Conroe's true 64-bit processors ?

curvey · 23 Jun 2006 at 21:32

My question may seem a little odd, but i was told to make sure these processors have true 64-bit achitecture (core or something) - otherwise there just 32-bit processors.

Could someone point me to the right direction ?

I understand they have *EM64T 64-Bit Technology* but thats all it says.

Also, previous LGA775 cpu's are 32-bit ? then how come the Conroe 64-bit cpu's are also socket LGA775 ?

Any help is much appreciate

Cheers.

str · 23 Jun 2006 at 21:38

EM64T is Intel's version of the 64-bit tech:

http://en.wikipedia.org/wiki/EM64T

Vertigo · 23 Jun 2006 at 22:27

Basically Intel had to make a major U-turn over 64-bit "on the desktop" and support AMD's 64-bit extensions.

Of course they couldn't bring themselves to use any of AMD's terminology when describing their extensions, despite them being forced to support the same instructions, so they called them EM64T and made no reference whatsoever to AMD when launching the new processors. This actually drew a fair bit of press at the time, the general consensus being that this was a very low thing for Intel to do and showing some humility and acknowledging AMD's lead in this area could well have ingratiated them with a lot of people.

As for the LGA775 socket, this has supported 64-bit for ages as there have been P4 processors supporting the extensions for quite a while. It was S478 that was 32-bit only.

Stelly · 23 Jun 2006 at 22:28

Vertigo1 said:
Basically Intel had to make a major U-turn over 64-bit "on the desktop" and support AMD's 64-bit extensions.

Of course they couldn't bring themselves to use any of AMD's terminology when describing their extensions, despite them being forced to support the same instructions, so they called them EM64T and made no reference whatsoever to AMD when launching the new processors. This actually drew a fair bit of press at the time, the general consensus being that this was a very low thing for Intel to do and showing some humility and acknowledging AMD's lead in this area could well have ingratiated them with a lot of people.

As for the LGA775 socket, this has supported 64-bit for ages as there have been P4 processors supporting the extensions for quite a while. It was S478 that was 32-bit only.

Yes and Microsoft also use AMD's 64bit extensions

Stelly

Corasik · 23 Jun 2006 at 23:22

Every LGA775 processor had the potential to be 64bit... even the socket 478 Prescotts had the 'earliest' version of EM64T on die, quite a few people managed to get a look at prescotts under the microscope (literally), and realised that compared to northwood, it had a complete duplicate ALU. on Pentium 4, EM64T basically links up the two ALU's to form a single 64bit system. (However like Hyperthreading on early P4's.... intel disabled all the 64bit modes on early prescotts, and only added them when they decided it was time to do it)

The side effect on the way 64bit was 'bolted' into Netburst, was generally poor performance whenever running in 64bit mode.

However Conroe is a totally different design, its 64bit from the ground up, and doesnt need any 'tricks' to work in 64 bit mode. There appears to be little or no performance loss when running in 64 bit mode. The chips are generally more powerfull than AMD64's in both 32bit, and 64bit execution speeds.

It always surprises me that AMD get so much credit for the 64bit extensions of the X86 platform. While they had the guts to actually produce the first X86-64 processor, it wasnt exactly revolutionary. Their extensions closely mimic the transition from 16bit, to 32bit, where the AL/AH/AX type registers were extended to 32bits as EAX. AMD extended again as RAX.

That said, AMD didnt stop their, as they added a number of new general purpose registers, which can considerably improve the performance software written to make use of them. AMD did an excellent job implementing it, but it was evolutionary, and not revolutionary even so.

