MPR analyst writes to Register

Memory report not relevant to PCs

Letter Microprocessor Report analyst Peter Glaskowsky sent us this email, which may be useful to readers (see Direct Rambus shows no gains over current memory technology). Glaskowsky writes: "This paper is not relevant to the PC market for the following reasons: 1) The simulated processor is not an x86 processor, and is not substantially like any x86 processor used today. 2) The simulated L2 cache uses a 128-byte line size, four times the size of the L2 caches on PCs. 3) The simulated L2 cache is 1M in size, 2-8 times larger than the L2 caches used in PCs. 4) The simulated processor bus is 128 bits wide, twice the width of PC processor buses. 5) The simulated conventional DRAM arrays are also 128 bits wide, twice the width of DRAM arrays in PCs today. 6) The simulated applications were not, and did not behave like, the applications and operating systems used on PCs. Factors 2-6 happen to work in favor of conventional DRAMs such as SDRAM and against DRAMs with more sophisticated architectures. (I don't know enough about factor 1 to decide which way it would skew the results, but it is very likely to have an effect one way or the other.) I can explain these factors in more detail if there is enough interest. This study will have no effect on Intel, Via, Compaq, or any other PC company capable of interpreting it correctly. The study should be ignored by participants in the PC market (or better yet, repeated with new conditions that match those that exist in PCs). A few disclaimers are in order here: I don't think there's anything wrong with the study itself. The authors defined and justified their conditions quite adequately. They never said the study was relevant to the PC market. I have no axe to grind with SDRAM, nor do I have any bias toward DRDRAM or other advanced DRAM architectures. I think it would benefit your readers to post this email (or a summary of it) on your site so they can get some perspective on your original story. Thanks in advance. Peter Glaskowsky Senior Analyst, MicroDesign Resources" Mike Magee replies: The interesting point here is Peter's point that it would be useful to perform a similar study for x86 processors. The authors of the report have emailed The Register and thanking us for showing an interest in the subject. Incidentally, we used to receive copies of the Microprocessor Report but have been dropped from the circulation list. Perhaps you'd like to add us again, Peter. Our address is on the front page. ®

Sponsored: Network DDoS protection