Pentium M beating P4 & FX?...????

[jon_k]

http://www.tomshardware.com/cpu/20050525/pentium4-10.html

I was pretty shocked when I read this and I almost find it impossible to believe. Apparently ASUS came out with a kit that allowed you to put a Pentium M processor into your desktop. At first I was like, "yeah yeah the Pent M is mobile and peforms like crap," but apparently putting it in a board with a decent chipset and memory makes one helluva difference! They show this baby beating the almighty Athlon 64 FX series and completely trashing the P4 series. Not only that, but it produces an EXTREMELY low amount of heat since it was designed for notebooks. Watercool this sucker, and you could achieve some INSANE overclocks.


Intel.... wtf were you thinking. Your could have owned the desktop market with this techlology.


Comments?

 

[liquidtrance123]

Yea they're great performers, even at 533FSB, but i hear when they jump them up to 800FSB and add all the P4 features it'll probably jump the heat up alittle more not sure by how much tho. It looks like the pentium3 architecture that the Pentium M is based on is the way to go

 

[Scali]

Intel.... wtf were you thinking. Your could have owned the desktop market with this techlology.

You know what they say about hindsight
Fact is that Intel DID own the desktop market with the P4 technology for a long time (and actually still does, even though it's no longer the fastest CPU).
Back in those days, the P3 had trouble keeping up with the Athlon, and Intel decided to design their new CPU to make sure it would scale up to incredible clockspeeds...
But it didn't quite work out that way, because 90 nm had some unpleasant surprises.
And since the P4 cannot reach the clockspeeds that were originally intended, the older architectures can catch up.
The question is now, where is the 'sweet spot'? The pipeline of the P4 is obviously too long at this moment, since it is physically impossible to reach the speeds it was designed for (somewhere around 5 GHz at least, Intel was even hoping to reach 10 GHz). The P3 pipeline is too short to go past 3 GHz... So what do you do? Stick to the P3's pipeline, and try to get more speed from extra cache etc? Or try to make the pipeline longer so it will reach higher clockspeeds, just not as high as the P4?
I wonder what Intel is going to do. We now only know that the P3/Atlhon pipelines are currently better than the P4 pipelines for most applications... we don't know on which end of the balance the P3/Athlon are... will you get more performance if you make their pipeline longer, or shorter? Or is it actually at the sweet spot already?

 

[jon_k]

I honestly think they should just ditch the P4 architecture outright. First thing Intel needs to do is get a mainboard out to the market that will accept Pentium M CPUs natively without the ASUS upgrade kit. Secondly, they need to start fine tuning the Pentium M for the desktop. The Pentium M has very low heat and voltage requitements, so there is a lot of room for performance improvement in the desktop arena. I mean, if Tomshardware can put up some jerry-rig Pentium M system that get's this kind of performance, I expect Intel can fine tune it to get about 25% more speed out of it.

 

[liquidtrance123]

If intel drops the P4 line and goes with the PM line they've still got maybe another 2 years before they max out the Pentium M architecture the way its built now. I'm talking just upping the cache, FSB, and instruction sets, hell maybe even HyperThreading. After they end up adding all these to the Pentium M they'll be at the end of the rope after reaching maybe 3ghz? Thats plenty performance out of a CPU however..... During this time they'll need to be really thinking hard on how to come up with another CPU architecture.

Bottom line: Going with the Pentium M will give intel the speed torch again, but they'll need to be thinking on what they need to do next. Hell maybe if they stick with the PM max it out with all the optimizations i mentioned above and add Dual and Quad cores with the right chipset.....maybe the Pentium M is the CPU of the future

 

[jon_k]

From what these benchmarks show, the PM is the fastest CPU on the market right now if you properly tweak it. You slap it in a multi core setup, and you're got your next line of processors for the next 3-4 years.

But you are right Intel needs to have something cooking for when the PM runs out of steam. I would say they ought to focus on improving single-threaded application performance with multi-core setups. Multicore is definatly the way of the future (all supercomputers these days runs lots of processors simultaniously - CRAY has one that has 1024 processors i think). It intel could come up with a technology called "Auto thread" or whatever that took single threaded apps and sent all the cores to work on them that would be awsome. Think SLI except for CPUs.

