Intel Core architecture vs future AMD K8

[gOJDO]

Technology improvements on Intel Core arhitecture chips are more than impresive. Meanwhile they are improving it and they are building a 45nm fab to produce new quadcore chips based on the same architecture. AMD is more than few steps behind, they are still stucked on 90nm SOI and old K8 with freq<3GHz.

We can only guess what AMD can change on its K8, but I think they won't do much. s939 K8 has a minor performance adventage(about 5%) over s754 K8 on the same clock with its double RAM bandwidth of the s754. So why should we expect performance boost from the same architecture with better RAM bandwith?

K7 and K8 architectures with their integrated DDR controler, have more benefit from low-latency rather than high-latency higher freqfency RAM. The DDR2 controler on the AM2 will not boost the performance of K8 architecture as long as the core waits for data from L2. Larger L2 cache per core will solve many data lags and AMD plans to implement 2MB insted of 1MB in their high-end performance AM2 chips. More memmory bandwidth is useless if the processing units are not capable to process the waiting available data. More processing units and higher freqfencies are they key for greater performace in this case. But AMD won't improve nither freqfency, nither the architecture of 90nm K8 in AM2.

And until 2008 we will have no K8(or maybe K9) on the 65nm SOI which probably should have higher clock, 4 cores on a chip as planed, more HTT links with more bandwidth(HTT2.0 and HTT3.0) and maybe will have its cores improved with extra FPUs(there are rumors on the inquirer), iSSE4(i guess there would be no problems to implement) and more 64bit instruction sets and registers. But all of these are only a unrealised guesses and better K8 is only on papers today.

AMD and Intel have their own philosophies about the value of the chip: "performace per wat", "performance per clock" and etc. Most users have different philosophy, it is "performance per dollar". Not all of us are buying the last released hardware available on market, and not all of them are buying it everytime it comes with some improvements. For me, better CPU is that which in combination with the mainboard and the RAM will perform better for the same money I plan to spend for CPU, RAM and mainboard together. I was buying Intel solutions until AMD K7 came in to play. I sence there would be Intels in my cases replacing the K8s in the folowing few years.

Intel were always a gigant and technology leader compared to AMD. AMD are copying and using intel inovations with some minor changes since they are producing chips based on Intel x86 architecture. Intel have money to discover and invent and they made a wrong step gaining MHz with its Netbrust architecture. But this time they made a huge jump with its Core architecture and they will knock out AMD like in the past before K7.

 

[Gazza]

well im putting together An
Opteron 180 2.4
Asus A8R32-mvp
ATI 1900XTX
still undersided on ram 2x512 or 2x1g
But what im getting at is as far as gamers are concerned 4 cores or 2x 2 cors on a 2 chip board wont be any faster or better
So i willl probably stick whith this configuration for quite a while
but as for the buisness sectore intell cannot go the same path as AMD with motiple cpu's per motherboard and i think they new that years ago and have been building for moltiple core cpu's 4and 8 cores to take back industry loss
there arcetectual desine is vastly different and very refreshing i never did like the P4 designe but i throught the P3 was faster but the new conroe
Will turn a lot of AMD heads even mine towards intel again
I say good one Intell finaly butting together a great chip and packege
😀 Gazza

 

[TabrisDarkPeace]

We can only guess what AMD can change on its K8, but I think they won't do much. s939 K8 has a minor performance adventage(about 5%) over s754 K8 on the same clock with its double RAM bandwidth of the s754. So why should we expect performance boost from the same architecture with better RAM bandwith?

Finally, a decent and well thought out post

All comes down to K8 having such a nice cache hit rate (around, 80% to 95% in most game software).... so you can only expect to see a jump in performance when cache misses occur.... eg: memory access

If memory performance doubles (at same latency) the performance during the 5% miss rate will double, thus performance rises 5% when memory clock speed is raised (all other things being equal, including processor clock speed, bus widths, latency, etc).

The fact most people on these forums denied this fundamental fact, until they saw the numbers, while you embrace knowledge has gained you respect from me, and likely many others.

For someone with a post count of only 3, that is pretty f-ing good IMHO. 8)

You took what you knew, and applied a scaling (proportional) technique. So simple, and so elegant.

