epsilon84 :
Exactly - a system is only as good as it's weakest point.
How exactly do you propose AMD 'fix' Phenom? I believe it is a doomed architecture right from the start. No matter what AMD does, it will always be behind Intel in terms of IPC and scalability. What will AMD achieve in Q2? A bug free Phenom 9900 at a 'whopping' 2.6GHz? Benchmarks have proven that a 9900 is STILL slower than a Q6600, so despite all AMD's efforts at 'fixing' Phenom, their flagship product by Q2 08 will still lag behind an 18 month old architecture.
I agree. I think AMD lost the desktop space big time. Unless AMD engineers can pull something out of the hat, I seriously won't consider Phenom to be a competitive processor against Core 2.
OK, some think there is hope in 45nm, but what are we expecting here, miracles? Long gone are the days when die shrinks meant dramatically improved performance, just look at AMD's latest 65nm shrink for an example. Even Intel's Penryn, merely achieved an overall 5% IPC improvement (SSE4 notwithstanding), and half of that is probably down to the larger cache.
The point of moving to 45nm is to reduce power consumption, at the same time increasing clockspeed. Unless the 6Mb L3 cache can provide any help, you won't see any increase in IPC.
Then the next question becomes, how will AMD's 45nm process benefit K10? From the scarce data we currently know, it seems like AMD
won't incorporate HK/MG in their 45nm process. AMD only said they could, but didn't make any promises. In addition to that, IBM won't implement HK/MG until 32nm, so it is likely AMD won't get any 45nm HK/MG from IBM. AMD doesn't have the resources, or the technical background to start everything from scratch. Currently they're still struggling on their 65nm process.
A best case scenario for AMD is if their 45nm parts match current Kentsfield in IPC, that in itself would take a 10% IPC improvement which is unheard of in modern times from a die shrink. But let's give AMD the benefit of the doubt, and assume they pull it off. So 45nm K10 will match a Kentsfield by Q4 08... umm, can anyone say 12 months late to the party? By then Intel will have fully ramped their own 45nm Penryns and will be on the verge of launching Nehalem, if it's hasn't in fact launched already.
Exactly. Higher clocked K10 has been delayed to Q2, and judging from AMD's past record, Q2 means end of summer. This will probably hit Intel's Nehalem demo. What if Nehalem is about 50% better than Core 2, plus the bonus of high scalability? What will we, as enthusiasts, or corporates do, if we know there is a killer product just about 2 quarters away? Will we spend the bucks and buy AMD's K10, or wait for Nehalem?
50% better is only my conservative speculation. If Inq can be trusted, they claimed about 100% faster clock-for-clock against the current 65nm Core 2. How is AMD going to counter that? Although I hate to say this, but it seems like AMD's fate has been determined.
K10/Phenom as a direct competitor to Intel's quads is as good as dead. It may end up being the bargain bin alternative (and it's heading that way fast) but nothing more. All their efforts at 'fixing' Phenom will come to nothing, because the competition is also advancing, and at a faster rate too. Phenom will spend it's entire lifespan being nothing more than a budget contender, totally at the mercy of Intel's pricing department. If Intel ever decides to launch a sub $200 Penryn quad... I'll leave you to ponder the consequences of such a move.
It is pretty ironic, that what AMD touted in the first place (40%?), became the budget solution for the market. In addition to the high manufacturing cost of K10 native quad, AMD also needs to pay off K10's research cost. In all honesty, I really don't know how they can do it.
But, maybe its just me being pessimistic.
Facebook
Twitter
Instagram