Is AMD going to be in trouble 2006?

[miketheidiot]

I have heard a while back that with k10 (successor to dualcore aka k9) will have some kind of signifigant change to the execution core, which is still almost identical to the oldest k7's.

anyways dualcore PM's in '06 is a long ways away, assuming they actually come out on the desktop by then. between then and now intel has to go through the 65nm change, and who knows how that will go.

 

[Duvie]

Originally posted by: miketheidiot
I have heard a while back that with k10 (successor to dualcore aka k9) will have some kind of signifigant change to the execution core, which is still almost identical to the oldest k7's.

anyways dualcore PM's in '06 is a long ways away, assuming they actually come out on the desktop by then. between then and now intel has to go through the 65nm change, and who knows how that will go.

exactly!!! They encounter leakage issues like they have with their disasterous 90nm change and this will all be a moot point...

I believe AMD has stated their first dual core will be on the 90nm process...P4 duak prescot should be as well, but I think from leaked power usage we can all see where that one is headed...

 

[xbdestroya]

I think 64-bit functionality will be a much bigger deal 12+ months out from now. The fact is that it can bring some pretty large speed improvements if designed for, and if Windows 64 gains any traction whatsoever, Dothan on the desktop is not going to cut it - especially if games and stuff start coming out 64-bit compiled and optomized. Now, obviously 64-bit is doing many Windows users any good right now, so that makes things a little harder to predict into 2006. I think the Dothans are great, but a great 32-bit processor may not cut it if things really do go 64-bit.

 

[imported_composer]

Originally posted by: Green Man
Dothan has 2M L2, so I would be surprised to see it get a huge benefit from dual channel. I don't think there is much benefit to going with the higher latency of DDR2 until DDR2 speeds get up high enough to negate the effect. By then AMD will have DDR2 as well. All of the information I have seen has shown Yonah without iAMD64 as well. I don't see how Intel can plan to move it to the desk top no matter how well it performs in 32 bit when the buzzword is 64 bit. I think A64 will continue to hold an IPC as well as a clockspeed advantage over the Pentium M successor, but I hope intel can find a way to push AMD into at least releasing some faster clock speeds.

OTOH, if you buy a 3000+ now and overclock it to 3800+, you are sure to have a near top of the line processor for the next year and a half +.

Agreed. I'm seeing the 3200 at 10x275 and up, crushing speeds, and LOW temps (39C) at 1.425 vcore vs my XP 2600 at 60C.

My predictions: Intel is behind in the dual core race to some extent, especially with 64 bit chips, however, MSFT delayed 64 bit OS to accomodate INTEL..which is ok, AMD 3200 512 cache chips are clocking way way way up there (.09 winchesters) and hit a wall at 10x280, yet notice that the WALL is not a TEMP wall at all, I belive it's some sort of lock somewhere in the chain, regardless, if a 3200 can do 10x275 and be 40C, and run near FX55 speeds, beating a 2MB P43.8 and be at 40C, expect AMD to release DUAL cores, 1MB cache and 512 cache, ALL FX .09's (if ever, single CPU's) will be 1MB cache as well and the mainstream will be 939 1MB cache FX/64, high end 4500/5000 FX single core 1 MB cache at 3.0 ghz, then the duals at 1 and 2 MB cache.

Where talking some serious power here. AMD has the heat under control, Intel does not, AMD is also ordering .065 machines, so exepct even higher speeds, lower temps.

And guess what? DUAL CORES ARE COMING SOON! What we have now in terms of power will be ecliped almost two fold (not games though) but in rendering, video audio plug-ins, and so on before APRIL.....can't wait, just hope it stays 939 pins, if not, then I hope to see the FX series end up on the .09 process, but what may happen is the FX line goes DUAL core/single core 2 MB cache, and the 939 line gets the 1MB cache, essentially turning all the 939's into FX (1mB cache) and all the FX to 2MB single and dual cores.

I wouldn't worry about AMD, if your holding intel (INTC) though, I may worry a bit.

 

[VirtualLarry]

Originally posted by: Green Man
Dothan has 2M L2, so I would be surprised to see it get a huge benefit from dual channel. I don't think there is much benefit to going with the higher latency of DDR2 until DDR2 speeds get up high enough to negate the effect.

On the other hand.. the larger L2 would help offset the latency penalty of DDR2, and the additional bandwidth of DDR2 could offset/negate the need for dual-channel. On the whole, I could easily see a dual-core Dothan sucessor on the desktop running with single-channel DDR2. (Well, ok, possibly dual-channel might be needed for dual-core, depends on how DDR2 speeds ramp in relation to processor development.)

