Will Intel's 45nm send AMD to the grave?

[dmens]

If you can't read chinese, the link shows a yonah @ 2.7ghz beating FX-60 stock in superpi 1M and 2M, and some of the synthetics. The commentary on that site is now "pro" than say, the faux gamepc paxville review which was so popular here, LOL.

vidtor: the one pro that is overwhelmingly the most important is number two, cost reduction and supply, and that is the one AMD sorely needs. To answer point 1 on total Z, there is only so much power savings you can squeeze from Z reduction. The uarch will save a lot more, and with merom, intel will be equal or better on perf/watt (imo). Point 3 on extra features, with the exception of cache, the amount of extra logic required for many new features is minimal. 64-bit support is maybe 2% extra. VT is basically nil.

As for the cons, neither point makes a difference. The process switch is an inevitability, the only question is how smoothly it is pulled off. The only concern at this point is if they can match intel's 65nm ramp and how soon they can do it. Because those trends I stated are time critical and intel 45nm production and 32nm dev isn't slowing down.

 

[fatty4ksu]

The end of AMD is here.

Intel has stopped joking around, it over AMD.

 

[TraumaRN]

Originally posted by: fatty4ksu
The end of AMD is here.

Intel has stopped joking around, it over AMD.

The troll has returned...:disgust:

Post some decent facts or dont post at all.

Just a thought I had, what type of volts would we be potentially dealing with in 45nm? 1.1? 1.15v? and are the volt tolerances going to be less as the gate width shrink? Like my Venice stock was about 1.38V and it's been running at 1.55V(yea i know it's high but i got a crappy OCer) silky smooth for nearly 6 months, with a shrink to 45nm or 32nm would the range of max volts be less? Anyone with decent knowledge of uarch that can answerthis?

 

[Viditor]

Originally posted by: dmens
vidtor: the one pro that is overwhelmingly the most important is number two, cost reduction and supply, and that is the one AMD sorely needs. To answer point 1 on total Z, there is only so much power savings you can squeeze from Z reduction. The uarch will save a lot more, and with merom, intel will be equal or better on perf/watt (imo). Point 3 on extra features, with the exception of cache, the amount of extra logic required for many new features is minimal. 64-bit support is maybe 2% extra. VT is basically nil.

As for the cons, neither point makes a difference. The process switch is an inevitability, the only question is how smoothly it is pulled off. The only concern at this point is if they can match intel's 65nm ramp and how soon they can do it. Because those trends I stated are time critical and intel 45nm production and 32nm dev isn't slowing down.

I agree that over the long run point 2 is indeed the most important. But, if we put the future release of Merom aside, it's the release of Cedarmill and Presler that was critical for Intel at this time...and the only advantage they have in PERFORMANCE is a reduction in power usage. Your point on uarch being king for power is well taken and I agree completely...though I am sure that you can't honestly know that NGMA (Merom uarch) will meet or exceed AMD's power/performance because you can't know what AMD will have when it's released.
As to 64bit, I believe the number is ~5%, but Intel will be doubling their cache as well (closer to 50%). It's quite possible that AMD will follow suit...not to mention quad-core.

While I agree that the process change is inevitable, timing on it is VERY important from a business standpoint. The longer you can afford to delay the change, the lower the cost on your previous process. This is critical to a balance sheet!
One other thing to keep in mind...
AMD has 2 new 300mm Fabs (Fab 36 and Chartered) which are shipping for revenue at the end of this quarter. It is imperitive that they properly estimate production to avoid the excess inventory problems that Intel experienced in 2004...
It's quite telling that AMD is planning on releasing their 65nm product at the busiest time of the year, when they will most likely require the greatest amount of inventory.

 

[dmens]

My prediction of merom's perf/power is based on the fact that AMD will continue to use the K8 core in the immediate future, compared to what I know about merom. Given the advantage AMD already holds in the memory interface, I don't think the switch to AM2 will gain too much. Since K8 is only going to ramp clock speed for a while, its performance can be extrapolated. When a new factor is introduced (K10, 65nm, whatever), then everything will change again.

