Intel Increases Pipeline stages towards 30 w/ Precott design?

[zephyrprime]

Hey, I already started a thread about this topic in the CPU group. Anyway, there's been quite a bit of discussion about this issue for the last few days.

Some have pointed out that the current P4 has a 28 stage pipeline if you include the stages before the trace cache and after the point where branches are determined.

Others have pointed out that Intel supposedly did some calculations some years back that showed that the theoretical optimal pipeline length is 40 stages.

With the P4, they could seed compilers with information on how to handle the longer pipeline, and more importantly, how to handle SSE2. Prescott however has just the longer pipeline, so there's a decent chance Intel won't be able to get back as much effeciency out of the Prescott, since the only thing going for it this time are more pipeline compiler optimizations

I don't think there's really a lot that needs to be done to optimize compilers for prescott that isn't already done for the P4. Really, you can only avoid branching to a certain degree.

 

[Rectalfier]

I am by no means an AMD zealot, considering I just bought a centrino laptop, and a P4 desktop. I may have been wrong on some of the points I brought up, but the fact remains that AMD has a better processor right now. If Prescott's pipeline is increased, I don't see it being any faster than AMD's current lineup. Regardless, Intel will remain the Godfather of processor production, as AMD does not have the capacity to overcome Intel.

 

[Wingznut]

Originally posted by: mikecel79

Originally posted by: Wingznut

Originally posted by: mikecel79

Originally posted by: Rectalfier
Here's the way I see it. Intel has already delayed the Prescot due to heat issues. I think they are trying to work around this by increasing the pipeline. It just seems that Intel has done this all of a sudden, since they couldn't produce their original Prescott design, and have thus delayed it. Could this be a massive trip for Intel? The release of the Pentium 4EE sure seemed like an act of desperation. Rest assured, Intel will have something good for us again, but it looks like 2004 will be AMD's biatch.

Extending the pipeline isn't something that can be done in 3 months. They would basically have to go back to the drawing board to extend the pipeline. This was planned well before Prescott was delayed. They couldn't do that in 3 months or even 6 months. Extending the pipeline would not help it deal with heat either.

Best post so far.

Thanks. I don't think people have any idea how much work and engineering goes into making a processor. I don't think even I know how much work goes into it. So many people talk about just extend the pipeline or just ramp it up 1Ghz. I don't think they realize how much validation and testing goes into a CPU before it's released.

I know... The first thing I thought about when I read aRCeNiTe's post (about raising the A64's clockspeed by 1ghz) was, "Hell, why stop at 1ghz??? They should just go for double the current clockspeed."

 

[thorin]

Originally posted by: arcas

Fast forward to today...the P4 design is currently chugging along at 3+ ghz and by almost any benchmark is far faster than the fastest P3 it supplanted. Likewise, in terms of actual performance, the deeply-pipelined P4 design and the shallow-pipelined Athlon design have largely kept pace with one another. If at any point in time one design were found to offer, say, 5x higher throughput than the other design, I think one could confidently claim that design is superior. But in fact that hasn't been the case. At no time over the past 2-3 years has the deeply-pipelined design had a significant performance advantage over the shallow pipeline design. Or visa versa. So these arguments are largely moot.

Ok so now we get Intel to re-engineer the P3 baed on current manufacturing techniques and it can hit 3GHz and blasts the P4 to pieces.....

Thorin

 

[Insomniac]

Originally posted by: Wingznut

Originally posted by: mikecel79

Originally posted by: Wingznut

Originally posted by: mikecel79

Originally posted by: Rectalfier
Here's the way I see it. Intel has already delayed the Prescot due to heat issues. I think they are trying to work around this by increasing the pipeline. It just seems that Intel has done this all of a sudden, since they couldn't produce their original Prescott design, and have thus delayed it. Could this be a massive trip for Intel? The release of the Pentium 4EE sure seemed like an act of desperation. Rest assured, Intel will have something good for us again, but it looks like 2004 will be AMD's biatch.

Extending the pipeline isn't something that can be done in 3 months. They would basically have to go back to the drawing board to extend the pipeline. This was planned well before Prescott was delayed. They couldn't do that in 3 months or even 6 months. Extending the pipeline would not help it deal with heat either.

Best post so far.

