Barcelona Availability Issues?

[TechnologyCoordinator]

http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2+50001028+40000343+1050930800&name=Barcelona

As of this posting NewEgg doesn't even carry the 2.0ghz model, and the only model that is in stock is the 1.7ghz edition. (1.8 and 1.9 are out of stock)

Anyone else seeing availability issues with Barcelona, especially at the higher frequencies?

 

[boduke]

A friend of mine is a VP of sales at a Tier 2 server manuf and they've had to cancel or offer 1.9ghz parts to people that have ordered 2ghz servers...the response he received from AMD is "sometime in november"...

No yield issues DELETED...everything thing they said at the earnings call was smoke to soothe weary investors....

 

[yomamafor1]

Well, as TC pointed out earlier, Dirk said their "defect density" is similar to their older 65nm models, then he proceeded on making the situation looks like the "yield" on Barcelona is similar to Brisbane.

If you plug the same defect density in for Brisbane, you get a yield of 60+%, which is realistic. As a result, I suspect that AMD can only yield Barcelona at 40% max. I also suspect that an increase of 6% in operational spending might also be the result of low yield.

 

[Kob]

There is no question about this - up to now AMD barely shipped out parts.
The big OEMs do not give you an option to select Barcelona in their on-line configurators, and SUN said that they would have Barcelona-based servers available only in December.
Newegg has had some parts for a few days, and then "out of stock" for a couple of weeks.
I also checked on some stores in Akihabara in Japan - and these guys always show first the latest and the greatest - no Barcelonas in stock up to now.

Dirk M. said in the conference call that they shipped "tens of thousands" units in the 3rd Q - namely in September.
Now APPRO, a 2nd or 3rd tier server OEM, is building a monster machine for the 3 US national labs with 48 thousands Barcelonas in it, for delivery in Q1 2008. They got the CPUs. I don't know if they got them all, but if yes, they sucked clean all of AMD production.

As I said in another thread here, Intel said that doing a native quad on 65nm is impractical. Intel knows production, and I trust them on that one. Dirk said that they need to "tune the design to the process". My take is that AMD is stuttering/stalling until they get a 45nm line running, and until then we are in for a rough ride that does not take off.

For a recent channel view on the situation, see
http://www.crn.com/hardware/202404792;jsessionid=QF1WQPJFA2Q50QSNDLPCKHSCJUNN2JVN

Noteworthy quote from the article: "When a company like Avnet, which has been [AMD's] premier distributor, says they're getting no answers, that's signs of a problem."

 

[dragonsprayer]

that's a pricing issue - they are try to create a market for higher end by cutting the price below a q6600

so they can sell a few higher end units for profit

its going to be hard for amd to profit at $200 quad cores this soon out of the box -

secondly - it may be the amd can not make the faster chips due engineering problems with making a "true quad core". if you think about a 3.2ghz limit for amd x2 chips then if you have 4 cores that's a 1.6ghz limit for all 4.

this may sound like simple logic, but amd lacks certain engineering techniques of high hafinium gates etc - that's why Intel pasted 2 chips together the silicon can only handle so much data flow until leakage occurs.

i think amd can make all the 1.7ghz chips they want but over that they may be having issues.

 

[dragonsprayer]

i will add - its like a bucket and water - with 4 cores you double the size of the bucket but the water still must flow though a single outlet.

while intel is critized for its not true core joined at the bus - the bus is like adding more holes to the bucket.

i suspect amd's problem is the interface of the 4 cores or - amd needs to stack slow small chips per ibm techniques join them 3 dimensionally as intel does with fsb or stacking them sideways

 

[itotallybelieveyou]

People would actully buy those???? Unless you are pure AMD and wants a quad core?

 

[TechnologyCoordinator]

Bandwidth really isn't an issue for either company.

There's probably a high demand for the Barcy chips, but I have a feeling there is virtually a non-existent supply.

 

[itotallybelieveyou]

Are they coming out with higher clocked models?

 

[TechnologyCoordinator]

Are they coming out with higher clocked models?

Yes; but I don't think they're coming soon, but others say there are coming in November.

 

[dragonsprayer]

if there was non existant supply they would not have $200 chips on newegg

 

[cbxbiker61]

As I said in another thread here, Intel said that doing a native quad on 65nm is impractical. Intel knows production, and I trust them on that one.

