sarmad

i think you are probably right. further aggravating the problem is that without hi-k/mg the 45nm process will be leaky and incapable of scaling in frequency.

so basically the *best* it can do is reduce the die size while doing nothing to improve performance. and it *is* possible to do worse...

gb

i'm highly skeptical of any useful amd 45nm output in 2008. i was merely repeating amd's words.

personally i think they'll go through some "restructuring" before they have a useful 45nm output.

gb

actually at one point in today's call they mentioned that they had 45nm cpus in fab "as we speak".

it's going to be interesting...

gb

i lost count of how many analysts started their questions with complements on amd's quarterly performance. don't these guys have any shame?

q4 should be interesting if/when intel qualifies 45nm and puts these into the saleable mix while running 65nm full bore.

frankly i'm a bit disappointed that intel didn't appear to keep amd's units or asps flatter. apparently they grew both.

gb

Intel/STM Numonyx gets HSR approval.

http://biz.yahoo.com/ap/071018/intel_stmicroelectronics.html?.v=1

gb

no mas

would it be a better board if no one responded to any of its posts? they're generally so outlandish that they don't seem worth the effort to rebut. furthermore i don't think any new visitor would be confused.

how about just pretending its posts don't exist?

just a thought

gb

dually core based celerons in q108

http://www.digitimes.com/mobos/a20071017PD218.html

that's really piling it on!!

gb

gbgc?

gb

ibm micro?

chipguy i think you posted the 15% decline in ibm micro sales. i would have thought q3 would be a good one for them as they would be shipping volumes of cell processor to sony for holiday console deliveries. am i off base or has wii cleaned their clocks?

also any news about the board's decision on the fate of ibm micro?

relevance of course is process development for amd...

gb

re: 1.7g

maybe that's where the part has the most yield. there might even be yield slower than that but the price to move them may not be worth it.

gb

take a look on www.analog.com at their captouch chips. lots of design info on how to design a variety of touchsensor interfaces including one similar to the apple "wheel."

of course no mention of legal difficulties that might present.

analog makes nice stuff! crazy dds, etc.

gb

just needs to be wrapped in 15 layers of pseudo-philosophical rambling and other obfuscation.

and accompanied by a fine brew...

gb

i'd put my money on bryant any day rather than on your mumbo jumbo.

he's a straight shooter. you just don't like what he has to say.

gb

are you referring to the 200 in ireland?

that's hardly meaningful given a 90K headcount worldwide.

gb

sounds like arrow bought the inventory from intel. i wonder what the one time dollar amount is?

gb

btw, i'd expect this sort of "scalability" of raytracing to be demo'd no later than fall '08 at idf.

gb

two interesting articles both readable:

http://www.pcper.com/article.php?aid=455

http://blogs.intel.com/research/2007/10/real_time_raytracing_the_end_o.html

an interesting point these articles didn't make is that while raytracing can be done on x86 in the main cpu(s), it can also be straightforwardly ported to larrabee which is also x86+.
i fully expect intel to produce libraries that make it easy to do so.

larrabee plugged into a nahalem (or later) based system through quickpath will redefine workstation and highend pc graphics but will also allow this stuff to scalably move down to mainstream pcs due to the portability of x86.

this is stuff that nvidia can't do and amd/ati probably can't afford with their allocation of scarce resources.

this will significanty devalue of nvidia and ati discrete graphics silicon in the mainstream and above for all but legacy software. with the lower ends of the spectrum being serviced by integrated raster graphics either in chipsets or on the cpu that puts a complete pincer move on the discrete encumbents.

as multicore cpus continue to move into lower price bands more and more of the "graphics" will be performed by the cpu and even the "integrated graphics" raster stuff will get squeezed more and more until it eventually disappears.

we'll be back to where we started with first pcs in the early '80s: the cpu does all the drawing. easy to see how those 32nm transistors will be spent.

gb

ps. discovered thru inq
http://www.theinquirer.net/gb/inquirer/news/2007/10/11/intel-man-touts-real-ray

ruiz probably won't get "roasted" until a few more 0.5B quarterly losses. the q3cc will likely be completely free of actual product roadmap questions since all the analysts care about is the bottom line rather than how it occurs.

unfortunately without competitive products the bottom line will continue to suffer greatly. and with every delay intel raises the definition of "competitive."

somehow that eludes the grasp of the analysts. they appear to think that magic will occur. i think the expectation of magic will vanish by the q108 cc.

i can hear that call now "just wait until our whizzo 45nm process ramps..."

gb

knowing what i know about pso i think amd will remain under pressure as long as they exist...

gb

actually i don't think he's even arguing. he's just stirring to pot to watch the reaction. it's his form of amusement.

mas and others like him make me want a paid subs just for the ignore feature.

otoh, when i get tired enuf i just go away for a while.

cheaper!

gb

it was mentioned several times in the q1 and q2 06 conference calls by bryant.

gb

Sarmad

I think there's an entry on the balance sheet for authorized distribution inventory. I don't recall what the precise name is.

