No 2.6GHz Phenom in Dec

[SexyK]

Originally posted by: spittledip

Originally posted by: Viditor

Originally posted by: 21stHermit
According to DigiTimes today, AMD will NOT have a 2.6GHz Phenom for sale in December, 2007.

Here's some of the article:

"Amid stiff barriers from 90nm and 65nm conversion, sources at Taiwan motherboard makers do not expect AMD to roll out the highest frequency 2.6GHz version of its new quad-core Phenom CPU family before 2008.

AMD declined to comment on the speculation, saying that it cannot reveal details about the upcoming Phenom launch.

The sources noted that AMD will only introduce two Phenom CPU at the November 19 launch, the 2.2GHz 9500 and 2.3GHz 9600 models. A 2.4GHz 9700 model is expected to launch during December, but a 2.6GHz one will not be introduced until 2008, they noted.

While expressing satisfaction about the new 65nm desktop CPU lineup, the motherboard makers explained the key reason for AMD failing to meet its roadmap on the launch of a 2.6GHz Phenom is because of barriers relating to conversion to 65nm."

One has to wonder if 65nm is a problem, how AMD expects to have 45nm next summer?

It would appear that IC fab IS Rocket Science!!!

Conversion to 65nm? AMD is now 100% 65nm production (there are no longer any 90nm lines now that Fab 30 is closed for the Fab 38 conversion), so I don't quite understand what they mean...

Haven't all AM2s been 65nm???

Just because they are producing 65nm chips at this point doesn't mean that they are getting their yields they want, or that they are not still experiencing issues with their 65nm process in general. I think that's what the author was insinuating.

 

[daveybrat]

All i hope is that when the Phenom's come out that they are compatible with my motherboard! I'd love to be able to upgrade to a quad-core chip without a new motherboard upgrade.

 

[myocardia]

Originally posted by: spittledip
Haven't all AM2s been 65nm???

No, the majority of AM2 cpu's produced were 90nm. Only with the fairly recent Brisbane cores, did AMD have a 65nm core.

 

[Viditor]

Originally posted by: myocardia

Originally posted by: spittledip
Haven't all AM2s been 65nm???

No, the majority of AM2 cpu's produced were 90nm. Only with the fairly recent Brisbane cores, did AMD have a 65nm core.

To be fair, those Brisbane cores have been out for a year now...but your point is a good one.

 

[Viditor]

Originally posted by: SexyK

Just because they are producing 65nm chips at this point doesn't mean that they are getting their yields they want, or that they are not still experiencing issues with their 65nm process in general. I think that's what the author was insinuating.

Thanks SK...
From what they've told shareholders, and considering they're selling them so cheap, I would say that yields have been excellent. However, going to the higher latency cache has certainly been a serious problem for them...I don't think it has to do with the 65nm process, but the 65nm chip design has certainly been an issue.
My guess is that they've been focusing all of their R&D resources on K10 and the 45nm process rather than reworking the 65nm K8 design. From their POV, this makes sense as the 65nm K8s have very little time left in this world...

Edit: I should also add that this is a good example of how powerful Intel's huge R&D budget really is...they wouldn't have needed to make the sacrifice.

 

[Keysplayr]

Originally posted by: Phynaz

Originally posted by: nachovidal2

Originally posted by: Arkaign
When AMD brought out their high-end 486/586 chips, they outperformed Intel
*dark days where K5 was very weak against Pentium/MMX*
When K6 came out, it was better than PMMX (P2 was too $$$ to really compare against)
When K7 came out, it was better than P3-Katmai
When Thunderbird came out, it was better than P3-Coppermine
When Athlon XP came out, it was better than P4-Willamette
*dark days where P4-Northwood was better than Athlon XP*
When Athlon64 came out, it was better than P4 (ALL)
When X2 came out, it was better than PD
*dark days where C2D/C2Q are far better than AMD (ALL)*

So, historically, AMD has been kicking butt in the performance area, and it's sad to see that status slip so severely. If AMD can at least bring out a product that gives 85% of the performance for less or equal to the same percentage in price, I think they are still in the game.

Intel 386s outperformed AMD 386 (ALL).
Intel 486dx/dx2/dx4 outperformed AMD 486/586 (ALL).
Intel P2 outperformed K6,6-2,6-3 (ALL), even celerons outclassed K6 in games.

AMD has "only" really outperformed Intel from K7 days to Intel Core gen. And never so clearly as Intel is doing now with C2D against X2.

Actually, AMD has never held the performace crown for an entire cpu generation.

