Is AMD planning a similar re-engineering of their line of CPUs

[beggerking]

Originally posted by: Regs
You will see that AMD is reconstructing their line of CPU's with a module approach. AMD is forming a brand new approach to x86. If programmers can't design a game with dual core in mind, AMD will facilitate this by adding parralism to the entire computer. .

umm....I thought preemptive multitasking has done that already...
sounds interesting though..

 

[Fox5]

Originally posted by: SlitheryDee

Originally posted by: harpoon84
Not in the short term I don't think. Further down the track, 2008-2009, definitely.

Netburst -> Core2 = Revolutionary. 50% higher IPC, 50% lower heat output. The only thing they share is the quad pumped FSB.

K8 -> K8L = Evolutionary. Not saying it won't bring decent performance improvements over K8, from what I've seen there are some major architectural improvements, but it's still based on K8 rather than being a completely new design.

Actually it's more like Pentium3 -> Pentium M -> Core Duo -> Core 2 Duo = Evolutionary

Just because it evolved from intel's mobile line doesn't mean it's not derivative.

P3 -> netburst was revolutionary (or at least a genuinely new arch)

K6 -> K7 was revolutionary

I'd call the transistion to Core 2 Duo revolutionary in much the same way K6 to the Athlon was. Remember the K6 was a Pentium Pro class (in features and structure) level processor, so if the move from K6 to Athlon was revolutionary, than the move from the Core Duo to the Core 2 Duo, the latter of which is superior to the athlon in features and structure, must be a revolutionary move as well. (since hey, everything from P3 to Core Duo was only evolutionary changes, right?)

 

[Duvie]

Originally posted by: Zebo
AMD's been offering the same processor for 4 years now only changing memory supports and adding a second core...754 first, then added dual channel mem in 939/940, then added DDR2 support in AM2 etc but it's all same core and instruction set. While A64 was wildly sucessful they have been asleep as far as I can tell and Intel is about to pound them into dust if they don't come up with something other than adding PCIe controller (K8L) to the CPU. 65nm some say, but that's still same core, may give them some increased clock speeds like 3.5Ghz+ but so can intel go that high and on a much faster per clock core. So AMD can not compete clock speed wise. Need new core by next summer.

I agree...However no redesign is necessary "like Intel" cause the current design was intended to hold many cores and it wasn't as bad a base as Netburst was for Intel....INtel needed a redesign cause they were through with Netburst. K8 is getting long in the tooth but the fundamentals are there for multicores, multi HTT links (even newer and improved HTT), added cache, etc...

 

[Zebo]

Duvie, Kensfield will extend it's lead compared to Conroe vs X2. No matter how many cores AMD adds when intel matches them the numbers get worse for AMD because each Intel core is so much more effcient than AMD's. So if Conore leads X2 by 20%, Kensfield it will lead 4x4 types chip by at least 40%. Bottom line: Intel owns AMD like a bitch until they get a more effceint core out.

 

[Brunnis]

Originally posted by: Zebo
Duvie, Kensfield will extend it's lead compared to Conroe vs X2. No matter how many cores AMD adds when intel matches them the numbers get worse for AMD because each Intel core is so much more effcient than AMD's. So if Conore leads X2 by 20%, Kensfield it will lead 4x4 types chip by at least 40%. Bottom line: Intel owns AMD like a bitch until they get a more effceint core out.

That makes no sense at all. You don't add the percentages. If one Core 2 Duo CPU is 20% faster than one A64 X2, then two Core 2 Duos (Kentsfield) aren't going to be 40% faster than two A64 X2s. Not unless the Core 2 Duo's scaling in multi core configs are superior to that of the K8, which is highly unlikely. It's more likely that the difference between them will become less when more cores are added, since K8 is expected to scale slightly better in this regard.

EDIT: The easiest way to think about it is probably like this:

Let's say Athlon64 X2 performance is 100 (on some scale) and Core 2 Duo performance is 120. When you add another X2 to the AMD system you get twice the theoretical performance (slightly simplified, I know) and end up with a performance index of 200. When you add a second Core 2 Duo to the Intel system you also get twice the theoretical performance over the Core 2 Duo and end up with a performance index of 240.

