE to spend more, they have more fabs.



<< 5. SSE2 will be supported widely soon. even now the performace is very very good in multimedia compared to Amd. (point,2) >>



Hammer series will also be supporting SSE2, but dont expect to see much SSE2 for atleast a year or more, and many programs still wont support it. However once Hammer supports it we may see alot of SSE2 software because its become like a standard like MMX.

 

[PliotronX]

GUTB debunked by Dulanic.. GG

 

[Sunner]

<< As will DDR which is lower latency, now all we need to see is if Dual >>


DDR SDRAM wont be able to play that game forever.
RDRAM's pincount is far lower than SDRAM's, a dual channel RDRAM config still has fewer pins than a single channel SDRAM config.

Then of course there is the problem with the high speeds RDRAM is running at, but thats another problem

 

[Degenerate]

Channel DDR will work with Crush.... if so its the Rambus killer. [/1]

Pin count. Price will drop. Why not quad pumped rambus? Much more band width




Not quite 2H buddy, but in AMDs press release we will see 1.5Ghz in 2Q 01, and 2Ghz sometime 2H, tho more likely Q4.

ok , i am wrong, intel p4 willl hit 1.7 and with much better overclockability. 2H clawhammer will be 2002 (start) ( it it a new design like the p4?)



And imagine what AMD is gonna do with .13, don't think Intel is the only one with it or something.

no. its just that intel has spent mor eon it and will deliver much earlier



Again AMD is already working on this. No Intel advantage.

Again. intel is first




In the end, amd will have to use the long pipline just like the p4 which will end up similar to the p4 unless there is a way around it. Again, Intel is the one who does something new first.

 

[Degenerate]

Channel DDR will work with Crush.... if so its the Rambus killer.

Pin count. Price will drop. Why not quad pumped rambus? Much more band width




Not quite 2H buddy, but in AMDs press release we will see 1.5Ghz in 2Q 01, and 2Ghz sometime 2H, tho more likely Q4.

ok , i am wrong, intel p4 willl hit 1.7 and with much better overclockability. 2H clawhammer will be 2002 (start) ( it it a new design like the p4?)




And imagine what AMD is gonna do with .13, don't think Intel is the only one with it or something.

no. its just that intel has spent mor eon it and will deliver much earlier



Again AMD is already working on this. No Intel advantage.

Again. intel is first




In the end, amd will have to use the long pipline just like the p4 which will end up similar to the p4 unless there is a way around it. Again, Intel is the one who does something new first.

 

[Dulanic]

<< << As will DDR which is lower latency, now all we need to see is if Dual >>


DDR SDRAM wont be able to play that game forever.
RDRAM's pincount is far lower than SDRAM's, a dual channel RDRAM config still has fewer pins than a single channel SDRAM config.

Then of course there is the problem with the high speeds RDRAM is running at, but thats another problem >>



Of corse, Im not argueeing that. DDRII will be serial, I never said Serial isnt the future..... but Rambus is not.




<< ok , i am wrong, intel p4 willl hit 1.7 and with much better overclockability. 2H clawhammer will be 2002 (start) ( it it a new design like the p4?) >>



Yes hammer is a totally new core. And how do you know about overclockability of Palomino and Hammer? Have you used them.... and Im mainly talking about Palomino.... T-Bird beats P4 clock to clock.... yet AMD is gonna be just behind in clock speed.



<< no. its just that intel has spent mor eon it and will deliver much earlier >>



Absolutely Intel spent more, they have alot more fabs. And not much earlier... 2-3 months earlier MAX... might be 1-2.



<< In the end, amd will have to use the long pipline just like the p4 which will end up similar to the p4 unless there is a way around it. Again, Intel is the one who does something new first. >>



Just because Intel had to cut performance to gain clock speed doesnt mean AMD has to. It may end up that way, but it doesn't look soon. Only time will tell which path is better. Same with server, Intel is going with a whole new architecture with Itanium, because they say x86 is dying.... AMD says hell no it isnt, and well prove it with the hammer series.

Only time will tell which path is better.

 

[Czar]

I just like to clarify one thing, the main reason why Rambus is dead. It already has 60ns latency or something, but the main thing is about Rambus

The more Memory the more latency

This does nto work whien you think about the future, Rambus is dead.

 

[rickn]

Degenerate: you wont get much support for Intel on these message boards, which is one reason I very rarely visit. I have always been an Intel supporter(not RAMBUS though) It is always just a bunch of penis envy arguements.

 

[Biggs]

Both are dead, the Palomino Pro will have 5GB of L1 cache!

 

[fkloster]

<< Rambus is dead >>



Oh boy...