If AMD is going to increase its cache and sell quads, then 65nm is ever more critical. I disagree with the wisdom of a delayed 65nm push once yield/quality criteria have been satisfied. There is absolutely no financial gain from holding onto a process which yields less revenue per area. afaik, the sooner the better.

 

[rgreen83]

I think AMD is in a good position now with 90nm and should just skip 65nm and go straight to 45nm around the same time as intel.

 

[n7]

Originally posted by: rgreen83
I think AMD is in a good position now with 90nm and should just skip 65nm and go straight to 45nm around the same time as intel.

Somehow i think that's a hell of a lot easier said than done...

 

[hans007]

Originally posted by: Viditor

Originally posted by: dmens
Yeah, the timing of the intel release is related to the Q4 earnings. But enough with press releases...

This brings back a point I made in another topic, that process is king. Just like how intel needs to deliver a new uarch, AMD needs to catch up on process. Given their current 65nm gap, I'd have to say the 45nm is just as lagged, if not worse. The issue of high volume manufacturing becomes more critical at smaller features, and that is something IBM/AMD never had a good track record with.

As process size decreases and the trend towards multicore continues, I'd argue that process generations will have even more impact than before based on the number of chips a single wafer can produce. It is amplifying the effect of economies of scale. I am looking forward to see the results of AMD's 65nm ramp and how soon they can pull it off, because it will be a good indicator of future trends, IMO.

Fair enough...just keep in mind that there is a difference between when it's POSSIBLE to release and when it's actually released. Let me break it down into pros and cons:

PROS
There are 3 main advantages to converting to a smaller node...
1. Reducing power requirements by decreasing lengths
2. Increasing overall yield/wafer
3. Adding things (i.e. cache or 64 bit extensions) to existing designs without increasing the footprint

CONS

1. Existing equipment must be discarded creating a Capital Loss
2. All new nodes must be ramped over time, reducing production


1. In the current situation, AMD has no need for "pros" 1 and 3 to remain highly competitive...
2. AMD uses 10% of their lines for R&D which allows them to improve a ramp before they initiate volume production
3. There is a mathematical point where the increased yield of conversion to 65nm outweighs the costs of the "cons". For AMD, this point is calculated by their APM 3.0 software.

I certainly have neither the math nor the data to know where that point is, but judging by the brilliant success of their "late" release of 90nm, I do have a great deal of confidence that AMD does. Also, keep in mind that the longer they delay, the greater their initial yield will be when they DO convert to volume on 65nm...

wrong buddy , you ALWAYS have to decrease your node size.

#3 pro, is adding things. that would be MORE cores. #2 the yield per wafer is too low, for dual cores now for amd. that is why they are having all thse shortages. if the node size were reduced the problem would be fixed since you could make dual cores for a little more than single core cost.

dual cores use more power so then #1 would have to be reduced too (their current dual cores dont use that much power, but the yonah/conroe are much lower)

adding things, they probably will want to eventually go with 1mb cache per chip, plus semprons are costing them too muchto make and are still a large portion of their sales.

with the current 65nm and 90nm mix, a sempron costs them nearly as much die space to make as a pentium D 920.

the main reason they have to move to a new node is intel is moving to a new node. if intel adds a ton of new features decreases heat, etc then their competitive edge will be erased.

that is why not moving to 65nm as soon as possible will always bite them in the ass. the published reports now, say volume production will not start until 2nd half 2006 or first half 2007. That is awful. it will totally screw them because intel is suddenly 1 process generation ahead.


the main thing that got them back in the game after nearly dying for the 2nd time is the total screw up that was the p4. the 90nm transition for intel was so bad, that it allowed amd to catch up and even exceed on process tech for a while. if intel pulls ahead again and executes amd will not die, but they could lose share or not grow at the rate they have been.

i mean no one wants AMD to die. i would have had to pay like $700 for my pentium D 820 and we'd have no alternatives.



also i'd have to say the reason amd has better chipsets is because nvidia is there. nvidias newest chipsets for intel are also great but they cost more and the reasons is that intel makes its own chipsets. there is less incentive fo rthe platform since its hard to get people to not buy an intel chipset unless you supply some niche (like ati has with the radeon express 200). most people and more importantly OEMs are not going to buy nvidia nforce chipsets for anything buy SLI, so that whole mainstream market is gone for the non sli nvidia chipset (which is why it just came out). sure its a bigger pie to get a small chunk of, compared to the amd side, but i think that is why nvidia prioritizes amd since they are top dog there.

nvidia has proven you can outdo intel in chipsets performance wise, but it just isnt THAT big a difference for anyone to really care . a i975x is only barely slower than a nforce intel sli as it is.

