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
And what rules would those be?
http://www.spec.org/cpu2000/docs/runrules.html
I am advocating the rules, not arguing against them.
Lol. You really should consider this one as your message signature.
K.
wmbw
Thanks. Chris has it wrong then, saying 2x2MB for Woodcrest - or I just got it wrong - or better said i expected exclusive 2MB caches because Woodcrest would perform a lot better with it than with 4MB shared cache imo.
K.
Chipguy
Could be. I suggest we leave it open until SUNs submission they referred to today is posted on Spec.org.
However, extrapolating from the submission you linked me to would lead closer to 3.4 than 3.6 - or should I say a 3.2 with Foster? Sorry could not resist .
K.
mmoy
Microsoft Office doesn't come packaged for MMX, SSE and SSE2.
It runs x86 at the lowest level that they support.
Ami, if code would be packaged for MMX, SSE and SSE2, how would it behave on a CPU that does not support each of these instruction sets?
Wrt Spec, the rule is a compromise, afaik, limited to use the same compiler for all apps within a set of Spec-benchmarks. Even if it produces poor code for one or two apps within the set.
K.
Chipguy
Increasing this performance to 2500 SPECfp_base2k from
faster hardware alone would require roughly a 3.6 GHz
Opteron.
I calculated roughly 3.2 GHz, assuming SUNs submission is Specfp2000 peak for 2.8 GHz Socket939 Opteron, from hardware alone.
K.
wmbw
...the only meaningful score is one that is run with the same compiler, OS, and methodology.
While I am not saying your take is without merits, why loose any breath talking about Spec-scores if you don't accept the rules of producing the benchmarks?
K.
wbmw
By extrapolating the data, it's quite clear that Conroe is no small boost from today's top designs.
This would depend on what you consider is today's top design, to start with. :) And then, which incarnation of Conroe you are refering to; the shared cache version or Woodcrest?
K.
wbmw
But I can't see AMD getting their SPEC scores up to the level of the recent digit-life estimate
Dunno which CPU SUN used for submission of Specfp(peak, I assume) of 2344. http://epscontest.com/presentations/05q3_sun.htm?slide=26&a, Slide 29.
However, from there to Specpf(base) 2500 before 3GHz Conroe appears would not look like a stretch to me. With tailwind from compilers one speedstep could be even enough, two will definitely do the job. :) The integer part is harder to make, though. It would take at least two steps (with a little help from compilers), and maybe three.
K.
Keith
Could be, albeit I understand Mikes remarks for all Semprons he tracks and the lower priced Athlon64 models as well.
It would need to look at pricewatch occasionally for the next couple weeks to see what the lower value parts do to confirm the assumption i posted.
K.
Mike,
thanks again for keeping us updated. From your comments, could be Henri "Floodgate" Richard is limiting outflow of Value-CPUs for a while again to inititate another step of trading up AMDs offerings in retail. I like what I saw wrt supply management for all brands after XP as much as I hated to see what happened to the latter.
K.
Keith
Thanks. Looks exactly like what you called for. Spells huge Opteron 1xx S939 volume. :)
From a technical stance, I like the airflow-design allowing for fanless CPU-heatsink.
K.
Keith
I've noticed some discussion what will come and and at what price but frankly did't give it any thought. We'll know better tomorrow anyway.
I'm looking forward for an interesting combination of Andi's technical and Jonathans marketing capabilities in today's presentations. :)
K.
Keith
Does that look quite strong to anyone else?
Yes. From a Taiwanese Tier-one point of view, that is. Not sure how much of the gain is market- and how much is mss growth.
But anyway, the numbers sure look pretty healthy.
K.
alan
Yes. However, as I said, it is not Intels style to assign such numbers for Cedar Mill products performing better than Prescott. The link implies Prescott (at least 2MB L2 lithography) has VT circuitry on die already. In essence, Cedar Mill really looks like a plain shrink - couple of *T additions for future use and changes on stepping-level notwithstanding.
