Conroe/Merom to be 20% faster than AMD's best

[Carfax]

Instead, Intel will count on its microarchitectural improvements and a faster front-side bus to deliver the 20 percent improvement in performance over AMD's chips, based on standard benchmarks, Eden said.


Source (http://marketwatch-cnet.com.com/Intel+strikes+back+with+next-generation+chips/2100-1006_3-6041120.html?type=pt&part=marketwatch-cnet&tag=feed&subj=news)

Interesting if true. AMD's K8 already has high IPC, and it's expected to go even higher once the AM2 socket debuts with it's DDR2-800 capability.

20% greater performance would be a helluva feat for Intel to pull off..

Especially if it's 20% clock for clock, which is what I think it is. :)

[dinos22]

so i guess if they are comparing current socket939 chips with ddr1 they are going to be in a loooot of trouble come AM2+65nm chips

[agenda2005]

Intel know quite well that AMD will have DDR2-800 with a dual core FX-64 chip running at 3.0GHz by Q1 2007 http://www.dailytech.com/article.aspx?newsid=850

If Intel can boast that Conroe will be 20 percent faster than this chip, then be sure that increasing clock speed alone will not cut it for the competitor.

Words are coming out from the horses mouth, the designer and leader of the P-M team as well as Conroe and Merom, that is quite scary. Time will tell though.
AMD probably have something up their sleeve, maybe 4MB L3 cache on the FX to close the gap, time will tell if they wake up from their complacancy due to the small win from X2 and Opteron.

[dinos22]

meh until these start rolling in we won't know

i guess what my point was also is the fact the RAM/newddr2 controler will give a big boost to performance

i'm looking forward to Intel stuff but wouldn't get carried away with their PR work just yet

[nn_step]

well they are talking big.. Lets see if their procs can do what they say thay will.

[agenda2005]

If you dont know, someone have AMD64 AM2 CPU with DDR2-667 and the conclusion was that infact it run slower than the current 939 CPU with DDR1-400 in many instances, so Intel must have had some way of measuring up the performance of that CPU even by merely overclocking it to DDR2-800 and comparing the performance with Conroe.
PM me if you want proof of that statement.

Trust me, that is probably the case. I like the way Intel is turning on the heat on the competition. Afterall, its good for the consumer as it help lower prices.

[dinos22]

If you dont know, someone have AMD64 AM2 CPU with DDR2-667 and the conclusion was that infact it run slower than the current 939 CPU with DDR1-400 in many instances, so Intel must have had some way of measuring up the performance of that CPU even by merely overclocking it to DDR2-800 and comparing the performance with Conroe.
PM me if you want proof of that statement.

Trust me, that is probably the case. I like the way Intel is turning on the heat on the competition. Afterall, its good for the consumer as it help lower prices.
hmmm what did you say...i really didn't get this post

are you saying someone tested AM2 with DDR2 ram?:slobber:

[nn_step]

Are you sure you don't mean Revison F (aka Socket F)?

[agenda2005]

hmmm what did you say...i really didn't get this post

are you saying someone tested AM2 with DDR2 ram?:slobber:

Absolutely YES!:toast:

[dinos22]

Absolutely YES!:toast:
where what how when links i want everything quick!!!!:banana:

can you link the site

[Carfax]

I'd like some more info on that aswell if you don't mind :D

I'm surprised that the K8 would perform slower with DDR2-667, provided that it was good DDR2, and not the high latency junk.

[zir_blazer]

Intel makes too much noise. Did anyone forgetted that they promised that NetBurst will reach 10 GHz with the Tejas Core in 2005 and that Prescott was gonna send AMD to the grave? The only that Prescott did is dig they own grave. Everything that they say is Vaporware until we see real numbers, not only that I don't expect anything worthy from Socket M2, but the expectatives about Intel are unsolid at best. Intel way of Benchmarking is always using some Synthetic Benchmark optimized specifically to favour Intel Architecture (Though being Architectures of Short Pipeline, optimizations for Conroe should do something good for AMD too when compared to optimizations for NetBurst) over whatever AMD got. Show me a Conroe and I will be the one that decides how much that theorical 20% is worth in real life.