Really? First off, it's unclear exactly what you mean and what you classify as a CPU generation. Because the way I see it, A64 had no competition performance wise.

 

[Phynaz]

Isn't A64 a K8 chip?

Then I'd say it has been getting smacked around pretty good the last 16 months.

 

[harpoon84]

Originally posted by: keysplayr2003Really? First off, it's unclear exactly what you mean and what you classify as a CPU generation. Because the way I see it, A64 had no competition performance wise.

K5 -> K6 -> K7 -> K8 -> K10 = CPU generation.
K5 I was too young to remember, not into PCs at that age, but I take it the Pentium was superior?

K6 never had the performance crown AFAIK, the P2 and P3 beat it convincingly.

K7 was faster than the P3 and Willamette P4s but was beaten by Northwood P4s towards the end of its cycle.

K8 layed a smackdown on P4s/PDs but is now getting creamed by Core 2.

K10? Will it even beat Core 2? I doubt it, but some people think it's possible.

 

[Cerb]

Actually, the K6 was pretty good until you got into floating point. Later in life, though, much like we have now, Intel was able to out-manufacture them and make up for the better integer performance. The K6 series was plagued by heat issues and late launches, though.

Also, to be a little more accurate, the K7 was beat once all the P4s were given proper bandwidth and HT support all around (why Intel waited so long w/ only the 3.06 HT, I'll never know). Athlon XPs could beat the As, and match or beat the Bs, but the Cs killed 'em dead, dead, dead.

AMD will have their days in the sun again, but probably not until after their upcoming CPUs are obsolete, or if the IBM-pioneered 45nm stuff pans out well and quickly.

 

[Arkaign]

Originally posted by: Cerb
Actually, the K6 was pretty good until you got into floating point. Later in life, though, much like we have now, Intel was able to out-manufacture them and make up for the better integer performance. The K6 series was plagued by heat issues and late launches, though.

Also, to be a little more accurate, the K7 was beat once all the P4s were given proper bandwidth and HT support all around (why Intel waited so long w/ only the 3.06 HT, I'll never know). Athlon XPs could beat the As, and match or beat the Bs, but the Cs killed 'em dead, dead, dead.

AMD will have their days in the sun again, but probably not until after their upcoming CPUs are obsolete, or if the IBM-pioneered 45nm stuff pans out well and quickly.

Spot on stuff. The K6 was a good bargain until Mendocino Celeron hit the streets, then it was all over. The only advantage to K6 was you could run it on a dirt-cheap mobo, but the only stable Super-7 board I ever used was actually one with a performance-robbing ALi chipset. I tried Epox, Asus, etc VIA Apollo Super-7 boards, but they were horribly unstable for me. My first Celeron was a 300a on an Asus P2B, and that thing ran circles around my K6s even at stock, but at 504mhz, it was smokin Even now it could be used for a basic web / retro gaming box with no issues whatsoever, running XP w/512MB.

 

[Cerb]

I think several chipset makers had driver issues. I never had any trouble (well, the AT mobo design of mine was crap, and Voodoo2 cards should have had some cooling), but my K6-II box ran NT4. Given that several VIA and nVidia based boards I've had trouble with within the last 3-7 years have run solidly as Linux-based servers (the firewall I'm going through to post this was at one point my desktop mobo, and I didn't get rid of it because it was slow), I think a ton of issues we thought were hardware back then were software based. It took way too long for anyone but SiS to get their act together.

Back on topic, I hope AMD prices their new parts 20+% lower than competitive Intel parts. Right now they have a good value PC edge, but with the 1333MHz chips (mainly their introductory pricing and the newer E21x0 CPUs), it takes DVI-based IGP to keep even that low-end value a strong one.

 

[formulav8]

You NEVER saw the TRUE potential of the K7 design until a good chipset came out. Either a nForce or 266a. Paired with either of those chipsets the K7 would walk all over the P3s. Sadly the original K7 chipsets were trash. :thumbsdown:


Jason

 

[spittledip]

Originally posted by: myocardia

Originally posted by: spittledip
Haven't all AM2s been 65nm???

No, the majority of AM2 cpu's produced were 90nm. Only with the fairly recent Brisbane cores, did AMD have a 65nm core.

o yeah duh

 

[spittledip]

Originally posted by: Phynaz
Isn't A64 a K8 chip?

Then I'd say it has been getting smacked around pretty good the last 16 months.

Yeah but 939 platform k8s killed P4s...

