What is AMD's next move?

[Hacp]

Originally posted by: Duvie

Originally posted by: Diasper
Single core = Intel with a slight win but <10% (except maybe encoding?)

Dual-core = Neck and neck or even AMD with a slight edge (except maybe encoding again?)

IF AMD implements a better CPU-cache system including just faster cache we'll see a much more interesting scenario emerge.

OK someone is smoking something...


Single Core = pretty much a tie....AMD is much better at gaming for liked price chips but behind 10% in encoding and such. AMD is faster inmath apps but behind in CAD apps that can take advanatge of HT...

Dual Core = not even close....Similar priced cpus are anywhere from 20-30%+ over Intel and even in multimedia....Wake up!!!

Since I seem to prove what I state start pulling some links to prove what you stated...

Maybe he means the new Pentium "V" and he's speculating on the performance compared to AMD?

 

[Gamingphreek]

This is ridiculous. I cant believe we have another one of these threads.

Please OP. You list me the "Promising specs" and tell me exactly what they do.

-Kevin

Edit: Also please explain to me how hyperthreading works and why in the world they would even consider using it in a 14 stage pipeline.

 

[Furen]

Originally posted by: Gamingphreek
This is ridiculous. I cant believe we have another one of these threads.

Please OP. You list me the "Promising specs" and tell me exactly what they do.

-Kevin

Edit: Also please explain to me how hyperthreading works and why in the world they would even consider using it in a 14 stage pipeline.

SMT (Hyperthreading, in this case) works by utilizing unused execution units to run two threads in parallel. This works very well when the pipeline is deep, since execution units will idle during cache misses, branch mispredictions, etc. It also works really well in CPUs that are so wide that they dont fill all their execution units at once. If you remember that AMD has stated that they believe the optimal issue width for x86 code is 3, meaning that they keep their execution hardware working pretty much most of the time. If this applies to all x86 code, not just code running on AMD's architecture then Intel's 4 issue Conroe should have some idle execution hardware that can be used with SMT. Adding SMT (of course) requires a higher transistor count because though the execution units are not duplicated, some parts of the front end have to be. This is why you don't just slap SMT on any processor, but rather you use it on a processor that can gain more from it than you would by just using those transistor for something else (like cache, for example).

 

[Lithan]

Personally I dont think Intel will give any more performance than they have to. Their new chips will probable put them about where they were with northy C's versus Xp's at best. But PRICE is what I want to see. Now that AMD has jacked the price of their modern socket chips up about 700% at release and 250% when they are "mainstream" over where they were when Northwood C's came out <$40(1700+ B) versus $320(3500+ pre-winny) versus $140(3000+ Venice)>, Intel can SURELY compete with them price-wise while still making money through the nose. They simply choose not to release modern procs (excluding Cellys) at price levels under ~ $200 to consumers.

If intel can offer me slightly superior performance for the same price and at the heat/etc specs they are bragging about, it's an obvious choice. But if we are looking at $250 starting price for their new 'evolved-dothan' chips without much in the way of improvements, Amd might still be a better choice at least in the single core arena. Dual core, unless Amd slams prices down sub $150-175, a dothan-esque Duallie @ $200-300 is an unchallenged threat.

 

[Gamingphreek]

Originally posted by: Furen

Originally posted by: Gamingphreek
This is ridiculous. I cant believe we have another one of these threads.

Please OP. You list me the "Promising specs" and tell me exactly what they do.

-Kevin

Edit: Also please explain to me how hyperthreading works and why in the world they would even consider using it in a 14 stage pipeline.

SMT (Hyperthreading, in this case) works by utilizing unused execution units to run two threads in parallel. This works very well when the pipeline is deep, since execution units will idle during cache misses, branch mispredictions, etc. It also works really well in CPUs that are so wide that they dont fill all their execution units at once. If you remember that AMD has stated that they believe the optimal issue width for x86 code is 3, meaning that they keep their execution hardware working pretty much most of the time. If this applies to all x86 code, not just code running on AMD's architecture then Intel's 4 issue Conroe should have some idle execution hardware that can be used with SMT. Adding SMT (of course) requires a higher transistor count because though the execution units are not duplicated, some parts of the front end have to be. This is why you don't just slap SMT on any processor, but rather you use it on a processor that can gain more from it than you would by just using those transistor for something else (like cache, for example).