200 / 240 = 1.2, which tells us that the performance lead over the AMD system remains 20%.

 

[Zebo]

NO it does not work that way. We saw how switching from single to dual core with Prescott vs A64 to Smithfeild vs X2, the leads were doubled in applications that utilized both cores. In otherwords what was a 5% lead became a 10% lead and so on.

 

[Brunnis]

Originally posted by: Zebo
NO it does not work that way.

Of course it works that way.

Originally posted by: Zebo
We saw how switching from single to dual core with Prescott vs A64 to Smithfeild vs X2, the leads were doubled in applications that utilized both cores. In otherwords what was a 5% lead became a 10% lead and so on.

The reason that the X2 gained a lot in certain benchmarks, compared to the PD, was that the PD lost the HyperThreading support that the regular P4 had. Slightly better scaling also helped the K8. What you're describing is impossible if you just add extra cores and all other factors and scaling being equal.

My previous post explains why your scenario isn't possible. The only way for the Conroe to Kentsfield change to extend the performance diff over the X2 would be if it scaled better (or added a feature like HT or something else that increased performance).

 

[A554SS1N]

Originally posted by: George Powell
A die shrink to 65nm will help them push clockspeeds up a bit, they certainly did well when they made the move from 130 down to 90.

What, 200MHz??!! That's hardly what I'd call "doing well" AFAIK the best 130nm FX did 2.6Ghz, and they haven't gone higher than 2.8Ghz at stock with 90nm?

 

[Bull Dog]

AMD has gone to 3.0GHz with 90nm (Opterons)

Originally posted by: Zebo
Duvie, Kensfield will extend it's lead compared to Conroe vs X2. No matter how many cores AMD adds when intel matches them the numbers get worse for AMD because each Intel core is so much more effcient than AMD's. So if Conore leads X2 by 20%, Kensfield it will lead 4x4 types chip by at least 40%. Bottom line: Intel owns AMD like a bitch until they get a more effceint core out.

Then by your math a Core2 Solo will only be 10% faster than AMD64s :disgust:

 

[aka1nas]

Originally posted by: Zebo
Duvie, Kensfield will extend it's lead compared to Conroe vs X2. No matter how many cores AMD adds when intel matches them the numbers get worse for AMD because each Intel core is so much more effcient than AMD's. So if Conore leads X2 by 20%, Kensfield it will lead 4x4 types chip by at least 40%. Bottom line: Intel owns AMD like a bitch until they get a more effceint core out.

I would think that the advantages of hypertransport over Intel's FSB will come into play more with quad-core and will level out the performance advantage somewhat with apps that can utilize that many cores.

 

[dexvx]

Originally posted by: aka1nas

Originally posted by: Zebo
Duvie, Kensfield will extend it's lead compared to Conroe vs X2. No matter how many cores AMD adds when intel matches them the numbers get worse for AMD because each Intel core is so much more effcient than AMD's. So if Conore leads X2 by 20%, Kensfield it will lead 4x4 types chip by at least 40%. Bottom line: Intel owns AMD like a bitch until they get a more effceint core out.

I would think that the advantages of hypertransport over Intel's FSB will come into play more with quad-core and will level out the performance advantage somewhat with apps that can utilize that many cores.

Nope. FSB issues are only relevant on Netburst. Even 4 cores on 1 x 1066 bus, Kentsfields scaling is still good. According to the guys at XS, in Cinebench and SuperPI, Kentsfield scales from 2->4Cores the same if not better than Opteron.

 

[Zebo]

Originally posted by: Bull Dog
AMD has gone to 3.0GHz with 90nm (Opterons)

Originally posted by: Zebo
Duvie, Kensfield will extend it's lead compared to Conroe vs X2. No matter how many cores AMD adds when intel matches them the numbers get worse for AMD because each Intel core is so much more effcient than AMD's. So if Conore leads X2 by 20%, Kensfield it will lead 4x4 types chip by at least 40%. Bottom line: Intel owns AMD like a bitch until they get a more effceint core out.