 

[daballard]

Originally posted by: Viditor
Thanks SK...
From what they've told shareholders, and considering they're selling them so cheap, I would say that yields have been excellent. However, going to the higher latency cache has certainly been a serious problem for them...I don't think it has to do with the 65nm process, but the 65nm chip design has certainly been an issue.
My guess is that they've been focusing all of their R&D resources on K10 and the 45nm process rather than reworking the 65nm K8 design. From their POV, this makes sense as the 65nm K8s have very little time left in this world...

Edit: I should also add that this is a good example of how powerful Intel's huge R&D budget really is...they wouldn't have needed to make the sacrifice.

I think you're mis correlating the reason AMD is selling their processors so cheap.

 

[Arkaign]

Originally posted by: daballard

Originally posted by: Viditor
Thanks SK...
From what they've told shareholders, and considering they're selling them so cheap, I would say that yields have been excellent. However, going to the higher latency cache has certainly been a serious problem for them...I don't think it has to do with the 65nm process, but the 65nm chip design has certainly been an issue.
My guess is that they've been focusing all of their R&D resources on K10 and the 45nm process rather than reworking the 65nm K8 design. From their POV, this makes sense as the 65nm K8s have very little time left in this world...

Edit: I should also add that this is a good example of how powerful Intel's huge R&D budget really is...they wouldn't have needed to make the sacrifice.

I think you're mis correlating the reason AMD is selling their processors so cheap.

Yep, ASP when you do not control the performance lead is based entirely upon where your products compete at. Judging by the price, it looks like they're going up against the Q6600.

Believe me, if AMD could get away with selling them at a higher price, they would. But they're smart enough to know that nobody wants to buy a $300 AMD chip that gets pwned by a $250 Intel chip.

 

[harpoon84]

Originally posted by: daballard
I think you're mis correlating the reason AMD is selling their processors so cheap.

Well, it couldn't possibly be performance related, therefore yields are excellent.

 

[Viditor]

Originally posted by: Arkaign

Originally posted by: daballard

Originally posted by: Viditor
Thanks SK...
From what they've told shareholders, and considering they're selling them so cheap, I would say that yields have been excellent. However, going to the higher latency cache has certainly been a serious problem for them...I don't think it has to do with the 65nm process, but the 65nm chip design has certainly been an issue.
My guess is that they've been focusing all of their R&D resources on K10 and the 45nm process rather than reworking the 65nm K8 design. From their POV, this makes sense as the 65nm K8s have very little time left in this world...

Edit: I should also add that this is a good example of how powerful Intel's huge R&D budget really is...they wouldn't have needed to make the sacrifice.

I think you're mis correlating the reason AMD is selling their processors so cheap.

Yep, ASP when you do not control the performance lead is based entirely upon where your products compete at. Judging by the price, it looks like they're going up against the Q6600.

Believe me, if AMD could get away with selling them at a higher price, they would. But they're smart enough to know that nobody wants to buy a $300 AMD chip that gets pwned by a $250 Intel chip.

You guys are missing the point...
My point was that Gross Margin is climbing very nicely despite lower ASPs.
The Gross Margin amount is Net Revenue minus COGS (Cost Of Goods Sold). This is the number prior to things like R&D, Admin charges, Amortization, Interest costs, etc...
GM also includes equipment charges, so the buildout at Fab36 and conversion of Fab38 is part of that cost.

BTW...does anybody have ANY indication that yields are anything but excellent? All the numbers I've seen (as well as everything AMD has flat out stated) indicates that Yields are superb...
If anybody has a shred of evidence to the contrary, I would be very interested in seeing it.

 

[21stHermit]

Originally posted by: Viditor
BTW...does anybody have ANY indication that yields are anything but excellent? All the numbers I've seen (as well as everything AMD has flat out stated) indicates that Yields are superb...
If anybody has a shred of evidence to the contrary, I would be very interested in seeing it.

If yields are excellent and AMD's market share is growing, albeit at the low end. Why is AMD losing so much money? Just a simple question, not trying to bait you.

Hermit

 

[Viditor]

Originally posted by: 21stHermit

Originally posted by: Viditor
BTW...does anybody have ANY indication that yields are anything but excellent? All the numbers I've seen (as well as everything AMD has flat out stated) indicates that Yields are superb...
If anybody has a shred of evidence to the contrary, I would be very interested in seeing it.

If yields are excellent and AMD's market share is growing, albeit at the low end. Why is AMD losing so much money? Just a simple question, not trying to bait you.

Hermit

It's an excellent question...the short version is that it's for 2 reasons, ATI and Capital Expenditures for Fabs 36 and 38.

