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 : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?

To: Petz who wrote (48551)	7/21/2001 1:10:45 PM
From: Ali Chen	Respond to of 275872

John:" define the "platform efficiency" of a system" First, there is no doubt that the "efficiency" is benchmark-specific. So it is a matter of agreement which class of benchmarks to consider - MS-application-type like BAPCO, of computational-type like SPEC CPU2000, - results may be somewhat different. Second, it does not matter how "efficient" and "balanced" is a CPU as compared to bus bandwidth or latency etc.. What does matter for all practical purposes is the task completion time. All other benchmarking metrics are derivatives. Then, for a given task/benchmark, the "platform [in]efficiency" can be defined as amount of time wasted in various I/O processes like memory walks, disk transfers, accelerated video functions, etc. In other words, anything that consumes the test completion time but cannot be speed up by CPU no matter how fast it could be is a "platform waste". The platform waste can be found by extrapolating the trend of completion time into the limit of an infinitely-fast CPU. In this approach the "platform efficiency" appears to be frequency-independent in most cases, and may serve as an inherent platform characteristic rather then a measure of relative advances in photolithography. Your definition of efficiency, "the % increase in benchmark performance divided by the % increase in clock rate", is deficient. For example, let's consider a group of STREAM benchmarks. They are designed to characterize the memory susbsystem, and by design must be CPU speed independent. So the "% increase" is minimal for this benchmark. Yet say a platform B can complete the same task of huge memory block transfers in 1/3 time of a platform A. Apparently the platform B is more efficient than platform A. Similar considerations are valid for any other benchmark. Regards, - Ali