 

[Stugots]

I doubt that intel could turn the Pentium M into a 64bit chip, they couldnt do it with northwoods.

 

[Scali]

I honestly think they should just ditch the P4 architecture outright.

They will, just not now (they just released a 6xx at 3.8 GHz, which is still beyond the performance level of even overclocked Pentium-Ms).
And they won't use Pentium-M, but a more future-proof next-gen architecture. One that is able to last for another 4 years or so, like the P4 itself, and the PPro architecture before it.

 

[liquidtrance123]

they could probably add 64bit extentions but you gotta think after they add all this stuff its gonna probably just just as hot if not hotter than the prescotts, granted it'll be much much faster, but we're back to obscenely hot processors. Also if they add all these optimizations and things is it gonna degrade baseline performance?

 

[jon_k]

They will, just not now (they just released a 6xx at 3.8 GHz, which is still beyond the performance level of even overclocked Pentium-Ms).
And they won't use Pentium-M, but a more future-proof next-gen architecture. One that is able to last for another 4 years or so, like the P4 itself, and the PPro architecture before it.

from what i've seen the new 3.8ghz falls short of the Athlon FX series which the Pentium M is able to beat. They could add stuff like cache and minor architecture improvements and keep the Pentium M series competitive on the desktop market for at least 2 or 3 years.

 

[liquidtrance123]

you also gotta think about the only way the Pentium M can beat them all right now is to hit clock speeds over 2.5-2.6Ghz. Technically the highest clocked Pentium M right now i think is the 2.13Ghz which is still nothing to scoff at. However in order to take the crown from even the AthlonFX is to come out of the gate full force and jump the clock speeds up at least to 2.6Ghz with maybe SSE3 and Hyperthreading. Otherwise, at stock speeds, the AthlonFX still takes the cake.

 

[jon_k]

From what the review said, Tomshardware hit 2.6 GHZ with ease. Stock cooling and everything.

edit- not only that, but at stock speeds the Pentium M 770 basically performed the exact same as the FX 53. With the exception of Quake III (which an outdated benchmark), it beat all the P4 in every single game benchmark at stock speeds.

 

[Scali]

from what i've seen the new 3.8ghz falls short of the Athlon FX series which the Pentium M is able to beat.

In some game benchmarks perhaps... but not everywhere. The P4 is still king in multitasking and audio/video encoding, for example.

They could add stuff like cache and minor architecture improvements and keep the Pentium M series competitive on the desktop market for at least 2 or 3 years.

This is Intel we're talking about. not AMD. They are not supposed to patch up old processors in the hope of remaining competitive for another few years. They are supposed to design revolutionary new architectures and be the performance standard.

 

[jon_k]

please don't turn this thread into a flame war. I'm having a good discussion.

As to your points Scali, yeah, with multithreading the P4 is definatly on top, but I am mainly a gamer so I head straight to the game benchmarks . But I would imagine that the Intel engineers could implement Hyperthreading into the M series just like they did with the P4. I'd also imagine they could come up with a dual core version as well.

 

[jon_k]

omg, wtf is up with this thread? I said this exact same thing in a couple of threads a week or two ago, and all the intel fanb0ys ran all over me.

Me : "In my opinion intel should have ditched netburst for pentium-m a long time ago, especially for dual cores since the p-m is so much more energy efficient"

Multiple Intel Fanb0ys : "Prescott is better than pentium-m, I think intel and their 3000 engineers know better than you!"

Now this guy comes along, very late I might add, praising the pentium-m and everyone falls all over him......its a conspiracy against me isn't it?

One reason might be becasue these benchmarks were just released a few days ago and you said you brought up the topic a few weeks ago.

 

[Erasmus354]

I dont think Intel should waste the time trying to add hyperthreading to the pentium-m. First of all there is the fact that of the shorter pipeline in the pentium-m. This means that the small performance gains from hyperthreading are even smaller. Meaning you are investing a lot of man hours into something that is of very little benefit.

