Certainly not
adzoox said:
The SonyPS3 and XBox are using the same processor (said loosely) with custom tweaks - both processors are derivitives of the PowerPC and both are RISC - like the Apple/IBM G5 ...
With all due respect, the Plastation 3 and XBox are not even remotely using the same processor. The most important characteristic of the Cell that makes it unsuited (at least at present) for a PC implementation is that it is an
in-order architecture. Current x86/PowerPC architectures have complex out-of-order hardware solutions that allow them to execute some operations while putting off other operations that require some input that is not yet ready. For instance, consider the following pseudocode:
X = 100
Y = X * 100
X = Y
Z = 50
The command to set z equal to 50 requires no input from the rest of the program. In current PC architectures, the CPU would take this code apart and rearrange it, so that setting Z equal to 50 could be executed independent of the X and Y manipulations. This may seem like a rather minute point, but imagine for a moment that the Y = X * 100 step instead a rather large, complicated computing task that will take a few moments to complete. Now Z is left waiting for Y to complete before it can complete, any free integer execution units go wasted.
Because of its in-line nature, the Cell architecture is actually far simpler in many regards than the current G5 or x86-64 architectures. The Cell has far shorter pipelines than your typical processor (as it does no reordering) and lack a significant around of hardware complexity that is generally needed for said reordering.
It is important to realize that these differences are not advantageous and were simply chosen to reduce size/complexity and increase yields. Even today they are having yield problems, as indicated by Sony's announcement that the Playstation 3's Cell's would contain one redundant SPE (read faulty). The ONLY similarly between the XBOX360's processor design and that of cell is the PPE (PowerPC Processing Element) is clearly based on the PowerPC (and is coincidentally more or less IBM's only significant contribution to Cell).
However, even Cell's PPE is highly modified. It is only a dual-issue core and, again, in-order. I suppose I better get to the point. Cell generally trades hardware complexity for software complexity. Cell has a tremendous amount of power at it's disposal, however, tapping into that power on a general-purpose desktop solution would be next to impossible. A purpose-built compiler could go a long way in compensating for the PPE's in-order nature (by reordering instructions in the code rather than in hardware); however, the SPE's would be next to impossible to make sufficient use of. They include no cache and have 256K of local memory. Unlike cache, this memory does not operate on it's own, which means the software must address the memory of an SPE it wants to utilize in order to send it it's instructions. So efficient use of a Cell processor would require software to be written to reorder instructions and keep track of the memory for 9 separate cores. This is a slightly ridiculous proposal for a flexible desktop solution, and one that the console market is largely immune to, as all console games are highly architecture specific, purpose built applications (though I still have my doubts as to the degree to which any software engineer will be willing to utilize Cell's power).
If you consider that many of the Apple's platform's popular apps are cross platform (Opera, AdiumX (libgaim), Skype, Limewire, Office, Adobe Creative Suite, and now even Quicktime and iTunes), it looks even more unlikely that anyone would be willing to optimize their software for an Apple Cell platform.
Now for the question of Intel over AMD. If Apple is seriously considering a switch to x86, I am not at all surprised to hear Intel. Without getting into why Intel makes sense from a corporate perspective, the Intel platform seems far more complimentary to Apple's. Intel's CPUs have consistently bested AMD's offerings over the past several years in multimedia tasks such as encoding/decoding video and audio. This is because despite AMD's hyper-efficient memory-controller-in-core design, there's simply no substitute for raw clock cycles when compressing or manipulating data. Indeed, Pentium D's would likely offer superior performance in Apple's self-defined market, the digital hub. iMovie, iDVD, iPhoto, iTunes, Quicktime and even most of Adobe's Creative Suite would more than likely perform better on an Intel solution. AMD's architecture, on the other hand, is far better suited for memory-bandwidth intensive tasks such as gaming, a niche where Apple is nearly nonexistent.
Moreover, AMD's price-advantage has purposely dried up. In fact, Athlon X2's look to be substantially more expensive than Pentium D's. Finally, Apple's mobile solutions represent a larger and larger portion of it's installed base. Intel's Centrino is the hands-down winner in this arena. AMD's Turion solution is a little half-baked. There are many characteristics of the Athlon 64 design which makes it better suited for a mobile platform than a desktop Pentium4. Turion exploits that fact, and tries to get away by simply sliding AMD's desktop processor into a notebook. I don't maintain that the Turion platform is bad. In many ways it is the near equal to Centrino, but in many ways it is not. This post is already too long for me to expound upon them.
I made this post for two reasons. First, please stop speculating about Cell's in Apples unless you have some technical solutions to Cell's shortcomings to offer. Cell was not designed as a desktop processor, and just because it has one small core with similar origins to Apple's current G5's says absolutely nothing about it's suitability for a desktop solution. Second, stop the hysteria over Apple-Intel rumors. If Apple was planning an x86-64 migration, I would fully expect it to be to Intel rather than AMD. And if Apple did migrate, it would be one of the best moves they have ever made. I can explain that further if anyone is interested.
-Hank Reardon