I'd like some more info on that aswell if you don't mind :D

I'm surprised that the K8 would perform slower with DDR2-667, provided that it was good DDR2, and not the high latency junk.
Seriously, who is expecting that K8s performs better using DDR-II when the extra Memory Bandwidth that they give is pretty much useless or of little impact in Single Core and Dual Core respectively (Anandtech did a Review with a Rev. E K8 with the new Divisors to test 533 MHz DDR Memory Modules)? On the other hands, they aren't Latency dependant neither, because almost in any Benchmark that I saw diferent Latencies showed little diference at best (I tested CAS 1T and CAS 2.5T in my system and results where depressing. Well, my system doesn't got a decent Hardware configuration to do such testing, but what I saw should be a valid figure anyways). Expect around the same performance or just minimally higher, not worth the wait, the money to migrate to that platform or anything else.

[DilTech]

If you're going to bring up Intel making alot of noise before a release, merely bring up the release of Williamette. We heard HUGE things, much like what we hear now about conroe from them, when in reality the P3 was faster at the time of release.

With Intel, I'm in a I'll see it when I believe it position these days.

For the record guys, Intel ALWAYS makes this much noise before a new core release, ESPECIALLY a brand new chip design.

Also, Agenda is talking about Anandtech, with the board they showed they also had a early prototype, and supposidly it ran slower than the 939 chips. Obviously, it was a prototype. Likely an unfinished memory controller.

DDR 800 is going to effectively DOUBLE the bandwidth for AMD, considering dual channel gave 5% up, that should be good for some kinda boost at the bare minimum.

[onewingedangel]

Are you sure you don't mean Revison F (aka Socket F)?

revision F will be used in socket f, socket am2 and socket s.

Just like revision E (90nm chips, winchester, venice, san diago, manchester and toldedo) were used on multiple sockets, and just as revious revisions were.

Socket F may in fact be amed after the fact that is designed for the f stepping server cpus, but thats not the only place f steppings will turn up.

[oozberg]

When I see it, I belive it.

[Cobalt]

So you really believe intel is going to give us accurate numbers? Yea I can just imagine it:
"Well we tested our brand new architecture but it seems that it just doesn't cut it compared to our glorious competitors" :rolleyes:

Intel will say that they get a 20% boost but over which AMD chip and in what benchmark? Maybe they do win by 20% in one benchmark that doesn't mean that it will be better in every benchmark. Intel also has no idea what the performance of AM2 will be like. Unless theres been some corporate espionage, conspiricay time!

[MaxxxRacer]

While I'd like to see more competition from Intels side, as others have said, I will believe it when I see it.

[ahmad]

Let us assume that Intel was right, and it was 20% faster than AMDs current architecture. Thats pretty freaking unimpressive :stick: . Especially considering that the A64 is about 3 years old now? 4 years old? And it took them this long to catch up....

:stick:
:stick:
:stick:

[MaxxxRacer]

The article says its 20% faster than AMD's architecture that will debut in late 06 / early 07... but how in the world would Intel know that when no one else knows how amd products will perform in q4 06/ Q1 07...

[agenda2005]

We do not have to argue back and fort, there are examples among us that will help us understand the situation. Look at the Yonah overclocked result http://www.xtremesystems.org/forums/showthread.php?t=88226.

That is 23sec Sp pi 1MB at 2760MHz for Yonah. It will take A64 FX a Whooping 3600MHz to make that cut. If you don't know search properly.

Conroe and Merom on the other hand are going to be a different thing upward from Yonah.

Intel expects its forthcoming Conroe and Merom chips to deliver a performance advantage of at least 20 percent over chips from Advanced Micro Devices that are slated to be released at the same time, an Intel executive said Friday.

The article infact say the performance difference will be AT LEAST 20 percent.

Take note of the word AT LEAST, if the reference is based on clock for clock comparison, then it will surely be difficult to catch up with a 3.33GHz Conroe.

"We believe we'll be able to open a major gap," with the new processors, Eden said. Eden led the design team that created the original Pentium M processor, which is credited as the inspiration for Intel's embrace of low-power design philosophies and the model for Merom and Conroe.


The combination of all those architectural changes will allow Intel to outperform AMD's planned offerings for the second half of 2006 without having to resort to adopting AMD's integrated memory controller design, Eden said. "It will take at least a year and a half to two years to close such a gap."

Take note of those DOUBLE QUOTATIONS in what the Intel executives said. These guys are putting their credibility on line. I will be lean more towards beleiving what they say this time, because they have good knowledge of what goes on in the industry.

[dinos22]

but you are merely comparing current A64 architecture to new Intel one............i can tell you right now without a shadow of a doubt that the new AMD/DDR2 architecure will definately outperform older generation......AMD benefits enormously from memory tweaking/timings/frequencies......we will find this out sooner than Intel crowd the way things are going anyways......really looking forward to AM2 vs Conroe........it will be one hell of a matchup

[[XC] Lead Head]

The combination of all those architectural changes will allow Intel to outperform AMD's planned offerings for the second half of 2006 without having to resort to adopting AMD's integrated memory controller design, Eden said. "It will take at least a year and a half to two years to close such a gap."