I know what it mens. I asked the OP if he could tell me what this is because i dont think he has a clue what he is talking about.

-Kevin

 

[Hard Ball]

Originally posted by: bunnyfubbles
Its probably been said before, but PV isn't something AMD has to step up to, the PV is Intel stepping up to AMD to match the A64/X2. However they'll need something better in terms of marketing, because now intel can brag they're just as fast

After the advent of Conroe/Woodcrest, AMD will be losing dollars, and lots of it, until K10 shows up sometime Q4 06 - Q1 07. I think you can put this one in the bank. Even is the performance is dead even (which is doubtful), The K8/K9 would not enjoy the same status as it does now in the enterprise and enthusiast community as the preeminent performance platform. Remember, intel sells by brand name and marketing, and AMD sells by performance.

 

[Viditor]

Originally posted by: Hard Ball

Originally posted by: bunnyfubbles
Its probably been said before, but PV isn't something AMD has to step up to, the PV is Intel stepping up to AMD to match the A64/X2. However they'll need something better in terms of marketing, because now intel can brag they're just as fast

After the advent of Conroe/Woodcrest, AMD will be losing dollars, and lots of it, until K10 shows up sometime Q4 06 - Q1 07. I think you can put this one in the bank. Even is the performance is dead even (which is doubtful), The K8/K9 would not enjoy the same status as it does now in the enterprise and enthusiast community as the preeminent performance platform. Remember, intel sells by brand name and marketing, and AMD sells by performance.

You're making a LOT of wild assumptions here...

1. You're assuming Conroe/Woodcrest will come out on time and as advertised...Itanium didn't, Prescott didn't...

2. You're assuming that Conroe/Woodcrest will be superior to K8...Prescott was supposed to be but wasn't even close.

3. You're assuming that AMD have no more improvements at the stepping level...remember the "E" stepping and how much that improved performance/power? We also have no idea how much 65nm will effect the K8...

4. You're assuming that a slight (theoretical) performance lead by Intel's next gen chips will completely stop AMD's momentum in sales...this is absolutely not true for a lot of reasons.

a) It takes at least a year to validate a new server platform, so widespread adoption of Woodcrest couldn't really take place until the end of 2007
b) The companies that are already converting to AMD and those that do so over the next 2 years have a vested interest in that platform (similar to what we've seen for Intel for the past 2 years). This means that AMD would have to "sit on their hands" and really screw up to actually lose those customers.
c) Everything we've seen on the nextgen Intel platform appears to make Intel only competitive with AMD, not greatly superior. This is not a compelling reason to shift platforms.

 

[Gamingphreek]

Originally posted by: Viditor

Originally posted by: Hard Ball

Originally posted by: bunnyfubbles
Its probably been said before, but PV isn't something AMD has to step up to, the PV is Intel stepping up to AMD to match the A64/X2. However they'll need something better in terms of marketing, because now intel can brag they're just as fast

After the advent of Conroe/Woodcrest, AMD will be losing dollars, and lots of it, until K10 shows up sometime Q4 06 - Q1 07. I think you can put this one in the bank. Even is the performance is dead even (which is doubtful), The K8/K9 would not enjoy the same status as it does now in the enterprise and enthusiast community as the preeminent performance platform. Remember, intel sells by brand name and marketing, and AMD sells by performance.

You're making a LOT of wild assumptions here...

1. You're assuming Conroe/Woodcrest will come out on time and as advertised...Itanium didn't, Prescott didn't...

2. You're assuming that Conroe/Woodcrest will be superior to K8...Prescott was supposed to be but wasn't even close.

3. You're assuming that AMD have no more improvements at the stepping level...remember the "E" stepping and how much that improved performance/power? We also have no idea how much 65nm will effect the K8...

