Carl,
One of the things I should note is that RDRAM is generally available in ECC form, while SDRAM is generally not. This is a place that the memory industry has visited before, alpha radiation used to cause bit errors in all DRAM, and engineers included ECC. (The problem was solved by switching materials in the package, if my memory serves correctly.) ECC is expensive, it increases memory costs by something like 10%. That is more than the profit margins of a lot of companies, to a design engineer, 10% is a hell of a lot. Engineers only include ECC for very good reasons, and the only good reason is that the memory technology needs it. SDRAM doesn't need it, RDRAM does.
This essentially proves that RDRAM (as used in PCs) drops bits.
Do you really expect us to believe that because "RDRAM is generally available in ECC form" this "essentially proves" that it is inherently less reliable than SDRAM?
In one of your posts last year you pointed out that many single bit errors will go unnoticed and do no harm. However, such errors may insidiously modify data or programs if non parity/ecc memory is used. Most people blame system crashes on flakey software, but that is not the only explanation. This is why I think that ALL memory should be ecc protected in systems that matter to a business. How do you know that you are not getting occasional single-bit errors right now from your SDRAM?
It is certainly true that improved packaging has improved reliability enormously, but there is nothing that we can do to avoid cosmic radiation.
John |
