AMD three-core Phenoms to ship March 2008?
Second, 45nm generation to follow a year later
Basic details have begun to filter through regarding AMD's three-core Phenom line-up, due to go on sale in Q1 2008.
That's AMD's official launch window, but according to motherboard-maker moles cited by DIgiTimes, they'll actually appear at the end of the that quarter, in March 2008.
The sources said the chips are - surprise, surprise - fabbed at 65nm, equipped with the Socket AM2+ interface and sit on a HyperTransport 3.0 bus. All of which, anyone could guess at based on what's leaked out already about the larger Phenom family.
The tri-core chips will contain 2MB of L3 cache shared among the three processing cores, each of which has its own, dedicated L2 cache, the size of which remains unknown, though 512KB is a good bet. We also suggest AMD equip its chips with the ability to support 800MHz DDR 2 memory.
More interesting are the moles' claims that we won't see second-generation tri-core Phenoms until H1 2009, when they will be upgraded to a 45nm production process, the HyperTransport 3-enabled Socket AM3 interconnect and the ability to use DDR 3 memory.
AMD announced the three-core CPU series in September. The chips are almost certainly quad-core parts that have failed to pass AMD's tests for full four-core operation. Rather than bin them, or release them as even lower cost dual-core parts, AMD gets to offer them as mid-range parts.
COMMENTS
It depends on the price
I worked out (while waiting for Eclipse to load one day) that I need exactly 3 cores for developing Java apps. If these things are only €20 more than an dualie, then I would buy one, but if they are only €20 cheaper than a quad then I would go without peanut butter for a week or two.
It's possible that only 3 cores could save the need to buy a noisier fan and bigger PSU, and I'm a very noise and price sensitive kinda guy. I'm running a 2200+ Sempron at the moment, even though I could afford a Barton at the time I bought it.
I'm certainly in favour, although I suspect you won't find many outside the homebuild market as the supply could dry up in an instant when AMD get their yields up to scratch.
A quick game of leap frog
I can see what AMD are doing here they are going to play a short game of leap frog in order to show that they are different and do not copy everything Intel does. I anticipate a 6 core on the horizon, perhaps they aim to get that out before Intels 8 quad. Then it may go 12 core AMD 16 core Intel, after that point the game will end. And by that point we will probably be on 8nm anyway.
The only difference...
..is that those chips are downgraded to existing production part lines. i.e. a P4 that gets downgraded to a Celeron or a Conroe that gets knocked down from an E6700 2.6Ghz to say an entry level E2140 1.6Ghz. AMD is creating an entirely new line of unplanned chips based on duds.
This is a huge set back in terms of marketing and PR. Everyone who pays attention to the CPU market (read OEMs, since Dell and the like are the real customers here) knows this move is 100% because AMD is having terrible fab problems with 65nm, and you're going to have to convince those people (again OEMs) that these tri's can be sold to Joe Sixpack. Can they take garbage and polish it shiny enough to fool the laymen? Only time will tell, but I'm not going to be holding my breath.
They could downgrade the "tricores" to dualies and probably make just as much money, even though they'd sell at lower prices, simply because you don't have to spend the money to basically create a new niche market in such a short time frame.
They're not going to make good margins on these chips, heck I'd be sursprised if they break even after all the advertising and sales pitches that are going to have to be made just to get OEMs to buy in and then convince Joe sixpack to buy one, and if Intel jumps into the marketing fray with ads bashing these gimped quads, then they're really going to be loosing money. If I was Intel's marketing department, I'd be pretty damn excited about rubbing AMD's face deeper in the mud over th next 4-6 months.
Also @ Kevin
Wouldn't it be interesting to find out how many Core 2 Duo CPUs are failed Quad parts?
I'm betting at least a fair amount... some probably have 3 cores that COULD function but Intel choose not to offer such an SKU. Perhaps it's not the case at all, but would be interested in finding out...
