Re: Reality Check
Originally posted by wumpus
Just a reality check:
-At 1.8ghz, which is the projected top speed of the initial 970, it performs like a 2.8ghz P4 on SPEC. Even if the wider memory bandwidth does some magic, we are looking at running at the equivilent of a non-HT 3.06ghz P4. Better hope they are cheap, Apple will need to use 2 in each machine to become competitive on performance. We will be up against 3.6ghz+ HT P4s running on 800mhz FSB, not to mention the Hammer brigade.
Actually, we will be up against Intel's Deerfield, which is Itanium 3 (Merced) for the desktop (much as 970 is Power4 for the desktop). On SPEC numbers I believe the 1.8MHz introductory speed of the 970 in theory will best the Itanium 2, or at least be right about the same; no word on the Itanium 3 or Deerfield. Note, of course, that Deerfield will have lower SPEC numbers than it's heavy-iron big brothers, just as the 970 is a bit weaker than the Power4 at that clock rate would/will be.
AMD's story is about the same: Opteron ("Sledgehammer") is set to best the 970's SPEC numbers by a decent amount, but the same may not be true for its "little cousin", the desktop-aimed Clawhammer chip.
In the 32-bit world, we will be placing ~2GHz G4+'s against 3.2GHz P5s (note that P5 is to be debuting later in the year at 3.2GHz, albeit more efficient per cycle than the P4).
While one may compare the SPEC numbers between 32-bit and 64-bit processors, there is a reason why servers pay for the lower-performing 64-bit processors, and why processors like Itanium cost way more than P4s and Xeons. 64-bit processors, obviously, have the definite advantage that they can run 64-bit code along with the 32-bit SPEC calculations. The key on the high end is that Apple capitalizes on this ability with 64-bit iApps et al that "sing" compared to their 32-bit cousins.
-ApplePI may have been the thing that helped to kill the 'G5' according to the reliable Architosh. It looks like a BIU to enable PPCs to run on HyperTransport, nForce-like mobos. If Apple did not need it, they would not have joined the consortium. It may enable Apple to rely more on partners like nVidia for mobos...Maybe Apple should have tried harder to get more out of Moto, even just real DDR support for the aged, inadequate G4.
Please stop this "fake DDR" nonsense. The PowerMac DDR is largely crippled by an undersized FSB to the CPU, but that does not make it "fake". Just not as effective as it should be.
Also note that RapidIO is Moto's entry into the "better FSB" world. Too little, too late, perhaps, but Apple is getting something from Moto too. Apple has choices, which as we all know is what it likes.
-Many firms in Apple's target markets will want to upgrade and will not want what Ars Technica rightly called the 'overpriced embarassment' that is the current 'Pro' lineup. They will not be excited about 1.4ghz duals with fake DDR. They will buy PCs and not go back.
Quite true. The only hope for the next batch of pro(sumers) on the fence is if Moto can get a better FSB technology out the door with their 7457 chips. I'm no sure what the FSB is in these, but there are rumored to be both FSB and cache improvements.
To be absolutely clear: the bottlenecks we can see right now are in the System Controler-to-CPU bus, and in the L2 cache (which just exacerbates the deficient bus problem). This is what keeps the latest DDR machines from knocking your socks off, and this is also what keeps dual PPC machines from really shining (because both CPUs
share that deficient bus). Given appropriate breakthroughs here, Apple's back in the game.
No FSB/cache breakthroughs, however, and fence sitters will be hoping off to Intel-land in droves.