Thanks. I don't think people have any idea how much work and engineering goes into making a processor. I don't think even I know how much work goes into it. So many people talk about just extend the pipeline or just ramp it up 1Ghz. I don't think they realize how much validation and testing goes into a CPU before it's released.

I know... The first thing I thought about when I read aRCeNiTe's post (about raising the A64's clockspeed by 1ghz) was, "Hell, why stop at 1ghz??? They should just go for double the current clockspeed."

But, but, all you have to do is increase the multiplier. I can't believe it never occurred to AMD to do that. No wonder Intel controls the CPU market.

 

[Insomniac]

Originally posted by: thorin

Originally posted by: arcas

Fast forward to today...the P4 design is currently chugging along at 3+ ghz and by almost any benchmark is far faster than the fastest P3 it supplanted. Likewise, in terms of actual performance, the deeply-pipelined P4 design and the shallow-pipelined Athlon design have largely kept pace with one another. If at any point in time one design were found to offer, say, 5x higher throughput than the other design, I think one could confidently claim that design is superior. But in fact that hasn't been the case. At no time over the past 2-3 years has the deeply-pipelined design had a significant performance advantage over the shallow pipeline design. Or visa versa. So these arguments are largely moot.

Ok so now we get Intel to re-engineer the P3 baed on current manufacturing techniques and it can hit 3GHz and blasts the P4 to pieces.....

Thorin

They already did that, it's the P4.

 

[zephyrprime]

Ok so now we get Intel to re-engineer the P3 baed on current manufacturing techniques and it can hit 3GHz and blasts the P4 to pieces.....

Thorin

They already did that, it's the P4.

If the P3 were made using modern processes, we could expect it to have a speed that were somewhat less than Athlon XP clockspeeds so it wouldn't really blast the P4 away. But if it were really re-engineered but retained the same pipeline length, it's hard to say how it would perform.

 

[Markfw]

But, but, all you have to do is increase the multiplier. I can't believe it never occurred to AMD to do that. No wonder Intel controls the CPU market.

OK, can somebody else other than me reply to this ? An Intel biased person will even do...............

 

[Duvie]

I am still wondering about why all this confusion on number of pipelines....It was dissected just months ago and the ppl stated a samll increase in pipeline but not liek the articles states...

So which are the real numbers???

P4 current pipelines????

Prescotts proposed pipelines???

 

[Ionizer86]

In a slightly different light, we can kind of think of P-M as P3 reengineered. Some pretty awesome IPC they've got there with a 1.6/1.7GHz chip blowing away P4's that are clocked like 600MHz to 800MHz higher.

 

[Wingznut]

Originally posted by: Markfw900

But, but, all you have to do is increase the multiplier. I can't believe it never occurred to AMD to do that. No wonder Intel controls the CPU market.

OK, can somebody else other than me reply to this ? An Intel biased person will even do...............

Umm... Insomniac was being sarcastic.

 

[Duvie]

Originally posted by: Ionizer86
In a slightly different light, we can kind of think of P-M as P3 reengineered. Some pretty awesome IPC they've got there with a 1.6/1.7GHz chip blowing away P4's that are clocked like 600MHz to 800MHz higher.

i dont know about blowing away...look behind the intel hype on them.....I have the 1.5 version centrino w/ 1mb of l2 cache and only a handful of things did it do faster then a 2.0ghz p4 I tested...Most of the time it was rather poor and acted like a 1.5-1.8ghz cpu.....Then throw it in a laptop that sucks it down a couple hundred mhz in comparison to a desktop PC with no dual channel DDR, 4200rpm harddrives, etc.....

 

[Insomniac]

Originally posted by: Duvie
I am still wondering about why all this confusion on number of pipelines....It was dissected just months ago and the ppl stated a samll increase in pipeline but not liek the articles states...

So which are the real numbers???

P4 current pipelines????

Prescotts proposed pipelines???

We'll know for sure when the CPU comes out. Even the person who wrote the article isn't sure.

The current P4's have a 20 stage pipeline.

 

[Insomniac]

Originally posted by: Wingznut

Originally posted by: Markfw900

But, but, all you have to do is increase the multiplier. I can't believe it never occurred to AMD to do that. No wonder Intel controls the CPU market.

OK, can somebody else other than me reply to this ? An Intel biased person will even do...............

Umm... Insomniac was being sarcastic.

Hehe, I thought it was obvious.

