AMD to try to take on Intel Core 2 with tri-core Phenom
It's the multi-blade shaver war all over again
IDF AMD's upcoming Phenom desktop processor family will include CPUs with three processing cores, the chip maker announced last night. Now that dual-core products are commonplace and four-core chips becoming more so, why not tri-core too?
AMD didn't provide any details of the Phenom series members' speeds and feeds, but it did state that the three-core versions will go on sale in Q1 2008 following the shipment of quad-core parts this year.
It also confirmed the chips will use HyperTransport 3.0 for connectivity and contain a bank of L3 cache that will be shared by however many cores a given CPU contains. It didn't say how much L3, however.
Of course, no one sets out to make a three-core product - AMD is simply being commercially canny. Of all the quad-core chips it punches out, some will have cores that are dead or not up to scratch. AMD can't sell them as four-core products, and while it might have once sold them as dual-core chips, by disabling one of the remaining cores, it now plans to offer them as three-core CPUs.
That allows it to not only sell chips it might otherwise have had to discard or offer as cheaper, two-core parts, but also allows it to position the three-core products as superior to Intel's rival Core 2 Duo chip.
Until the three-core Phenoms are independently tested, we can't say how they will line up against the Intel chips, but there's no doubt that three cores provide more processing resources than two, to the benefit of anyone running CPU-intensive tasks like games, media manipulation applications and so on.
The problem for AMD is that Intel is gearing up to release CPUs based on its 45nm 'Penryn' architecture - the Phenoms will be 65nm chips - and if it can get half-decent yields will push quad-core chips into the mainstream.
Indeed, here at the Intel Development Forum, company CEO Paul Otellini will today speak on the subject of "extreme to the mainstream". Core 2 Extreme is Intel's four-core gaming product line, and even if Otellini will be using 'extreme' in a broader sense, he clearly means to drive four-core products further down-market.
Intel has the advantage here that its quad-core chips are made from two, dual-core parts. AMD's four-core chips are based on a single, monolithic four-core design. Inevitably, that means Intel is going to get better yields than AMD, since statistically a given 300mm wafer is going to yield more working dual-core chips than working quad-core parts.
Factor in the greater production density afforded by 45nm production - smaller cores mean you get more of them out of a given standard-sized wafer - and Intel will potentially have the volumes to allow it to push down prices.
Why then buy a three-core CPU from AMD when you can get a four-core chip from Intel for the same price? Yes, an AMD chip may perform better than an Intel one with the same number of cores, but an extra core will surely give Intel the lead.
Intel hasn't said how it will price up its 45nm desktop chips, but there's no doubt it will be very aggressive, and AMD will have to follow suit.
COMMENTS
3 cores faster than 4?
@Steve: The fur-core benchmarks are all on systems where there are two dual-core processors in two sockets or two dual-core processors on one chip but still using the frontside bus. Three or four cores with a single crossbar (as AMD uses and Intel is rumored to use soon) could be much faster. I'll wait for benchmarks of actual systems to find out for sure.
Number of the Beast
So, now AMD has a chance of making a 6-core chip (two 3-cores in the same package), and by extension, they should naturally look into making a 6x6x6 platform for devilishly fast performance.
Nice
Nice , now with these babies and multiple motherboards , I can cheaply assemble and create a very cheap low cost desktop version of the mid nineties super computers , now that rocks !
why more cores?
The analogy to a core for DRM, 1 for apps, 1 for antivirus isn't too far off...
Having tested multicores for quite a while now, even single threaded apps can see a great improvement when run with an SMP OS. Each new process started will be placed on the least busy core, and while a core is tied up doing I/O other processes that aren't awaiting I/O can run at full speed.
Classic tightly coupled SMP tests show a significant roll off after 8 processors without SMP aware apps, but up to there you get almost a multiplier effect for each new added core so long as memory and I/O bottlenecks are avoided.
This *does* depend on your OS being good at multiprocessing.
@ Anon
"Why do I suspect that these will be duff or crippled Quads..."
Because, clearly, you read the damn article.