4. You're assuming that a slight (theoretical) performance lead by Intel's next gen chips will completely stop AMD's momentum in sales...this is absolutely not true for a lot of reasons.

a) It takes at least a year to validate a new server platform, so widespread adoption of Woodcrest couldn't really take place until the end of 2007
b) The companies that are already converting to AMD and those that do so over the next 2 years have a vested interest in that platform (similar to what we've seen for Intel for the past 2 years). This means that AMD would have to "sit on their hands" and really screw up to actually lose those customers.
c) Everything we've seen on the nextgen Intel platform appears to make Intel only competitive with AMD, not greatly superior. This is not a compelling reason to shift platforms.

QFT. Everyone here is thinking this is a miracle chip and AMD is just going to just throw their hands up and call quits.

-Kevin

 

[Hard Ball]

Originally posted by: Gamingphreek

Originally posted by: Viditor

Originally posted by: Hard Ball

Originally posted by: bunnyfubbles
Its probably been said before, but PV isn't something AMD has to step up to, the PV is Intel stepping up to AMD to match the A64/X2. However they'll need something better in terms of marketing, because now intel can brag they're just as fast

After the advent of Conroe/Woodcrest, AMD will be losing dollars, and lots of it, until K10 shows up sometime Q4 06 - Q1 07. I think you can put this one in the bank. Even is the performance is dead even (which is doubtful), The K8/K9 would not enjoy the same status as it does now in the enterprise and enthusiast community as the preeminent performance platform. Remember, intel sells by brand name and marketing, and AMD sells by performance.

You're making a LOT of wild assumptions here...

1. You're assuming Conroe/Woodcrest will come out on time and as advertised...Itanium didn't, Prescott didn't...

2. You're assuming that Conroe/Woodcrest will be superior to K8...Prescott was supposed to be but wasn't even close.

3. You're assuming that AMD have no more improvements at the stepping level...remember the "E" stepping and how much that improved performance/power? We also have no idea how much 65nm will effect the K8...

4. You're assuming that a slight (theoretical) performance lead by Intel's next gen chips will completely stop AMD's momentum in sales...this is absolutely not true for a lot of reasons.

a) It takes at least a year to validate a new server platform, so widespread adoption of Woodcrest couldn't really take place until the end of 2007
b) The companies that are already converting to AMD and those that do so over the next 2 years have a vested interest in that platform (similar to what we've seen for Intel for the past 2 years). This means that AMD would have to "sit on their hands" and really screw up to actually lose those customers.
c) Everything we've seen on the nextgen Intel platform appears to make Intel only competitive with AMD, not greatly superior. This is not a compelling reason to shift platforms.

QFT. Everyone here is thinking this is a miracle chip and AMD is just going to just throw their hands up and call quits.

-Kevin

Not really, I don't think anyone is thinking that. Conroe does not need to be a miracle chip to better the K8; all they need is about 15% increase in performance from the Yonah to take over the lead from K8/K9. And that's why AMD is probably working fast and furiously on K10, trying to make sure that it isn't late, or it could even make an early appearance.

It is clear that the M2 CPUs, along with Taylor, are a perfectly OK lineup to counter the Presler/Yonah; but it simply will not be on the same footing as any later Intel generation; and is almost certain to lose the performance crown.

 

[Gamingphreek]

You have no basis for this performance speculation. You are simply assuming that since it is a new architecture it will beat everything. We cannot accurately speculate or comment on this new architecture. Even Intel doesn't know how well it will work.

Additionally, 15% to take the lead? Ok, what if power consumption were to remain the same which would mean the leakage, and heat would remain the same also. Thankyou, i would rather have a cool running chip that is neck and neck or slightly behind.

Simply too much speculation right now, not enough fact.

