I'd argue that SB graphics are good enough for anything but gaming. The only reason many folks get a discrete card is that most of the OEMs shove them down your throat. It's their way of making a couple of extra bux.

I've often tried to configure laptops with the features I want knowing full well that the integrated graphics are more than sufficient for my needs and find that the set of features I want only comes with a discrete card.

For business I can't imagine any corp desktop buyer getting a discrete card unless it's a workstation and that's a whole different matter.

So I expect IB to continue to reduce the real need for discrete which is not to say the OEMs won't try to continue their practice of bundling.

gb

power rulz. the things intel is planning for the platform, not just the cpu, are going to make people upgrade once more. power, ease of use and security features are going to be wowzers.

i sure hope one of the ubooks is supersized to 17". my wife would love an ubook that size for her embroidery field trips.

gb

people get all warped and twisted if it isn't passed thru the ieee or similar. with 80%+ (and rising) of the market if intel licenses it to all their oems then it's a standard in my book.

someone will p*** and moan about whether amd gets to play of course. my guess is intel will say "sure go ahead". nothing to lose.

gb

KNC is focused on DP not SP.

gb

IB is typical of what it takes to launch new silicon. IB is now in production but not yet qualified. That occurred in Q3 and results in a charge. Once the material qualifies and ships for revenue the charge is reversed. That is expected in Q4. I'm sure we'll hear about whether than happened in the Jan CC if not before.

gb

IIRC TSMC tried 32nm HKMG gate first per the IBM process and it was a flop. The dropped it like a hot rock and began developing their own HKMG gate last on 28nm. They never went production on 32nm.

gb

please clarify

gb

i suppose.

Assuming that Oracle's actions are permanent may be premature.

gb

Had an Intel fab yielded at what he thought was outstanding, it would have been shut down and a major investigation would have been under way to see what was wrong.

+1
gb

I lost track of the "club" members. Is there anyo who is particularly interested in a process that competes with Intel's fast CPU process other than AMD? Lots of folks are interested in low power stuff.

I suppose fast programmable logic folk would be interested in finding a foundry that has Intel's highend CPU process but generally they want something that has flash memory for intialization rather than being loaded from off chip at powerup.

Do you know what happens to malls that lose their anchor tenant? They get bulldozed! Ironic, no?

gb

I don't think CapEx is the real problem. Chemistry and physics are the problems. They got a half-baked (literally) process from IBM and it hasn't worked and it isn't going to work.

Gate first presumes that you won't damage the incredibly fragile gate layer while building upon it with multiple depositions involving chemical, mechanical and thermal process steps. It is just a brain dead approach.

I've likened it to making a beef stew atop a delicate souflet in the same pan. How about holding a piece of sheet metal in your hand while activating a stamping mill? How about doing your nails with a chainsaw? Diamond cutting with a jackhammer.

We could start a whole "dozens" thread on this.

gb

It will be interesting to see how the existing 32nm gate first process plays out. With no additional customers, as far as I know, for this process and AMD's dwindling share and likely reluctance to buy a bunch of inop wafers, rather than good die, this looks like a death spiral.

gb

That probably explains the bulk of the $44M for contract cancellations mentioned in the reorg announcement in the Nov 3 8K.

gb

Great explanation.

Thanks!

gb

With no transistors used to implement 64bit instruction decode, no 64 bit registers, no 64bit capable ALU, etc.

gb

He did reply to my email regarding alchemy asking if I meant turning lead into gold. I clarified with "turning sand into gold". He didn't respond to that.

Perhaps he's working on something cleverly negative to write related to the other definition of alchemy. That would be consistent.

gb

Let me reprhase my question, in light of your answer.

Assume two Intel chips, one with 64bit x86 capability but unused due to only being able to run 32bit code (for whatever reason) and the other with no 64bit transistors at all (presumably to save die space)

Would the second be significantly lower power than the first? Or would all of the cache and uncore make the difference a too small percentage to worry about?

thanks

gb

I wasn't trying to be combative. I just was looking for clarification.

gb

I think I posted my guesses to the "new plan" here but at the risk of being repetitive (and repetitively wrong) I think they'll drop new highend servers in favor of low power blades. I think they'll drop mainstream desktop and highend workstation new development. I think they'll focus on whatever low(er) power solutions they can deliver with the process tech they're stuck with rather than what they wish they had. At the low power blade level they can probably leverage the mobile client silicon cores.