Then by your math a Core2 Solo will only be 10% faster than AMD64s :disgust:

That's exactly right if C2D lead by X% in a dual threaded app they will only lead by 1/2X in the same bechmark single vs. single core from respective companies. Watch and see. Numbers will not change in single threads.

 

[Zebo]

Originally posted by: Brunnis

Originally posted by: Zebo
NO it does not work that way.

Of course it works that way.

Originally posted by: Zebo
We saw how switching from single to dual core with Prescott vs A64 to Smithfeild vs X2, the leads were doubled in applications that utilized both cores. In otherwords what was a 5% lead became a 10% lead and so on.

The reason that the X2 gained a lot in certain benchmarks, compared to the PD, was that the PD lost the HyperThreading support that the regular P4 had. Slightly better scaling also helped the K8. What you're describing is impossible if you just add extra cores and all other factors and scaling being equal.

My previous post explains why your scenario isn't possible. The only way for the Conroe to Kentsfield change to extend the performance diff over the X2 would be if it scaled better (or added a feature like HT or something else that increased performance).

As I said HT was disabled in the becnhmarks on 630's and AMD's multithreading leads doubled with the trasition to dual core. And no I'm not going to go looking for that one article that showed this.. Duvie and I and others talked about it at the time how A64's subble leads al of a sudden became exponetial ones with dual core trasition by both companies.

 

[atom]

Originally posted by: Zebo

Originally posted by: Bull Dog
AMD has gone to 3.0GHz with 90nm (Opterons)

Originally posted by: Zebo
Duvie, Kensfield will extend it's lead compared to Conroe vs X2. No matter how many cores AMD adds when intel matches them the numbers get worse for AMD because each Intel core is so much more effcient than AMD's. So if Conore leads X2 by 20%, Kensfield it will lead 4x4 types chip by at least 40%. Bottom line: Intel owns AMD like a bitch until they get a more effceint core out.

Then by your math a Core2 Solo will only be 10% faster than AMD64s :disgust:

That's exactly right if C2D lead by X% in a dual threaded app they will only lead by 1/2X in the same bechmark single vs. single core from respective companies. Watch and see. Numbers will not change in single threads.

That is fuzzy math at best............

Multi-threaded apps don't necessarily distribute processing load evenly through all it's threads.

And your 40% assumption is....ridiculous. How did you even come up with that? So a theoretical 8-core core 2 server will have and 80% processing advantage over an 8 core opteron system? That's just silly.

 

[Duvie]

Originally posted by: atom

Originally posted by: Zebo

Originally posted by: Bull Dog
AMD has gone to 3.0GHz with 90nm (Opterons)

Originally posted by: Zebo
Duvie, Kensfield will extend it's lead compared to Conroe vs X2. No matter how many cores AMD adds when intel matches them the numbers get worse for AMD because each Intel core is so much more effcient than AMD's. So if Conore leads X2 by 20%, Kensfield it will lead 4x4 types chip by at least 40%. Bottom line: Intel owns AMD like a bitch until they get a more effceint core out.

Then by your math a Core2 Solo will only be 10% faster than AMD64s :disgust:

That's exactly right if C2D lead by X% in a dual threaded app they will only lead by 1/2X in the same bechmark single vs. single core from respective companies. Watch and see. Numbers will not change in single threads.

That is fuzzy math at best............

Multi-threaded apps don't necessarily distribute processing load evenly through all it's threads.

And your 40% assumption is....ridiculous. How did you even come up with that? So a theoretical 8-core core 2 server will have and 80% processing advantage over an 8 core opteron system? That's just silly.

We have seen reported several times that that last part is definitely not true...in 4P system the woodcrest will not keep or "grow" its leads....I think the problem is mainly the platform but nonetheless....

we have to be careful to extrapolate leads based on number of cores...I have seen many multithreaded apps...Some that gain 100% by adding a core to some that only gain 10%.....I have also seen quite a few apps with my quad core that become IO limited and thus added power wont gain you much more of anything....