Just like a sniper measures range by taking something they know and multiplying. (Because most rifles are zeroed for a given range, and the sight is totally off at closer or further distances - Much unlike the majority of computer games on the market where the sight is always zeroed at every range, in real-time without the player taking significant time to do it... they simply don't).


(I am looking for your other posts now btw)

However, I hope you compared equal clock / cache S754 to S939 processors, instead of by their part numbers. (As S754 is clocked 200 MHz higher for equal part number, to help make up for lower throughput from memory).

That, and 65nm AMD CPUs are likely for Socket AM2, but perhaps when Socket AM2 is released the first batch will still be using 90nm SOI. That die size reduction (90nm -> 65nm) could gain them no more than 40% in clock speed, and likely only 25% - 30% when everything else is factored in.

Also K9 is likely to be skipped, and a K10 architecture pushed in its place.

 

[gOJDO]

@TabrisDarkPeace, 10x

Well, I see a lot of "fights" about who is better than who when Intel and AMD fans are talking about them, and in most cases I don't get the point of those discussions. Most of them as arguments are using marketing numbers,labels and specs("64BIT", "4GHz", "dual core", "ultra super fast", "4MB L2") rather than the facts of overall perforamnce and adventeges in the reality.

It is a kind of rediculous, considering the supported features of the current hardware as any kind of adventages or performance boosts, when almost always they are not supported and are useless for the current software.
How many of us are running our 64bit processors on 64bit OSes and how many 64bit apps do we have today(count the unreleased 64bit drivers as a software also)?
How many % of the current apps that we are using have a benefit from multicore computing and how many of them are compiled for multicore parallel processing?
There are a lot of questions in this context that demants the statements of the hardware producers. What one crazy man can do, 10 smart men can't.

What about the software? huh.....We are in the era of unoptimized computing. Software was never unoptimized as well as today. Make an retrospective what programers did on Z80, Commodore64, Amiga, Atari, 8086, etc. and what hardware resources did they have.
Why to lose time making an sofisticated and improved application and sell it once, when you can sell the same app more times in different versions while you are developing it? (damn, i hate security patches and critical updates as much as the new splash-screens)

It is clearly simple, both software and hardware producers care only about the profit. They work uncordinated for the functionality of the computer system (current hardware is not compatible to the current software and vice-versa current software is not compatible to the current hardware). But both of them are earning money from the old allready seen technologies and products(while there are sheeps, there would be wool).
They both can do better, but they don't want to! They will stop making money that way. So why should we say that some one of them is good and other is better?
Lets say that all of them are bad(and greed), but some of them are less.

 

[TechnologyCoordinator]

What about the software? huh.....We are in the era of unoptimized computing. Software was never unoptimized as well as today. Make an retrospective what programers did on Z80, Commodore64, Amiga, Atari, 8086, etc. and what hardware resources did they have.

You are very correct. Software coding is not done nearly as efficiently as it was back in the day when programmers had to worry about kilobytes of RAM, not megabytes or gigabytes. I would love to see software benchmarking, which is much less feasible than hardware benchmarking, but possible in some cases!

 

[TechnologyCoordinator]

Intel were always a gigant and technology leader compared to AMD. AMD are copying and using intel inovations with some minor changes since they are producing chips based on Intel x86 architecture. Intel have money to discover and invent and they made a wrong step gaining MHz with its Netbrust architecture. But this time they made a huge jump with its Core architecture and they will knock out AMD like in the past before K7.

I do not fully agree with your last paragraph there. I don't think of AMD as simply copying Intel's products, that is not true. AMD products MUST be 100% software compatible with Intel products, otherwise you end up with another Apple.

I agree that Intel will have the best products at this year's end, but I think AMD will snap back sooner than what you expect.

 

[quantumsheep]

very interesting case presented here, it's nice to see someone putting forward some reasonable unbiased thoughts.

 

[HA_ZEE]

AMD is trying to offer the consumer (you) a universally compatible product, that is why they use the X86 platform. Now, if AMD has been copying Intel how come they had the first 64 bit processor on the market, the first Dual core processor, the first on die memory controller which gives them a huge memory bandwidth advantage and a large advantage in performance per Watt? If the new processor from Intel does perform as well as the initial tests show it is only because their hand was forced by AMD to compete. Intel would have been happy to keep pumping the Prescott's up our butts without changing anything (my house is warm enough thank you). The market place is always driven by competition. I currently use an AMD system for gaming because it was simply the best choice (AMD 4000). Now, if that Conroe is as good as claimed and it is not too expensive it would make a nice gaming system UPGRADE from what I currently have (market place competion is a win for us).