Some say Intel didnt think the world was ready for 64bit desktop processors, and at the time it was probably correct, the more sceptic would say that intel had spent too much money on IA64 to abandon it to the X86 processors so soon. Either way, Its good that AMD were able to give intel a firm kick up the ...., and force them back ontrack, and come up with Conroe

Vertigo · 23 Jun 2006 at 23:52

Stelly said:
Yes and Microsoft also use AMD's 64bit extensions

Stelly

And now history is repeating itself as Intel, publicly singing the praises of the new Conroe processors, are quietly admitting that the high clock speed Netburst architecture was a waste of time all along and AMD had it right with their lower clock speeds but higher work-per-cycle Athlons. Of course they'll never admit that.

Corasik · 24 Jun 2006 at 00:05

I would hardly say that netburst was a waste of time. The end of the P3 lifecycle, intel were struggling to break 1Ghz.. The 1.13Ghz P3 was a disaster which intel had to recall. Netburst got them back on track, with a processor that scaled up to 3.8Ghz.

During that time, intel put a lot of development dollars into massively improving the manufacturing processes, while the P4 core went through a few evolutions from Willemotte to Presler.

Now because their processes are good, and finally Intel seem to be focusing on the design and performance of their CPU's again, with Core 2 about to be released and the next 2 generations of Core already on the roadmaps. (Nelhalem, and Gesher)

Dolph · 24 Jun 2006 at 10:55

Vertigo1 said:
And now history is repeating itself as Intel, publicly singing the praises of the new Conroe processors, are quietly admitting that the high clock speed Netburst architecture was a waste of time all along and AMD had it right with their lower clock speeds but higher work-per-cycle Athlons. Of course they'll never admit that.

Well, that's one way to look at it

Of course the actual situation was that intel tried to be revolutionary with Netburst whereas AMD carried on with what was basically a copy of the P6 design with some additional execution units...

I love the way some people try to paint AMD as some plucky revolutionaries or something. They weren't, they carried on with the same old thing while someone else tried something revolutionary. It failed, and that was the reason AMD got ahead. Intel's gamble failed.

NathanE · 24 Jun 2006 at 13:22

Spot on there, Dolph.

K8 = K7 derived + IMC + HTT + AMD64.
K7 = P6 derived + better FPUs than a K6.
K6 = a hunk of junk really. Barely had OOE.

Think of the P6 and the K7/K8 as a regular car engine. Now think of NetBurst as one of those Honda VTEC engines... loves to rev' but doesn't actually have much torque any of the time. Still a valid and acceptable engine design however. Intel's problem was that they couldn't wring out enough clocks (uh, revs) to remain competitive.

NetBurst will be back. Probably sooner than we all think right now.

its 64bit from the ground up, and doesnt need any 'tricks' to work in 64 bit mode

So was NetBurst... underneath those macro-ops are micro-ops operating at 128-bit. It's the same with the K8. Processors have been 64-bit ever since MMX really. It's just only "recently" have they decided to expose more 64-bit power at the macro-op level. There have also been optimisations to the pipelines in modern processors to allow 64-bit instructions to complete in one clock cycle.

curvey · 24 Jun 2006 at 13:58

Corasik said:
However Conroe is a totally different design, its 64bit from the ground up, and doesnt need any 'tricks' to work in 64 bit mode. There appears to be little or no performance loss when running in 64 bit mode. The chips are generally more powerfull than AMD64's in both 32bit, and 64bit execution speeds.

Wow thanks for all the feedback, the above quote and others seems to some up the conroe being 64-bit all the way. I originally had s478 (3.4EE) and had to use an intel driver for this to run 64-bit applications - running 2 cycles for a 64-bit instruction.

Vertigo1 said:
As for the LGA775 socket, this has supported 64-bit for ages as there have been P4 processors supporting the extensions for quite a while. It was S478 that was 32-bit only.

I was'nt sure about this LGA775 always being a 64-bit socket, i have been living in the past for a long time with my s478.

Just ordered the new conroe - Thanks guys for the information.

Vertigo · 24 Jun 2006 at 17:51

Dolph said:
Well, that's one way to look at it Of course the actual situation was that intel tried to be revolutionary with Netburst whereas AMD carried on with what was basically a copy of the P6 design with some additional execution units...