In the future though, I wonder if the way that the Pentium/Celeron lines will be delineated, will not be by cache sizes, but rather by whether or not the chips are dual- or single-core. (Single-core chips either being a seperate line, or also comprised of chip with one defective core that is disabled during mfg.)

A question though - will dual-core Dothan sucessors, have individual duplicated L2 caches, or will both cores feed from one shared L2 cache?

 

[Zebo]

Originally posted by: stratman

Originally posted by: Zebo

Originally posted by: CraigRT
competition is good! but this will only help AMD in the long run! :evil:

There will be no compitition in rendering or science apps. I mean the M just gets crushed...
http://www.gamepc.com/labs/vie...asp?id=gmso&page=7

http://www.gamepc.com/labs/vie...asp?id=gmso&page=9

Is there an easy way to fix this to make the dothan a well-rounded proc?

And could Intel throw on a on-die ddr controller?

I think so.. I'm well aware of Dothans glaring holes. However..add more registers.. more FPU units etc. Like I mentioned and everything should be better. Maybe PM can comment on it's flaws while at the same time why it does so darn well, better than A64 in some game tests and Lame.

 

[VirtualLarry]

Originally posted by: Zebo
You guys that think memory won't do much for dothan, fine. However there were some performance snips to the core they had to make for lower power in it's intended function, laptops. When they bring a derivative to desktop they are more free to add features, ramp etc which can only increase dothans already excellent performance since they are no longer restricted to 20-28W envelope.

Well, maybe, but... the biggest advantage of the Dothan over Prescott, besides the IPCs of the architecture, is the thermal envelope. If Intel suddenly decides to abandon those thermal-design principles once they produce a seperate desktop-only variant of the Dothan.. well, then they're looking at the same sort of thermal brick-wall that they hit with Prescott. I think that doing so would be extremely unwise. I haven't heard anything about the power-saving features of the Dothan having major negative effects on performance, although I could be wrong. I'll go read that link.

I think that they (Intel) are starting to think in terms of cycles/watt, rather than cycles/sec, and the trend from hereon out will be "wider, not faster", so if Dothan sucessors never clock faster than 3.0Ghz, will that really be a problem? If they've got an 80-120W thermal envelope to work with in today's desktop chassis (on the high-end of air), then they could plunk down 4 Dothan-like cores with 30W each then, and probably have higher effective performance, without scaling clock speeds again into the stratosphere.

 

[VirtualLarry]

Originally posted by: Marlin1975

Originally posted by: DrMrLordX

Originally posted by: Sc4freak
On Die? DDR? Intel? I think that if Intel ever includes an on-die memory comtroller it'll be for DDR2.

Intel won't go for on-die memory controllers if they continue to push BTX.

That and they make a PILE of money from their chipsets. Look at AMD's chipssets in the market and most offer about the same performance, so for a oem you would choose on price. That would hurt intels $$$ to much. But for AMD that only makes server chipsets, putting the mem controller on die adds performanmce and they don;t lose any money in the chipset war.

For as much money that Intel makes in chipsets, they still make more off of their CPUs I'm pretty certain, and if the market demands that Intel integrate a memory-controller into the CPU in order for their CPUs to remain performance-competitive in the market, you can be assured that they will do so, even if it means a slight hit to their chipset revenues, although I honestly don't see how that will follow. Whether or not the DRAM controller is in the chipset or the CPU - you still need a DRAM controller, DRAM, and a system chipset, in addition to the CPU itself. They'll just add more chipset peripheral features in order to differentiate their chipsets in the market. Just think, with the reduction in chipset pin-count from moving the DRAM controller off of the system chipset, it might open up the possibility of increasing the number of PCIe lane signals instead, etc. It could be a major bonus on the whole.

 

[Zebo]

Originally posted by: composer

Originally posted by: Green Man
Dothan has 2M L2, so I would be surprised to see it get a huge benefit from dual channel. I don't think there is much benefit to going with the higher latency of DDR2 until DDR2 speeds get up high enough to negate the effect. By then AMD will have DDR2 as well. All of the information I have seen has shown Yonah without iAMD64 as well. I don't see how Intel can plan to move it to the desk top no matter how well it performs in 32 bit when the buzzword is 64 bit. I think A64 will continue to hold an IPC as well as a clockspeed advantage over the Pentium M successor, but I hope intel can find a way to push AMD into at least releasing some faster clock speeds.

OTOH, if you buy a 3000+ now and overclock it to 3800+, you are sure to have a near top of the line processor for the next year and a half +.

Agreed. I'm seeing the 3200 at 10x275 and up, crushing speeds, and LOW temps (39C) at 1.425 vcore vs my XP 2600 at 60C.

