Rambus, in case you didn’t know, are more or less the creators of all the major releases in PC memory (like “all your RAM are belong to BUS” style). They are working to up the number of bits passed per clock cycle to 32, which is up from the current 2 bits per clock cycle.
That alone would allow up to 16 gigabits per second to be transferred on a chip running at 500MHz, that’s around 16 times that of DDR2. now you throw a bunch of those chips together, up the clock speed to something more modern, and you are hitting that terabyte per second number they are aiming for.
That type of performance would smoke anything, even Rambus’ own 4.8GHz XDR memory. They say that even high-end graphics cards don’t hit higher than 128 gigabytes per second of memory usage. No word when we can expect to see the technology, but it looks like they are working to keep Moore’s Law a law, as in not being disproved.
Rambus prepping 1TB/sec computer memory [via macnn]