Microsoft's Xbox 360, Sony's PS3 - A Hardware Discussion

Final Words

Game consoles have always been different, architecturally.  The PlayStation 2 was very different from the original Xbox, and thus it is no surprise to see that the two platforms continue to be quite different this time around. 

Given what we’ve discussed thus far, there are a number of conclusions we can draw:

The most important thing to keep in mind is that the revolution in physics engines and collision detection isn’t going to happen over night.  The first games for both consoles will, for all intents and purposes, be single threaded titles.  More adventurous developers may even split up execution into two concurrent threads, but for the most part don’t expect to see a dramatic change in the quality and reality of the physics simulation of the first titles, especially when compared to titles like Doom 3 and Half Life 2. 

However, a change is coming and by the end of next year multi-threaded game engines should be commonplace on both consoles and PCs, which will hopefully lead to much more entertaining experiences.  The approach to that change will be different according to the platform; without a doubt, developers will have their work cut out for them.  

The transition to multi-threaded development alone will increase development time 2 or 3 fold.  Not to mention that the approach to architecting game engines will differ whether you are porting to the Xbox 360 or the PlayStation 3.  The Xbox 360 is clearly going to be the easier of the two to develop for once a game engine is multi-threaded, just because of the general purpose nature of its hardware.  That being said, it won’t be impossible to get the same level of performance out of the PS3, it will just take more work.  In fact, specialized hardware can be significantly faster than general purpose hardware at certain tasks, giving the PS3 the potential to outperform the Xbox 360 in CPU tasks.  It has yet to be seen how much work is required to truly exploit that potential however, and it will definitely be a while before we can truly answer that question. 

Cell’s on-die memory controller is a blessing for game performance; it most definitely will keep the PPE fed far better than the Xbox 360’s external memory controller.  Even the cache size advantage of the 360 won’t be able to offset the reduction in memory latency thanks to an on-die memory controller. 

The on-die memory controller is not all an advantage however, a big part of its inclusion is out of necessity.  Remembering back to our discussion about the SPEs as being in-order with no cache, threads run on these processors only have access to 256KB of local memory, which is reasonable for a cache, but not much in the way of memory.  So these SPEs will depend on having low latency access to memory in order to keep their pipeline filled and actually contribute any useful performance to the system.

At the end of day 1, when running mostly single threaded code, the performance difference from a CPU standpoint between the Xbox 360’s Xenon and the PS3’s Cell processor is basically a wash.  The 360 has more cache, while the Cell has a lower latency path to main memory.  In the end, the first generation or two of games will mainly be a GPU battle between the two consoles, and both will offer significant improvements over what we have with current consoles. 

Graphics-wise the 360’s Xenos GPU and the PS3’s RSX are fairly different in implementation, but may end up being very similar in performance.  Treating Xenos as a 24-pipe R420, it could be quite competitive with a 24-pipe RSX despite a lower clock speed.  The unified shader architecture of the Xenos GPU will offer an advantage in the majority of games today where we aren’t very geometry limited.  The free 4X AA support offered by Xenos is also extremely useful in a console, especially when hooked up to a large TV.

If the PS3’s RSX isn’t much more than a higher clocked G70 then at least we have a good idea of its performance.  NVIDIA has mentioned that by the time the RSX launches we will have a faster GPU on the PC, which leads us to believe that the performance advantages of the RSX are mostly clock speed related.  At 550MHz, the RSX GPU should have no problems handling both 720p and 1080p resolutions, although the latter won’t be possible in all games, mainly those that are more texture bandwidth bound.  We do think it was a mistake for Microsoft not to support 1080p, even if only supported by a handful of games/developers.  At the same time, by not imposing strict AA implementation regulations like Microsoft, Sony does open themselves up to having some PS3 games plagued by jaggies despite the power of the console.  Given the amount of power in both of these consoles, we truly hope that their introduction will mark the end of aliasing in console games, but some how we have a feeling it won’t.  Aliasing has plagued console games for too long for it to just disappear, that has to be too good to be true. 

With at least 5 months before the official release of Microsoft’s Xbox 360, and a number of still unanswered questions about the PlayStation 3, there is surely much more to discuss in the future.  The true nature of NVIDIA’s RSX GPU, the real world programming model for Cell, even final hardware details for both consoles has yet to be fully confirmed.  As we come across more information we will analyze and dissect it, but until then we hope you’ve gained more of an understanding of these consoles through this article.