So...If that is true then. Intel will have to use one of AMD's technologies to get
a head in performence again? Well then Good Job intel! Having to resort to use your competitors technology.

[agenda2005]

So...If that is true then. Intel will have to use one of AMD's technologies to get
a head in performence again? Well then Good Job intel! Having to resort to use your competitors technology.


You seems to misunderstood that quote. Read the quote carefully. Intel will get all those lead WITHOUT USING AMD Integrated Memory Controller.

What is the fuss about copying technology anyway? AMD has copied more of Intel technology than the other way. The very existence of AMD in rooted in copying Intel x86 from the onset.

[Xenogias]

merom is something to look out for...most of you can remember how starved for data dothans were. Now they have DDR2 with yonah. Also, I like their pricepoint with Yonah. We'll see if Intel maintains that with Conroe/merom. most likely not.... but in terms of price/perf. Intel wins right now.

[mdzcpa]

As far as Intel executives making statements like this....so what. They've done it before and they'll do it again. It means nothing. Its just the PR team on overdrive.

Worse yet, its an age old marketing tactic. Tell the consumer NOW, that your product LATER will be better than the competitor's product LATER so that you can keep your channel partners happy NOW. Same old tired PR 101 stuff.

The bottom line is that Intel HAS to say this. They are getting thier azzes handed to them and they are loosing market share left and right. And, it isn't current CPU sales that concern them. They are boasting about what they have coming in order to save as many of thier strategic partnerships they can (eg Dell and a host of others). Strategic partnerships are long term relationships...they take a long time to build, and they help you when you are in a weak point in the cycle (like Dell still selling Intel only).

A PR move like this tells me Intel is worried. It tells me they have some relationships on the line right now. I bet Michael Dell told them to put something out that stomps azz or he'll be looking elsewhere.

I say :yawn2: :yawn2: :yawn2: Wake me up when the Intel chips do arrive. And when they do we'll know for sure how they perform.

[agenda2005]

Mike,
Don't go to bed sleeping easy that your AMD stock will keep rising. My advise is that you sell it now and get the maximum profit. Intel stock is as low and cheap as it can get. It will do you good if you can buy some now.

[Carfax]

Let us assume that Intel was right, and it was 20% faster than AMDs current architecture. Thats pretty freaking unimpressive :stick: . Especially considering that the A64 is about 3 years old now? 4 years old? And it took them this long to catch up....

:stick:
:stick:
:stick:

Thats an insane comment :slap: Do you know that the K8 has an approximate 20% lead over the K7 in IPC, and that it is MOSTLY due to an ODMC?

Without the ODMC, the K8 is probably a little over 5% faster than the K7, due to better prefetch, branch prediction and some other stuff..

If Intel can manage to get ATLEAST 20% boost over the K8s which are due out at the end of the year without using an ODMC, then that is freaking AMAZING!

Getting higher IPC out of x86 based architectures is like trying to squeeze blood out of a rock. So if Intel managed to get this high of an increase without using an ODMC, then I have to hand it to them.

[rozzyroz]

Intel will have to use one of AMD's technologies to get
a head in performence again? Well then Good Job intel! Having to resort to use your competitors technology.

actually, intel developed the on die memory controller in it's timna chips in the late 90's (according to the attacted artical from the op). once rambus got to be too expensive, thay scapped the chip... from the way the artical reads, amd used one of intel's technologies to get a head in performence....

the use of the large cache makes up for the on die mem controller. thats why we see the intel chips getting such a nice boost from cache.

i think conroe will be a monster of a chip.

[agenda2005]

I've started saving money and looking for low latency D9 fat body DDR2 because Conroe is going to be a BEAST. Imagine running 400*9 = 3600MHz at DDR2-800 3-2-2-4. Oh my goodness!!!

[CPLB]

I'm excited for Intel's new stuff too. As much as I like my AMD setup I'm tired of this one sided battle. Can't wait to finally see some competition!

[K.I.T.T.]

Any idea when conroe debuts? Is it Q4 this year or Q1 2007?

[Cybercat]

Thats an insane comment :slap: Do you know that the K8 has an approximate 20% lead over the K7 in IPC, and that it is MOSTLY due to an ODMC?

Without the ODMC, the K8 is probably a little over 5% faster than the K7, due to better prefetch, branch prediction and some other stuff..