Intel should concentrate harder on making a dual core pentium-m. That would provide much more benefits as well as giving their dual core chips a fighting chance at competing with AMD's dual core chips. (As we all know Intel simply cant compete with AMD in dual core atm thanks to their heat limitations)

 

[Erasmus354]

One reason might be becasue these benchmarks were just released a few days ago and you said you brought up the topic a few weeks ago.

Look back, toms did articles on this awhile ago, not to mention the adapter has been out for some time, and someone already set a Pifast 1m world record with a pentium-m. There have been benchmarks showing that the pentium-m is a very good architecture for quite some time, these new benchmarks dont change anything.

 

[Scali]

I dont think Intel should waste the time trying to add hyperthreading to the pentium-m. First of all there is the fact that of the shorter pipeline in the pentium-m. This means that the small performance gains from hyperthreading are even smaller. Meaning you are investing a lot of man hours into something that is of very little benefit.

Please explain why IBM uses SMT on their PPCs then.

 

[lithium726]

erasmus is right, this has been discussed many a time - and while the pentium-m is a great performer in gaming and other specified application, it falls dismally to the ground in things like SIMD operations (SSE1/2) and makes it weak in streaming apps and encoding (which is the very thing intel has been known for in teh past 4 years). it is not ready to be a desktop CPU yet, which is why intel has not made it a desktop CPU yet... i feel like we're beating a dead horse here - the pentium-m is NOT the end all solution to intel's problems, but it is a step in the right direction.

they could probably add 64bit extentions but you gotta think after they add all this stuff its gonna probably just just as hot if not hotter than the prescotts, granted it'll be much much faster, but we're back to obscenely hot processors. Also if they add all these optimizations and things is it gonna degrade baseline performance?

no. the penitum-m has had power saving optimizations added to it that cannot be added to the pentium4 due to architechtural differences. seeing as how the pentium4 had twice the TDP rating of the highest end pentium3 when they came out, i severly doubt the pentium-m will ever be able to get as hot as a prescott.

This is Intel we're talking about. not AMD. They are not supposed to patch up old processors in the hope of remaining competitive for another few years. They are supposed to design revolutionary new architectures and be the performance standard.

intel does not always define the performance standard - AMD had them bested in teh thunderbird days(vs. coppermine, tualatin and willamette), and at the current moment (on the desktop segment, intel rules hte mobile world) and intel had AMD bested in most of hte northwood days.

Intel should concentrate harder on making a dual core pentium-m. That would provide much more benefits as well as giving their dual core chips a fighting chance at competing with AMD's dual core chips. (As we all know Intel simply cant compete with AMD in dual core atm thanks to their heat limitations)

i whole-heartedly agree.

I'm talking just upping the cache, FSB, and instruction sets, hell maybe even HyperThreading.

none of those things would benefit the pentium-m in any sort of amazing way.
1. the pentium-m does not need huge amounts of cache to keep competitive, its pipelines and parallel architechture dont require it - unlike the pentium4 and its one long pipeline, which needs to be flushed and reloaded in the instance of a miscalculation.
2. due to the large amounts of cache already present on the dothan core (2m) and simply to its architecture, it, like its pentium-3, 2, and pro ancesters before it, are not bandwidth limited. most of the bandwidth limitation for the dothan is alleviated by that huge 2mb l2 cache, and as benchmarks show, the 533FSB on dothan does very little for perforamnce, while hurting battery life.
3. theyll need something other than SSE (SIMD) instruction sets for that, as the pentium-m performs dismally here
4. the pentium-m cannot support hyperthreading, its pipeline just isnt long enough to warrent such a technology. unlike the pentium4, where the pipeline is long enough to hold two threads at once(28 stages for NW, 31 for prescott), the p-m just cant handle that, with its 10 stage pipeline.

Please explain why IBM uses SMT on their PPCs then.

please explain the implementation of both. i am willing to bet it is much different than intel's hyperthreading.

and if it is the same, please tell me why AMD hasnt implemented this techonology onto their chips, as intel and AMD share patents, and AMD has access to the technology.