-Kevin

 

[andyliu88]

I hope Conroe beats the current K8 to death, so I can buy my X2 dirt cheap. ^_^

 

[MemberSince97]

Originally posted by: meson2000
I agree with Hacp. That is what I have been reading on the net too. 2006 will just bring DDR2 and the 65nm process. If AMD releases the 3ghz FX 59 on the 90nm process in Qtr 1 2006, then I think they will be around 3.4 by the end of 2006/1st Qtr 2007. No significant changes to the core in 2006. I think the reason why no one has heard anything about K10 (K9 has been skipped to bring in dual core K8's) is because I think AMD's new push is just adding more cores to the CPU. 2007 will bring a quad core Athlon with probably a shared L3 cache and maybe (but highly unlikely) a next gen DDR3 memory controller. I really think that AMD thinks they have found the sweet spot with the K7-K8 core and are only going to make minor tweaks going forward and just increase clock speed and add more cores. I seriously doubt the 'mythical' K10 will be a major redesign considering the small amount of changes they made to the core going from K7 to K8.

This seems to me the most plausible and least radical approach...

 

[MDme]

we all honestly don't know how the P-V will perform. but one thing i notice is this:

intel is trumpetting: performance per watt as their mantra. if that is their selling point, i PRESUME that intel's chips won't be beating the current AMD line by much (if even) since they would be marketting PERFORMANCE if this were the case. My guess however would be for intel to really shine in being low power high performance chips - unlike the high power consuming P4s.

I think intel will beat AMD on a clock for clock basis.

as to whether they will seize the performance crown, I think that will depend on the scaling (clock) of their chips. If they can achieve sufficient clock speed they may beat AMDs top chips. remember, AMD will still be able to squeeze out clockspeed speed grades from the current architecture especially once they move to 65nm.

 

[Valkerie]

Like it'll matter.

AMD or Intel, one or the other, you really only have two companies to choose from.

 

[Gamingphreek]

I miss the days with Cyrix. I remember my first non-intel chip PR-166+ Good times...goodtimes.

-Kevin

 

[Diasper]

Originally posted by: Hacp

Originally posted by: Duvie

Originally posted by: Diasper
Single core = Intel with a slight win but <10% (except maybe encoding?)

Dual-core = Neck and neck or even AMD with a slight edge (except maybe encoding again?)

IF AMD implements a better CPU-cache system including just faster cache we'll see a much more interesting scenario emerge.

OK someone is smoking something...


Single Core = pretty much a tie....AMD is much better at gaming for liked price chips but behind 10% in encoding and such. AMD is faster inmath apps but behind in CAD apps that can take advanatge of HT...

Dual Core = not even close....Similar priced cpus are anywhere from 20-30%+ over Intel and even in multimedia....Wake up!!!

Since I seem to prove what I state start pulling some links to prove what you stated...

Maybe he means the new Pentium "V" and he's speculating on the performance compared to AMD?

I did, which is what this thread is about - or if you read my other threads my position would also have been clear. I'm not nuts to say that about things currently

Only speculation and I guess this speculation goes along the lines that for starters we know Yonah will be very competetive with AMD, yet Intel is coming out with a newer supposedly more modern architecture. Now presumably as they are choosing to use this over Yonah it is presumably better - which may be one or a number of these fields: scalability (mhz), scalability (in making 'many' multi-core), increased performance in oneor a number of areas, offers cost savings benefits with scaling.

Also obviously, speculation is based on what we know of the specs and archiecture and comparing them to current and past archectures and what likely effects it should then have bearing in archiectural problems/challenges/efficiencies and inefficiencies in mind.

This is all speculation and we'll only know once we get our hands on them.

On that note a perhaps interesting bit of news: it appears a German website computerbase.de already has a very early benchmark of Yonah. The banchmark doesn't say much to me except that Yonah won't be blowing AMD away - probably as their dual-core archiecture isn't as efficient as hypertransport and direct core-to-core communication of AMD.

 

[glugglug]

PV is slower than a 3800+. Why would AMD need a new chip?