 

[Maximilian]

Originally posted by: coldpower27

Originally posted by: RichUK

Originally posted by: akcorr
Just wondering since Intel's effort to rethink/re-engineer thier line of products looks to have been a great move?

Do you mean by Intel moving away from Netburst?

If so, Intel have just followed the ideals of AMD, but at present have a much better design with Core 2 against K8. AMD are going to be tweaking the K8 core which will then inturn be called K8L. So i dont think they will be reengineering anything major anytime soon, maybe with K10 but who knows.

Or you could say when back to their old ways of doing thing from back in 2000, with their power efficient and high performing Pentium 3 processor. Or taking a page out of their mobile line and just losening the grips on the thermal constraints there.

As well I agree in that it will be H1 2008 before we see anything of the magnitude of what Intel just did on the desktop for improved performance from a new architecture.

My thoughts exactly.

 

[ChunkyMonkey]

Originally posted by: atom

Originally posted by: Zebo

Originally posted by: Bull Dog
AMD has gone to 3.0GHz with 90nm (Opterons)

Originally posted by: Zebo
Duvie, Kensfield will extend it's lead compared to Conroe vs X2. No matter how many cores AMD adds when intel matches them the numbers get worse for AMD because each Intel core is so much more effcient than AMD's. So if Conore leads X2 by 20%, Kensfield it will lead 4x4 types chip by at least 40%. Bottom line: Intel owns AMD like a bitch until they get a more effceint core out.

Then by your math a Core2 Solo will only be 10% faster than AMD64s :disgust:

That's exactly right if C2D lead by X% in a dual threaded app they will only lead by 1/2X in the same bechmark single vs. single core from respective companies. Watch and see. Numbers will not change in single threads.

That is fuzzy math at best............

Multi-threaded apps don't necessarily distribute processing load evenly through all it's threads.

And your 40% assumption is....ridiculous. How did you even come up with that? So a theoretical 8-core core 2 server will have and 80% processing advantage over an 8 core opteron system? That's just silly.

you guys need to get out of the house more often & get laid..seriously.. :roll:

 

[jiffylube1024]

Originally posted by: ChunkyMonkey
you guys need to get out of the house more often & get laid..seriously.. :roll:

Thanks for taking the effort to sign up and post your well thought out opinion!

------------

One of the big problems with AMD right now is that they're stuck on 90nm, so Intel on 65nm can either add even more cache to their 4-core design or just keep die sizes smaller with 4MB of shared cache vs AMD's much larger 90nm design which will use 4 X 512K caches on each chip.

Obviously this primarily affects the bottom line of each company's balance sheet, not actual performance, but Intel has a lot more wiggle room when it comes to what kind of "extra" they can throw into new designs by being one manufacturing process step ahead. And when AMD makes the jump to 65nm in '07, Intel will be a year off from 45nm. Next year is gonna be a hot one for mainstream dual core processors and the introduction of quad-core, though!

Now that AMD's had a couple of years to fatten itself at the trough, the transition to thin profit margin cores won't be so harsh for them to take...

 

[Duvie]

Originally posted by: jiffylube1024

Originally posted by: ChunkyMonkey
you guys need to get out of the house more often & get laid..seriously.. :roll:

Thanks for taking the effort to sign up and post your well thought out opinion!

------------

With quality first post like that, we can only assume the rest wont be worth waiting to see....

 

[ChunkyMonkey]

Originally posted by: jiffylube1024

Originally posted by: ChunkyMonkey
you guys need to get out of the house more often & get laid..seriously.. :roll:

Thanks for taking the effort to sign up and post your well thought out opinion!

------------

One of the big problems with AMD right now is that they're stuck on 90nm, so Intel on 65nm can either add even more cache to their 4-core design or just keep die sizes smaller with 4MB of shared cache vs AMD's much larger 90nm design which will use 4 X 512K caches on each chip.