 

[savantu]

There are a few smarty-pants around here , who shout that Intel should ditch Netburst and HT.Sadly , their arguments reflect either a low IQ or shallow understanding of the principle "how can I do more with less "

Intel analyzed in the mid '90s the trend in software and hardware and came to a simple conclusion :

-IPC is limited and hard to get , close to 0,6-0,9 for most programs , far from the theoretical 2-3 chips could do.
-most of the applications of the future will involve multimedia stuff

The answer was simple : Netburst.It embodied the best of 2 worlds : excellent for multimedia and easy to scale frequency wise.Why bother with extracting IPC ? It is easier to just build up the frequency.

Netburst handled x86 in a very elegant way , just look at how many execution units a P4 has compared to a K7/8.Less than half and close to a third if you don't like to check it.

Process improvement couldn't keep up with Netburst so the architecture is now stuck.Maybe on 65nm or 45nm an improved Netburst variant will likely rule again performance and frequency wise.

Anyway , the future is definetly based on things learned from Netburst's revolutionary aproach : Cell is over 4Ghz , Power 6 will also be over 4 Ghz.

Pentium M is a combination of Netburst and P6 , but is inherently flawed in many ways which make it less than ideal for desktops right now.Yonah will change that.

 

[Scali]

intel does not always define the performance standard - AMD had them bested in teh thunderbird days(vs. coppermine, tualatin and willamette), and at the current moment (on the desktop segment, intel rules hte mobile world) and intel had AMD bested in most of hte northwood days.

I never said they did, I said they should, given their size and resources.

please explain the implementation of both. i am willing to bet it is much different than intel's hyperthreading.

There's not all that much room to vary the implementation of SMT really. The basic idea is to just have a second set of whatever it requires to run a thread (program counter, selector registers, etc). Also, even if there are slight differences (priority mainly), I see no reason for Intel to just ditch HyperThreading. It makes much more sense to just develop it further, and if possible, take a leaf out of IBM's book.

and if it is the same, please tell me why AMD hasnt implemented this techonology onto their chips, as intel and AMD share patents, and AMD has access to the technology.

For the same reason why Intel doesn't have on-chip memory controllers and hypertransport etc yet: they didn't get round to it yet, their competitor just applied it a generation earlier. These things are not trivial to just add to a processor, you need a redesign of a large part of the core. Therefore it doesn't make sense to try and patch it into a current product. But I am quite sure that AMD will eventually follow suit... They are going to have to lengthen their pipelines aswell, eventually. You can't remain competitive at ~2.6 GHz forever. I bet that will also mean they will adopt Netburst's tracecache,

 

[SSE4]

I think I remember there being an interview with Webber where he said they aren't impressed with Hyper-Threading and don't intend to implement it. At least not for a long time.

Although the Pentium-M being a great performer is no secret, they're working on a brand new core. Will it have Pentium-M aspects? Most likely. We'll just have to see with Merom and Conroe.

 

[Erasmus354]

Netburst handled x86 in a very elegant way , just look at how many execution units a P4 has compared to a K7/8.Less than half and close to a third if you don't like to check it.

Process improvement couldn't keep up with Netburst so the architecture is now stuck.Maybe on 65nm or 45nm an improved Netburst variant will likely rule again performance and frequency wise.

Anyway , the future is definetly based on things learned from Netburst's revolutionary aproach : Cell is over 4Ghz , Power 6 will also be over 4 Ghz.

Pentium M is a combination of Netburst and P6 , but is inherently flawed in many ways which make it less than ideal for desktops right now.Yonah will change that.

1) I prefer to say that netburst handled x86 in a very primitive ala "brute force" way while the other solutions are the "elegant" ones.

2) Intel is taking a few things from netburst, but they have largely realized that there are fundamental flaws to its approach with netburst, and is therefore making an entirely new core.

3) I fail to see how cell relates at all to netburst

4) Pentium M is actually more of a fine tuning of the P3 than anything else

 

[Scali]

1) I prefer to say that netburst handled x86 in a very primitive ala "brute force" way while the other solutions are the "elegant" ones.

