KZ, Intel is likely going to post benchmarks that truly crunch memory bandwidth, but they might not be the same benchmarks that the hardware and gaming web sites use. Part of the problem is that some of those popular benchmarks really don't exercise the memory bandwidth all that much. For example, Business Winstone 99 is very light on memory accesses. (In BW99, the processor bus is occupied less than 10% of the time.) Quake II crunches memory bandwidth more (especially the Crusher benchmark), but not enough to exceed the limits of 440BX and PC100 SDRAM.
Intel is trying to predict the future based on historical trends. When 440BX and PC100 SDRAM was released, very few applications really needed the added bandwidth. But over time, as new applications, games, and graphics controllers came to the market, the 440BX was able to handle the increasing memory bandwidth needs. Based on that, we could see the same thing happen with Camino and RDRAM. At first, Camino and RDRAM may be deemed useless. In fact, I wouldn't be surprised if a few benchmarks here and there made Camino look slower than 440BX. But over time, newer applications and benchmarks will definitely show that Camino and RDRAM hold a marked advantage over 440BX and PC100 SDRAM.
The only difference this time is the high profile of Rambus and RDRAM. We never saw this much attention focused on 440BX and PC100 SDRAM. Why? Thanks to the marketing of Rambus, combined with the "anti-marketing" of the Anything But Rambus coalition, people's views have become incredibly distorted and short-sighted. Look how quickly people jumped to conclusions once those pre-release RDRAM benchmarks were released by outside web sites like Tom's Hardware Guide or Apex. All of a sudden, everyone considers themselves an expert, when in fact, they really know squat.
As for EETimes, Bert McComas, and the others who will supposedly publish the real truth behind the benchmarks, I'm almost positive they won't be telling Intel's side of the story. They'll most likely harp on how RDRAM doesn't provide the improvement for applications today, how DDR SDRAM provides the same improvement at a lower cost, and so on. But they won't mention anything about Intel's point-of-view, which is that RDRAM will likely be much better going forward into 2000 and beyond than the alternatives like DDR SDRAM. And when you have to plan projects that won't hit production until two or three years down the road, and when you consider that a lot can change during that time, you'll understand why Intel is more concerned about tomorrow than today.
Tenchusatsu |