 

[gOJDO]

I do not fully agree with your last paragraph there. I don't think of AMD as simply copying Intel's products, that is not true. AMD products MUST be 100% software compatible with Intel products, otherwise you end up with another Apple.

I agree that Intel will have the best products at this year's end, but I think AMD will snap back sooner than what you expect.

I was expecting an disagree for that part becouse I were thinking very similar like you now, a few years ago, after the K7 arived. Intel implemented the x86 architecture in 1978 in their 16bit 8086 iAPX86 and 4 years later the same did AMD. In the folowing years AMD is trying to catch up with technology and to share more % of the personal computers market mostly held by Intel. AMD were always late with the technology implementation and their technolgy is a almost all done by Intel, starting with its x86 architecture.
Intel started producing socket for processors, AMD did that latter.
Intel switched socket with slot, AMD did that latter.
Intel started to ship processors with heatsink on the die, AMD did that latter.
Intel implemented DDR2, AMD are still on the way.
Intel implemented LGA processors, AMD are still on the way.
It is the same for the MMX, SSE, SSE2 and SSE3 instructions, first are implemented by Intel and latter by AMD. The AMD 3D Now! were implemented in the K6-II(Chompers) as a try to beat Deschutes. The SSE were not developed by Intel, but they first implemented them in Katmai, adding much more multimedia instructions set and more registers than the AMD 3D Now!.
MIPS Tech in 1991 produced the first 64bit processor. Intel implented them many years ago, before AMD did in their mainstream K8 processors.
The SOI...I am not sure, but I think that and the first dualcore CPU are from IBM.
The PCIe standard was developed by Intel(I think that was the case for AGP also).

See, when first Athlons came, that was the first time when AMD was runing shoulder to shoulder next to Intel. The cheaper Athlon(Thunderbird) 1400C 180nm 256kB L2(fullspeed) FSB266 with DDR 266MHz was overperforming Pentium III(Tualatin) 1200MHz 130nm 256kB L2(fullspeed) with SDRAM 133MHz. I was supriced also when K7 AthlonXP(Palomino) on lower clock knocked out Pentium4 Willamate s423 256kB L2. The K7 were equiped with more FPUs, integrated DDR controler, multimedia 3D Now! and wider software supported and more efficient SSE.

Intel Pentium 4 absolutly disapointed me and I started thinking that AMD are getting better than Intel even they have less than 10th of Intels budget and production capabilites. Their try to achieve performance troguh the freqfency was unsucessfull becouse of the less instrcutions per cycle isue of the stage-extended Netburst architecture. The heat was another enemy of P4 and AMD solved that a bit with turning on SOI. The K8 was another success of AMD, not becouse of its great architetcutre or its top-models performance, but with the low prices on the market.

Now Intel strikes back with nice technolgy improvements and that will lower the prices of the current CPUs and will push its rival to make a better chip.

 

[TechnologyCoordinator]

Also, don't forget that Intel is using a copy of AMD64 extensions, they just call it something different.

However, getting back on topic, Intel architecture is now using some of the same strategies as AMD, they are getting much more performance with Conroe at lower frequencies. I always thought that one day I could buy a 10ghz processor, but with Conroe it looks like ghz escalations may be over.

Will we ever get past 4 and 5 ghz? Or is it just impossible?

 

[Ycon]

Well, Intel could easily release 4 GHz CPUs but they dont seem to want to.

 

[LithiumSunset]

It's very well possible and may happen. The problem before is, when you increase clockspeed, you increase heat and add more stress to the processor. With Intel using 65nm techology, they can clock higher and run cooler. I for one, really don't think that's where they want to go though.

I think the thing both manufacturers want is more for less. More performance for less clocks and power. Intel has proven that they can achieve that now with Conroe and family.

 

[Zoron]

You forget something very important in your ramblings. Back in the early days of the 8086 AMD actually made processors for Intel. AMD couldn't have possibly been "late" as they actually made the damn things for Intel.