Actually, AMD is the primitive and bruteforce one here. As said, the P4 has very little execution units really. Only 2 ALUs for example, and one FPU/SIMD unit. Even less than the P3 had, actually. AMD has 3 ALUs, 2 FPU/SIMD units and such.
Also, you can't really deny the elegance of tracecache, can you?
And the double-pumped ALUs were quite sophisticated aswell.
The only thing about the P4 that is bruteforce, is the clockspeed, really.

3) I fail to see how cell relates at all to netburst

It scales to reasonably high clockspeeds, far more than a P-M or Athlon64 for example.

 

[FanATIc]

I dont think Intel should waste the time trying to add hyperthreading to the pentium-m. First of all there is the fact that of the shorter pipeline in the pentium-m. This means that the small performance gains from hyperthreading are even smaller. Meaning you are investing a lot of man hours into something that is of very little benefit.

Intel should concentrate harder on making a dual core pentium-m. That would provide much more benefits as well as giving their dual core chips a fighting chance at competing with AMD's dual core chips. (As we all know Intel simply cant compete with AMD in dual core atm thanks to their heat limitations)

Im quite sure the dothan is based on a longer pipline but fewer of them, compared to a P4. The Athlon 64 series work this way. Longer pipelines, less clocks but more work per clock. Shorter pipes would mean you'd have to constantly kick up the mhz such as how the P4 works.

 

[SSE4]

Im quite sure the dothan is based on a longer pipline but fewer of them, compared to a P4. The Athlon 64 series work this way. Longer pipelines, less clocks but more work per clock. Shorter pipes would mean you'd have to constantly kick up the mhz such as how the P4 works.

Longer pipelines allow higher core frequencies, but also require more steps. Since each instruction must go through these steps, it can take longer for an instruction to be executed. The Pentium 4's have increased the pipeline length and increased the number of steps (21 for the Northwood to the 30-something of the Prescott I think) in the pipeline itself. The Dothan and AMD64, although having to run at lower frequencies because of the shorter pipeline, take less time to execute their instructions, and therefore get work done faster.

 

[Scali]

The reason for the longer pipelines at high clockspeeds is that transistors cannot be made faster. Transistors require a certain 'propagation time' to take on a new state. This propagation time hasn't decreased much over the years, so you need other ways to increase the clockspeed.
Each stage in the pipeline is a set of operations that has to complete in at most one clockcycle. The higher the clockspeed, the shorter the clockcycles are. Which means that you have to reduce the amount of work done in a single cycle. So you cut up the workload in more stages. The problem with the P4 is that they used so many stages that the clockspeed could easily go well beyond 5 GHz... but they cannot make CPUs that fast, because the transistors 'leak' at 90 nm. This is because the transistors are now so small, that the insulation between transistors is now only a few atoms thick, which allows electrons to pass through the insulation.
The result of this is that a larger current flows through the CPU, even when it is not being used. And this current causes the CPU to heat up. And heat is the transistor's worst enemy. The hotter it gets, the higher its propagation time becomes.
And that's why the P4's pipeline is 'too long'.

The Athlon64 and the P-M face the opposite problem. Their pipelines are designed to operate at a maximum speed of about 2.5-3.0 GHz, and that limit is pretty much reached. They won't be able to run at higher clockspeeds unless either transistors can be made faster, or the pipeline is extended.

 

[Dan_D]

The reason for the longer pipelines at high clockspeeds is that transistors cannot be made faster. Transistors require a certain 'propagation time' to take on a new state. This propagation time hasn't decreased much over the years, so you need other ways to increase the clockspeed.
Each stage in the pipeline is a set of operations that has to complete in at most one clockcycle. The higher the clockspeed, the shorter the clockcycles are. Which means that you have to reduce the amount of work done in a single cycle. So you cut up the workload in more stages. The problem with the P4 is that they used so many stages that the clockspeed could easily go well beyond 5 GHz... but they cannot make CPUs that fast, because the transistors 'leak' at 90 nm. This is because the transistors are now so small, that the insulation between transistors is now only a few atoms thick, which allows electrons to pass through the insulation.
The result of this is that a larger current flows through the CPU, even when it is not being used. And this current causes the CPU to heat up. And heat is the transistor's worst enemy. The hotter it gets, the higher its propagation time becomes.
And that's why the P4's pipeline is 'too long'.

