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.
LOL. There was some different K9 thing that got scrapped, but it is clear from what AMD has said that Alsup is talking about the same design that AMD is attempting to bring to market.
When I say "cache" I'm not counting the increased LSD rumor. Loop Stream Detector is part of front end.
Yeah, I know, I just meant as another possible source of IPC gain.
Anand claims the L3 of Lynnfield measures at ~35 by "cachemem" but is actually 42 cycles, per Intel, or something. And that the same tool pegs SB at ~26 cycles
http://www.anandtech.com/show/2960/2
The kids at amdzone are in full anxiety mode over the Sandy Bridge benchmarks.
Yeah, the canardPC leak had L1 @ 3 cycles (from 4), and L2 @ 9 cycles (from ? in Westmere)
But Anand says he measured 4 and 10... but who knows.
I suppose the "trace cache" (Loop Stream Detector) being bumped to ~1000 entries might help some, if that's what happened.
It has to help that L3 latency appears to have been substantially improved, right? Or would prefetchers be mostly hiding that anyhow?
I wonder where they found the ~20% additional IPC? Seems like it would be getting to slim pickings by now...
Yes, and so the overall rate should be lowered, but applied to all earnings, foreign or not.
And consider that the other CPUs were allowed to turbo (the i7 880 up to 3.73GHz for a single thread), and that the tested sample had only 6MB L3...
And yet, on average it was 10% faster.
Correct for the above, and it looks like on the order of 20% IPC improvement... which is stunning, if correct.
And that iGPU is neither full speed (1100 vs 1350, assuming its turbo was even on) and has only half the execution units (6) of the top (12) model!
All in all, "a home run".
I'm not sure you've grasped the context of the discussion. ;)
Did you notice that the 4-wide decoder has to supply 2 cores? ;)
Erm, no. :) It means that a portion of the 50% gain with 33% more cores is not due to IPC, but due to higher clocks. MC could not turbo cores under low-threading conditions. To the extent that these applications are averaged in, the IPC gain is lower, relative to MC.
It turns out from another JF post that part of that "50% gain from 33% more cores" across a selection of server loads is relying partially on the effect of low-threaded server loads that benefit from BD having turbo (per module) where MC has no turbo at all.
Which I think means that the gain is significantly lower if one is looking at fully-loaded work.
Why not go after nVidia? Or ARM?
ARM? No way that gets approved. antitrust-o-rama.
nVidia... might get by the antitrust types... but their business is already cratering as Intel (& AMD) put GPUs on-die.
Oops. :)
Yeah, I think most people think of their (really bad, honestly) consumer AV product for Windows. That one does indeed blow chunks, compared to the alternatives like MSE & Avast. Even *norton* is arguably better (shudder).
But I suspect that is not Intel's main focus in buying them.
AMD server share plummets to 6.5% (!)
http://arstechnica.com/business/news/2010/08/bad-news-for-amd-as-intel-gains-server-share.ars?utm_source=rss&utm_medium=rss&utm_campaign=rss
According to a new market share report from IDC, Intel managed to take critical server market share from AMD, with the former company seeing a year-over-year jump from 89.9 percent last year to 93.5 percent this quarter. Meanwhile, AMD's market share dropped from 10.1 percent to 6.5 percent.
Hmm, that's not bad... so other than currency fluctuations, are there any risks, or odd tax implications?
AMD fusion is hanging out with IBM's hik/mg process tech
on the boulevard of broken dreams.
But, but... Dirk said the 32nm hk/mg yield problems were of the completely standard variety one *always* encounters when ramping a process, but that he just felt like mentioning them publicly on the CC for the benefit of any interested parties.
Yes, but there'd be the constant worry that dingos might steal them.
You must be crazy if you think Fusion isn't going to take a huge bite out of intels ass.
Pssst. I have a little secret for you...
Intel has been shipping "Fusion" (iGPU on CPU package) since Q4 2009.
What's that? On package isn't good enough?
Intel will ship iGPU on die in this Q4 2010.
Meanwhile, AMD has not yet shipped ANY "Fusion" parts-- their CPUs have neither on-die nor even on-package iGPUs. Their mainstream fusion part (Llano) is in fact delayed due to GlobalFloundering's problems with 32nm. (Who could have predicted that one?) They promise to launch a low-end TSMC bulk 40nm fabbed device with on-die graphics in Q1, as they continue to struggle along with Llano.
So... "Fusion" and "APUs"... AMD picked the buzzwords, but Intel's been doing it for nearly a year now, with AMD still MIA.
There's this whole cloud/saas area as well:
http://www.mcafee.com/us/cloudsecure/
But that's a generic argument that can be made against *any* acquisition.
I will say that I hope Renee James is merely uncomfortable in front of a camera, rather than the completely incompetent buzzword-bingo moron she appears to be in this video on the deal:
IMO, many of these requirements will not actually accomplish what the FTC and the casual reader thinks they will. Put another way, Intel can meet the letter while avoiding the spirit.
I don't think these two really have the technical background necessary to correctly evaluate the settlement-- I'd trust that Intel's lawyers knew what they were doing.
Looks like the AMD / AAPL rumor may only amount to this:
Ontario in a new version of AppleTV.
http://blogs.barrons.com/techtraderdaily/2010/08/09/apple-digitimes-sees-new-ipads-cdma-iphone-new-apple-tv/?mod=yahoobarrons
Yay?
Other than the extension of the Via x86 license for another 5 or so years.
3rd quarter earnings will be another shot in the face of these doomsayers, imo.
If Intel met his earnings estimate and had a P/E of 15, it should be $30...
See definitions I and U. Basically, anything with a small screen ( < 7" ) is not a "Computer Product", and "Mainstream Microprocessors" are only those used in "Computer Products".
Yes, he's out to lunch. LOL @ Intel packaging an nVidia gpu in an MCM with one of its CPUs.
I seem to recall that basically nothing applied to anything with a 7" screen or smaller. So, no worries there?
Ir doesn't actually say that. It says:
Under change of control, they agree to offer to not to sue over the course of the next year while "good faith" patent license negotiations take place with the acquirer, if the acquirer also agrees not to sue during that year of negotiations.
Doesn't say the negotiations have to succeed.
JF claims the 50% is not "up to" but "an aggregate *estimate* of server workloads", but hey, if they throw in workloads that benefit linearly from doubled FLOPs, that can boost up an average real quick.
I have found one good thing for it: "words with friends", the online scrabble game. It's just the right size for that.
And not enough apostrophes.
And not enough apostrophes.
Elmer, Romans, countrymen, lend me your ears.
I come to make allusion, unrecognized.
Can you dispense with the fake, broken English?
If you think Intel "is getting crushed today" (down ~2%), then you should see DELL, AMD, NVDA, STX, CSCO, AAPL, etc.
Makes one doubt it's anything to do with some Intel-specific deal... or anything, for that matter.
And as of today, of course, Apple refreshed the entire iMac & Mac Pro lines with the latest Intel 32nm parts. Maybe he should just retract the entire article.
The author is blissfully unaware that Intel will also be offering Sandy Bridge in an IGP-less version in the same segments that Bulldozer will attempt to address. So... there goes that argument.