Just like Intel knew RAMBUS made sense back in the day? Just like long execution pipes coupled with extreme clock frequencies ala P4 made sense?

Hmmm, I don't know that I'd trust Intel's take on their competitors production methods/capabilities. Any company who isn't delivering a particular type of product will minimize it. Of course when they do have it, all of a sudden it becomes a must-have. Who delivered 64-bit x86 first? Who delivered native dual core x86 first? (hint, both answers starts with an A)

Personally I figure by the time I actually need a quad (or tri) core the bugs will have been worked out by those guys building super-computers. I'm in no hurry.

 

[TechnologyCoordinator]

if there was non existant supply they would not have $200 chips on newegg

Let's get this straight:

Newegg only has 1.7 ghz chips.

Newegg is out of stock of the 1.8 ghz with an ETA of 10/26/07

Newegg is out of stock of the 1.9 ghz with NO eta

Newegg doesn't even have the 2.0 ghz part listed


AMD used the 2.0 ghz part in all of its benchmarks and it's not even available in ANY quantities. The big vendors don't have systems out and they won't even take orders.

This smells like a paper launch.

Sometime next year we'll start getting all the defectorons they're making right now.


AMD needs to fix the supply issue before launching Phenom. Don't kid yourself, this isn't an excessive demand issue. Demand is probably healthy for their new products, but the issue here is a trickle of LOW END processors into the market; 2.0 ghz isn't shipping and 2.5 ghz is merely the punchline of jokes on online message boards.

I don't say this because I hate AMD, I say this because they are royally screwing themselves and if they go under we're all screwed.

 

[yomamafor1]

Just like Intel knew RAMBUS made sense back in the day? Just like long execution pipes coupled with extreme clock frequencies ala P4 made sense?

Hmmm, I don't know that I'd trust Intel's take on their competitors production methods/capabilities. Any company who isn't delivering a particular type of product will minimize it. Of course when they do have it, all of a sudden it becomes a must-have. Who delivered 64-bit x86 first? Who delivered native dual core x86 first? (hint, both answers starts with an A)

Personally I figure by the time I actually need a quad (or tri) core the bugs will have been worked out by those guys building super-computers. I'm in no hurry.

Actually, it is already proven that native quad approach is not very smart in the first place. AMD insisted on doing native quad, and now they're plagued with low clock speed, horrible yield (40% max), and high cost. AMD's operational expenditure is up 6%, while they are reducing head counts and expenditures. I'm guessing the low yield is starting to show its ugly side. In addition to that, they still do not have the performance lead AMD promised us a few quarters ago.

Yes, AMD invented both x86_64 instruction and "native" dual cores, but those are the past. At the moment, AMD has no performance lead, and bleeding hard.

I understand your sentiment of supporting the underdog, but the hard cold fact is there: AMD's native quad core venture might be a failure.

 

[pausert20]

Yomomma, I believe you last sentence should read: AMD's native quad core venture might be a failure built on 65nm process.

 

[Slobogob]

Yomomma, I believe you last sentence should read: AMD's native quad core venture might be a failure built on 65nm process.

Actually if it really fails on 65nm there won't be any 45nm quads by amd and thus the statement is true.

 

[cbxbiker61]

I'm not sure I would call a current 4x1.7 low-end. Low end quad yes.

If you've really got a job that demands the extra 17% boost a 2.0 would offer vs a 1.7 I feel your hurt. I'm guessing that 17% would cost an extra $200-$300 a chip.

I'm sure a lot of folks are happy to stuff two 2344's in a server board to get 8x1.7 at say $450. Now that's a deal! And it is available now.

BTW, super computer users don't think too hard about swapping chips say 12 months from now, when faster chips are available for less. They can cost-justify their compute time.

I'd expect that the people that are bitchin the most about faster chips not being available are probably not the people that have compute jobs that peg 4 cores 24x7x365.

 

[yomamafor1]

To add to TC's comment, this is from Ars Technica.

http://arstechnica.com/journals/hardware.ars/2007/10/24/amd-barcelona-finally-showing-up-in-channel-but-at-sub-2ghz-clockspeeds

About a week ago we covered AMD's reported plans to launch Phenom at the end of November and noted specific indications that Barcelona parts were still in short supply and unavailable in channel as of Oct 12. Since then, Barcelona availability appears to have increased somewhat, as low-power (HE) 1.7GHz and 1.8GHz parts have appeared in Akihabara at prices of $245 and $300 respectively. The 1.7GHz HE part is also available in the US for $219 at NewEgg, but not the 1.8GHz.