K.
p.s: After I replied to mas' posting I realized I problably misunderstood him - if he was just indicating Cedar Mill will be able to clock higher or support higher FSBes, that is. Which is likely to be the case. Albeit, as Anands table indicates, not initially.
Alan
..."not only the Merom die (2M L2?) but also the Woodcrest die".
Yes.
In terms of the spec_int debate, a simple linear extrapolation of the past two years puts AMD at about 2450 in late 2006. It really should be an exponential, but a linear fit seems pretty close for the past few years.
In this ballpark, yes.
K.
mas
Well, that's what most of us expected for Prescott as well.
And then there is this nomenclature tidbit: http://www.siliconinvestor.com/readmsg.aspx?msgid=21688588
K.
wmbw
Yeah. Not much difference between our assessments, give or take couple of points, couple of months.
Btw, if Mike's on spot with his headline in http://www.theinquirer.net/?article=26086.
Intel will bring Woodcrest lithography to High Performance Pentium-Ms as well. It's somewhat plausible, although I don't remember Intel indicating it up to now. But well, if so, it would probably just be the top-end product for benchmarking reasons.
K.
wbmw - edit: link updated, tx j3pflynn
Yup. This refers to the Woodcrest-Lithography (which will appear as High-End-Conroe as well), where the author hints Intel goes back to exclusive Caches. On this basis the expectations for performance sound reasonable, maybe whith a questionmark wrt to memory-starving of a 3Ghz part to be resolved by means of future platforms.
However, for any conclusions on competitive impacts it would need a timeline - and a priceline as well. From the above lithography considerations i would not expect such an animal sooner than K10.
And costly. By this time I am pretty confident K9 will be already beyond the Spec benches the author forecasts for Woodcrest/Conroe 3GHz, as well. K9 is just three speedsteps away from 3GHz.
http://www.computerhq.com/AMD_Socket_940_Opteron_Dual_Core/OSA280FAA6CB/products/partinfo-id-561938.....
link from neubiene/w:o
K.
wbmw
It sure does, given that Intel expects the next version of their WiFi (codenamed Golan) to dissipate substantially less than a watt.
Aren't current state of the art Wi-Fi solutions specced in this ballpark already?
K.
Pete
If it acts like all the other Intel CPUs under this test, it will have a far higher power usage than the specified TDP.
Unless kept within TDP by means of OCDM, no?
K.
mas
Fab30 still has a long 90nm life. Actually, it will reach its full capacity when the 150M tooling-investment of this year comes online. Albeit, I am not sure lithography changes allowing for what Jerry Moench suggested in the presentation you linked to will ever be implemented into Fab30. (E.g. he indicated two additional metal layers; iaw there is changes well beyond what you would call a stepping)
IIrc, Jerry Moench last year indicated he was working on a second generation 90nm process (not 65nm, not K9). In fact I believe to remember he introduced the term 2nd gen 90nm.
In essence, I expect this process materializing in Singapore next year. And hopefully Jerry and his folks found a solution for the problems scaling cache he elaborated on in his talk.
K.
mas
What you suggest is maybe independent from 65nm. There is a 90nm process in the works for 300mm - and new sockets are in the cards for next year as well.
K.
Keith
A fantastic result, the best yet.
Yes. Four to five Watts for the WiFi sounds like a whole lot, doesn't it?
K.
Keith
Yes. Actually, I see more Dell weakness than strength of other players reflected in the picture; along the line Dell admitted its offerings are trading down and otoh we see AMDs offerings trading up at other vendors, somewhat moderating the general (tyrolean thesis) trend.
However, at the end it's just the flip side of the very same medal.
K.
Keith
Apologies. I missed your picture form yesterday's walk down wall street. :)
K.
Chipguy
Well I did not even comment on what could be behind the picture Buggi took this morning.
Neither would I dare to comment on the reasons of this:
http://biz.yahoo.com/prnews/050908/sfth177.html?.v=1
As humble as I am, I only try to see the pictures.
K.
Darbes
Buggi walked along Wall Street this morning and pictured this writing on the wall:
http://ichart.finance.yahoo.com/z?s=HPQ&t=6m&q=l&l=on&z=l&c=DELL&p=s
K.