I love the way some people try to paint AMD as some plucky revolutionaries or something. They weren't, they carried on with the same old thing while someone else tried something revolutionary. It failed, and that was the reason AMD got ahead. Intel's gamble failed.

Sorry but "revolutionary", utter rubbish imho.

Intel's only priority with the Netburst architecture was to produce a processor which clocked as high as possible. They already knew that most people associated clock speed with overall performance and so they decided to exploit this by designing an architecture which would offer over twice the clock speeds of the previous generation. AMD stuck with a tried and trusted architecture and focused on optimising and improving it rather than doing a radical re-design for no real reason. If it ain't broke, don't fix it.

I don't see anything revolutionary about the Netburst chips at all tbh. They brought a few minor improvements such as SSE3 to the party but nothing radical. Even the much anticipated move from Northwood to Prescott was a total flop, producing chips which were barely any faster yet produced ridiculous amounts of heat. Even when they went dual-core with the Pentium D, AMD's X2 processors were generally regarded as better performers.

AMD, OTOH, forged forward with their visions of 64-bit processors for the average consumer, convinced that this was the future despite Intel's insistence that 64-bit was for servers only and the average consumer had no need or desire for it. Intel was eventually forced to concede that AMD were right and humbly add support for AMD's 64-bit extensions into their processors.

I think that, over the past few years, AMD has proven itself to be a far more forward thinking and "revolutionary" company than has Intel. They've had the performance edge for several years now, both with single and dual core. With the advent of Conroe, I think Intel will retake this performance lead and probably hold on to it for quite a while but I still believe that they've taken more than a brief look at the technology of the AMD processors when designing theirs.

And before I get labelled as an AMD fan-boy, I should point out that I've had various processors from both Intel and AMD over the years, choosing each time on the tangible merits of each option rather than any brand loyalty. I'm currently running a P4 Prescott.

dafloppyone · 24 Jun 2006 at 18:02

competition = good

Corasik · 25 Jun 2006 at 01:09

Really... it is true that Prescott has twice the number of ALU's compared to P4, which is how intel converted P4 from 32bit, to 64bit integer operation. And its why the EM64T implementation on Prescott wasnt such a high performer.

Here's the details, and diagrams of the various netburst processors.

http://www.chip-architect.com/news/...or 64 bit processing (not for multithreading)

PS. Its a very early version of Prescott, the site writer clearly believed it was a set up above P4, to deserve the name P5, but the rest of the page has some good info.

As for LGA775, as with the previous socket 478, the FSB between motherboard and processor is already 64bits wide, and because its quad pumped, it can transfer 4x64bits on every clock cycle.

In every way Conroe is definatly a 'real 64bit integer processor', with an integrated 128bit floating point/sse units.

Dolph · 25 Jun 2006 at 07:31

Vertigo1 said:
Sorry but "revolutionary", utter rubbish imho.

Intel's only priority with the Netburst architecture was to produce a processor which clocked as high as possible. They already knew that most people associated clock speed with overall performance and so they decided to exploit this by designing an architecture which would offer over twice the clock speeds of the previous generation. AMD stuck with a tried and trusted architecture and focused on optimising and improving it rather than doing a radical re-design for no real reason. If it ain't broke, don't fix it.

I don't see anything revolutionary about the Netburst chips at all tbh. They brought a few minor improvements such as SSE3 to the party but nothing radical. Even the much anticipated move from Northwood to Prescott was a total flop, producing chips which were barely any faster yet produced ridiculous amounts of heat. Even when they went dual-core with the Pentium D, AMD's X2 processors were generally regarded as better performers.

AMD, OTOH, forged forward with their visions of 64-bit processors for the average consumer, convinced that this was the future despite Intel's insistence that 64-bit was for servers only and the average consumer had no need or desire for it. Intel was eventually forced to concede that AMD were right and humbly add support for AMD's 64-bit extensions into their processors.