If Intel can manage to get ATLEAST 20% boost over the K8s which are due out at the end of the year without using an ODMC, then that is freaking AMAZING!

Getting higher IPC out of x86 based architectures is like trying to squeeze blood out of a rock. So if Intel managed to get this high of an increase without using an ODMC, then I have to hand it to them.
I'm sorry, what? What about benchmarks and applications that do not depend that much on memory? What about Hyper Transport? Where do you pull that 5% number from? The lower end your digestion system?

It must not be that hard to get more IPC out of x86 processors as you say. I mean all Intel had to do is go back to the Pentium 3 with the mobile line...

[Carfax]

I'm sorry, what? What about benchmarks and applications that do not depend that much on memory? What about Hyper Transport? Where do you pull that 5% number from? The lower end your digestion system?


What about hypertransport? If you think hypertransport has anything to do with IPC, then you're clueless..

I was talking about changes to the core itself. The microarchitectural enhancements of the K8 contribute barely over 5% increase in IPC (average in 32-bit operations)

It's a fact that the ODMC contributes to the majority of the performance boost that the K8 got over the K7, and AMD's Bob Colwell said so himself in his Stanford presentation if I recall.

The reason for the large boost from the memory controller is perhaps because the K7 architecture was bandwidth starved, and giving the K8 an ODMC helped alleviate this tremendously.

Not to mention, having an ODMC helps with memory latency, something which is even more of a system bottleneck for PCs.

It must not be that hard to get more IPC out of x86 processors as you say. I mean all Intel had to do is go back to the Pentium 3 with the mobile line...

It's hard to get higher IPC period from any architecture. x86 is the hardest, because it is so archaic..

Thats why AMD created x86-64, to remove or address most of the bottlenecks inherent in x86-32.

For example, the meager amount of registers, and the horrible stack based x87 floating point unit foremost comes to mind.

x86-64 doubled the amount of registers, and uses SSE2 rather than x87 for math.

The fact is, it is very difficult to get higher IPC out of an x86 based microarchitecture, which is why if you go and read some of the more tech oriented forums, alot of engineers, programmers etc are skeptical that Intel will manage to pull it off.

However, Conroe will be the first modern 4 issue x86 core ever created, so it's definitely setting a precedent.

[vapb400]

Any idea when conroe debuts? Is it Q4 this year or Q1 2007?
Q3 2006
~July

although I wish they could get it out by the M2 release @ Computex

[Shadowmage]

If the best Conroe chip is only 20% faster than the FX-60, then that is very unimpressive.

It is true that creating faster CPUs is immensely difficult due to dependencies (unlike the massively parallel graphics procecssors), but also remember that they are on the 65nm process, which by itself should raise frequencies quite a bit. The Conroe die is also very large, so those extra transistors better be used for something good :)

[vapb400]

Not the FX-60, it says "Conroe and Merom chips to deliver a performance advantage of at least 20 percent over chips from AMD that are slated to be released at the same time."
So that'd be FX-62...and to have a 20% lead over your competitor is always impressive.

[Carfax]

Um guys, that 20% figure is most likely clock for clock performance!

It doesn't take into account any clockspeed advantage that Conroe XE might have.. The XE version is going to be 3.33ghz, and the fastest AMD will have at that time is 2.8ghz.

When you consider that Conroe has higher IPC than the K8, aswell as higher clockspeed (atleast on the XE), it should be way ahead of the K8.

[biohead]

http://www.xtremesystems.org/forums/showthread.php?t=89853

:o

[mdzcpa]

Mike,
Don't go to bed sleeping easy that your AMD stock will keep rising. My advise is that you sell it now and get the maximum profit. Intel stock is as low and cheap as it can get. It will do you good if you can buy some now.

LOL...I already sold half of my holdings at $42 a share. Not bad when I picked it up at $16 :banana: What's funnier is that had I listened to others' advice, I wouldn't even had bought it.

Like I said, as have others, we'll beleive it when we see it. There hasn't been a x86 cpu release that offered a 20% over the current benchmark.

I'm not going to lose any sleep over it. If and when Intel finally delvers something worth having, I'll be there. Until then, its only conjecture and marketing hype.

[agenda2005]

LOL...I already sold half of my holdings at $42 a share. Not bad when I picked it up at $16 :banana: What's funnier is that had I listened to others' advice, I wouldn't even had bought it.

Like I said, as have others, we'll beleive it when we see it. There hasn't been a x86 cpu release that offered a 20% over the current benchmark.

