I have some exciting news for everyone. For a long time it's looked like AMD's quad core K8L would be server oriented in it's initial introduction while Intel's quad cores in the K8L launch timeframe would be dual die types. However, it looks like there will be true single die quad core competition in Q3 2007 between AMD and Intel.
http://www.google.ca/translate?u=http%3A%2F%2Fwww.hkepc.com%2Fbbs%2Fnews.php%3Ftid%3D678537&langpair=zh-CN%7Cen&hl=en&ie=UTF8
http://www.google.ca/translate?u=http%3A%2F%2Fwww.hkepc.com%2Fbbs%2Fnews.php%3Ftid%3D678534&langpair=zh-CN%7Cen&hl=en&ie=UTF8
How are they accomplishing this feat you may ask? Well in the case of Intel, it looks like they are very aggressively pushing toward 45nm and it's certainly impressive that they seem to want to debut the process immediately with full size quad cores rather than smaller dual core shrinks. All we know about Yorkfield besides 45nm and single die quad core is that it'll use a 1333MHz FSB (please get it to 1600MHz). It'll probably have a 12M shared L2 cache.
On the AMD size, it seems AMD is pushing back AM3 and DDR3 support until H2 2008 or about 6 months. Instead of launching quad core K8Ls on desktop for AM3 with backwards compatibility for AM2, they will be launching on the new AM2+ in Q3 2007 with backwards compatibility for AM2 and forwards compatibility with AM3. Given the year gap between AM2+ and AM3, it may well be that the intial cores won't have a DDR3 IMC in addition to the DDR2 IMC in order to save space so it won't be as much the cores that are forward compatible than the K8L architecture itself. People wouldn't be putting 1 year old chips in a brand new platform anyways so it isn't a huge difference.
Now the difference between AM2 and AM2+ is HT3.0 support and individual core power clocking support in the later. Current motherboards are obviously AM2 and so although you can upgrade to quad core K8L you'll probably have slightly lower performance and higher power consumption than on a modern platform. The good news though is that even AM2 should support future 45nm chips so AMD is certainly keeping their upgrade paths open.
What is interesting is that K8L is now longer called Rev H but is now called Rev B. I suppose that is why all the "Hound" names were cancelled. These desktop quad core K8L's are called Altair. Now the bad (less good) news. People who were expecting miracles with AMD's 65nm process will be disappointed. First, there will be no clock speed increases initially. While Altair will be available in FX 4x4 configurations and mainstream X4 AM2+ configurations, they will be topping out around 2.9GHz (thanks to the new half multiplier in Rev G). What's worse, when AMD said quad cores within the same thermal profiles as current chips they meant it. Altair TDPs will be 125W in both FXs and high-end X4s. No doubt lower clocked X4s will have a lower 89W TDP, etc., but that's not quite the same thing. L3 cache will remain at the previously announced 2MB so the large cache versions are still targetted for 2008. (Z-RAM is not likely on the horizon at all).
Now for those of you who want to jump on 4x4 when it's released in a month, it may not be a good idea. Why? Like AM2, current Socket F based 4x4s will be replaced by Socket F+ 4x4s in a little more than 6 months. Now people who already have an AM2 then K8L is a great upgrade option, but if you are buying a yet to be released high-end platform that's already outdated it may not be as inviting. While HT3.0 support isn't important in a 1P environment of AM2+, it'll definitely be useful in the 2P 4x4. The whole point of going to HT3.0 afterall is to assist in cache-coherency. The current 4x4s lack of support for K8Ls individual core power planes will also be a killer since those 2z125W TDP numbers will end up being a lot higher without power saving features.