You talk about going from socketed chips to a slot form factor as if it was an advancement. It wasn't. At the time, neither Intel nor AMD could use cache memory that operated at the same speed as the core. Therefore, the cache had to be placed off-die. The true advacement came once full-speed cache was implemented on-die and chips once again went back to a socket form-factor.

AMD's 3D Now! was technically superior to Intel's MMX... but software vendors didn't support it. If they had, Intel probably would have been forced to introduce SSE sooner.

Intel's first 64-bit chip was the Itanium. (AKA Itanic). Hardly the success story of Athlon64. Just because Intel had a 64-bit chip before AMD, didn't make it a better processor. AMD designed a better 64-bit chip because it could do 32-bit just as fast (or even faster) than it's old XP line of processors while building in 64-bit capability. AMD didn't "copy" Intel, they took the initial idea and made something better.

 

[quantumsheep]

i remember in pczone magazine in december 2001 reading an article about intel showcasing "the future of CPUs" which had a story about a 1 thz CPU.Now it seems we will never get there though!

 

[apache_lives]

Technology improvements on Intel Core arhitecture chips are more than impresive. Meanwhile they are improving it and they are building a 45nm fab to produce new quadcore chips based on the same architecture. AMD is more than few steps behind, they are still stucked on 90nm SOI and old K8 with freq<3GHz.

We can only guess what AMD can change on its K8, but I think they won't do much. s939 K8 has a minor performance adventage(about 5%) over s754 K8 on the same clock with its double RAM bandwidth of the s754. So why should we expect performance boost from the same architecture with better RAM bandwith?

K7 and K8 architectures with their integrated DDR controler, have more benefit from low-latency rather than high-latency higher freqfency RAM. The DDR2 controler on the AM2 will not boost the performance of K8 architecture as long as the core waits for data from L2. Larger L2 cache per core will solve many data lags and AMD plans to implement 2MB insted of 1MB in their high-end performance AM2 chips. More memmory bandwidth is useless if the processing units are not capable to process the waiting available data. More processing units and higher freqfencies are they key for greater performace in this case. But AMD won't improve nither freqfency, nither the architecture of 90nm K8 in AM2.

And until 2008 we will have no K8(or maybe K9) on the 65nm SOI which probably should have higher clock, 4 cores on a chip as planed, more HTT links with more bandwidth(HTT2.0 and HTT3.0) and maybe will have its cores improved with extra FPUs(there are rumors on the inquirer), iSSE4(i guess there would be no problems to implement) and more 64bit instruction sets and registers. But all of these are only a unrealised guesses and better K8 is only on papers today.

AMD and Intel have their own philosophies about the value of the chip: "performace per wat", "performance per clock" and etc. Most users have different philosophy, it is "performance per dollar". Not all of us are buying the last released hardware available on market, and not all of them are buying it everytime it comes with some improvements. For me, better CPU is that which in combination with the mainboard and the RAM will perform better for the same money I plan to spend for CPU, RAM and mainboard together. I was buying Intel solutions until AMD K7 came in to play. I sence there would be Intels in my cases replacing the K8s in the folowing few years.

Intel were always a gigant and technology leader compared to AMD. AMD are copying and using intel inovations with some minor changes since they are producing chips based on Intel x86 architecture. Intel have money to discover and invent and they made a wrong step gaining MHz with its Netbrust architecture. But this time they made a huge jump with its Core architecture and they will knock out AMD like in the past before K7.

AMD does what they do best - minor tweaks here and there - they have never jumped like conroe, prolly cause conroe was where the P4 SHOULD have let off and, and AMD prolly also got behind waiting for some competition with P4.

What we dont realise here is Intels P6 design behind conroe and conroe its self aswell is Intels greatest work and is prolly the best architecture in the world (fast, efficent, cool running).

Im guessing it will take AMD 2 years to catch up atleast on the single/dual/quad cpu front - there designs take 9 months to make and test and they aint gona cancel there investements, nor could extra cache, faster memory or faster HT help speed the K8 up.

 

[conroe]

Also, don't forget that Intel is using a copy of AMD64 extensions, they just call it something different.

Will we ever get past 4 and 5 ghz? Or is it just impossible?

Intel traded SSE3 for 64bit. I have seen a 660 over 7ghz benched.