In an i850 configuration, latency plays no part in the violent spanking DDR implemtations get from properly configured rambus implemtations. Not now, not in even the distant future. How about we get that statement etched (sp) in stone. Anyone with a DDR setup want to compare memory benchmarks w/Nox &amp; I? Serial (sp) memory IS the future, and current (Rambus + i850) is one hell of an example of serial memory BLUDGEONING anything parallel (DDR) in existance period. (...its actually kind of embarrasing )

Int ALU/RAM bandwidth/s 1582
Float FPU/RAM &quot;&quot;&quot;&quot;&quot;&quot;&quot;&quot;&quot;&quot;&quot; 1618

...and degenerate, a 20 stage pipeline is not a liability, its an asset. Coupled with superior branch prediction will allow Intel procs to scream...&amp; they will need to scream because of the longer pipeling...

 

[Mule]

fkloster flexing it's muscles again!

 

[MGMorden]

I don't see why so many people get into an uproar about which company/technology is gonna be ahead. All I care about is that we get faster stuff and that it gets cheaper, which means that all companies and technologies serve the good of the cause. AMD or Intel? Pick one. I personally kinda do crop rotation, going with one or the other every year or so when I upgrade. I don't have any strong preference for either (I actually hope both of them do great with their chip designs, and though it doesn't look like they will, I wish Via could become a strong force in the market). More competition is good for us. Any strong one-sided brand loyalty kills competetion.

 

[Degenerate]

<< And how do you know about overclockability of Palomino and Hammer? Have you used them.... and Im mainly talking about Palomino.... T-Bird beats P4 clock to clock.... yet AMD is gonna be just behind in clock speed. >>




Palomino, is just the reduced (in size, power to reduce heat) versioj of the Tbird. I am saying the P3 is also getting an upgrade, and that both are coming to an end soon.

Any way, only time will tell how the Hammer will do.

 

[MGMorden]

Palomino is just the code name. The chip is gonna be called the Athlon just like the T-Birds are called Athlons, and so the Athlon line of chips should go to 1.8-2.0Ghz, which should last at least another year and a half, which means it ain't gonna end soon. (of course P3's aren't ending soon either. After P4 becomes mainstream they'll just become Intel's Celeron-type processor taking over the value side of the Intel camp).

 

[RagingGuardian]

Degenerate, I like the way you carefully outlined your points about the P4's future being really bright but do you suppose that AMD is going to sitback and twiddle their thumbs while Intel ramps P4 production?

 

[Dulanic]

<< Palomino, is just the reduced (in size, power to reduce heat) versioj of the Tbird. I am saying the P3 is also getting an upgrade, and that both are coming to an end soon. >>



Actually AMD has no plans of coming to an end with the Athlon Core soon.... it will last into past 2002, We have Palomino and Thouroughbreed (Have no clue how to spell it) Coming too. Hammer will start as a server chip. And actually Palomino will have some changes besides heat and power, AMD said so at the Q4 conference call... nothing huge like a whole redesign but it will have some changes to enhance performance. You have to remeber, just because Intel does something like resign the whole chip like in P4, AMD doesnt have to follow, they have their own plans.

Intel went with the P4 style because they cant get much more out of P3... even with .13 Micron. AMD plans to take the Athlon core past 2Ghz. There is no reason for a total redesign, clock for clock the Athlon core can easily beat a P4... so if AMD can keep within 30% clock speed of the P4 there is no reason at all to do a redesign.

 

[Soccerman]

wow, people don't think both ways..

1st, why does the P4 succede in running well when it has inferior processing power? BANDWIDTH.

how much more bandwidth does a P4 have then an Athlon? almost 2x as much, both RAM and FSB bandwidth.

so, all things being equal when AMD releases thoroughbred at the beginning of next year, along with an increase in FSB speed I'm guessing, (seeing as they can tell that the P4 thrives on it, and the Athlon chokes on it's limited cache bandwidth, let alone the FSB and Memory bandwidth), the Athlon spanks the P4 silly.

however things are not equal (nor will ever be). the P4 and Athlon are both great solutions, but one is better in many ways.

why choose an Athlon?
-you are not contributing to RAMBUS's annoying business practices, and you are supporting competition.
-the Athlon is currently as fast, or faster, even when the P4 is running at higher clock speeds, in most applications (save SOME games, like Quake 3, which is dependant on FSB speed and RAM speed).
-currently the Athlon upgrade path is clear, Socket A all the way (thoroughbred will probably support both a 133mhz FSB and a higher FSB, like there will be 100 and 133mhz fsb Palomino chips).

about this RAM issue, there's no question that companies must reduce pin count, or be faced with RAMBUS wasting their product in terms of speed.

in other words, serializing will be a partial solution.

the unacceptable levels of latency cannot be avoided as clock speeds increase.

it's simple, the higher the CPU clock speed is, the more clock cycles that the CPU waits for RAM.

being capable of reducing the hit is only a partial solution (it's the reason why we have cache, and RAM, hard drives aren't fast enough, but higher speed Mass Storage devices are one of the biggest priorities in the computer industry today. BANDWIDTH BANDWIDTH BANDWIDTH.

so, what kind of RAM do we need? lower pin count, without having to increase clock speed so high (thus using DDR, Quad Data rate etc) to accomplish the same goal.