Prior to the ATI acquisition, AMD's "nut" (the amount of revenue required to break even) was ~$1 Billion/quarter...it's now closer to ~$2 Billion/quarter.
It's not all ATI however. Added to that is the huge expense of building out 2 Fabs...Fab 36 and the conversion of Fab 30 to Fab 38. Building a Fab now costs close to $3-4 Billion, and the refit of Fab 30 is close to $2 Billion.
Fab 36 is now paid for (as of Q2), but Fab 38 isn't yet...

Edit:Keep in mind that even after the Fabs are paid for, CapEx continues for things like 45nm conversion. However, this is much less expensive as the equipment they have/are installed is designed to make the conversion without being replaced. IIRC, they will be able to use these lines down to 32nm, though they will then have to do another full conversion.

 

[Phynaz]

Originally posted by: 21stHermit

Originally posted by: Viditor
BTW...does anybody have ANY indication that yields are anything but excellent? All the numbers I've seen (as well as everything AMD has flat out stated) indicates that Yields are superb...
If anybody has a shred of evidence to the contrary, I would be very interested in seeing it.

If yields are excellent and AMD's market share is growing, albeit at the low end. Why is AMD losing so much money? Just a simple question, not trying to bait you.

Hermit

To expand on Viditor's response a little, keep in mind there is little to no money at the low end. It's probable that AMD is selling their low end chips below cost. There has been more than one report published stating that AMD needs an ASP over $100 to break even. Add in that in their last earnings statement where they said they need $2B in revenue to break even, that number makes sense.

As Viditor stated, but in another way, as volume goes up so does margins.

Think of it this way (made up numbers):
Pay 100 employees $1,000
Pay fixed costs (land, taxes, etc) $100,000
Pay amortization and depreciation $100,000
Produce 100,000 chips, bill of materials $100,000.
Sell those chips for $2.00 each = $200,000.
That's a $200,000 loss.

Now sell 200,000 chips and it works out:
Pay $100,000
Fixed costs $100,000
A&D $100,000
BOM $200,000
Revenue $400,000

Well, you have still lost $100,000, but a couple of things happened.

1. You reached break even for cash flow - this is a good thing.
2. Your loss per chip made dropped, therefore your margins increased. This isn't exactly how gross margins - the number to look at - is calculated, but it will do for our purposes (we haven't accounted for inventory for one).

This is the position AMD is in right now. If I remember correctly, they were cash flow positive on operations, but once all their expenses got added in they still were spending more than they made. I don't pay much attention to the writedowns they are taking on ATI, as those aren't cash.

Hope that explanation isn't too obtuse

 

[GFORCE100]

The very fact AMD stated it would sell tri-cores was as strong indication as any that AMD is having yield problems.

Look at the die size of the Phenom, 283mm2 is it? Even Intel would have an above average yield problem on such large dies.

AMD should have played smart and do the 2x2 die approach like Intel and later when they get 45nm rolling, think about making it one die.

Beats me why they pay their tech guys so much money when such decisions seem obvious given the type, performance, and history of AMD.

AMD as a company got over confident in itself and felt the time is right to get cocky, and they failed badly. It's now costing them when if they had gone with the 2x2 die approach, they would be selling Phenom today.

Not only have they got poor yields but also the yields they do have aren't that good if they can't make enough 2.6GHz capable chips. If you need to pump ample voltage through a chip in order to get it running at some frequency then that's your yield right there.

I don't know why it is, is it the people at AMD, the coffee they drink, or their incompetence but AMD has a history for making bad decisions that cost them. AMD had about 3 years or strong sales with the Athlon 64's/X2's and this made them feel like God.

I find it wrong to get paid multi-million wages and not be held responsible for steering the company up the wrong tree. Hector and his friends at AMD get massive bonuses despite AMD losing money, it's just how you NOT do business. 3Dfx also splashed money left right and center and where did they end up? If you take millions, give millions back into the company.

 

[Viditor]

Originally posted by: Phynaz

Originally posted by: 21stHermit

Originally posted by: Viditor
BTW...does anybody have ANY indication that yields are anything but excellent? All the numbers I've seen (as well as everything AMD has flat out stated) indicates that Yields are superb...
If anybody has a shred of evidence to the contrary, I would be very interested in seeing it.

If yields are excellent and AMD's market share is growing, albeit at the low end. Why is AMD losing so much money? Just a simple question, not trying to bait you.

Hermit

To expand on Viditor's response a little, keep in mind there is little to no money at the low end. It's probable that AMD is selling their low end chips below cost.

I can assure you that this is incorrect...if for no other reason than it's illegal (Predatory Pricing).

There has been more than one report published stating that AMD needs an ASP over $100 to break even.

