Replies to post #4633 on Rambus (RMBS)

[calbiker]

granularity equals bus width times prefetch is a simple FACT.

LOL, is that coming from our resident mathematician? I would think a mathematician would break down this equality to the smallest common denominator.

What I don't like about this equation is that 'Bus Width' is a dependent variable.

For example, we start with a graphic card design and it has some given memory bandwidth requirement. After you choose what type of memory to use, the bus width falls into place. The bandwidth requirement is the independent variable while the bus width is the dependent variable.

So here's the math to break down elixe's equation to the smallest common denominator:

Granularity = Bus Width * prefetch

Where: Bus Width = bandwidth / (core clock * prefetch)

Substituting the 'Bus Width' quantity into the first equation we get:

Granularity = [bandwidth / (core clock * prefetch)] * prefetch
Granularity = bandwidth / core clock

LOL, would you call that a simpler FACT?

You may want to suggest to elixe to give it up. Ha.

[pauls59]

Hi Cor,

Also I find the whole discussion whether Intel would go for XDR with microthreading very theoretical. Now we speak not only of a memory type which is in low supply (XDR), but also of one which is still on the drawing board (XDR+microthreading).

I agree that it is theoretical, but for different reasons than you say. In low supply, but what of its competition, is there anything running at 3.2GHz or better that has a better supply? XDR + microthreading is on the drawing board, but I expect to see it soon because it will boost efficiency enough to provide up to 4 times the performance.

The requirements to implement microthreading means more threads in the core and a more complex control bus. I imagine that there is a market for it though. That market will likely be Game consoles, servers, and HDTVs. It won't take long for something with that kind of performance gain to be used. 3 years perhaps. That will boost production.

Multicore Microprocessors, each core with 2 pipelines and hyperthreading enabled, are being produced. There are presently 4 processes that can be working simultaneously. That number will grow. FSB Bandwidth requirements will grow. The trend is more processes happening simultaneously.

FSB speeds are increasing faster than memory bus speeds. Granularity is becoming an issue. XDR with microthreading is the best cure for that.

Look at the trends. The trends, more simultaneous processes, FSBs beginning to outpace DDR memory busses, and growing granularity, mean that we are on the verge of needing better solutions than DDR is coming up with. XDR with microthreading is a solution. Workstations using XDR with microthreading by 2008 or 2009 is what I predict.

The theoretical part is not whether there will be a need for it. The theoretical part is whether Rambus and its customers can bury the hatchet that they have with litigation and the desire of some manufacturers to keep Rambus out of the market.

It may very well be that we will see a roadmap from Intel in a couple of years showing XDR for workstations as early as 2008.

JMHO

Paul