My predictions: Intel is behind in the dual core race to some extent, especially with 64 bit chips, however, MSFT delayed 64 bit OS to accomodate INTEL..which is ok, AMD 3200 512 cache chips are clocking way way way up there (.09 winchesters) and hit a wall at 10x280, yet notice that the WALL is not a TEMP wall at all, I belive it's some sort of lock somewhere in the chain, regardless, if a 3200 can do 10x275 and be 40C, and run near FX55 speeds, beating a 2MB P43.8 and be at 40C, expect AMD to release DUAL cores, 1MB cache and 512 cache, ALL FX .09's (if ever, single CPU's) will be 1MB cache as well and the mainstream will be 939 1MB cache FX/64, high end 4500/5000 FX single core 1 MB cache at 3.0 ghz, then the duals at 1 and 2 MB cache.

Where talking some serious power here. AMD has the heat under control, Intel does not, AMD is also ordering .065 machines, so exepct even higher speeds, lower temps.

And guess what? DUAL CORES ARE COMING SOON! What we have now in terms of power will be ecliped almost two fold (not games though) but in rendering, video audio plug-ins, and so on before APRIL.....can't wait, just hope it stays 939 pins, if not, then I hope to see the FX series end up on the .09 process, but what may happen is the FX line goes DUAL core/single core 2 MB cache, and the 939 line gets the 1MB cache, essentially turning all the 939's into FX (1mB cache) and all the FX to 2MB single and dual cores.

I wouldn't worry about AMD, if your holding intel (INTC) though, I may worry a bit.

Everyone but Kristopher knows 2005 is AMD's year. I can't wait to see those benches P4 typically does well on like TMPGEnc get stomped by AMD.. I'll be really surprised if there is a single real-world benchmark intel P4 wins afer dual core/SSE3 AMD64 comes.

 

[VirtualLarry]

Originally posted by: xbdestroya
I think 64-bit functionality will be a much bigger deal 12+ months out from now. The fact is that it can bring some pretty large speed improvements if designed for, and if Windows 64 gains any traction whatsoever, Dothan on the desktop is not going to cut it - especially if games and stuff start coming out 64-bit compiled and optomized.

Exactly. If anything, AMD needs to strongly consider 64-bit enabling even their "value" Sempron line, so that their entire AMD64-derived line is in fact, 64-bit capable. Performance of AMD64 is comparable or better to Intel's fastest Netburst-based chips, and even if Intel 64-bit enables those chips on the desktop, they really can't clock any higher. Pentium-M is Intel's only savior here, and I'm speculating that retrofitting the design to add 64-bit capability would require a pretty major core overhaul, that could take as long as two years to see mass-production. Meanwhile, AMD is busy busting out their newest dual-core 64-bit enabled chips by then, and cleaning-up the high-end of the market.

Originally posted by: xbdestroya
Now, obviously 64-bit is doing many Windows users any good right now, so that makes things a little harder to predict into 2006. I think the Dothans are great, but a great 32-bit processor may not cut it if things really do go 64-bit.

Which, for the desktop, really comes down to one company - Microsoft. You've seen how they've been dragging their feet on XP64 for AMD64.. right now, MS is Intel's strongest ally in the CPU arms race. But if customers increasingly demand it, MS will eventually have to release it, methinks. Currently, any business that wants to run Windows on 64-bit, has to go with an Itanium solution instead of AMD64, currently.

I do see a small niche-market remaining for dual-core Netburst chips, for servers, but with the thermal output and long pipelines, they really aren't all that well suited for either rackmount blade servers, or in terms of transaction IOs/sec, respectively. AMD64/Opterons crush them in nearly every way. But Intel needs that small niche in order to transition the Xeon and the Itanium onto the same basic platform infrastructure. If not... well, IMHO, Itanium will be dead in two years. There's just not a lot of traction for that platform, outside of a few gov't-built supercomputer clusters, which may have been as much of a "buy american" ploy/corporate-welfare handout as anything.

 

[zakee00]

Originally posted by: miketheidiot
I have heard a while back that with k10 (successor to dualcore aka k9) will have some kind of signifigant change to the execution core, which is still almost identical to the oldest k7's.

anyways dualcore PM's in '06 is a long ways away, assuming they actually come out on the desktop by then. between then and now intel has to go through the 65nm change, and who knows how that will go.

yeah, i bet intel will basicly have to re-design their 90nm cores from 130nm, then the 65nm cores from 90nm.

 

[dmcowen674]

Originally posted by: Zebo
Topic Title: Is AMD going to be in trouble 2006?

When have they ever been out of trouble???

Only because of great technology has their bacon been saved and still around.