Keith
Yup. Hope these guys in Tokyo hired last year are helpful in both respects.
K.
Rink
Thanks for elaborating.
so take this for what you think it's worth
Don't worry. I believe there is not many around here being more sceptical about Spansions prospects, or better said about NOR prospects.
Btw, i believe the adoption rate of Mirrorbit is a just a function of its manufacturability. The market does not buy touted cost advantages. It buys the lower price modules. Actually I am not sure Spansion is beyond the inflexion point in terms of Mirrorbit and Floating gate yet - although we have seen the inflection point of prices recently. :)
K.
CJ
But it does give a huge cost advantage over Intel
If NROM quadbit will ever become manufacturable down the road we probably should watch out for Quadbit NAND instead of Intel. :)
[Heck, scratch Quadbit above. And while you are at it, scratch the first part of the statement as well. Ed.]
K.
CJ
Something AMD will find out when they try to do MLC on Mirrorbit
While I share your take on Strataflash, Spansion will face other challenges for Quadbit. (However, from the tone of the following excerpt these are not necessarily smaller).
There is a potential to generate 4-bits per cell instead of 2-bits in some of the advance work attempted by Spansion for the ‘QuadBit’ product line. This variant of ‘ORNAND’ MirrorBit technology combines random access, high speed writes and reads, with low cost and high density that generates 4-bits per cell with only 4 levels of voltage sensing. (Sorry, no link).
I believe what you said previously about 16 states is accurate, though. Four voltage levels are enough to accomplish this, right?
K.
mas
Linpack score of Red Storm as from August 1st submission is exactly as expected for single-core Opterons (albeit expected for an earlier deployment, and iirc on the basis of 2.2GHz parts.) Bill Camps said a year ago planned replacement of CPUs by Dualcores will boost scores by a factor of 2.5 by end of this year. (Or was it 2.25? its long ago, I'd need to look it up).
K.
avatar
Interestingly, these results of (discrete grafics based books) battery-runtime-tests are diametrically opposed to recently published results for UMA-grafics-Turion and Centrino books for low and high performance utilizations, respectively.
K.
Keith
While it could be there have been spot sales at .4 somewhere, somewhen, none of the 15 4-Mbit modules tracked by FRMA had contract prices below .56 USD last quarter. The lowest reported OEM price was .52 USD. (sorry, no link available :) ).
K.
erratum
Takes a Dollar of ASP to make it these eight M income.
K.
Darbes
While ascending ASPs from direct consumption of the folks I was talking about get lost in the roundings, these chaps are opinion leaders. Their buddies taken up the ladder to double ASPs will show in AMDs numbers: Just couple cents in ASP probably - but then, a dime in ASP makes eight point something million bucks, very close to AMDs net income last quarter.
K.
Combjelly
Traditionally the "price/value fraction" of the overclocking community favoured AMD products in the 75 Dollar price range, the last two ones i remember were Barton 2500+ and 1700+ Thoroughbred.
AMD still offers products at this pricepoint, but has built similarly compelling value propositions at the double and quadruple of this pricepoint now.
I begin to understand what AMD really means when it says there has to be freedom of choice out there.
K.
Keith
Is there a way to copy and paste or to get it in such a format?
1. Download the file
2. Open it with Acrobat
3. Choose Select Tool
4. Mark anything
5. Move the mouse inside the marking, then on the broken picture icon. Choose "Recognize text using OCR". This will open a dialog box from where you will be fine on your own, I guess.
This is for Acrobat 7 Pro. If it does not work like this on your version, let me know, I will mail you an OCR-ed version then.
K.
Chipguy
SUN is working on these kind of things for Solaris under the Janus moniker. Afaik not yet for Sparc-Code though. Actually I am not sure it would work efficiently enough to make sense for the time being at all: Code for massively parallel architectures would probably not run well even if you could make it run on X86 hardware. Maybe in couple of years when X86 will be more parallel.
K.
Chipguy
why bother with the SPARC boxes at all?
For the virtually only reason Sparc-boxes are sold nowadays: To run existing application code.
K.