Dear Elmer:
After the MTH fiascos, I do not believe any workstation user would risk a MRH even if, Intel had one. I have not seen any motherboard with one. Besides, the latency costs would double, even triple, if it was used. 1GB of PC133 Cas 2 (512 x 2) memory costs at most $1,100. 1GB of PC800 (256 x 4) costs at least $3,520. That is more than three times as much. Abit's KA7 can use 4 DIMMs at 512MB each. I think a 1GHz Athlon with 2GB of memory is still 40% cheaper and performs better than 1GHz Cmine. And for the SMP case of 2 Cmines, two of the above is still cheaper in a Beowulf configuration. They would be three times as fast due to the four times memory, double the i/o bandwidth and their inherent FPU horsepower.
Within three months, the 760MP comes with 2GB PC2100, and dual Tbirds at somewhere over 1.2Ghz each (possibly Mustangs at 1.46GHz) at a price still less than the SMP Cmine board. It may even have a third party optimizing compiler and triple the performance. Then, Intel is in a world of hurt. It will finally dawn that RMBS is only good for the area it was designed for, embedded applications where one or two chips in a chain makes sense, and Intel will finally give up on it for PCs. The later Intel realizes this the worse it will become. Of course, AMD partisans will cheer all the way to the bank until then.
Pete |
