Continuing my debate with
apoppin from the GT200 thread, concerning the failure of R600, Fusion and the AMD-ATi acquisition...
apoppin said:
. . . and i would say "what other reason" would ATi go with hat in hand and ask to be swallowed up by AMD if they did NOT share their Fusion Vision? - unless you believe, as some of us do [i did] that Fusion was a "ploy" to unbalance nVidia while they scrambled to fix Phenom and r700. But i don't think so anymore.
Do we know for a fact that ATi went to plead to AMD to be acquired? I'd say it was AMD who initiated the merger talks, it would make more sense. If they *did* share AMD's Fusion vision, they could work together on it while remaining two separate companies. But perhaps ATi didn't want to do Fusion, and had to be bought to change its mind. nVidia was too big to swallow, AMD probably didn't trust in VIA & S3's capability to deliver good chipsets and GPUs (although, for the low-power market where AMD's aiming all the time, buying VIA and S3 could have been a better choice), so the only one left was ATi...
apoppin said:
i believe AMD using ATi's know-how, has managed to get a prototype of fusion and they have diverted their engineers away from lesser projects [note the conservative specs of r700]
If you think about it... Fusion is not anything revolutionary. The GPU just moves from the northbridge to the processor and everything else stays the same. Just like moving the memory controller to the CPU, except that is much more logical. And what do you need to create a Fusion processor? Just put a PCIe controller on the CPU and an RV620 on the other end of the lanes. HyperMemory will do the rest.
As for the conservative specs of R700, I think there's a different reason for that. AMD is broke and needs to trim costs, so they focus on the most profitable segments and dual-GPU solutions, because that way they only need to design one chip instead of two.
apoppin said:
and are content to attack the midrange and IG while they prepare their real weapon against intel and nVidia [who is also eyeing Via for a SiS division to make their OWN CPU-GPU] while Intel struggles to take everyone on at once. My analysis is that intel has the most to lose and will do so without clear vision - Larrabee is insufficient for at least 5 years, imo
It's not about having a Fusion-like CPU, it's about having a complete platform. nVidia's running scared because AMD has its own chipsets and GPU's and Intel is working on Larrabee and has its own chipsets as well. Of course, it you put a GPU into the CPU, then you can be absolutely sure that nVidia won't sell its IGP chipset together with your CPU... but then again, not all CPU's will be Fusions.
Intel is doing Larrabee because they're not short-sighted. Just look at R580, G80... those are essentially the many-core chips that Intel sees as its tomorrow, except they're here *now*. Intel saw that getting to its goal will be bumpy if they try to upgrade their CPU, but easier if they start with a GPU. Larrabee is in fact primarily aimed as a GPGPU solution, but if Intel tried to market it as a specialized co-processor, it would cost too much. So they'll make it render pictures as well, make profit with it as a GPU, and then offer it as a specialized co-processor for a ridiculously low price.
My fingers grow weary, so let's leave the R600 for a later time, shall we?
Also, my impression remains that the 8/12-core SKUs will use a new 256-bit socket, but we shall see. If that isn't even the case, then get ready for yet another debacle of epic proportions.