I think that, over the past few years, AMD has proven itself to be a far more forward thinking and "revolutionary" company than has Intel. They've had the performance edge for several years now, both with single and dual core. With the advent of Conroe, I think Intel will retake this performance lead and probably hold on to it for quite a while but I still believe that they've taken more than a brief look at the technology of the AMD processors when designing theirs.

And before I get labelled as an AMD fan-boy, I should point out that I've had various processors from both Intel and AMD over the years, choosing each time on the tangible merits of each option rather than any brand loyalty. I'm currently running a P4 Prescott.

That's because you're looking at it knowing it's failed.... It's very easy to look back and say "Well, that was never going to work...." when you already know it didn't.

When the Netburst core was first announced and first launched, the architecture was revolutionary for a desktop processor, it was like nothing else that had been seen before. The fact that it failed doesn't do anything to change that. And yes, I am aware that in part it was demanded by the marketers to have high clock speed. That doesn't mean it wasn't a revolution in chip design at the time.

WJA96 · 25 Jun 2006 at 08:04

Would I be correct in thinking that none of this 64-bit stuff actually matters if I'm not running a 64-bit OS eg. Windows XP 64-bit or the Linux equivalent?

Vertigo · 25 Jun 2006 at 09:58

Dolph said:
That's because you're looking at it knowing it's failed.... It's very easy to look back and say "Well, that was never going to work...." when you already know it didn't.

When the Netburst core was first announced and first launched, the architecture was revolutionary for a desktop processor, it was like nothing else that had been seen before. The fact that it failed doesn't do anything to change that. And yes, I am aware that in part it was demanded by the marketers to have high clock speed. That doesn't mean it wasn't a revolution in chip design at the time.

It was certainly a different approach but that doesn't make it revolutionary.

My point isn't that Intel didn't try to do something different, as I accept that they did. What I'm saying is that the Netburst architecture was designed solely to raise clock speeds. If the marketeers hadn't demanded far higher clock speeds to help sell the processors to naive customers, then I don't think Intel would have gone the way they did and would instead have followed a similar path to AMD, optimising and improving what they already had.

A revolutionary processor should bring something genuinely new to the table and offer real benefits to the end user as a result. I'd class the Althon 64 and Opteron processors as "more revolutionary" than Netburst for their introduction of 64 bit extensions and integrated memory controllers. The Netburst architecture offered no such radical improvements over the previous generation, it was designed solely to raise clock speeds.

Remember when the first P4 processors appeared, either at 1.4 or 1.7Ghz (I can't remember). Everyone laughed and pointed because they were slower than the fastest P3 processors, which were clocked far lower. Intel said that the speeds would soon ramp up far beyond 2Ghz, which in time they did, but the improvements were not solely down to the architecture but also to the reduction of the process size, which would also have benefited the older P3 technology as well.

Witness what happened with the Pentium-M processor. This processor is basically a Pentium 3 with some improvements, but with twice the clock speed that the Pentium 3 line ended with due mainly to the reduced process size. A 2.13Ghz P-M was easily the equal of the fastest P4 which clocked at well over 3Ghz. All the P4 gave you was increased clock speeds, more heat and more power consumption. This would be like a car manufacturer designing an engine that ran at 20000rpm, used twice as much fuel and wasted twice as much energy in heat, yet produced exactly the same performance as a "regular" engine. Very revolutionary that!

NathanE · 25 Jun 2006 at 10:26

Vertigo1 said:
The Netburst architecture offered no such radical improvements over the previous generation, it was designed solely to raise clock speeds.

You are blinding yourself into believing what you want to think.

NetBurst... revolutionary multi-media capabilities. SSE and SSE2 turned the once strenuous task of playing a DVD (often needing a seperate hardware MPEG decoder) into something that barely tasked the CPU at all. Revolutionary.

