New Research shows that Intel Corp’s preferred next generation Rambus DRAM memory chips are 25% slower at running applications in a ‘real world’ environment than current PC100 synchronous DRAM chips. The paper, produced by Inquest Market Research, was collated from benchmarks produced by Dell Computer Corp at the Intel Developers Forum.
Dell ran Microsoft Office 2000 applications – Word, Powerpoint and Excel – on a Rambus system and a PC100 system. Using a benchmark called Office Bench; Dell’s engineers contrasted a BX chipset platform using 100MHz SDRAM against a Camino platform with 800MHz Direct Rambus. Other than the DRAM and chipset, both systems were configured identically with 500MHz processors. According to the research, performance figures indicate that under normal conditions Rambus slows the performance of Microsoft Office 2000 applications by an average of 25% when compared to PC100. In total, the Dell tests found PC100 to be faster than Rambus in five out of six cases. The PC100 performance advantage for these five tests was 34% on average. A separate research paper from Inquest – testing VIA Technologies Inc’s Apollo PC133 chipset – suggests that the gap would widen with the higher bus speed system.
So where does this leave Intel’s claims about the superior performance of the Rambus technology? The chips do offer a performance hike for graphics-intensive applications such as 3D games, which require a high bandwidth system. However, Intel wants Rambus to become the memory of choice for mainstream PCs, not fall into a gaming and workstation niche. Bert McComas of Inquest describes Intel’s own bandwidth benchmarks for Rambus as ‘broken’. The bandwidth test in particular produces results that defy reason and logic – exceeding the theoretical peak burst bandwidth of DRAM in a number of cases. The AGP test is not necessarily broken, but it doesn’t seem too enlightening or valuable. Why would Intel release it? Intel needs a tool to market excess bandwidth, or rather ‘Headroom’. When used on a few specific platforms, this test conveniently spits out results that support Intel’s claims about the effective throughput of PC100 versus Rambus, he said in a paper called Platform Tests v1.1.
The damaging results of the first independent benchmarks of Rambus chips calls into question Intel’s claims about the minimal performance benefits of using PC133 SDRAM chips and infrastructure. And shows why Dell’s Jay Bell was onstage to welcome Intel’s tentative acceptance of the PC133 standard at IDF. á