 

[Viditor]

Originally posted by: dmens
My prediction of merom's perf/power is based on the fact that AMD will continue to use the K8 core in the immediate future, compared to what I know about merom. Given the advantage AMD already holds in the memory interface, I don't think the switch to AM2 will gain too much. Since K8 is only going to ramp clock speed for a while, its performance can be extrapolated. When a new factor is introduced (K10, 65nm, whatever), then everything will change again.

If AMD is going to increase its cache and sell quads, then 65nm is ever more critical. I disagree with the wisdom of a delayed 65nm push once yield/quality criteria have been satisfied. There is absolutely no financial gain from holding onto a process which yields less revenue per area. afaik, the sooner the better.

Fair enough, mate. My own predictions are that the embedded SiGe process will give the K8 a much better power/performance, but we will have to wait and see.
As to the values of delaying the release of 65nm, we shall have to agree to disagree...
I guess that considering our backgrounds (yours as an engineer who understands business and mine as an investor who understands engineering), it's inevitable that we would disagree on this point...

Good discussion...thanks!

 

[Viditor]

Originally posted by: hans007


wrong buddy , you ALWAYS have to decrease your node size.

I never disputed that...the question is over timing. Reducing the node size doesn't have to be done as soon as is humanly possible, there are many other considerations. Look at what a disaster it was for IBM and even Intel at 90nm...do you remember when Dell was forced to delay their Prescott rollout and substitute Northwoods due to lack of inventory from Intel? By delaying for 2 turns (6 months), AMD had NONE of these problems and ramped at an almost fully mature yield.

#3 pro, is adding things. that would be MORE cores. #2 the yield per wafer is too low, for dual cores now for amd. that is why they are having all thse shortages. if the node size were reduced the problem would be fixed since you could make dual cores for a little more than single core cost.

AMD has already solved their supply shortages by bringing 2 new 300mm Fabs on-line at 90nm. They will have more than double their current capacity by the end of this quarter.

dual cores use more power so then #1 would have to be reduced too (their current dual cores dont use that much power, but the yonah/conroe are much lower)

adding things, they probably will want to eventually go with 1mb cache per chip, plus semprons are costing them too muchto make and are still a large portion of their sales.

with the current 65nm and 90nm mix, a sempron costs them nearly as much die space to make as a pentium D 920.

the main reason they have to move to a new node is intel is moving to a new node. if intel adds a ton of new features decreases heat, etc then their competitive edge will be erased.

that is why not moving to 65nm as soon as possible will always bite them in the ass. the published reports now, say volume production will not start until 2nd half 2006 or first half 2007. That is awful. it will totally screw them because intel is suddenly 1 process generation ahead.

You've touched on a few points here, allow me to respond...
1. AMD is reducing heat and increasing power performance by changing to a new embedded SiGe strain process.
2. Presler has a die size of 162 mm2, Sempron's is 84 mm2...so it's actually the other way around. The 940 is about twice as much to make. It's actually quite a bit more as yields on a 920@65nm are certainly much lower than a Sempron@90nm (on a percentage basis)...but of course the Sempron uses SOI (not quite the equivalent of the yield differential, but it's something).
3. AMD has already said they are planning on going to 65nm in H2 06, with full conversion planned to be by mid 07. This is perfectly timed for any additional cache required to compete with Conroe/Merom.

the main thing that got them back in the game after nearly dying for the 2nd time is the total screw up that was the p4. the 90nm transition for intel was so bad, that it allowed amd to catch up and even exceed on process tech for a while. if intel pulls ahead again and executes amd will not die, but they could lose share or not grow at the rate they have been.