RDRAM does some of this. it drastically reduces pin count, and uses DDR tech to minimize the insane clock speed required to compensate this, however even 400mhz is too high on today's die processes (too much heat, too much power draw).

what else would we need then? a totally new memory concept. magnetic RAM? perhaps, but electonic memory storage is really up against a wall unless you are willing to throw massive amounts of power, and cooling at the problem.

 

[Midnight Rambler]

<< << no. its just that intel has spent mor eon it and will deliver much earlier >>

Absolutely Intel spent more, they have alot more fabs. And not much earlier... 2-3 months earlier MAX... might be 1-2. >>

This is not true. Intel isn't spending more because they have more fab plants, they are spending more to add mostly new fab plants, with some upgrades to existing sites as well. So the truth is that Intel is significantly increasing their fab capacity advantage over their competitors (by at least 3-4 additional, ie. new, plants).

 

[polar]

IMO, athlon is gonna dry up soon, as price of RDRAM getting lower and more developers start to optimise for P4's new architecture. 80x86 architecture has reached its limit long time ago, running harder in a wrong path will only bring one to the deadend sooner.
through a painful year, intel has successfully(but not soomthly) transfered to a whole new platform. IMO, if amd can't make a smooth transaction into a new architecture within 1 year, they will be in serious trouble, since they dont have the resource like intel has to handle the periodical pain.

 

[OneEng]

P4's future looks pretty bleak to me. Here is my rationale:

First let me address RAMBUS memory. Serial memory while having lower pin count has much much higher latency (bad bad bad). If this isn't bad enough, it must also run at higher frequencies makeing it consume more power, run hotter, and lowering yields in production (BAD BAD BAD BAD). I fully expect DDR and QDR to make RDRAM obsolete within 3 years. Pin count is an issue, but not that much of one. It doesn't cost THAT much to add a few traces to a board.

As an aside, since MS, Intel, Micron, Infineon (Seimens), and many other large corporations are aligned against RAMBUS, they are assured a loss in their legal battles over DDR patents.

Now for the P4.

In its current encarnation (RAMBUS only, 217mm2 .18um aluminum interconnects, 1.5Ghz), Intel is most likely losing money on every P4 they sell (since they give away 2 sticks of RDRAM with every P4 right now). P4 is not much of a solution for Intel. Even if they could convince consumers that they need a 1.5Ghz machine and everyone went out and bought one, Intel would suffer financially due to the cost factors of this chip and its platform.

Both Intel and AMD are faced with a downturn in PC demand. People simply do not need to upgrade their machines yet in order to run their favorite applications well. This hits Intel's P4 particularly hard. Do I really need a 1.5Ghz P4 ..... well maybe not since I can get one of these 1Ghz do-dads for half the price.

OK, lets fast forward a couple of quarters and Brookdale PC133 SDRAM is now availible for P4. Although this may be a saving grace for P4's platform cost along with raising the clock speed up to around 2Ghz, using SDR with P4 will ABSOLUTELY KILL the performance! Here on Anands, it has been shown that a 25% reduction in RAM bandwidth (800Mhz RAMBUS vs 600Mhz RAMBUS) in a dual channel configuration reduced performance by 17%!

Using PC133 memory would be a 300% REDUCTION IN MEMORY BANDWIDTH and would result in horrendous performance.

Back to reality. The consumer may well not know or care that the P4 1.8Ghz he is buying is not even fit to compete with a 1Ghz PIII at less than half the cost, so as far as purchasing goes, it may do alright if the media fanfare is sufficient (which I am sure Intel will see to).

This gives Intel about 6 months of superiority over AMD (in marketing only of course since the 1.7Ghz Palimino will be 2 times as fast as a SDR 2Ghz P4). AMD is due to release their SOI .13um copper Hammer series Q1 2002. This chip should be more than a match in both clock speed and performance to the poor P4. It will have SSE2 support also. This should eliminate the benchmark bias we currently have to endure with all this proprietary compilation and make SSE2 standard throughout the industry (finally).

With a roadmap like this, I seriously don't see how anyone can think the future of P4 is all that bright. It certianly isn't going to die and dissapear or anything stupid like that, but P4 faces enormous competition.

AMD's business model makes them hugely profitable at $100.00 ASP. Intel needs $200.00 to pull the same margin. P4 is a very poor marketing tool for Intel right now and costs way too much for Intel to produce. Intel would have been better off evolving the PIII core for the short run while they re-designed the P4 to better cope with todays market. I don't think Intel had any idea that P4 would be facing such stiff competition by the time it was released or that RDRAM would simply be rejected in the market.

 

[viper007]

OneEng:

Have you been reading my mind again? They were exactly my thoughts!

Cheers..