Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
wbmw
at 90nm there is a much steeper trade-off between power and performance.
I generally agree.
By shrinking the process and maintaining the same frequency, the lower power benefits will seem more distinct than they've been in the past.
Well, yes, I believe this can be said for products designed and fabbed with a sound understanding of the underlying physics and adressing these appropriately. However, it certainly cannot be said for every 90nm product. In some cases, power went up considerably, maintaining the same frequency. I believe another trade-off is getting steeper here, as well: Power/performance versus diecost.
K.
wbmw
Venice may look impressive in that it cuts real world power dissipation in half over its predecessor
Indeed. Enough to impress me. And I am not one particularly easy to impress. I follow this industry since 82, certainly there was always improvements towards the end of a node in this respect. But cutting the power within a node in half for a particular perfomance level is nothing I saw often.
Btw, the top bins is not where the music really plays nowadays imo. Skewing process parameters towards Turion-feature-yields makes a lot more sense (read: dosh) for AMD.
K.
wbmw
They released a 2.6GHz 130nm Athlon 64-FX, didn't they?
I thought 130nm topped out at 2,4GHz (?)
Nevermind. 130nm was actually not all that impressive. They had working Silicon in 2001 already, then launched in spring 03 parts between 1,4 and 1,8 GHz, and topped out 18 months later at 2,4 GHz, while working down power envelopes as well. I would not be overly impressed about binsplit scaling of their 90nm node as either, wouldn't it be for the fact they work down power envelopes simultaneously at the pace I see it happening.
Wrt to one-time power gains from their 90nm process, I agree on it for their first 90nm Winchester-Cores. But I see considerable continous (not-one-time!) improvements beyond these over the last year as well. I believe the ability for extend learning over the pilot-mode into volume manufacturing is indeed something APM makes a difference for. Maybe not the only one, though: Another contributing factor in this respect is engineering ressorces: Fab30 was operated pretty much as a foundry within its first four years of operation when it had two architectures and several lithograpies in this fab and needed to migrate nodes while maintaining volume production. As soon as the fab is fully migrated to 90nm, processing only three lithographies in one node and no migration to come, the concentration of process-engineering is certainly beneficial for the remaining processes.
K.
How do you figure??
Looking at the trends of AMDs and Intels 90nm products wrt characteristics. It's obvious that binsplits are going up and power is going down at an impressive pace in AMDs offerings since the very beginning of its 90nm node.
I never ever said Intel has poor yields (while I frequently said I believe there is common miscomprehension about world class yields nowadays on this board). In fact, it is clear Intel has significantly lower diecost than AMD. But raising sweetspot of characteristics does a lot of good things for AMDs economies. Yield improvement is one implication of it; but well, Intel being a literally standing target in this respect currently helps a lot in this respect.
K.
wbmw
Their engineers are just now learning these processes, while the processes have been at place at Intel for a decade or more.
Now that would really neither explain the characteristics learing-curve AMD demonstrated over the last year nor the fact we did not see anything similar for Intels offerings.
But well, put it this way, if what you say is right AMDs engineers are just learning a lot quicker than their counterparts at Intel nowadays. However they do this.
K.
deleted - obsolete
alan
I believe the centrino wireless chip (802.11 radio and guts) is built on 90nm
200 or 300mm network?
K.
alan
I suspect of the four 65nm factories about 1.5 of them will be used for lower power "other" stuff and about 2.5 of them for CPU's.
Which other stuff do you believe will be on 65nm anytime soon? I'm not saying it is impossible, but Intel never used a leading edge node for anything else than CPUs afaik.
K.
rupert
Yes I remember. I assume radio-shack pays about that much for it, why I find its retail-pricetrag impudent. But then, it is a business decision. If they believe they can't sell twice as much for 199, it's reasonable from their stance as long as they have a USP for PIC in the U.S.
Btw, Solectron assembles it in Mexico if I remember right.(?)
K.
bobs
The CPU in the box radio shack will sell is a Geode GX. I'd appreciate any correction, but I would put its capabilities at 366MHz rather in the ballpark of a Pentium2 /K5. When I look at GeodeGX, I still think of it as a Cyrix on steroids. I used one of these, not exactly sure how it was clocked, because these chaps invented the Pentium-Rating back then. And the plus behind it as well. The part I had was a PR 166+.
Anyway, I believe PIC is rightly positioned in the U.S. and in every other developed country in a greater scheme:
I mean whatever one might think about taking people in developing countries by the hand to lead them into the glory land of googled wisdom - there is plenty of people around us who deserve the same treat.
Enough sarcasm for today. Or maybe not. 300 bucks for a PIC is impudent - but well, the people targeted with it are the usual victims of rip-offs. Now it's enough.
K.
Andi B podcast-interview
Sorry if posted already, I could not follow most of the thread lately.
Part 1
http://blogs.sun.com/roller/page/rgiles/Weblog/i_o_podcast_0004_andy?catname=
Part 2
http://blogs.sun.com/roller/page/rgiles?entry=i_o_podcast_0005_andy
buggi pointed to it on w:o
K.
p.s: Albeit i am aware it is an offtopic-request, I'd appreciate your takes on the podcast-format in general. Tia.
mike
Frankly, I'm not sure, and I did not get a hold of Andy immediately for clarification.
In the balance sheet, it is accounted for as "Deferred income on shipments to distributors". The part I am not sure about is if and how it is accounted for in the P&L.
