Slashdot Mirror
Intel Begins Shipping 64-bit Prescotts
Rucas writes "With a minimum of fanfare, Intel has begun shipping a version of the Pentium 4 with 64-bit instruction set extensions. The news came to light not via an Intel press release, but rather through the spec sheet for a new server from IBM. In the midst of the new IBM eServer xSeries servers based on the recently released 64-bit Xeon is a blade server powered by the 64-bit Prescott. This marks the first product appearance of the new CPU."
Because the Opteron has an on-die memory controller. That can boost things up to 20% in some cases. It also makes designing motherboards easier because you don't need both a north and southbridge. It makes it harder to upgrade to a new memory technology, but it can be disabled allowing you to do that (I think). If they switched to that buffered "FB-RAM" or whatever (there was an article on the idea a while back on a big hardware site) that would fix that.
But anyway, Intel is stuck in a hard place. Because of the memory controller, their chips perform slower because of the extra latency, so they must ratchet up clockspeeds. The solution? An on-die memory controller. So why don't they do it? They CAN'T.
Intel has been pushing BTX for a variety of reasons (although most people blame Prescott's heat for it). But the way BTX is designed Opteron boards can't be made into a BTX form-factor because the memory is too far away from the CPU (there is too much electrical noise, IIRC). This means that Intel can't switch to an on-die controller without either changing BTX (what I think will happen because of AMD), or finding a way around the noise problem (little faraday cages?).
If you add in things like that the Intel chip only supports 36-bit address (I believe) while the Opteron handles 64-bit addresses (the actual bus is smaller right now, but that could easily be changed) and other performance factors (the top P4EE is outperformed in Doom 3 by a chip that costs more than $800 less, see the Inquirer) and Intel is in hot water.
All of this should be interesting to see what happens. Intel seems to be in trouble (performance wise, at least in the short term).
Comment forecast: Bits of genius surrounded by a sea of mediocrity.
Intel has been, in reality, behind AMD for at least two years. Now it just gets confirmed.
Intel has been publishing some phenemonal research on new processor architectures recently. For example, "Continual Flow Pipelines" appearing in ASPLOS of this year shows some awesome potential. It is a novel new technique for a superscaler out-of-order processor that does not use things like reorder buffers which don't scale well with instruction window size. Surely Intel has patented this technique before publishing in an academic conference.
Intel will catch up rather quickly.
I wonder why the corporate take-up of AMD has been so slow?
As with microsoft, a lot of it has to do with politics, arm-twisting and inertia.
Also, people like to pay more to get the same (or inferior) thing because, of course, in the corporate mind paying more = better product.
Treehugger? Treehugger... Treehugger!