 

[Viditor]

Originally posted by: MemberSince97

Originally posted by: meson2000
I agree with Hacp. That is what I have been reading on the net too. 2006 will just bring DDR2 and the 65nm process. If AMD releases the 3ghz FX 59 on the 90nm process in Qtr 1 2006, then I think they will be around 3.4 by the end of 2006/1st Qtr 2007. No significant changes to the core in 2006. I think the reason why no one has heard anything about K10 (K9 has been skipped to bring in dual core K8's) is because I think AMD's new push is just adding more cores to the CPU. 2007 will bring a quad core Athlon with probably a shared L3 cache and maybe (but highly unlikely) a next gen DDR3 memory controller. I really think that AMD thinks they have found the sweet spot with the K7-K8 core and are only going to make minor tweaks going forward and just increase clock speed and add more cores. I seriously doubt the 'mythical' K10 will be a major redesign considering the small amount of changes they made to the core going from K7 to K8.

This seems to me the most plausible and least radical approach...

There are certainly other speculative changes on AMD's roadmap...
How about putting the Southbridge on the chip as well as the Northbridge?
Increasing the bandwidth of HT?
Support for FB-DIMMs allowing 256GB of Ram on a single server board?
Increasing the number of actual memory controllers on the chip?

Those are a few possibilities off the top of my head...the point is we just don't know.
At this point in time, Intel is in the position of promising anything they can to keep customers from continuing to make the change to AMD...once that change occurs, it's very difficult to get them back (at least in business it is...).
Will Conroe/Woodcrest actually show up on time and as promised? Who knows...it would be good for us users if they did as competition is always a good thing (assuming of course AMD can get an injunction against Intel's present "marketing" practices).
Obviously at the moment, it's an AMD world for those who understand them...but the Utopia for us all is a 50/50 split between the 2 where they have to compete for our dollars every day!

 

[Furen]

Originally posted by: Viditor

There are certainly other speculative changes on AMD's roadmap...
How about putting the Southbridge on the chip as well as the Northbridge?
Increasing the bandwidth of HT?
Support for FB-DIMMs allowing 256GB of Ram on a single server board?
Increasing the number of actual memory controllers on the chip?

Well, I dont think there's much point to keeping this thread alive any longer but I'd like to add this:

First, latency to the southbridge is not a performance bottleneck right now (which is why it's the farthest away from the CPU) and bandwidth could be increased so easily (if it ever does become a bottleneck) that there's no point to getting it on the CPU (and increasing die size for no good reason). In fact, having multiple northbridges (like on the nForce4Pro) is probably better since motherboards can be costumized to fit the consumer's needs.

I wouldnt count on having more memory channels because the pin-count on these CPUs is already insane, no need to make packaging more expensive, plus DDR2 mitigates the possible need for it in quad core CPUs.

I'm sure we'll see more HT bandwidth on socket F (since HT bandwidth does become a bottleneck once you go over 2 CPUs) and I hope we'll see an FB memory controller with lots of channels (since the pincount is lower I'd hope 4, but 2 might be enough, since these CPUs will always be at least in pairs). I have read that FB dimms using DDR2 and DDR3 should be socket compatible, which would be a great boon to AMD (then again, it'd probably mean that intel will jump into the integrated mem controller bandwagon).

 

[Vee]

Some nice points have been made in this thread, and some points that hint at a false perspective of things. I'm not going to revisit any of that, even if there may be reasons to.
(I'm feeling this great reluctance to post these days. I'm sure the "content" of some threads here lately have something to do with that...:disgust: )

Instead I'm going to limit myself to a couple of points I think have yet to be made here:

Intel have not come up with much new. The Conroe seems basically a souped up and refined K8 Athlon64. (And no, I'm not missing 'code-fusion'. Very clever and yet another nail in the coffin for compiler dependant ISA design, VLIW, RISC, EPIC, Cell.)

AMD have every reason to be content and satisfied with what Intel have revealed:
- There are no surprises!



Then a point that has been made previously, but deserves to be stressed:
Conroe have been cooking for a while, but don't forget that it will continue to cook for a good while yet. - It seems some are thinking Intel have launched this chip?
2H 06 would typically mean available '07. Wouldn't surprise me if some people on this forum are running AMD OpteronII or Athlon X4 by about then.