The Athlon64 and the P-M face the opposite problem. Their pipelines are designed to operate at a maximum speed of about 2.5-3.0 GHz, and that limit is pretty much reached. They won't be able to run at higher clockspeeds unless either transistors can be made faster, or the pipeline is extended.

That is by far the best explanation of what is wrong with Netburst. By wrong I mean with the 90nm process and the Prescotts heat issues. Just because its different and has to run at twice the clock speed as an A64 to get the same performance, isn't necessarily a bad thing. We view it as such typically. Running a Pentium 4 at 4GHz is just fine. If it can do it.

Its just like with cars, you can make alot of power with a 4 cylinder or a V8. The difference is in how you accomplish this goal. The end result is the same. Of course each approach has natural advantages and disadvantages.

 

[Scali]

That is by far the best explanation of what is wrong with Netburst.

But it seems like I'm the only one who sees what's 'wrong' with the Athlon64.
In the case of the P4 the architecture is not the limit, the manufacturing technology is. Intel has some tricks up their sleeve, so with 65 nm it could be possible that the P4 breaks that 4 GHz barrier after all (they have already demonstrated 65 nm dual-core Xeons at 3.73 GHz and 1066 MHz FSB). It could also be possible that the P4 architecture will just be 'put on ice' for a while. The next architecture will probably have a shorter pipeline to get more of the clockspeeds that can be reached today... But once the manufacturing improves enough, they may well bring the P4 back (just like they brought the P3 back for the P-M architecture?).

The Athlon64 on the other hand is pretty much at the end of its lifecycle. AMD will have no choice but to come up with a new design. The scary thing is that we have not heard anything at all yet about a new architecture from AMD. So that could mean that they cannot hold on to their current lead much longer.
The P-M is also pretty much at its end, but it will be replaced by the same architecture as the P4 next year, so no need to worry there

 

[robberbaron]

B
The Athlon64 on the other hand is pretty much at the end of its lifecycle. AMD will have no choice but to come up with a new design. The scary thing is that we have not heard anything at all yet about a new architecture from AMD. So that could mean that they cannot hold on to their current lead much longer.
The P-M is also pretty much at its end, but it will be replaced by the same architecture as the P4 next year, so no need to worry there

You don't know how well the next Intel architecture will perform, and I'm sure that AMD also has tricks up their sleeve. There seems to be an artifical clockspeed limit on the newest chips, except for the FX's that hit disproportionately higher numbers. This would make sense, as it would be useless for AMD to blow its load all in one "generation." By gradually releasing higher performing chips, they can make an architecture last a very long time. Intel did the same thing with the Netburst. They managed to get good performance out of it and still drag it on for a long time.

 

[Scali]

You don't know how well the next Intel architecture will perform

It's safe to assume that it will perform better. Intel would be stupid to put a new architecture on the market that's worse than the current one. It would only damage their image. They can just keep the P4 on until they have something better.

There seems to be an artifical clockspeed limit on the newest chips, except for the FX's that hit disproportionately higher numbers. This would make sense, as it would be useless for AMD to blow its load all in one "generation." By gradually releasing higher performing chips, they can make an architecture last a very long time. Intel did the same thing with the Netburst. They managed to get good performance out of it and still drag it on for a long time.

Apparently you have not been listening. There is a hard limit on the speed of any architecture, related to the number of stages, and the maximum propagation time of these stages. AMD can't just release a 4 GHz Athlon64, it's just physically impossible (die shrinks aren't going to help either, since they won't decrease the propagation time). Sure, they may be able to squeeze a few hundred MHz out of it, and they may want to spread that out over a longer period of time, but that's just details. I'm talking about the bigger picture here.
The P4 is different, as we've seen... put enough cooling on it, and it will go past 6 GHz.

 

[chrisf6969]

4) Pentium M is actually more of a fine tuning of the P3 than anything else

Yeah, Pentium M is Pentium 3 refined with more up-to-date features.

Quad pump bus of the P4, bigger L2 cache, better power management, etc..