From my general understanding of schemes, I believe it would be accounted for at least with respect to valuation: High-end CPUs in the inventories is manufacturing cost, stuff shipped out is market value (or better said what is invoiced, independent from when it is due). Insofar it would make a difference for the calculation of GM as well.
Needless to say, any assistance appreciated. My accounting skills are rusty nowadays. :(
K.
Rupert
Many thanks for going after it. However, this part in Axelias PR does not look like the tools are used in Intels fabs:
AMD was the first device maker to select Axcelis' single wafer implant technology last year.
K.
rupert
I wonder if somebody in the know could confirm Optima's Implant Precision and Angle Control is in Intel's 300mm toolset as well.
I'm asking because since Chipworks has found the geometrial precision of Intels dies from 300nm node impressive.
K.
duplicate, disregard eom
Mike
Many thanks for clarifying.
I was referring to the Pentium 8xx DC and 6xx 2MB parts. Why are so many in stock and the sell rate looks low, and they keep ordering more.
Ah ok. I just phoned up Andy Bryant to ask what this is about. He said look, it's quarter end and I need to come up with numbers in three weeks. When I look at what I have, revenues soso, and I have this chipset bogus out to explain why I am coming in a tad below midpoint and below the market. Problem is I am a tad short on gross margins as well, and I have nothing much to cover this part. So one of the things that at least help a bit is stuffing the channels with higher end parts, which I have a pile of because OEMs don't buy less of these than I want.
Hth.
K.
Mike
Many thanks for the update. :)
"A lot of stockpiling going on… "
Just to let you know people are reading your comments and look at your data: I see a whole lot of minus signs in the column of sequential change in jjs sortable version, albeit no signs of significant stockpiling across the board. (?)
K.
tecate
Oh. Putting other people's laurels on its head is good tradition at Intel. So it did not come at any surprise when Intel suggested it is the inventor of Wireless LAN in its Centrino-campaign and nobody else can make it work as good as Intel as well. It did not disturb much Intel could not even make the client silicon of it initially for Centrino but had to buy it from Philips.
K.
JJ
Lots of minus signs in stock plus orders versus last week. I'd expect this to continue next week as well, before the price round looming at the beginning of Q4.
K.
buggi
I think its the first time, that we could see, the inventory split between Flash and CPG at AMD.
Spansions inventories were in S-1 in January already.
Looking at CPG inventories, next step is assessments on the structure in terms of raw materials/WIP/finished goods. WIP consists of wafers in Fab30 and test- and packaging facilities plus die inventories.
For the valuation side, there is probably not much finished goods at all in AMDs inventory, these are rather in the deferred income on shipments to distributors position, where the valuation is obviously sales price. In the WIP valuation is manufacturing cost or market value, whichever is lower.
K.
rupert
I actually see Spansion rather up on units than on ASP
this q. Couple of upticks for lower density modules in spot markets do not change the environment significantly imo.
I do not share the presumption Intel is the cause for the NOR-mess, btw.
K.
Keith
Yes I did. I edited for clarity. Just wanted to put it in two dimensions qoq and yoy to point out it is bad, but probably not getting worse.
K.
Keith
It could, theoretically. But NOR seems to stabilize now - on a sadly low level, though. Anyway, Spansion should be up qoq. But I see it still down yoy for Q3.
K.
Keith
wouldn´t this be substantially below what AMD has indicated in their Q2 CC?
That was for CPG. It does not contradict the cited sales growth necessarily.
K.
Buggi
Although everybody is free to change his mind whenever he wants, the reasoning sounds rather poor than standard.
K.
mmoy
Pathscale is considered the fastest compiler suite for AMD64 currently, at least for scientific apps. Can't tell how they are in general though, let alone answer your detailed questions.
K.
Keith
Well. As long as sliding market share and Intel-only coincide, AMD has a lucky problem with it.
K.
Good catch - thanks. K. eom
Keith
If FS would use their 2-4P (especially Opteron) blades
I don't think it would work this way. They need egeneras blade-plane (i believe that's what they call what DELL calls blade-center), their middleware and management software as well, and certainly their Xeon-blades as well to offer their customers choice.
K.
Windsock
Thanks. Never saw this symbol. Is it Nas100 or Composite?
K.
windsock
ami, what's IXIC?
Tia
K.
Keith
I've seen egenera blades displayed a while ago, and had somebody explaining it to me. IIrc you can use Xeon- and Opteron-blades in their system. I believe indeed what FUSI will announce tomorrow is about blades, from the article you posted. The nine months lead would refer to SUN, as a counterstrike on Galaxy. I see no lead over HP. Their systems works pretty similar. I don't know how competitive IBMs Opteron blades are.
K.
Chipguy
HP claims over 5000 apps on HP-UX/IPF while Sun claims
about 7900 apps on Solaris/SPARC.
What would be your guess for the relation of developer-years for these apps? :)
K.
Mike,
many thanks.
K.
Keith
Yes. And it does not support any closed floodgate musings in the lower end. :)
Plenty of die-inventories to supply the value for a while down the road, apparently.
K.
Mike
Thanks. Am I reading it right K8 coming down to Sempron 2500?
K.
mmoy
Many thanks for outlining the options.
Do you happen to have an idea of which one is used most commonly nowadays for CPU-intensive apps nowadays, e.g. games?
K.
joe
One of the nice thing about AMD64 is that when generating the AMD64 code, there are certain assumptions you can make about the microprocessor, one of which is that it has MMX, SSE, SSE2.
Indeed. Many thanks for dissecting this one out.
K.
wbmw
Apologies I have hurt your feelings. My excuse: What intended to be funny crossed the line to become offensive.
K.