Originally posted by BartoThat's correct, and also COMPLETELY irrelevent to Apple's various G4 architectures.
Believe me, that's totally obvious reading your post
No CPU other than those with onboard memory controllers (eg POWERs, Athlon64s) ever sends and recieves directly to and from memory. In between the CPU and the memory is a northbridge (or system controller, or system intergrated circuit). The CPU makes requests to the northbridge for data from the memory, and the northbridge fetches the data from memory and sends it to the CPU.
It IS true that less data can be sent to the G4 alone than can be fetched from memory, but it is NOT half. System busses aren't that simple, you can't just look at the MHz and make a judgment. The G4's bus is very efficient for it's speed, and DDR is highly inefficient.
Now guess what. The Athlon is the same. The majority of Athlon computers in the world have a CPU bus slower than the memory bus. Guess what, PC user's don't call them "hacks" or "false". Do you know why? Because there's nothing wrong with asyncronous busses.
But wait, there's more.
In modern architectures, more and more devices have DMA. That's Direct Memory Access, the ability to ask for data from the memory, without going through the CPU. On the PowerBook, DMA devices include gigabit ethernet, the hard drive and most importantly the graphics chip. Using Quartz Extreme, the Radeon 9600 in the PowerBook G4 transfers buttloads of data from meory. So it's not JUST the CPU that consumes the memory bus, it's the networking, drives and graphics chip too.
Thanks for the post. A lot of consumers don't realize the RAM subsystem services more than the processor.
As far as FSB goes, while it would be nice to have a faster FSB, you have to ask yourself why? What will it really do for me? On the 7455 Processors, for almost all Office Productivity tasks, the FSB speed is invisible. The instructions and size of the data being worked on is within the performance envelop of the FSB. The question is where does a higher FSB make sense. Well Video editing, complex graphics work, SIMD operations on huge data sets. You get the point. For most of us what is there is more than enough, though it would be nice to start at 400MHz FSB to future proof for Digital Lifestyle apps.
But, one of the biggest problems with most computers is that they don't use what they got now. Software isn't developed very efficiently either. With simple techniques where Load operations are executed in parallel with Integer, FPU and SIMD instructions, a significant performance increase can be appreciated. Too, Apple continuing to optimize the GCC compiler, we are seeing significant performance increases with memory transfers as borne out by xBench with Panther compared to Jaguar.
Using Quartz Extreme many of the Aqua operations no longer use the FSB for display operations, but do DMA from the GPU. So a lot of the OS work that can clog up the FSB has been offloaded to the GPU subsystem, like shading, moving, resizing, window drawing, transparencies, etc. Panther uses even more of the GPU for many of its operations than Jaguar.
So what you have with Apples design is a better balanced system, were all the components are contributing to consistently good performance. The FSB will not be as bottleneck if less cars are driving down its road.
There are other advantages to the Apple implementation of the FSB. It is a fully duplexed FSB, which means that a Load and a Store operation can be performed simultaneously. So 167 MHz in either direction which would consume the entire DDR bandwidth.