I want to see a Pentium M with an 800-1066 FSB, dual channel DDR, PCIe, etc.... and see what it can do.

All of the times I've seen them on desktop boards whether 855 or 865/875 with adapter, they've been maxing around 600 FSB. Maybe the new desktop 915 boards with 479 socket will fix that and allow you to lower multiplier and really crank up the FSB!

 

[Ollin]

Lemme be the first one to say................wow these chips can get really expensive really fast.

 

[Russ]

I heard (many months ago, perhaps 6) that some guy used this ASUS kit to get his Pent M to 5.0 Ghz...

 

[thecoldanddarkone]

Actually the limit is like 3.x on the asus thing. And it is the world record for pi 1 million.

 

[Obi_Kwiet]

Question: Are we ever going to see a Pentium V? A name change might be refreshing once and awhile.

 

[lithium726]

pentium 5 is too repetitive, which is i think why they are going with pentium D

 

[robberbaron]

pentium 5 is too repetitive, which is i think why they are going with pentium D

Yeah, but the D is just two pentium 4's. The Pentium 5 would have to be a new architecture, not two Prescotts on a die.

 

[pxc]

Question: Are we ever going to see a Pentium V? A name change might be refreshing once and awhile.

Merom is the next totally new core, not based on Netburst (P4) or the P-M. Strangely, it's a multi-core EM64T mobile chip.

The name Pentium V would be kind of funny though. five five

 

[Dan_D]

It's safe to assume that it will perform better. Intel would be stupid to put a new architecture on the market that's worse than the current one. It would only damage their image. They can just keep the P4 on until they have something better.

Here's my various thoughts on the subject(s).

Well, while I am inclined to agree, Intel has released a worse architecture to succeed a previous processor. The Pentium 4 @ 1.4GHz was handily thrashed by Tualatin Pentium III's at 1.13GHz. Plus the Tualatins were SMP capable. The Pentium 4 wasn't. Sure after a long term evolution and the revamped socket 478 Northwood came out, things changed.

The physical MHz barrier of the processor cores is exactly why everything is going multi-core. They have to do it. Its finally getting too hard to ramp up clock speeds. While the implimentation of the AMD 64's dual core processors is better than two processors "stuck" together, which is what they laugh at Intel for, is pretty much what a A64 X2 really is. At least in practicality.

Sure the bridge between the cores, is a more elegant solution, but still the result is the same.

I think we won't see anything truely innovative anymore for awhile. The architectures won't evolve the same way we are used to seeing. Look at the Pentium 4, the suckers had about the longest life of any core, since the Pentium Pro (P6 Micro-architecture). Which of course still works today in some way as even a strong performance processor in its own way.

Intel has a funny way of rehashing old ideas in a way that works well. Not to mention, milking an architecture for along time, while gaining alot of performance out of it. The P6 core went from 150MHz all the way to 1.4GHz and now with the Dothan and Banias, they've gotten to 2.13GHz. Granted the Dothan and Banias really have alot of differences and quite a few improvements over the previous incarnation of P6.

Look at Pentium 4, from 1.3GHz to 3.8GHz and now with multi-cores. Not bad really. With new manufacturing technologies like 65nm it could go closer to that 10GHz mark they said it could hit. Which % wise would be the same leverage they got out of the P6 core.

AMD on the other hand has had to make more drastic changes to keep up. While I must say the Athlon 64's memory design is truly the most innovative thing I've seen in processors in the last 8 years I've been in the business.

The Opterons scalability is amazing. I think that the Athlon/Opteron 64 might just have alot more innovation and life left in it. Even if it might not scale as far as Intel can make their chips. Memory performance alone can carry that processor along way. If anything 90nm parts are shaping up to be quite the OC'ers. The clock speeds, should scale well up to 3GHz and maybe beyond on air cooling. I think they can do it.

Really until the Athlon 64 came out I've never been impressed with AMD. I must say the Opteron platform as a whole impresses me even more. The way they''ve designed scalability into the whole platform is amazing. Wheras Intel included SMP as more of an after thought because higher end machines had more CPU's and there is a market for that in x86.

Oh well I am done ranting on about various processor things.