 

[MadRat]

The longer pipeline gives off more heat? Interesting. If Prescott really introduced 10 stages then perhaps this is the problem with thermal disipation. I still tend to believe the theory that its a 2-stage increase, relative to adding the trace cache...

Originally posted by: zephyrprime
Some have pointed out that the current P4 has a 28 stage pipeline if you include the stages before the trace cache and after the point where branches are determined.

If it was a simple 2 stage increase then Intel has some excellent results from 90nm sampling to believe 4GHz is possible by year's end! Who cares if thermal disipation increased to 90-100W for a 4GHz processor when raw computational power will have increased a shocking high 25%! People will dump their 3GHz P4's by the droves if this is the case. Remember that gaming and media encoding performance has steadily improved with Intel's clock increases. These processes don't have as many snarls and performance traps as other programs seem to run across, making a 4GHz processor seem awfully advantageous, in these respects, over a 4000+ PR rated processor.

Originally posted by: Duvie

Originally posted by: Ionizer86
In a slightly different light, we can kind of think of P-M as P3 reengineered. Some pretty awesome IPC they've got there with a 1.6/1.7GHz chip blowing away P4's that are clocked like 600MHz to 800MHz higher.

i dont know about blowing away...look behind the intel hype on them.....I have the 1.5 version centrino w/ 1mb of l2 cache and only a handful of things did it do faster then a 2.0ghz p4 I tested...Most of the time it was rather poor and acted like a 1.5-1.8ghz cpu.....Then throw it in a laptop that sucks it down a couple hundred mhz in comparison to a desktop PC with no dual channel DDR, 4200rpm harddrives, etc.....

That 1.5GHz Pentium M of yours is hampered by:

1. 400fsb (The P-4M is at 533MHz, the P4 at 800MHz) - This means the Centrino, even if it could use the memory bandwidth (say for its SSE/2 and FPU units) is severely crippled in comparison.
2. Weak memory support (single channel, DDR266)
3. L1/L2 architecture is P!!!-like, not P4 like (No trace cache, lower speed L2 cache)
4. Over-aggressive Speed Step usage (Theme of Pentium M is to save energy at expense of performance)
5. Weak periphery (Currently no choice for desktop hard drives; CPU-dependent integrated modem, sound, and NIC; poor video cards available; etc)

C'mon, if Intel uncorked the Pentium M by adding 800fsb, gave it true P-4 socket interchangability (where it would pick up dual-channel and DDR400 support), gave it performance ratings that aggressively compared it to the competition, added trace cache and P4-esque L2 design, and actually true desktop support then people would find it useful. It lacks raw MHz, so Intel's percieved desktop edge over the competition cannot be used, making this CPU unworthwhile for the marketeers as a weapon of choice in the future desktop wars.

Somoeone should slave an external SCSI hard drive array to their Pentium M system to see how much a set of 15k drives in RAID would give it a boost.

 

[arcenite]

Originally posted by: Insomniac

Originally posted by: Wingznut

Originally posted by: mikecel79

Originally posted by: Wingznut

Originally posted by: mikecel79

Originally posted by: Rectalfier
Here's the way I see it. Intel has already delayed the Prescot due to heat issues. I think they are trying to work around this by increasing the pipeline. It just seems that Intel has done this all of a sudden, since they couldn't produce their original Prescott design, and have thus delayed it. Could this be a massive trip for Intel? The release of the Pentium 4EE sure seemed like an act of desperation. Rest assured, Intel will have something good for us again, but it looks like 2004 will be AMD's biatch.

Extending the pipeline isn't something that can be done in 3 months. They would basically have to go back to the drawing board to extend the pipeline. This was planned well before Prescott was delayed. They couldn't do that in 3 months or even 6 months. Extending the pipeline would not help it deal with heat either.

Best post so far.

Thanks. I don't think people have any idea how much work and engineering goes into making a processor. I don't think even I know how much work goes into it. So many people talk about just extend the pipeline or just ramp it up 1Ghz. I don't think they realize how much validation and testing goes into a CPU before it's released.

I know... The first thing I thought about when I read aRCeNiTe's post (about raising the A64's clockspeed by 1ghz) was, "Hell, why stop at 1ghz??? They should just go for double the current clockspeed."

But, but, all you have to do is increase the multiplier. I can't believe it never occurred to AMD to do that. No wonder Intel controls the CPU market.

