First is the name K8L, which we first outed here, followed by some specs for the new AMD chips here. Yay. In a cleansing spirit, much akin to chugging bleach, we will start with nomenclature. The first thing is that when we wrote the AMD hounds pieces, we did not know what 'Rev' the parts corresponded to. The code names and specs are right, but the Rev nomenclature may be off, use it at your own risk.
Next up is the term K8L, something that was not in the AMD literature, or anywhere else until we posted it. It has since crept into the lexicon, and has appeared in many AMD documents, several of which contradict each other. What we were referring to with the name was the doubled FP parts that kicked off the 'hound' series of code names.
So, up to that point, there was no K8L, and several high ups, engineers, and generalised personnel at AMD have asked me where that name came from. Short answer, Intel. I don't know what should worry AMD more, that Intel knew about the hounds before we did, or that AMD now has dozens of Powerpoints out there with a Santa Clara inspired code name all over them. Time to go on a document scrubbing session....
So, that brings us to the chips themselves. There was much ado about the K8L/quad core parts taping out a week or so ago, once again, go team. A week of intensive digging has lead me to believe this part is not the native quad core dual FP beast commonly referred to as K8L. That part looks like it was delayed to Q1/08. No yay team there.
This has good and bad sides to it. The good is that the QC parts look to be an easy target to hit for Q2 if they are based on 65nm shrink of F, be it called G or something else entirely. Since AMD's architecture is fairly modular, and the cores themselves are a known quantity, it should make validation and testing much quicker. I am confident that the Q2 date is doable, barring any major impromptu shotgun podiatry.
The down side is that AMD is not going to have a new core in 2007, or at least have it until the waning days of 2007 if everything goes swimmingly. This means that Intel will pretty much own the 1S 2C market, where the meat of the sales lie. At 2S 4C, Intel will have a slimmer lead, but should still have a lead. At 4S, AMD all the way.
Now, you notice that covers 2C chips, what about QC/4C? Intel has a clean two quarters to ramp up and hold the mindshare high ground, but the AMD response, even with the older cores, should be more than adequate. Come Q2, it is game on all over again.
It looks like Intel has pulled off an engineering coup and will have dual-die QC parts on a 1333MHz FSB after all, but it won't be enough. AMD, but virtue of its overarching architecture, won't be an elephant drinking through a soda straw when it comes to memory accesses.
I expect AMD to win at all socket counts on QC chips, by a little on 1S, more as the socket count climbs. It will be benchmark dependent of course, and both sides will put out very favourable numbers. But, overall, I think AMD will win the majority of the benches. It will still have an uphill climb, Intel will have quite the entrenched userbase by then, so the great white hope for the green team is how hard Intel can ramp.
That brings us to power. Once again, due to the AMD infrastructure, the higher the socket count, the better AMD can scale without incurring excessive power overhead. Caneland may be a fast platform, but how many watts is the chipset going to consume? You can also say with confidence that a Conroe at 2.6GHz on a 1066FSB takes about 65W since the next step up is in an 80W thermal envelope. AMD has a little more leeway, if you want more details on performance per watt, look at this excellent article on Lost Circuits.
The take-home message on power is Intel has a lead on 2C chips right now, but performance per watt is still quite a toss up. On 4C chips, AMD potentially has the advantage, Intel needs two complete chips and a bus speed boost, AMD needs two more cores. I think AMD will have an easier time meeting their goal of putting the 4C parts in the same thermal envelope as the current 2C parts. Intel has amazing process engineers, but again , they are a bit hamstrung by infrastructure.
Overall, there are a few things to think about, not the least of which is Intel named an AMD processor. For 2007, some of AMDs chips will be early, some a little late, but nothing much more than an extension of what is there now. Performance over the next year will be won by both sides depending on socket count, core count, and alignment with several universal forces most people don't think exist(*). AMD will have a little easier time on power, but don't count Intel out. Clear as mud yet? µ
(*) To misquote the grand philosopher Rodney, it all has to do with the earth's rotation and its position around the sun. If it is lined up so the solar crashion particles pass easily through a rack of PCs, AMD will win the benches handily, 90 degrees off, it is all Intel. That is science for you. Microsoft has not figured this one out, and are blue because of it.
Sign up for INQbot – a weekly roundup of the best from the INQ