Hyperthreading. Although done before in other markets, it was the first time joe-home-consumer got to experience SMP behaviour and all its advantages whilst sitting at their desk at home. Revolutionary. Perhaps even more importantly it brought about the multi-core race that is now just getting started.

Power distribution and clock rates. As a side affect of NetBurst's design Intel spent billions of research of improving clock rates and balancing power distribution. The spent billions on researching new transistors, new fabrication processes. Some of the fruits of that spending are only just reaching the market. The Low-K transistors as are used by Conroe for example. It was also how they figured out how to clock a P6 derived design (uh, Conroe) upto near NetBurst levels. Revolutionary.

Vertigo · 25 Jun 2006 at 11:24

NetBurst... revolutionary multi-media capabilities. SSE and SSE2 turned the once strenuous task of playing a DVD (often needing a seperate hardware MPEG decoder) into something that barely tasked the CPU at all. Revolutionary.

SSE and SSE2 could easily have been implemented on any architecture and weren't dependent on Netburst. I've never said SSE wasn't a worthwhile advance, but as it's not dependent on the Netburst architecture itself, this is irrelevant. SSE, in all its versions, has been easily integrated into numerous non-Netburst architectures.

Hyperthreading. Although done before in other markets, it was the first time joe-home-consumer got to experience SMP behaviour and all its advantages whilst sitting at their desk at home. Revolutionary. Perhaps even more importantly it brought about the multi-core race that is now just getting started.

Hyperthreading is a different issue however and I'll conceded that it was the Netburst architecture which made this possible but I don't believe it offered any "revolutionary" benefit to the end user. Whether it gave rise to the current crop of dual-core processors is highly speculative and un-provable. With Conroe, it would appear that Intel are abandoning HT as presumably dual-core (and in future multi-core) processors render HT superfluous. You could argue that, far from HT "bringing about" the advent of dual-core, AMD realised that dual-core was the ultimate way forward and that HT was a merely a stop-gap measure unworthy of support.

As a side affect of NetBurst's design Intel spent billions of research of improving clock rates

Sorry but that's rubbish. The battle to improve clock rates was in no way a side-effect of the Netburst design, it was a fundamental requirement in order for it to produce the desired performance. There was much speculation when the architecture was first introduced that Intel envisaged it scaling to 4Ghz and beyond and, when this failed to happen, they found themselves a bit stuck.

You could almost say they've "lucked out" a bit as, because of their desperate need to ramp Netburst's clock speed, they have learned quite a few tricks which they've been able to use with Conroe but the fact remains that Netburst itself is being abandoned and Intel themselves are publicly stating that it's not all about clock speeds any more, obviously smarting from their previous mistakes.

NathanE · 25 Jun 2006 at 12:05

Vertigo1 said:
SSE and SSE2 could easily have been implemented on any architecture and weren't dependent on Netburst. I've never said SSE wasn't a worthwhile advance, but as it's not dependent on the Netburst architecture itself, this is irrelevant. SSE, in all its versions, has been easily integrated into numerous non-Netburst architectures.

NetBurst started it though. NetBurst brought the first highly scalable vector execution units to the desktop. That was my point. Before SSE we had MMX which was crap really. It doesn't matter that SSE has since been added to other processor designs, the point remains that NetBurst started the process off.

Hyperthreading is a different issue however and I'll conceded that it was the Netburst architecture which made this possible but I don't believe it offered any "revolutionary" benefit to the end user. Whether it gave rise to the current crop of dual-core processors is highly speculative and un-provable. With Conroe, it would appear that Intel are abandoning HT as presumably dual-core (and in future multi-core) processors render HT superfluous. You could argue that, far from HT "bringing about" the advent of dual-core, AMD realised that dual-core was the ultimate way forward and that HT was a merely a stop-gap measure unworthy of support.