I am sure you know what I meant... Per clock, AMD is obviously ahead of the game.

 

[Wingznut]

Originally posted by: aRCeNiTe
I am sure you know what I meant... Per clock, AMD is obviously ahead of the game.

But, just like actual clockspeed, "per clock" is only part of the equation.

 

[Duvie]

Originally posted by: MadRat
The longer pipeline gives off more heat? Interesting. If Prescott really introduced 10 stages then perhaps this is the problem with thermal disipation. I still tend to believe the theory that its a 2-stage increase, relative to adding the trace cache...

Originally posted by: zephyrprime
Some have pointed out that the current P4 has a 28 stage pipeline if you include the stages before the trace cache and after the point where branches are determined.

If it was a simple 2 stage increase then Intel has some excellent results from 90nm sampling to believe 4GHz is possible by year's end! Who cares if thermal disipation increased to 90-100W for a 4GHz processor when raw computational power will have increased a shocking high 25%! People will dump their 3GHz P4's by the droves if this is the case. Remember that gaming and media encoding performance has steadily improved with Intel's clock increases. These processes don't have as many snarls and performance traps as other programs seem to run across, making a 4GHz processor seem awfully advantageous, in these respects, over a 4000+ PR rated processor.

Originally posted by: Duvie

Originally posted by: Ionizer86
In a slightly different light, we can kind of think of P-M as P3 reengineered. Some pretty awesome IPC they've got there with a 1.6/1.7GHz chip blowing away P4's that are clocked like 600MHz to 800MHz higher.

i dont know about blowing away...look behind the intel hype on them.....I have the 1.5 version centrino w/ 1mb of l2 cache and only a handful of things did it do faster then a 2.0ghz p4 I tested...Most of the time it was rather poor and acted like a 1.5-1.8ghz cpu.....Then throw it in a laptop that sucks it down a couple hundred mhz in comparison to a desktop PC with no dual channel DDR, 4200rpm harddrives, etc.....

That 1.5GHz Pentium M of yours is hampered by:

1. 400fsb (The P-4M is at 533MHz, the P4 at 800MHz) - This means the Centrino, even if it could use the memory bandwidth (say for its SSE/2 and FPU units) is severely crippled in comparison.
2. Weak memory support (single channel, DDR266)
3. L1/L2 architecture is P!!!-like, not P4 like (No trace cache, lower speed L2 cache)
4. Over-aggressive Speed Step usage (Theme of Pentium M is to save energy at expense of performance)
5. Weak periphery (Currently no choice for desktop hard drives; CPU-dependent integrated modem, sound, and NIC; poor video cards available; etc)

C'mon, if Intel uncorked the Pentium M by adding 800fsb, gave it true P-4 socket interchangability (where it would pick up dual-channel and DDR400 support), gave it performance ratings that aggressively compared it to the competition, added trace cache and P4-esque L2 design, and actually true desktop support then people would find it useful. It lacks raw MHz, so Intel's percieved desktop edge over the competition cannot be used, making this CPU unworthwhile for the marketeers as a weapon of choice in the future desktop wars.

Somoeone should slave an external SCSI hard drive array to their Pentium M system to see how much a set of 15k drives in RAID would give it a boost.

I know that...I am well aware the limitation that have always existed in laptops in terms of older chipsets, slower devices.....However the claims intel makes at its site regarding p4m performance to that of desktop cpus several hundred mhz ahead if not ran on a laptop is obviously flawed....the P4m is the mobile chip and except a handful of enthusiast who buy one and throw it in a desktop it is meant to be that of a laptop....

I am not arguing taking that added cache and the other features and coupling it with the superior DC DDR chipsets of now, and 800mhz bus, and 15000rpm scsi would not be tremendous...the fact is any claim made by intel about this chip need to be in context of a mobile platform...

Are there mobile platforms with DC DDR chipsets and 10000rpm drives like my raptor?? I don't think so and since intel has had those misleading charts on their website for over 8 months now I am certain a platform like I listed above did not exist at the time or I would have had it.....