I think they'll continue to put whatever liptstick they can on their current pigs but redirect development and marketing efforts on lower performance mobile clients. I suspect they'll get an ARM license.

I also expect that to get where they want to go they'll consider mergers if necessary to get sufficient "mass". With the egos of the Sander's Years gone there's no one left to screw those up. Note that at their current net worth and market cap they might not be the senior or surviving name on such a merger.

gb

FTA:

just looked him up and did it myself.

gb

send Enderle a link to OED definition of "alchemy."

gb

So you think just having some never used transistors wouldn't affect power?

gb

Since so many of the skus are 64 bit now it's probably cheaper to give away fused off silicon free rather than design and validate another 32 bit core unless there is a power hit.

gb

I wonder if there is a significant leakage or other power cost to having 64 bit stuff there but disabled rather than a 32bit only core.

gb

Yup...mostly. Perhaps this is a phone chip focused on min power.

gb

AMD reinventing

http://www.mercurynews.com/business/ci_19358655

gb

The only way for Oracle to get a return on their Sparc investment is to kill Itanium. There isn't room for three non x86 architectures any longer. If this goes to trial I'd be surprised if Oracle doesn't lose and expensively.

gb

I'll be quite surprised if PCIe is the only means of connecting these. They may have shown it on a PCIe card for convenience and that may be a way it is offered commercially by Intel or others but in a big HPC environment PCIe is likely to be a hinderance.

gb

Makes sense. I don't expect KNC to hit the streets until Sept IDF or there abouts. It needs validation much like a server part.

Just keep your source code available.

BTW, have you tried any of the Intel tools? They have some impressive (by all accounts) MP and threading tools. Might be worth trying before bashing your head against GPGPU.

If it works sufficiently well, it would likely carry over to KNC when available.

gb

Every once and a while this board software goes a little crazy and refuses to mark items as read. It's doing it now with the current most recent 5 messages. Strange.

gb

No idea as to packaging, pricing or availability. I retired in 2000 and am far removed from any info. The package shown the the pictures on the web appears to be far too large to be just a packaged die on a carrier. I'd be surprised if it was PCIe only. It has a full virtual memory management unit and cache coherency so I suspect it has the ability to sit on a coherent mp bus as a full partner.

The ORNL presenter did not mention performance. He focused on the fact that they were able to port 10's of millions of lines of complex code and it functioned correctly in a very short time. That's a stunner IMO. He gave the names of the codes involved and specifically said this was complex stuff rather than just linpac.

So to answer your question somewhat indirectly: how long would it take to get your problem running correctly and performant on a GPGPU?

gb

Amusing thread on S|A:

http://semiaccurate.com/forums/showthread.php?t=4993&page=3

Seems that no academic claim can be trusted in front of a group of their peers but anonymous S|A posters should be respected for their name-calling skepticism.

gb

what board is that?

the usual reason intel cuts capital is either:

1. the economy is so bad everything is going to hell everywhere and inventory is predicted to go through the roof. not the case currently.

2. the factory geniuses have pulled another rabbit out of the hat and figured out how to do it cheaper. often the case.

it's never due to taking the foot off the technology pedal. covello apparently doesn't believe the last dozen years of presentations by intel where they predict that the next node ramping as fast as possible saves them money even when the following year they point to the past year's savings as proof.

gb

forgot to add:

tick-tock will mean that ib celerons will no doubt be appearing in q4 of next year. if the sb celerons didn't completely over run yawno and bobcat the ib ones surely will.

atom and arm will trample the other end of bobcat.

gb

At Intel we used to call that a difference without distinction. Yawno is a tweener with little to no additional utility other than to keep some reality challenged fans cheering.

With SB cores now all the way down to celeron space they'll soon be raging rather than cheering. I expect the volume and vitriol to increase.

gb

I've yet to find a situation where the supposedly suckie Intel graphics comes up short. It plays video fine including HD video. No artifacts at all that I can see even on my 2011 55" XBR 3D/HD Sony. I don't game at all nor does anyone I know.

Of course I do Win7 which Chuckles hates. I could care less about *ix. Life's too short for that. I support a dozen mostly Win7 computers for family and a few friends and have hobbies outside of computing.

I think about another year of Intel pulling away will leave S|A and a bunch of other ABI sites with nothing but memories to write about. They'll be nostalgia sites rather than fan sites.

gb

or a bunch of disgruntled shareholders after the amd's annual meeting. the timing of the initial inquiry would seem to make that possible.

gb

rimshot

that one i had to read three times!

gb