From: JAYSANJOSE Mar-23 5:23 pm
To: DANYK9 unread (39 of 43)
31732.39 in reply to 31732.31 prev next
Actually you hit that one on the head. Intel has been battling the electro migration/yield problem since .25u. The problems they are trying to overcome, heat and power use (and associated awful performance), are directly related to the number of transistors on die and the core clock speed. If they can get .18u to work it makes it possible to increase the number of transistors on die while simultaneously decreasing heat (and die size).
IBM started working on the copper interconnect process when they realized the physical dimension limits of Al at .18 (it's physics people, the laws of physics can't be broken). When you try using Al to get to .18u you about always end up with some electro migration, and that's not just cache (SRAM) I'm referring to. I'm talking about wafer yields. The number of processor cores that function drops to the 30% range when made with Al interconnects at .18u. Intel and IBM and the industry knew this years ago. AMD knows about it.
I was baffled and wondered if Intel had developed a new process when I first heard about them going to .18 with the PIII ("Coppermine"). If they hadn't found a way around migration (they were still working on yield problems at .25!), and they were sticking with Al (they never licensed the .18 Cu process from IBM), they were headed for serious yield problems. And from what we've seen from their Coppermine production in the last 6 months, it looks like they still haven't found a way around the low yields using Al at .18.
I have some contacts at AMD who claimed they were making .18u K6-3 cores, this was last year around the time they were finishing up their new Dresden fab. According to one source in Austin, AMD was then getting yields of over 70% making K6-3 cores at .18 in Dresden. I found that hard to believe. She was talking about their debate over releasing a 600 MHz .18u K6-3 before launching the Athlon. Of course the next thing I knew they launched the Athlon, which we'd heard had been delayed till September due to bugs they were working out. This was in June of last year. The next time I talked to my contact she said Dresden was already making .18u Athlon cores running at 950MHz. This was in November last year!
I don't know what to think of AMD anymore these days. They are starting to remind me of how Intel was a few years ago. And now Intel is reminding me of how AMD was (baffling, truly baffling). Last month my contact at AMD told me Dresden's got working Athlon silicon running with on die L2 cache running at 1.1 GHz, and it's .18u, and the yields are in the 60% range. Now what do I think about that? Well it seems hard to believe.. However, so far, every time this engineer has told me something new they've accomplished, right after she tells me AMD releases the chip into their channel. And so far every time the performance and specs of the chip they've released have actually surpassed what she'd told me.
As far as the Intanium benchmarks and specs, it seems ironic now the more I think about it, but the stuff that's been leaked out about Itanium, Merced, and McKinley is pretty accurate. It's not 100% completely accurate, but from what I've seen of Merced and Itanium, it's damn close. So it seems Intel has a lot of whispering engineers in addition to their big-mouthed marketing department. I know at HP it was a pretty tight ship. Now if I search the web for info on HP's version of Merced, the specs and performance I can dig up on it are pretty much nil compared to what I can find about Itanium and McKinley (but aren't those IDFs pretty much meant to hand out specs and info on whatever Intel currently has in the works?). I remember at HP we started getting queasy when we saw how loose Intel was about handing out specs on works in progress. Around that time (1997) the decision was made at HP to go at Merced development in parallel with Intel, not truly a joint development. We were figuring if we made significant breakthroughs they'd have less chance of getting right into the public domain this way. If we handed them over to Intel, we knew they would end up in an IDF somewhere down the road..
You guys probably aren't going to believe this, and I still can't figure out exactly how they are doing it, but the goddamn Register is actually the best (if I can use that term while referring to them) source for info on the IT industry. I have my suspicions on how they are doing it, but I saw first hand at Cupertino a traveling group of Brits who constantly came and went, and I know for a fact that there were a bunch of engineers at Cupertino who were in regular contact with email pen pals in England. I don't know how many times I've flipped onto the Register and read something and thought to myself "How in hell did they find that out??" They have some amazing sources, and what's amazing about their sources is, they are remaining uncaught! Stuff is slipping out of these companies and getting put right into the pages of the Register. Sometimes I read stuff in there and it's almost verbatim of a meeting I was at! The bastards are good, I'll give them that.
I was quite impressed by your posts on the K8. I actually followed the thread on it (I usually never read chat groups). I checked the pages you pointed to and sure enough, the pages backed up what you were saying. I'd appreciate any further digging you do on the K8 development. Whatever you post will be appreciated.
Like I said before, I got here after checking out Sharky's review of the 866 MHz PIII and spotting that backpedaling by Intel about .18u. I usually read the hardware review sites on the web, especialy when it's about a new chip. Believe it or not but a few times those guys have gotten chip samples before we have. They are not in-house lab tests by any stretch, but Sharky and Thomas Pabst know their stuff (Tom is amazing), and when I want a quick impression of the latest greatest chip on the market I'll go right to their sites. When I want to find out what management changes or product changes have been made at my company I go to the Register (big globs of sarcasm attached).
If you post here a lot and you keep posting about K8 here, I'll keep an eye on this forum. Sounds like AMD is working on an aptly code-named 64-bit chip. I'm going to email somebody in Austin about it later today..
delphi.com |