i mean no one wants AMD to die. i would have had to pay like $700 for my pentium D 820 and we'd have no alternatives.



also i'd have to say the reason amd has better chipsets is because nvidia is there. nvidias newest chipsets for intel are also great but they cost more and the reasons is that intel makes its own chipsets. there is less incentive fo rthe platform since its hard to get people to not buy an intel chipset unless you supply some niche (like ati has with the radeon express 200). most people and more importantly OEMs are not going to buy nvidia nforce chipsets for anything buy SLI, so that whole mainstream market is gone for the non sli nvidia chipset (which is why it just came out). sure its a bigger pie to get a small chunk of, compared to the amd side, but i think that is why nvidia prioritizes amd since they are top dog there.

nvidia has proven you can outdo intel in chipsets performance wise, but it just isnt THAT big a difference for anyone to really care . a i975x is only barely slower than a nforce intel sli as it is.

I think you are forgetting that AMD makes a large number of chipsets themselves. Most Opteron servers use AMD chipsets...

 

[dexvx]

Originally posted by: Dadofamunky

Originally posted by: dmens
Didn't an overclocked yonah do well against FX-60? yonah @ 2.7ghz comes close on power consumption to a stock FX-60.

http://www.oc.com.tw/article/0601/readocarticle.asp?id=4895

Yeah, that's a great link, doesn't display in Explorer without crashing and probably installs a malware program on your PC. Sorry, I'll wait until the pros wade in with their assessments.

Works fine in my IE6.

And obviously, any non-English site is automatically amateur because anyone without a first-hand comprehension of English can't make a hardware review site.

 

[Maximilian]

Originally posted by: themusgrat
We AMD fanboys will be around until Intel does somehing good. Duh. But they have made a good start with their newest P4s, the ones that OC really well.

EDT: And when AMD's CEO was asked the same question as the OP, he said that people assume that AMD will never make a better product than the current ones. Intel has a way to go until hey are ready for mass production.

If he had said anything other than that, shre prices would have sunk. He has to put a smile on things. His company could be falling apart, civil war w/e and he would still smile and make out everything is ok.

Theres the reason why only he knows how much "resources" they have for 45nm compared to what they need.

 

[Hard Ball]

Originally posted by: JS80

Originally posted by: 13Gigatons
It seems that AMD has only gotten stronger in the last 2 years where it was predicted they would leave the CPU market by several analyst that would like to remain nameless. Intel for it's part has stumbled and run into walls that it did not predict like netburst not reaching 5ghz or even 4ghz for that matter. Intel had to scrambled and get the Pentium M out and Core Solo and Core Duo after just to keep up. Desktop will have the same approach as well, it will interesting to see how Intel markets the newer chips though now that netburst is being phased out.

I guess what I'm asking is since Intel has woken up and decided to put some money into trying to beat AMD, if AMD has the resources to catch up. Sort of like Cold War where Russia didn't have enough money to keep up with US and lost.

Wait, JS80,

I think I may know who you are; by your writing style and your claims;

Aren't you the poster from notebookforums named Brett VanKirk that claimed to be from IDF (for a time, and later recanted the story); and tried to trash AMD at every turn??

 

[pm]

Originally posted by: Soviet
If he had said anything other than that, shre prices would have sunk. He has to put a smile on things. His company could be falling apart, civil war w/e and he would still smile and make out everything is ok.

Theres the reason why only he knows how much "resources" they have for 45nm compared to what they need.

You are saying that since he is CEO he has to deliberately mislead his stockholders for the purpose of making sure that stock prices stay high? Are we talking Enron here or AMD? What you are describing is illegal on multiple levels. If the CEO makes a statement about something, you can assume that it is being "spun", but I doubt that he would "still smile and make out everything is ok" if his company was falling apart... I don't think any CEO would be willing to risk massive fines and jail time nowadays when the papers are full of other CEO's who are sitting in court.


As far as the original question, I believe that 45nm is going to a challenge for the entire industry. Having seen the 45nm process information at Intel, I think Intel is well poised for it, and will likely be the first in the industry to move the new process technology. But I think the assumption that the rest of the industry won't be able to follow is highly unlikely. Intel will blaze the trail and the industry will be close behind. 45nm is not a "new paradigm", it's merely an evolutionary challenge.

