jackokring wrote:Eventual additions to the memory chip on top of the broadcomm chip.
The interface needs to be made so as chip 2 can be matured too.
A few rough paper calculations, imply that with one 32 bit float multiplier per 4KB of memory, would achieve about 40+ Gflops. If it's a smaller 16 bit multiplier with an assembly pipe to make 32 bit results, then 10+ Gflops of extra compute could be provided instead of doubling the memory.
Another reason for the memory chip as a reason for the location is the high volume of memory chip production, and design regularity.
I'm missing something here.... The registers are in the CPU or GPU. The memory is allocated to either the CPU or the GPU.
The PCB design, functionally, dictates that any processor chip (GPU+CPU) must have the same pinouts and very, very similar voltage requirement as the BCM2835, and not too much difference in current requirements.
The memory POP module has to have pinouts compatible with the processor package.
From what has been stated in various threads, this means that, at present, we are limited to the BCM2835. Whether or not there are higher clocked versions hasn't been determined (I can't tell from Broadcom's website...I simply haven't found enough specs to tell). It has been stated that a 512MB memory package exists--but it's too expensive to use and keep the Pi's price point. That is something that has the potential to change...the module price may come down, the cost margins may improve (due to the volume of demand), or--least likely of all--a higher priced Model "C" might be introduced.
Since this thread is concerned with MINOR improvements to make a "Model C", I could see three, or possibly four changes that could be realistically considered (in no particular order):
1. Faster model of BCM2835 (if it exists).
2. The 512MB memory module
3. A better SD card connector.
4. Some sort of upgrade to the polyfuses to handle MODEST additional current to both the system generally and the USB ports in particular (for example, raising the allowed current to each USB port to 200mA from the present 140mA, and total system current to a max of 1A).
Bear in mind that there are--apparently--issues with the polyfuses at present, and those issues are being looked at.
I believe that all four of those upgrades is consistent with the goals for the Pi, in that they would make the Pi more robust (the SD connector), easier to program (more memory and faster processor), or less finicky to work with (the SD connector and the polyfuse/power revision).
The only reason I wouldn't just suggest those items for a revised Model "B" is that they would drive the cost too high to hold the current price point, since the price point is a deliberate choice on the part of the Foundation.
On the other hand, if one were given at least a semi-free hand, after going for the above changes, I would start with replacing the voltage regulators with switching regulators, drop the composite video out connector in favor of an audio in connector, and add some form of system bus connector that would provide the ability to attach an expansion card that could carry all the other connectors people keep asking for (say, more USB ports, VGA, SATA, and so on).