Obviously this primarily affects the bottom line of each company's balance sheet, not actual performance, but Intel has a lot more wiggle room when it comes to what kind of "extra" they can throw into new designs by being one manufacturing process step ahead. And when AMD makes the jump to 65nm in '07, Intel will be a year off from 45nm. Next year is gonna be a hot one for mainstream dual core processors and the introduction of quad-core, though!

Now that AMD's had a couple of years to fatten itself at the trough, the transition to thin profit margin cores won't be so harsh for them to take...

I can appreciate a decent review , even a discussion about the pro's and the cons..
"yeah,intel is faster then amd again" ,or :
"I wonder what amd's next step will be"

But after reading this thread..it's like a bunch of guys trying to be rocket scientists !
I mean, jezus..just get over the fact who's faster & move on..

This is a forum,you're not going to win a pullitzer prize with your comments..
Maybe try talking in plain english,so that everyone can understand it
(and yes, I do understand,but there's others out there)

Want to make a difference ? Go share your thoughts with the companies &
maybe you'll get to work there

And no problem guys, glad to have given my opinion

On a side note ; I do have something to contribute..
I'm running on a new AMD X2 system , which I bought this week.
If you're going to have to want the fastest pc out there,
then you'll have to buy new parts every month..

Personally, i'm an amd fan & although intel may be ahead of things
at this time, amd won't just sit aside..they've come a long way..
i'm pretty sure they were already working on new concepts for the next line of their cpu cores even before Core2 came out..
(see, no fancy mumbo jumbo here )

greets !

 

[jpeyton]

AMD is only going to re-engineer their CPU architecture when they plan on introducing major elements from ATI.

I think AMD realizes that Core 2 is only going to be beaten if they change not only CPU design, but general PC architecture as well (we have been using derivatives of the same bus architecture for quite some time now).

 

[Brunnis]

Originally posted by: Zebo
That's exactly right if C2D lead by X% in a dual threaded app they will only lead by 1/2X in the same bechmark single vs. single core from respective companies. Watch and see. Numbers will not change in single threads.

So, if you have two CPUs and theoretically double the performance of each, the faster CPU will double its lead over the slower one (all other things being equal)?

I'm not into magics, so why don't you explain where this performance increase comes from? We know for a fact that both the Core architecture and K8 scale almost equally well when increasing the number of cores. So, how on earth can the performance difference change like you're saying?

Also, as I said, you can't really use the P4 -> PD vs A64 -> X2 as an example of this, since HT was disabled on the PD. The Core architecture doesn't have HT and as far as I know, everything will remain the same featurewise on the Kentsfield. Your theory should be easy enough to prove with some simple numbers. Please enlighten me.

EDIT: BTW, you aren't confusing the absolute performance difference with the relative difference? The relative difference in performance will of course be about twice as large. For example, if the Core 2 Duo performed 100 calculations more per second, the Kentsfield would perform about 200 calculations more than the quad core K8. The relative difference between the two would still be about the same, though (in this case roughly 20%).

 

[shady28]

"I think AMD realizes that Core 2 is only going to be beaten if they change not only CPU design, but general PC architecture as well"


And that is exactly right.

Those that have been around a while, if you think back to the mid 1990s and the advances made then, you should realize that this 'major' performance increase from Intel Core Duo 2 is really not all that revolutionary. We are talking a 20-30% boost in performance at the top end. Back in the mid 90s, we had a 100% boost in top end performance roughly every 18 months.

It's amazing to me that so much hype is made about such a paltry increase these days. It just goes to show how performance starved the market is, and has been. Lets keep in mind that the 3ghz P4 was introduced in 2001 - 5 years ago. That is still a fine CPU to this day, quite capable of running most any application or game without causing any distress for the user. I can't think of any other time period when that was the case.

The point being, increases in individual CPU performance are so relatively paltry now that it's almost a complete non-factor. Multi-core is where it's at, and whoever can get the most cores with the architecture that makes the best use of them out to market the fastest, wins. Intel definitly has a leg up on AMD now, and the next 12 months probably won't be AMD's best. But lets be real - most people are still out there buying $150 CPUs - a casual look in the forums will show a lot more people buying x2 3800+ CPUs (a $115 cpu now) than buying anything much over $200.

