AMD’s Next-Gen Microprocessor Still Does Not Exist – Company.

[SRTDodge05]

Advanced Micro Devices said that while its next-generation code-named Shanghai processor is scheduled to be unveiled in the second half of next year, the world’s second largest chipmaker still has not produced a single working sample of it. As a result, it now depends on luck whether the firm is on-track to release its 45nm chips not much later compared to Intel Corp.

“We have 45nm on the way. We will have initial samples also in January. I’m fairly confident that those puppies are going to boot,” said Mario Rivas, executive vice president of computing products group at AMD, in an interview with CRN web-site.
AMD Opteron quad-core and dual-core processors code-named Shanghai, which are expected to be released in the second half of 2008, will use AMD’s K10 micro-architecture that will be present already in the code-named Barcelona processors, but will contain some performance enhancements. Thin 45nm process technology will allow AMD to install 512KB of L2 cache per core as well as 6MB of unified L3 cache, which is a significant enhancement compared to Barcelona.

X-bit labs - AMD’s Next-Gen Microprocessor Still Does Not Exist – Company.

[k0NG0]

The link is wacky.

[SRTDodge05]

Originally Posted by k0NG0

The link is wacky.

Fixed

[Sandog]

Kongo, didn't you mention before that AMD didn't even have a 45nm prototype yet and it would be a year after that? Having Barcy on 45nm in 2008 seems more like wishful thinking.

45nm Penryn quad core dropped power consumption down by %30 under load and %50 at idle from the C2Q I think. The QX9650 has a TDP (max) of 130W but reviews calculate it to be around +80W under full load. AMD's Phenom is higher that C2Q to begin with so how much will they need to drop power to catch Intel and will 45nm get AMD this far? Also, the QX9650 is ran at 3.0GHz. The 2.6GHz Phenom already has a TDP of 140W, higher than any desktop chip I have ever read about. AMD has it's work cutout.

[XJ.]

The process AMD will be using is not as good as Intel's 45nm process wrt/power efficiency. I've read that there will be about a 20% power drop, so unless the designers managed to do some things that drop power usage, that'll be about it. I don't know how much time and effort has gone into developing the 45nm part.

It is possible that the "K10.5" (45nm Barcelona) will be shipping 4Q08 - if everything goes very well. The process is all ready well developed, so it depends on how well AMD's implementation works on that process - and whether they can build up a 45nm manufacturing node in time (they would need to have ordered the new litho equipment already, for example).

AMD might have time for one respin, but two or three will easily put them into 2009.

[k0NG0]

It's usually about a year from tape-out to retail (often includes some time to build stock). If you have working silicon at tape-out, then you're in pretty good shape of course. What AMD is talking about is doable, but as XJ said, margins aren't too big.

I'd say the 45nm machinery is installed already (or at least before the year is out) since they are sensitive items and require quite a bit of calibration and testing before you can start producing silicon. I think the foundry people at Dresden will be very happy to knock this one out of the park seeing as how the 65nm shift wasn't very pretty. That sort of motivation is usually a good thing.

[Sandog]

From the news on Phenom, AMD is getting the known bugs out and then it appears they will attempt a die shrink. It looks like it will be a year before this happens. With this news I bet a lot of people will wait for that to happen and go with Intel in the time being. It looks like AMD will be competing with Nehalem by the time this new chip arrives so it will still be an uphill battle. AMD was making progress with their reduced power K8 chips. These chips actually offer better power consumption than Intels C2D I believe and if they can push these chips it might help the coffers. It will be competing with Wolfdale next year though and not Conroe and both Conroe and Wolfdale are faster. Nobody really needs a quad-core anyway but it will be my next chip.

[Sandog]

Originally Posted by k0NG0

It's usually about a year from tape-out to retail (often includes some time to build stock). If you have working silicon at tape-out, then you're in pretty good shape of course. What AMD is talking about is doable, but as XJ said, margins aren't too big.

I'd say the 45nm machinery is installed already (or at least before the year is out) since they are sensitive items and require quite a bit of calibration and testing before you can start producing silicon. I think the foundry people at Dresden will be very happy to knock this one out of the park seeing as how the 65nm shift wasn't very pretty. That sort of motivation is usually a good thing.

Hopefully we won't see a repeat next year of the Phenom release which was seeing only a handful of the parts at the very end of the year with no supply. AMD can work parallel with getting K10 fixed and K10.5 (45nm) on a die. They should be able to do both in a year since K10 looks 99% finished. Intel had very good success with 45nm and I bet it will be a very forgiving die shrink for AMD (as far a power loss goes). AMD willl have to do some miracle financial accounting to stay afloat next year while the progress continues.

[k0NG0]

The 45nm stuff means AMD will have better margins as well. Meaning they can compete on price more easily. (Though Intel can match pretty much anything, if they really want to.) But the economies of scale a 45nm shrink brings with it can only be a good thing for AMD.

Also, a 45nm LV Turion X2 coupled with a mobile version of the new chipsets (that are incredibly low-power even with umpteen PCIe lanes and lots of features) may be a force to be reckoned with in the mobile sector. I have to say I've been more impressed with AMD's chipsets than I have been with Intel's lately. The platform looks quite healthy still.

[Sandog]

I think AMD and Intel are on a level playing field in chipset die size. Intel has mentioned %25 decrease in power for some of their next chipsets (Nehalem). AMD is certainly going to come out of their rut a much better company. If they get to 45nm quickly maybe they will keep up better with Intel. Intel as you know is working on 32nm as well (Sandy Bridge).

[XJ.]

Originally Posted by Sandog

Intel as you know is working on 32nm as well (Sandy Bridge).

IBM and it's partners have done test runs on 32nm (sram caches and small logic cores as per usual). IBM's 32nm (high K, immersion litho) process should be more competitive with Intel's 32nm process than IBM's 45nm process is visa vi Intel's 45nm. AMD is likely to see a more significant improvement in energy efficiency and clock gains going from 45nm to 32nm than Intel, though Intel will likely be at least 6 months ahead.