I think Intel has done a pretty good job over the years managing down the grey market which would be the majority of the rest of the distribution inventory. The historical primary source of this has been OEMs overordering then dumping it into the secondary market, sometimes for a profit, often not. By working with the Tier1 to precisely and timely deliver their needs the Tier1 have mostly stopped this practice.

The Tier2 can be more unpredictable but there is less likelihood of Tier2 overordering due to flattened pricing for volume and frequent price moves making it more risky to do so. Also with plenty of supply there is no need to double order from authorized distribution.

Keep in mind that the primary source of the inventory rise in 06 was the desire to have Netburst in reserve in case C2D didn't qualify. That would have been much worse than an inventory bubble. Intel has had that problem before (a decade ago) and they have sworn never to have that jeapordy again.

Of course the analysts never understand that. They're a pretty dense lot.

I don't think we'll know until the Q3 call in a couple of weeks.

gb

Sarmad

First I have to be clear that I'm no longer an Intel employee, having retired in 2000. I've also moved away from where I was working for them so I've pretty much lost contact with my Intel sources.

I don't know any more than what I heard at the various quarterly calls. I don't recall hearing a "poof." I think the combination of strong branding and price moves caused it to move all around the world. Of course that upsets all those who claim Netburst is/was of less value than worm sweat residue.

Your guess of current production vs inventory is probably a good one.

As far as distribution meaning "authorized distributors" Intel has an excellent understanding. As far as a more general definition of "distribution" meaning anywhere other than Intel and AD warehouses, Intel has a pretty good understanding of Tier1 inventory levels based on exec conversations. For Tier2 and below as well as grey market it has always been very hazy.

gb

it's actually quite possible that intel ships in one month more memory controllers than amd has ever shipped in its pathetic life.

to characterize imc as difficult for intel is just amd projecting their own inadequacy.

gb

I find it interesting that the SI thread has degenerated into "Hector mode", i.e. bashing Intel rather than reveling in the latest AMD product or roadmap.

Perhaps they're not too keen on AMD's prospects at this point?

gb

just shows how little the some of the semi analysts understand the technology.

what's amazing is that people actually invest based on their advice.

gb

wait until they hear the other shoe drop:

no useful 45nm process within the "solvency window"...

gb

i think amd management has just realized that the bloom has come off the rose.

hector devotes an entire bizweek interview to whining rather than products. espp canned, forced vacations rumored, bonuses likely canned.

penryn sampling everywhere, nahalem showing 16 threads, g0 stepping kicking butt, 32nm sram wafers visible, silverthorne running.

quite a contrast...

gb

i saw in one of the presentations (dunno which) a brief mention of some of the features. the one that caught my eye was the link training. it was probably in one of the keynotes since i think gelsinger was complementing the team on how quickly they got a system up and running with it.

nothing in detail however.

gb

you don't need their tools to use the libs. they're a separate package. of course i'm sure the tools would be very useful.

renee james presentation has a ringing endorsement for the threading and profiling tools from a game developer who used them for their fps.

gb

interesting that they called intel vague on "sse5". gelsinger was anything *but* vague in saying "no."

gb

mmoy:

on the renee james keynote she mentioned that intel has open sourced their threading libs. don't know if you've seen that.

gb

Knut is a techy not a product manager and wasn't making a product announcement except to say that Intel would be offering both client and server optimized SSDs early next year. Gelsinger mentioned it in his presentation as well.

If you're interested at all watch the presentation. It's really well done. Good material, well presented.

I haven't seen any foils available yet although MASS002 presenters might eventually post theirs.

Knut's presentation webcast is at the bottom of this page:
http://www.intel.com/pressroom/kits/events/idffall_2007/webcasts.htm

gb

Intel also had a tech presentation, I don't think it was a keynote, by Knut Grimsrud. I've seen him present internally and externally several times; he knows his stuff and is a very straight shooter. His presentation is techy but very understandable. Worth watching. It's at the bottom of the IDF Keynotes page.

My take away summary is that Intel is likely to do very well with its SSD offerings.

gb

I watched several of the IDF keynotes today. Toward the end of Ratner's presentation was a demo of QuakeIV at 1280x720 raytraced at 100fps.

I'm not a gamer but that appeared pretty impressive.

gb

WT: Any plots of share history over a few years?

It would be interesting to see the trajectories.

gb

Anyone found any IDF presentation downloads other than the Day-0 stuff?

tnx
gb

my first reaction is that they are likely to have many more "candidates" for this than they like and still chew up a lot of wafers in the process. and as you've pointed out before, all in a single 65nm fab shared among mobile, desktop and server.

i'm suspicious of the validation and configuration issues; those can probably be mitigated by making them single socket only.

marketing has never been one of amd's strong suits. marketing and pricing this is going to be their biggest challenge. someone else suggested that tri-core might have been the real reason richard left. i suspect they're right.

so who is the marketing whiz kid that's going to make this fly? are they going to find a dennis carter clone?

gb

it will probably require redefining what "good" is.

gb

At 283mm barcelona will *never* yield well.

And without something akin to HK/MG 45nm will just be a leaky fiasco so that makes a "shrink" useless.

Rock and a hard place...

gb