 

[keldog7]

QUOTE:
Will Intel's 45nm send AMD to the grave?
_________________________________________
That depends on whether or not their new chips SUCK as bad as their recent ones.

Other than that, the only thing really holding AMD back right now, is their inability to scale up production. With even their latest fabs soon to be operating at capacity, AMD won't be able to increase market share, because they won't be able to make them fast enough. Also, depending on how they handle their inventory, there may be a bit of a PR fiasco.

 

[Fox5]

Originally posted by: DeathBUA

Originally posted by: fatty4ksu
The end of AMD is here.

Intel has stopped joking around, it over AMD.

The troll has returned...:disgust:

Post some decent facts or dont post at all.

Just a thought I had, what type of volts would we be potentially dealing with in 45nm? 1.1? 1.15v? and are the volt tolerances going to be less as the gate width shrink? Like my Venice stock was about 1.38V and it's been running at 1.55V(yea i know it's high but i got a crappy OCer) silky smooth for nearly 6 months, with a shrink to 45nm or 32nm would the range of max volts be less? Anyone with decent knowledge of uarch that can answerthis?

Yes, the voltage tolerances will be lower.
At 45nm or 32nm, I'd expect we'll be below 1V at that point.

 

[RichUK]

I don?t see why people think that a manufacturing process is going to send any company to the grave . Its not as if that is a sales point, it?s a necessity for a company to evolve within the processor industry.

It?s a business! Companies like Intel are not out to provide a somewhat service, but to make a hell of a lot of money. Its all to do with how well a product sells and how its sold (not to mention the politics and successful marketing behind a given product). Lol I haven?t even seen an advert in which AMD was trying to advertise any of their products, ever. Have you? (and this is just consumer level not Corporate/OEM level)

I would have thought the Pentium 4 (or netburst) against A64 'performance/sales' proved this point, and the point being Intel sold more ?Netburst? processors even though at the time it didn?t offer better performance against its leading competitor 'AMD'. Now ask the question why.

If you look at the marketing potential netburst could have had, and if Intel had ever reached their target of 10Ghz (as indicated in several roadmaps), you can see the potential earning they could have had, going by more Ghz = More money. Basically periodically increasing frequency and possible feature set additions, to milk money over a given time period, until netburst cant be improved no longer. Even though the obvious evolution path is and has always been parallelism, aka dual, quad, multicore processors, also for the uninformed end-user, they would be under the impression they are purchasing a superior product (Due to the marketing hype etc).

Did you know that half the money spent on a box office film is spent on marketing. Point being if no one knows about your product and its competitiveness against the opposition, no one is going to buy your product or know it exists to some point (last point is aimed at a consumer level end-user).

 

[Zebo]

Originally posted by: hans007

Originally posted by: FoBoT
not without real dual core cpu's

the conroe is a true dual core with a real shared L2 cache that should be far more efficient than the x2.

i even bought intel stock because of this.


amd is pretty behind intel in process technology. their 65nm chips will not even be available until the very end of 2006 or in some leaked reports the 1st half of 2007.

intels have been out since the end of 2005 and by the time amd gets 65nm out intel will be doing possibly quad core on 45nm using dual buses for 4p servers.


i suppose amd will always have its niche. but with intel owning a cost advantage they already own the

"i just need a computer, ANY cpu" market since they are cheaper to make , and they will soon retake the mainstream desktop if conroe is as good as speculated (not to mention probably 1/3 less cost to build than an x2 and lower heat dissipation to boot).

anyways, thats just my opinion. ibm and amd better get 65nm up and running well, because the xbox360, ps3, revolution and amd's dual core chips all depend on it.

i know most of the anandtech community is pretty high on AMD. but their cpus are not even that cheap anymore , and they are just riding a huge wave of euphoria right now.

i mean i have owned basically every amd chip ever and have held their stock in the past etc. but for the coming year, it looks like intel is really getting its act together compared to years past.