Oops.. doesn't look too good...

Rumor has it that Phenom is scheduled to debut at 2.2 and 2.4GHz, but for all intents and purposes, Phenom is simply a re-badged Barcelona. If AMD can't deliver 2GHz parts in-channel six weeks after launching Barcelona, there doesn't seem to be much reason to expect the company will be able to launch 2.2GHz+ chips in wide availability anytime soon.

Ouch...

 

[yomamafor1]

Yomomma, I believe you last sentence should read: AMD's native quad core venture might be a failure built on 65nm process.

Actually, the statement might hold true by itself. Although AMD's ramping the 45nm process in H108, we probably won't see 45nm CPUs from AMD anytime soon until H109. Aside from that, AMD will likely to use SOI for its 45nm process, and delay HK/MG implementation for 2nd iteration of 45nm. Given that SOI is already giving AMD headaches in increasing Barcelona's clockspeed, the late HK/MG implementation on 45nm will not help with Barcelona's leakage. Bear in mind that 45nm is a lot more leaky than 65nm, and if AMD can't master their 65nm to reduce leakage, going 45nm won't help them at all.

 

[Kob]

Just like Intel knew RAMBUS made sense back in the day? Just like long execution pipes coupled with extreme clock frequencies ala P4 made sense?
...
Who delivered 64-bit x86 first? Who delivered native dual core x86 first? (hint, both answers starts with an A)

Those were design decision, not manufacturing/production, and Intel has had its missteps in design on numerous occasions.
Come to think of it, it was their design center in Israel that saved them - first with the Centrino and then with the Core architecture. The Pentium IV was designed in the US.

But in manufacturing, they know what they are talking about.

 

[boduke]

Bear in mind that 45nm is a lot more leaky than 65nm, and if AMD can't master their 65nm to reduce leakage, going 45nm won't help them at all.

It could, in point of fact, make things worse...but since I'm not buying off on the 45nm ramp BS spewed to soothe investors, I highly doubt we'll see anything on 45nm from AMD in any part of '08, and nothing of serious consequence in 1H09 either...

 

[rodney_ws]

secondly - it may be the amd can not make the faster chips due engineering problems with making a "true quad core". if you think about a 3.2ghz limit for amd x2 chips then if you have 4 cores that's a 1.6ghz limit for all 4.

I agree it appears AMD is having some serious problems... but what you said right there... either I'm reading it it wrong or you're VERY wrong.

 

[TechnologyCoordinator]

secondly - it may be the amd can not make the faster chips due engineering problems with making a "true quad core". if you think about a 3.2ghz limit for amd x2 chips then if you have 4 cores that's a 1.6ghz limit for all 4.

It doesn't work that way.

I'm going to throw out imaginary numbers here:

Let's say any barcelona core has an 80% chance of reaching 2.0ghz

HOWEVER

When you combine four 80% chances you together you get a much lower chance of four random cores reaching 2.0 ghz.


Intel's way is different, because they only have to combine two cores and two chances of reaching a certain level of performance. Intel can then pair high peforming dual cores together to make a high performance quad core. And also pair low performing dual cores to make a lower performing quad core.

AMD is just stuck rolling the dice hoping to get those four good cores. That's why they are doing the tri-core for the cases where they have three good performing cores and one not so good or defective one.

 

[KyleSTL]

It doesn't work that way.

I'm going to throw out imaginary numbers here:

Let's say any barcelona core has an 80% chance of reaching 2.0ghz

HOWEVER

When you combine four 80% chances you together you get a much lower chance of four random cores reaching 2.0 ghz.

0.80^4 = .4096 = 40.96% (chance of 4 random cores working at 2.0Ghz frequency) - oversimplifed stats analysis

or if they're lucky (90%, hypothetically)

0.90^4 = .6561 = 65.61%

Just for funnzies.

 

[homerdog]

How is it that AMD had the first native dual core? I thought the Core Duo was the first native dual core processor with its shared L2 cache. What is the technical reason why the Athlon64 X2 is native. Is it the fact that the two cores share the same memory controller? I know the X2s don't share any cache between cores. People seem to be pretty sure that the X2 is a native dual core, so I must be missing something.