 

[Idoxash]

AMD and Intel both needs to rethink their ways. I think it's about stupid for cpus to suck that much dang energy and throw it at really nothing. Some of you will argue that it's good for games ect... but that's crazy no games really needs that what needs to improve for games is the software, the coders that makes the games, the gpu, vpu, wutever you wish to call them. Heck even thoes are becoming crazy now days with the power they need just so you can have few fps more. To be honest VIA is doing it all right. Their new cpus might not be power house but they dang won't cook ya your steaks. Their new gpu chip looks dang nice too. It maybe not the best out there but what it eats up a few watts? Guess it's not just AMD and Intel thoe huh. I hear that IBM is got a hot chip too with that PPC 970. I must say thoe my older Duron 800 mhz with a small heatsink and a 2000 rpm fan only get's as hot as 80F.

--Idoxash

 

[Wingznut]

Originally posted by: Idoxash
AMD and Intel both needs to rethink their ways. I think it's about stupid for cpus to suck that much dang energy and throw it at really nothing. Some of you will argue that it's good for games ect... but that's crazy no games really needs that what needs to improve for games is the software, the coders that makes the games, the gpu, vpu, wutever you wish to call them. Heck even thoes are becoming crazy now days with the power they need just so you can have few fps more. To be honest VIA is doing it all right. Their new cpus might not be power house but they dang won't cook ya your steaks. Their new gpu chip looks dang nice too. It maybe not the best out there but what it eats up a few watts? Guess it's not just AMD and Intel thoe huh. I hear that IBM is got a hot chip too with that PPC 970. I must say thoe my older Duron 800 mhz with a small heatsink and a 2000 rpm fan only get's as hot as 80F.

So.... You're saying that an 800mhz Duron or a Via C1 is sufficient to run all of the latest software?

 

[dexvx]

FYI, the Pentium-4 Northwood is a 28 stage design.

 

[Idoxash]

To be honest thoes cpus do ok on their own and yes my old 800 mhz does dang good by the way. The duron is run with ddr ram (333mhz) even thoe the cpu it self might not have the bus of that speed I asure you it does really well for it's age. Runs most games ok with a decent video card and most software. What your trying to do is pin me in by saying it can't run software and games. However what your loseing here is the fact 90% of computers are not market to ppl that goes nutz over making movies, audio, 3d anime, ect ... Their marketed to ppl that uses them for basic stuff with the pre install software. NO ONE needs the kind of power you have now or in the future for them things. So why pay all that money for that kind of power and have a hot running computer that sucks more energy then a dang oven? When I sed VIA was on the right track what I ment was VIA may not have any speed crown on top of their heads but their tech has been proven to be cool running, cheap, easy on the power bill, and in the end does dang good for 90% of the users that uses computers these days.

If you wish to throw 1000s of dollars into a computer that you will not use 30 or so % of the power it can give then not only is that wasted money but it's wasted energy and in the end it let's INTEL, AMD, other chip makers continue to think like mad men and put our ever increasing energy draining costly computer parts. However thoes are my opinions and we all have our own choice I just voiced mines that's all. I understand some ppl needs that power and kewl for them.

--Idoxash

 

[Idoxash]

Also because VIA is spending their time and money in making great lill chips that does dang good for what there are and use very lill energy it means that in the future VIA will only get better and able to improve enought they can make even more powerful chips that runs as cool as their older ones. INTEL, AMD, spends their money on making the most powerful chips but forgeting that they going to need some 220 voltes in the near future.

--Idoxash

 

[Wingznut]

But what you don't realize Idoxash is that "90% of the users that uses computers these days" don't use the top of the line cpu. People on these msg boards (for the most part) are hardcore enthusiasts who aren't interested in Durons or C1's. They want to talk about the latest and greatest.

Not so different from car enthusiast boards... Who wants to talk about the latest Huyndai? After all, 90% of the people on the road don't need anything more than a 120hp car right?

 

[Idoxash]

You have a very good point there but what I probly forgot to stress my comments not aim at you guys there kinda aim at Intel and AMD. Would you love it if them two companys can make a really small powerful chip that sucks 20 or less watts? Yah everyone would but the fact is these compansy won't ever do it because there to worry about one another coming out on top. In the end when thoes poor normal ppl goto HP, Dell, wutever and buy a computer they know one thing "mhz, ghz" and "intel/amd" and they look for the fastes they can afford but wut they dun know is they dun need the power and it sucks more energy then their ovens and because of this Intel/AMD won't ever improve their chips for energy/cool running. Thoes are my points.

--Idoxash

PS) Once again my opinions were never ment to target the ppl of this forum I know what this forum is about as sed there about Intel and AMD not improving where it counts the most.