I agree 100%. Instead of AMD paying out fat bonuses to thier execs like some kid who got $50 for Christmas and it's burning a hole in his pocket AMD should have spent those profitable quaters on processes technology. Only when you get billions a quarter in profit like intel has can you make everyone rich that works for you. AMD will pay for thier spend thrift come late this year early next and definity when 45nm rolls from intel. Those chips cost 1/4 to make than 90nm. $$profit$$ Add in the fact intels newer chips will be fast and cool creating more demand I see AMD selling $40 chips again like t-bred days in order to survive.

 

[Zebo]

Originally posted by: keysplayr2003
Boy OP, you sure chose the "right" forum to ask this question. This is the AMD fanboy capitol of the internet!! . I like AMD, but people can get pretty crazy about it around these parts.

Check around - every enthusiast forum is... Techreport - OC forums - xtremesystems etc. Anyone that has a clue uses AMD ATM. It's not even close. Most forums divide the AMD and intel catagories and you can see immediatly how many are viewing each and thus popularity. EX:
http://www.xtremesystems.org/forums/forumdisplay.php?f=61

Notice 10x the number viewing AMD as intel as I type this.

I've asked in FI to divide both AMD/Intel and nV/ATI to no avail - which i think would save a lot of bickering.

 

[imported_michaelpatrick33]

Originally posted by: fatty4ksu
The end of AMD is here.

Intel has stopped joking around, it over AMD.

:laugh: :roll:

This coming from someone who stated his PIV 1.8 Willamette PIV encoded faster than his 3000+ AMD64. I love it. You are either a very clever satirist or ... dun de dun de trolling for Intelia.

Edit: So Intel has been "joking around" with corporate and consumers these last couple of years with their Netburst 31 pipeline fiascos (or whatever actual pipeline count the Prescott's have)? Are you kidding me? When Intel releases their next gen chips at the end of the year and AMD responds we shall see but until then AMD still rules the roost.

 

[scottish144]

If AMD dies, than Intel will be split up on monopoly charges. Who else would be left without AMD? Via? Transmetta? Lol

 

[AnandThenMan]

Originally posted by: ZeboAMD will pay for thier spend thrift come late this year early next and definity when 45nm rolls from intel. Those chips cost 1/4 to make than 90nm. $$profit$$ Add in the fact intels newer chips will be fast and cool creating more demand I see AMD selling $40 chips again like t-bred days in order to survive.

That's one of the most stupid things I have ever read in this forum.

 

[dev0lution]

I don't see any focused strategy from Intel at the moment except to dump product for cheaper to retain marketshare. You don't have to rush to 45nm if your architecture is efficient at 90 or 65nm except for the cost savings. I think they'll get better with the chips launching in the latter half of the year, but I don't think it'll do AMD in.

Besides, the fact that joe consumer may even consider AMD now is a big blow to Intel. 2 years ago your average PC buyer outside of these boards wouldn't have known that an AMD processor could run windows and all their familiar programs just like "Intel Inside" could. Now at least some do...

 

[Diasper]

You do realise cost is not just about what nanometer process level but what yields of maturity you have.

I remember reading an article that AMD always have very high yeilds - and that they always aim to launch into a new process with mature yields.

Intel I suspect will just launch into a new process flinging as much $ at it as possible - of course that doesn't necessarily mean it's cost-efficieient or mean that you are producing low cost chips if yields aren't as great. Perhaps taking a more cynical attitude moving to a lower process and trumpetting it is at least good for your stock price and all its manipulations.

 

[dexvx]

Half the posts in this are fanboism fiction and the sad thing is they don't even realize it.

Originally posted by: Diasper
You do realise cost is not just about what nanometer process level but what yields of maturity you have.

I remember reading an article that AMD always have very high yeilds - and that they always aim to launch into a new process with mature yields.

Intel I suspect will just launch into a new process flinging as much $ at it as possible - of course that doesn't necessarily mean it's cost-efficieient or mean that you are producing low cost chips if yields aren't as great. Perhaps taking a more cynical attitude moving to a lower process and trumpetting it is at least good for your stock price and all its manipulations.

Such as this statement. No company will put an immature process into mass production. No company.

Some people just cannot accept the fact that Intel has better fab technology.