I'm not going to lose any sleep over it. If and when Intel finally delvers something worth having, I'll be there. Until then, its only conjecture and marketing hype.

Cool! Hope things work fine for you. I know you are not a fan boy, neither do I, but we have all been rooting for AMD to excel because of Intel domination, for if it's not because of AMD in x86 arena, we will still be paying high prices to buy celerons from Intel today.
Competition is good, not only for consumers, even for the stock market. Look at What NVIDIA did to ATI recently. It's a healthy competition and I like it.
I beleive Intel has learned their lesson and they will likely do something worthwhile this time around.

[Cobalt]

I'm still not sure where intel are getting the figures for the AM2 procs. To use cybercat's phrase: the lower end of their digestive systems quite probably, although intel is better qualified than any of us to make those judgements. If Intel can bring a competitive processor to the market then I'll buy it simple as that. I loved the P3 days but things changed for the worse and frankly intel will have a hard time recovering.

Now I'm not really that knowledgeable of the new architecture but what about it gives it a 20% boost over what AMD has. It would certainly be a record breaker defeat (I think) if they can make that jump from one generation to the next and I just don't see it happening. If dothan is anything to go by it will at least equal AMD but 20% is just too big a margin and smells of PR.

I've basically said the same thing about three times in this post but I tend to do that when I'm tried.

[mdzcpa]

It's a healthy competition and I like it.
I beleive Intel has learned their lesson and they will likely do something worthwhile this time around.

I hope so. Cause you're right, I'm not a fanboy. I go with whatever is best. I don't beleive in brand loyalty. Heck, I still have my blockbuster 3.0 P4C in my server cause it was such an awesome chip in its day.

Don't take my skeptical comments as negative Intel, or pro AMD. I put little value in any company's PR machine. I like hard results!!

Anyway, I'd love to see a more competitive product from Intel. As you said, we all win that way :)

[agenda2005]

GUYS,
Did INTEL deliver the 20 percent performance advantage they were saying with 2.67GHz Conroe vs FX-60 at 2.8GHz from the benchmarks?
Add on top of that, DDR2-800 that will help improve the I/O latency.
Intel have so many possibilities and room for tweaking this BEAST. Talk about L1-L1 bridge, DDR2-1066 and CSI.

[zabomb4163]

mind if I throw the same bit of reason into the mix? i'll address each of the issues 1 at a time

1) changing the HT on amd64s does not increase performance. as many have found increasing it in some cases actually decreses performance

2) the most relevant comparison for socket changes and impact of memory bandwidth would be the conversion from socket 754 to 939.

most would agree that the difference between 939 and 754 clock for clock was ~5%. the amd64 architecture just doesnt get the kind of gains from meory bandwidth increases that people are expecting. the onboard memory controller already gives the processor enourmous bandwidth

3) major clock speed increases will overcome IPC advantages conroe has over amd64.......the 2.66ghz conroe has a 20% gain on improvement over a fx60 at 2.8ghz. to balance these 2 the fx60 would need 3.36ghz (at current IPC) or 3200 (if they get a 5% improvement using M2).

Few dual core processors are passing 2.9ghz on air cooling. however i am willing to assume amd can bring this peak up 400mhz (17%) through various core changes.

According to FCG on xtreme systems the top end conroe will be clocked at 3.33ghz. assuming performance scaled linearly we are looking at a 4.168ghz fx60 to match it in terms of performance or 3.9ghz with our assumed 5% increase in clock for clock performance with the new socket. Still an unreachable number.


Feel free to debate

[nn_step]

Here is an idea what if M2 is a hell of alot more than just a Memory controller change...
as for what I speak of.. you'll just have to wait..

[zabomb4163]

Here is an idea what if M2 is a hell of alot more than just a Memory controller change...
as for what I speak of.. you'll just have to wait..

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=1884&p=11i=1884&p=13

amd xp > AMD64 was a HUGE change. correct? what is the IPC difference between the two. honestly.

the amd64 needs an IPC improvement of at least 25% to match conroe.

[nn_step]

Honestly it has more to do with what it can do in a single clock cycle than anything else...

[zabomb4163]

Honestly it has more to do with what it can do in a single clock cycle than anything else...

thats what IPC is...

[nn_step]

I know.. but I can't share details.. yet... :p:

[zabomb4163]

number of threads that can executed simulutaneously?

[krille]

DDR3? ha...! i'm so original

[dinos22]

I know.. but I can't share details.. yet... :p:
you need to tell us who you work for :p:

[Drow]

@all
Waits for AMD AM2 un judges then. Still nobody can say details.