AMD has several things on the cooker right now, but it will probably be next year before we see any of it. 4x4 (quad CPU, quad GPU) would be my bet for the next major revolution. If AMD can get the architecture working efficiently (ie, its complex getting multiple cores to scale up well), then that would be my bet for the next 'revolution' in desktop processing power.

 

[ZerthX]

Originally posted by: shady28
But lets be real - most people are still out there buying $150 CPUs - a casual look in the forums will show a lot more people buying x2 3800+ CPUs (a $115 cpu now) than buying anything much over $200.

Originally posted by: ChunkyMonkey
On a side note ; I do have something to contribute..
I'm running on a new AMD X2 system , which I bought this week.
If you're going to have to want the fastest pc out there,
then you'll have to buy new parts every month..

Greetings, I'm new here. I read this forum from time to time for quite long, but I usually don't have enough time to contribute.

Although I've been computer fan for more than 20 years now, I've always been "real" - I've never spent too much money for that hobby of mine, at least not unreasonable amount.

Especially with processors, OCing was always pretty good way to get beter performance from lower-end components. Last 10 or so years I've always chosen cheaper processors that are good overclockers - preferably 50% (was easy with CeleronA and Northwood). I don't care about extreme overclocks, just moderate ones, attainable with air and with minor voltage increases, and as silent as possible.

I really look forward to Conroe, being GREAT overclocker, but nevertheless I bought last week A64 X2 3800+ and Asrock 939Dual-sata2 - great combo, VERY cheap upgrade since I could use my old DDR and (still) decent AGP card.

X2 OCed 2,5ghz on stock voltage (1,35v) and runs pretty cool and quiet with Freezer 64 Pro. Not quite 50% but nice nevertheless.

Conroe system can wait. Cheaper mobos, stable bioses, faster (and cheaper) DDR2... Maybe around Christmas will be the right time!

Originally posted by: ChunkyMonkey
Personally, i'm an amd fan & although intel may be ahead of things
at this time, amd won't just sit aside..they've come a long way..
i'm pretty sure they were already working on new concepts for the next line of their cpu cores even before Core2 came out..

I think Conroe is maybe the best procesor ever launched (not just the fastest currently), Intel or AMD, no matter. I've never been "fan" of either of them, just happy that there's a healthy competition. They were always pretty close, performance wise. Now Intel has the clear winner - C2D, and AMD can't do anyting about it, at least not in the near future. I don't feel sorry for AMD because they were so self-satisfied last few years and totaly unprepared for Conroe. And I'm VERY glad that they were FORCED to lower their X2 prices...

 

[lopri]

Originally posted by: ChunkyMonkey
I can appreciate a decent review , even a discussion about the pro's and the cons..
"yeah,intel is faster then amd again" ,or :
"I wonder what amd's next step will be"

But after reading this thread..it's like a bunch of guys trying to be rocket scientists !
I mean, jezus..just get over the fact who's faster & move on..

This is a forum,you're not going to win a pullitzer prize with your comments..
Maybe try talking in plain english,so that everyone can understand it
(and yes, I do understand,but there's others out there)

Want to make a difference ? Go share your thoughts with the companies &
maybe you'll get to work there

And no problem guys, glad to have given my opinion

On a side note ; I do have something to contribute..
I'm running on a new AMD X2 system , which I bought this week.
If you're going to have to want the fastest pc out there,
then you'll have to buy new parts every month..

Personally, i'm an amd fan & although intel may be ahead of things
at this time, amd won't just sit aside..they've come a long way..
i'm pretty sure they were already working on new concepts for the next line of their cpu cores even before Core2 came out..
(see, no fancy mumbo jumbo here )

greets !

There is some forum called HardOCP where you can feel more comfortable and excited at the same time. It won't take 5 mins to realize how perfect that forum is for you. Go take a look if you haven't already. --> http://www.hardforum.com