Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Rambus (RMBS) - Eagle or Penguin -- Ignore unavailable to you. Want to Upgrade?

To: Mohamed Saba who wrote (40986)	4/26/2000 8:50:00 PM
From: mishedlo	Read Replies (1) \| Respond to of 93625

Take This Tom's Hardware csaresearch.com The Truth About i820/RAMBUS Performance Life has been rather unkind to Intel's i820 chipset designers. Charged with developing the first reference platform the the controversial new RAMBUS Direct RDRAM memory architecture, these poor souls walked right into a political buzz saw as industry analysts and PC enthusiasts alike took aim at the expensive new RAM standard. That there would be resistance to a costly shift towards a new and unproved memory architecture was never in doubt. What is surprising is the level of vitriol evidenced by the architecture's detractors. For these people, the RDRAM vs. SDRAM debate has become a kind of "jihad," an opportunity to take aim at the industry leader and unload with both barrels. The fact that the arguments against RDRAM are flawed - most detractors quote dubious benchmark numbers derived from outdated performance testing methodologies - doesn't seem to phase them. The blood is in the water. Let the frenzy begin! Those familiar with our work here at CSA know full well our disdain for traditional linear benchmarking tools. Most often executed against "bare iron" configurations (OS and drivers only), and featuring just a few applications, these antiquated "test suites" fail to exercise the full range of PC subsystems. They're one-dimensional solutions measuring a complex, multi-dimensional world. Given the above, it should come as no surprise that we responded with skepticism to the initial claims of poor i820-based system performance. After all, how can an testing body judge a new platform when they don't have a clue how to measure its benefits? Clearly, these early benchmark results were not telling the whole story. So in an effort to set the record straight on i820 performance, we decided to conduct some benchmark tests of our own, using a much more realistic "constant computing" testing methodology. We considered two basic premises going into the benchmarking process. First, according to our contacts at Intel Corporation, the i820 platform was designed specifically for "constant computing" environments where multiple, concurrent tasks actively compete for the CPU's time. It's a high-bandwidth architecture targeted at complex client loads like those detailed in our Windows NT/2000 Performance reports (see the CSA Reports Library for more information). As such, i820's benefits are only fully realized when supporting a substantial client-computing load. Second, i820 - and its derivatives, the i840 and related platforms - were designed with GHz+ performance levels in mind. The RAMBUS memory subsystem, in particular, shows a significant benefit only after the clock frequency has reached a certain level (typically 800MHz or higher), and even then it needs a substantial workload in order to have a positive impact. The good news is that the RAMBUS architecture is scalable - the higher the clock frequency, the greater the real-world benefit. Of course, this all sounds great on paper. The real question is: How will i820 hold up under a next generation "constant computing" load? To find the answer we conducted comprehensive testing across i820 and i440BX-based PC platforms, using elements of the "worst case" loading scenario from our recent report on 2GHz Client computing. To make the comparison fair, and to focus the test on chipset (as opposed to peripheral) performance, we used common components - Audio, Video, CPU, NIC - wherever possible. The results speak for themselves: After executing the OfficeBench 1.2 Multitasking script against our demanding "constant computing" loading scenario, the 800MHz i820 system came out the clear winner with a completion time of 96 seconds - 17% faster than the 800MHz i440BX system, which completed the same script in just over 115 seconds. When we repeated the tests at 600MHz, the gap narrowed to 10%, demonstrating that in fact i820 performs better at higher frequencies. Note: A more detailed account of the test above scenario, including specifics on hardware components used and the structure of the "constant computing" loading environment, will be published as part of a more comprehensive Special Report on i820 platform performance later this quarter. The Bottom Line: When deployed in a complex, "constant computing" client environment, and when equipped with a high-frequency CPU, i820 - and by extension, RAMBUS and the rest of Intel's intelligent hub-based chipset architecture - shines. All of which begs the question: If the new design is so good, why did so many mainstream testing organizations miss the boat when it came to measuring i820 performance? We have our own ideas (i.e. an outdated methodology), but as always we'd like to hear your take on the matter. Whether you agree or disagree, let me know by dropping me a line here at CSA. - Randall C. Kennedy

To: Mohamed Saba who wrote (40986)	4/26/2000 10:24:00 PM
From: Victor Lazlo	Read Replies (2) \| Respond to of 93625

Mohamed, there is no question that the AMD supporters who spend so much of their time on this thread feel very threatened by RMBS. Victor

To: Mohamed Saba who wrote (40986)	4/26/2000 11:06:00 PM
From: Paul Ma	Respond to of 93625

They have no credibility Ha ha, keep up the sophistry! Paul Ma