Then you should find those reports and abuse the person who wrote them...it makes no sense at all. ASP (by the way, AMD has never been at $100 ASP that I know of) is a function of revenue. In other words, Revenue = ASP x Unit Sales. If you increase either function (ASP or unit sales) you increase revenue...however, to break even you must also maintain a good GM (in other words, you must keep COGS down).

Add in that in their last earnings statement where they said they need $2B in revenue to break even, that number makes sense.

As Viditor stated, but in another way, as volume goes up so does margins.

Sort of...if volume goes up, ASP stays static, and COGS go up at the same rate, then margins remain static.

In other words, %GM = (Revenue - COGS) / Revenue

So, in the case of AMD's Q3, the numbers were:

($1.632 B - $.963 B) / $1.632 B = 41%

Their Gross Profit was $669 Million...

Then comes the bad news...Expenses:

R&D = $467 M
Marketing and Administrative = $352 Million
Integration with ATI (things like severance pay) = $76 Million
Net Interest expense = $76 Million
Loss from sale of Spansion shares = $57 Million

There are a few other minor bits and pieces, but the Net Loss ended up as $396 Million.

Now, a few of those expenses are once-off (loss from Spansion and ATI Integration) as Phynaz explained...
But, as you can see from the numbers, the key is the Gross Margin.
This boils down to the components of GM:
ASP - how much they get for each part on average
COGS - How much it costs to produce each part (including the manufacturing equipment and raw materials)
Revenue - the number of parts they sell times the ASP

Now for the good news...

Much of the cost for manufacturing is decreasing (as illustrated by Q3's higher GM), and Q4 is generally a higher revenue quarter in terms of unit sales.
The decrease in COGS is from
1. All chips are now 65nm so they will yield more chips/wafer
2. much of the CapEx has already been paid for or will be in the next 2 quarters.

[/quote]

 

[Viditor]

Originally posted by: GFORCE100
The very fact AMD stated it would sell tri-cores was as strong indication as any that AMD is having yield problems.

Not at all! Both Intel and AMD have been following this practise at the beginning of every ramp for the last 7 years...the only reason it's a tri-core is because these are also the first native quad-core chips ever. AMD has already stated that Tri-Cores will have a very short lifespan for sales.
Think of it this way...let's say that they begin their ramp at a very high yield (say 70%).
That's ~217 candidate dice/wafer, so ~152 good dice. If they can salvage even 40% of the defective dice, that's 26 Tri-cores/wafer, or 780,000 Tri-cores/month...and that's on an amazingly high yield! The lower the yield of course, the more Tri-core candidates they have.

Look at the die size of the Phenom, 283mm2 is it? Even Intel would have an above average yield problem on such large dies.

AMD should have played smart and do the 2x2 die approach like Intel and later when they get 45nm rolling, think about making it one die.

Beats me why they pay their tech guys so much money when such decisions seem obvious given the type, performance, and history of AMD.

AMD has already stated that they couldn't do the MCM chips because they couldn't afford the R&D to design them for such a short lifespan. Since they are to be replaced so quickly by the more efficient native quad cores, AMD ate the lost sales for the last year.

AMD as a company got over confident in itself and felt the time is right to get cocky, and they failed badly. It's now costing them when if they had gone with the 2x2 die approach, they would be selling Phenom today.

HUH? Considering that Phenom X2s are coming out AFTER the X4s, I don't follow your logic.

Not only have they got poor yields but also the yields they do have aren't that good if they can't make enough 2.6GHz capable chips. If you need to pump ample voltage through a chip in order to get it running at some frequency then that's your yield right there.

Ummm...your mixing manufacturing yields up with chip design and bin splits now.

 

[Phynaz]

Selling something below cost is NOT illegal. Goes on all the time.

Pick up the local Fry's add, see those $5 DVD's? They are below cost.

There was a time in the early 90's when Ford was selling Escorts for $3,000 less than it cost to make them. They needed the volume to make the CAFE mandates.

And of course we know that Xbox's and Playstations are sold below cost.

You don't seriously think AMD is making a profit on a $50 cpu? The most basic math shows that it's impossible. Really, my 12 year old could figure it out.

Sort of...if volume goes up, ASP stays static, and COGS go up at the same rate, then margins remain static.

Ummm....NO.

I would even provide you the quote from the AMD conference call stating that volume increases margins, but it's rather tiring having to constantly quote them for you. You can look it up yourself.

Just quit doing the disservice to other people of providing them wrong information.

BTW, I'm still waiting for that link about Henri Richard. You know, when you said he had attempted to leave AMD a year before he did leave? Need a link to your post?