On a side note, I've noticed that the Kentsfield samples that have been floating around at IDF are stepping 7 up from the stepping 5 B1s that we've seen lately. Stepping 7 seems to be the next to final stepping. I just wonder if 7 implies it'll be B3 or C0. It may be C0 with launch at C1 especially if what The Inquirer said about Intel correcting some B2 Conroe errata in a new stepping is true. I doubt there's major performance increases in stepping 7, probably just steps refine power consumption and in Cloverton to enable 1333MHz FSBs.
http://www.google.ca/translate?u=http%3A%2F%2Fwww.hkepc.com%2Fbbs%2Fnews.php%3Ftid%3D678537&langpair=zh-CN%7Cen&hl=en&ie=UTF8
http://www.google.ca/translate?u=http%3A%2F%2Fwww.hkepc.com%2Fbbs%2Fnews.php%3Ftid%3D678534&langpair=zh-CN%7Cen&hl=en&ie=UTF8
How are they accomplishing this feat you may ask? Well in the case of Intel, it looks like they are very aggressively pushing toward 45nm and it's certainly impressive that they seem to want to debut the process immediately with full size quad cores rather than smaller dual core shrinks. All we know about Yorkfield besides 45nm and single die quad core is that it'll use a 1333MHz FSB (please get it to 1600MHz). It'll probably have a 12M shared L2 cache.
On the AMD size, it seems AMD is pushing back AM3 and DDR3 support until H2 2008 or about 6 months. Instead of launching quad core K8Ls on desktop for AM3 with backwards compatibility for AM2, they will be launching on the new AM2+ in Q3 2007 with backwards compatibility for AM2 and forwards compatibility with AM3. Given the year gap between AM2+ and AM3, it may well be that the intial cores won't have a DDR3 IMC in addition to the DDR2 IMC in order to save space so it won't be as much the cores that are forward compatible than the K8L architecture itself. People wouldn't be putting 1 year old chips in a brand new platform anyways so it isn't a huge difference.
Now the difference between AM2 and AM2+ is HT3.0 support and individual core power clocking support in the later. Current motherboards are obviously AM2 and so although you can upgrade to quad core K8L you'll probably have slightly lower performance and higher power consumption than on a modern platform. The good news though is that even AM2 should support future 45nm chips so AMD is certainly keeping their upgrade paths open.
What is interesting is that K8L is now longer called Rev H but is now called Rev B. I suppose that is why all the "Hound" names were cancelled. These desktop quad core K8L's are called Altair. Now the bad (less good) news. People who were expecting miracles with AMD's 65nm process will be disappointed. First, there will be no clock speed increases initially. While Altair will be available in FX 4x4 configurations and mainstream X4 AM2+ configurations, they will be topping out around 2.9GHz (thanks to the new half multiplier in Rev G). What's worse, when AMD said quad cores within the same thermal profiles as current chips they meant it. Altair TDPs will be 125W in both FXs and high-end X4s. No doubt lower clocked X4s will have a lower 89W TDP, etc., but that's not quite the same thing. L3 cache will remain at the previously announced 2MB so the large cache versions are still targetted for 2008. (Z-RAM is not likely on the horizon at all).
Now for those of you who want to jump on 4x4 when it's released in a month, it may not be a good idea. Why? Like AM2, current Socket F based 4x4s will be replaced by Socket F+ 4x4s in a little more than 6 months. Now people who already have an AM2 then K8L is a great upgrade option, but if you are buying a yet to be released high-end platform that's already outdated it may not be as inviting. While HT3.0 support isn't important in a 1P environment of AM2+, it'll definitely be useful in the 2P 4x4. The whole point of going to HT3.0 afterall is to assist in cache-coherency. The current 4x4s lack of support for K8Ls individual core power planes will also be a killer since those 2z125W TDP numbers will end up being a lot higher without power saving features.
On a side note, I've noticed that the Kentsfield samples that have been floating around at IDF are stepping 7 up from the stepping 5 B1s that we've seen lately. Stepping 7 seems to be the next to final stepping. I just wonder if 7 implies it'll be B3 or C0. It may be C0 with launch at C1 especially if what The Inquirer said about Intel correcting some B2 Conroe errata in a new stepping is true. I doubt there's major performance increases in stepping 7, probably just steps refine power consumption and in Cloverton to enable 1333MHz FSBs.