 

[Viditor]

Originally posted by: Furen

Well, I dont think there's much point to keeping this thread alive any longer but I'd like to add this:

First, latency to the southbridge is not a performance bottleneck right now (which is why it's the farthest away from the CPU) and bandwidth could be increased so easily (if it ever does become a bottleneck) that there's no point to getting it on the CPU (and increasing die size for no good reason). In fact, having multiple northbridges (like on the nForce4Pro) is probably better since motherboards can be costumized to fit the consumer's needs.

I wouldnt count on having more memory channels because the pin-count on these CPUs is already insane, no need to make packaging more expensive, plus DDR2 mitigates the possible need for it in quad core CPUs.

I'm sure we'll see more HT bandwidth on socket F (since HT bandwidth does become a bottleneck once you go over 2 CPUs) and I hope we'll see an FB memory controller with lots of channels (since the pincount is lower I'd hope 4, but 2 might be enough, since these CPUs will always be at least in pairs). I have read that FB dimms using DDR2 and DDR3 should be socket compatible, which would be a great boon to AMD (then again, it'd probably mean that intel will jump into the integrated mem controller bandwagon).

Latency hasn't been an issue before, that's true...but with the advent of PCIe that appears to be changing...remember that both Intel and AMD have been discussing placing the PCIe bus directly on the CPU recently. Both video and things like 10GBe would be helped immeasureably by this...

Vis a vis pin count...the Socket F is a 1207 pin...most of the chip architect guys I know have been strongly speculating that this is for a second memory controller on the Quad Cores...I'm not sure why you think that DDR2 mitigates the need for a second controller though. Could you explain that for me please? Cheers...

FBDIMMs are socket compatible across DDR, DDR2, and DDR3...

 

[thepingofdeath]

The whole core redesigns and clock speeds and what not are reaching their practical limits. All they can really do now is sharpen their hardware algorithms and add more cores. What's really going to make a difference is the advent of new technologies, most of which are being developed by IBM. The first one of these that will hit the market (3 years probably) is fiber-on-silicone products. No more copper inlay--cpu-transistor size fiber signal translators are already here, they're just expensive to produce at the moment. Once that happens, latency is hardly an issue. The next big thing is taking advantage of more than the 0 and 1 states of electrons, made possible by our better understanding of quantum mechanics. This means rethinking computers from the ground up. It's possible, and it's getting to be practical, but that's about 10 years out. After that, we get the BIG, BIG thing: instantaneous transmission. IBM can now literally teleport information (the quantum states of electrons, not matter, per say) from one location to another. The hardware required to do this is beginning to reach a reasonable size. Once that happens, there is no latency. That's about 15 years out.

 

[Pabster]

I could see a pair of memory controllers on-die with the quad-core. Makes perfect sense. DDR2 indeed increased bandwidth but I think all 4 cores feeding off the same controller might be a bottleneck.

Now how about dual DDR2 memory controllers AND PCI-E on the die. More wet dreams I suppose...

 

[Wahsapa]

duvie... tejas was cancelled so you think they came up with conroe between then and now...?

you seriously think that...?

i just stopped reading their because the thread obviously had to get worse...

 

[Duvie]

Originally posted by: Wahsapa
duvie... tejas was cancelled so you think they came up with conroe between then and now...?

you seriously think that...?

i just stopped reading their because the thread obviously had to get worse...

That is THERE not THEIR.....

Actually no I dont but with Tejas it was clear their DESKTOP platform was not going in that direction...It is perfectly possible the conroe was still going to be a mobile platform variant....Basically with all the issues of heat and power INtel had its best answer cooking "on the backburner" so to speak and moved to it...With that in mind think of how the laptop and mobile market is designed...NOT FOR PERFORMANCE PERSAY....hence why we may be seeing nothing earth shattering here except for very nice rumored wattage numbers....

I didn't mean they couldn't have been working on it, I am just saying this is and was not the plan for desktop up until a point as early as last year....That is all...

Edit: I dumbed it up for you....dont be so arrogant...looked like a fool with your THEIR spelling....