 

[apache_lives]

Also, don't forget that Intel is using a copy of AMD64 extensions, they just call it something different.

Will we ever get past 4 and 5 ghz? Or is it just impossible?

Intel traded SSE3 for 64bit. I have seen a 660 over 7ghz benched.

7.3ghz single core (prescott now 65nm's) and 5+ ghz for 65nm pentium d's

 

[MasterLee]

What was that 7Ghz cooled with ?
I'm sure CPUs 5Ghz and up are possible, but why ? We're not using the hardware we have to it's fullest yet. I'd love to have a 64 bit OS complete with 64 games and programs, prolly could'a saved another 50 bucks if they'da left that out since it's not being used (well not fully).

 

[conroe]

Dry Ice.

 

[Grimmy]

Do you think perhaps they could end up using a light wieght oil to transfer heat?

Would be better for a pump to keep it lubricated.

:? . o O (hang on, let me get my catchers mask on for the bashin)

 

[conroe]

Some AMDs(very few) can go near 4ghz, I know one has gone faster. AMD has a cold bug and will not work in low sub-zero temps.

 

[halbhh]

why is AMD supposed to stop innovating? That doesn't seem wise. why are we presuming we know everything they are going to do? Somehow that would not make sense. It would be rather self-destructive for AMD to reveal to a much richer competitor all it's strategies. If you want to know how the AMD processors of 1 year from now will stack up against the Intel processors of 1 year from now, I'm guessing you'll have to wait and see. Speculation is fun though, and this speculation based on the idea AMD does basically very little is entertaining for a moment and then just leads me to the above questions.

 

[endyen]

Well, lets see. Amd came out with the Socket athlons, that had an extra fpu. Intel started loosing inbenchmarks of the day. Amd came out with the first 1ghz chip. If you were not running extremely intel optimized progs, all the time. Amd gave better perf.
Intel in a fit of insecurity came out with the P4s. They ran fast, but just couldn't do the same work as the athlons.
Intel tried to catch up. They increased thier fsb to 133/533, and as long as you were running heavily floptimized code, you could get just about the same perf as an XP.
Finnally Intel brought out the 200/800 fsb, and mostly took the lead. Not for long though. The A64s just creamed them. That's how it has remained for over two years now.
Perhaps , in that time, Hector spent all his R&D money on a new beer for his Dresden employees, but I doubt it.
If Conroe is not just another Intel lie, Amd has had ample time to come up with something. After all, they have been working the IPC game all along. Intel on the other hand, only knows how to get that job done in Isreal.
The ball may be in Intel's court , but it looks to me like Amd has the home field advantage.

 

[apache_lives]

Well, lets see. Amd came out with the Socket athlons, that had an extra fpu. Intel started loosing inbenchmarks of the day. Amd came out with the first 1ghz chip. If you were not running extremely intel optimized progs, all the time. Amd gave better perf.
Intel in a fit of insecurity came out with the P4s. They ran fast, but just couldn't do the same work as the athlons.
Intel tried to catch up. They increased thier fsb to 133/533, and as long as you were running heavily floptimized code, you could get just about the same perf as an XP.
Finnally Intel brought out the 200/800 fsb, and mostly took the lead. Not for long though. The A64s just creamed them. That's how it has remained for over two years now.
Perhaps , in that time, Hector spent all his R&D money on a new beer for his Dresden employees, but I doubt it.
If Conroe is not just another Intel lie, Amd has had ample time to come up with something. After all, they have been working the IPC game all along. Intel on the other hand, only knows how to get that job done in Isreal.
The ball may be in Intel's court , but it looks to me like Amd has the home field advantage.

P1 age war was mostly won by the P1 (k5 was late)
P2 age war was won by the P2
P3 age initially was won by the P3 but later the coppermine and the thunderbird were even, no winner unless price is involved.
P4 sucked untill the northwood and the C finally pulled infront but not for long
Conroe - will beat atleast the first gen AM2's and maybe second gen, who knows.

AMD has never made a performance jump like that before, only minor updates wich work fine, im thinking conroe is where P3 would have left off and either intel did there research or amd didnt.

Perhaps when AMD finally comes up with an answer intel will use CSI so we mightnt see AMD on top for years until Intel gets lazy and gets taken over by AMD... we shall see.

 

[MasterLee]

No I didn't