Their Marketing has never helped them. Should say lack of Marketing.

 

[frootbooter]

Originally posted by: dmcowen674

Originally posted by: Zebo
Topic Title: Is AMD going to be in trouble 2006?

When have they ever been out of trouble???

Only because of great technology has their bacon been saved and still around.

Their Marketing has never helped them. Should say lack of Marketing.

Marketing would hurt them more than help them at this point. They can barely produce enough processors to satisfy the (unmarketed) demand. Now what happens when they advertise? Demand goes up by quite a bit... that'll lead to pretty high prices because they can't make enough processors to satisfy the demand.

 

[Green Man]

Originally posted by: VirtualLarry

A question though - will dual-core Dothan sucessors, have individual duplicated L2 caches, or will both cores feed from one shared L2 cache?

People seem to think that Yonah (dual - core Dothan ) will have 2M of shared L2.

While that's great for performance/watt, it will add latancy as the cache has to arbitrate between the two cores.

 

[VirtualLarry]

That makes sense if the L2 is shared, both from a die-size/cost and power perspective, but also from a mfg-cost perspective, as if one sector of the cache is defective, it can be mapped-out via laser-cuts or whatnot, without having to also disable one of the two cores as well. So they could also sell the chip as dual-core capable, but with 1M of L2, for example. (Possibly as a different or special OEM model.)

 

[VirtualLarry]

Originally posted by: VirtualLarry
That makes sense if the L2 is shared, both from a die-size/cost and power perspective, but also from a mfg-cost perspective, as if one sector of the cache is defective, it can be mapped-out via laser-cuts or whatnot, without having to also disable one of the two cores as well. So they could also sell the chip as dual-core capable, but with 1M of L2, for example. (Possibly as a different or special OEM model.)

Edit: Found an article mentioning that Yonah was indeed going to have dual-core with a shared L2, and when switching between performance and battery modes, the second core would be disabled to save power. That's even more clever.

Edit 2: Sorry, I guess I can't tell the diff. between "quote" and "edit" these days.

 

[boatillo]

--> 2005 is AMD's year = agreed.

--> Dothan looks promising = agreed.

--> AMD dual cores are gonna put the SMACKDOWN on Intel's = agreed.

Shot an email to an engineer friend who works for Intel and he also agreed people are dragging their feet around the office...things aren't looking chipper.

Now where does Cell figure into all of this?

 

[ribbon13]

Socket 939 dual cores won't see the light of day until a good while after the dual core Opterons are out. As for DDR2 on AMD, that's a bad guess. AMD would have to change the architecture of the memory controller for DDR2 which has high latency, and would sooner go for DDR3. Either way it would require a new socket. I don't see that happening.

Cell.... wtf? The computer world is dictated by results, not marketing hype directed towards moronic console gamers that would love to fellate a Sony exec. IBM's PowerPC 970s are selling well in the form of G5's, so at least they have nothing to worry about.

 

[gobucks]

personally, i don't care if AMD waits till Q2 2006 for dual core 939's, cause that's about the time i'll be looking to upgrade my 3000+, anyways. I just hope that the dual cores overclock well, cause if i have to chose between a 2.6GHz 3000+ and a 2.2GHz dual core overclock, i think i'll stick with my 3000+. Then again, the improved memory controller (supports 4 sticks at 1T/DDR400) and SSE3 might just be enough to make up for it.

 

[Amaroque]

Banias, and Dothan were built (from the P6 core) for mobile usage. There was a specific power envelope objective. MHz wasn't the objective of the development team.

It will not scale well in MHz. The future of any Dothan derivative, is with dual (multi) cores. By the time Intel were to get a Yonah on the desktop, it will have already have been too late for Intel to do any damage to AMD. Intel will need an entirely new core at this point.

Hopefully Intel will bring a good CPU design to the table in the coming years. Without competition from Intel, AMD will start slacking (raising CPU prices, infrequent speed bumps).

My 2¢

 

[BEL6772]

So, what about the dual-core EE? That is going to have a Smithfield core with HT turned on. I wonder what benefits we will see from running 4 threads at once ...

 

[VirtualLarry]

Originally posted by: BEL6772
So, what about the dual-core EE? That is going to have a Smithfield core with HT turned on. I wonder what benefits we will see from running 4 threads at once ...

Should be good for "blade" servers aimed at the transaction-processing market. That would likely be Intel's best weapon against Sun's new multi-core/multi-threaded chips aimed at the same market segment. Although I think that Sun probably has a slight leg up on Intel there, both in terms of vertically-integrated OS support and in terms of power-consumption (since this is still a NetBurst chip we are talking about here).