They aren't abandoning HT at all. Where did you hear that? HT will be back, probably in the next wave of Core 2 chips. And it will be back in a bigger way than it ever was before in NetBurst. Again, HT is a pervasive technology that should be present on all CPUs. It never was and never will be a "stop-gap" measure. If it really was a stop-gap measure why are AMD actively researching it and why are Intel being so secretive about their disabled HT logic in Conroe? HT was sparked off on the desktop by NetBurst. Revolutionary? Check.

Sorry but that's rubbish. The battle to improve clock rates was in no way a side-effect of the Netburst design, it was a fundamental requirement in order for it to produce the desired performance. There was much speculation when the architecture was first introduced that Intel envisaged it scaling to 4Ghz and beyond and, when this failed to happen, they found themselves a bit stuck.

Correct. Clock rate was and still is a fundamental variable in high performance CPU design. So why are you saying NetBurst was not revolutionary in this field? Just because it didn't have a very good IPC rate does not mean the lessons Intel has learnt about producing high clock rate designs was not valuable. Do you really think Conroe would have existed if it wasn't for Intel's engineers sweating and arguing for months on end on trying to tape-out the Northwood or Prescott designs? I'll answer that: No. The lessons they learnt will live on in the industry for decades to come. Revolutionary.

You could almost say they've "lucked out" a bit as, because of their desperate need to ramp Netburst's clock speed, they have learned quite a few tricks which they've been able to use with Conroe but the fact remains that Netburst itself is being abandoned and Intel themselves are publicly stating that it's not all about clock speeds any more, obviously smarting from their previous mistakes.

So because a design has since failed (or in NetBurst's case, ran out of scalability) that makes it unrevolutionary? :confused:

Vertigo · 25 Jun 2006 at 19:44

NathanE said:
It doesn't matter that SSE has since been added to other processor designs, the point remains that NetBurst started the process off.

Netburst may have "started the process off" but the point I'm making is that it wasn't and isn't dependent on the Netburst architecture and would have come about (or something similar) if Netburst had never existed.

They aren't abandoning HT at all. Where did you hear that?

Well, from an initial glut of HT in all the 800FSB single core Northwoods and Prescotts, HT was then relegated to the high-end Extreme Edition processors and has now been omitted from the first batch of Conroe processors. Looks like abandonment to me although this is merely my own perception and from your other comments you obviously have more info on this than do I.

HT will be back, probably in the next wave of Core 2 chips. And it will be back in a bigger way than it ever was before in NetBurst

So HT, like SSE, isn't dependent on the Netburst architecture and would probably have come about without it.

HT was sparked off on the desktop by NetBurst. Revolutionary? Check.

That might make HT revolutionary but not Netburst.

So why are you saying NetBurst was not revolutionary in this field? Just because it didn't have a very good IPC rate does not mean the lessons Intel has learnt about producing high clock rate designs was not valuable.

What I'm saying is that the Netburst architecture itself cannot be considered revolutionary simply because, in an attempt to get it running properly, Intel happened to learn a great deal about processor design.

So because a design has since failed (or in NetBurst's case, ran out of scalability) that makes it unrevolutionary?

The point I'm trying to make is that the Netburst architecture itself was a fundamentally flawed design which brought nothing revolutionary to the table. Please note that I'm making the distinction between Netburst and the other technologies that it may have supported or even given rise to, such as SSE and HT.

The fact remains that SSE and HT are not inherent parts of the Netburst architecture and would have come about regardless. Their presence and success (or failure) cannot therefore be taken into account when considering the Netburst architecture itself.

As for clock speeds, yes Intel may have learned a great deal about high clock rates through their efforts to get Netburst running "properly" and this knowledge will no doubt prove useful in the future but this doesn't change the fact that Netburst itself was a flawed design which deliberately sacrificed IPC rate to gain clock rate as a marketing ploy. Just because they learned a lot off the back of it doesn't make it revolutionary in itself. The solutions they came up with to solve the problems Netburst presented may have been revolutionary but not the architecture itself. That's the distinction I'm making.

(Something tells me we're never going to agree on this one

)