SI
SI
discoversearch

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.
Politics : Formerly About Advanced Micro Devices -- Ignore unavailable to you. Want to Upgrade?


To: Bill Jackson who wrote (88060)1/19/2000 2:19:00 PM
From: Petz  Read Replies (1) | Respond to of 1572269
 
Bill, re:On die L2 performance increase
<I suspect Scumbria will add to the gains as there may be a latency gain due to the speed and proximity>

I've already factored in the latency gain due to higher clock rate (speed), i.e., I've assumed that latency in nanoseconds is 2 1/2 times faster for on-die full speed cache vs. off-die 2/5 speed cache.

Just to summarize the analysis, it showed that an Athlon 800 with full-speed on-die cache would have performance equal to a 900 MHz CuMine assuming
1. The on-die cache of the Athlon was 512K
2. The latency of the cache improved proportionate to the clock speed of the cache. Actual latency of on-die L2 may be either better or worse than this. (Anyone remember where that article on various CPUs' L2 cache latencies and bandwidth is?)
3. Both systems using PC100 memory. If the Intel system uses high speed RDRAM, the 800 MHz Athlon would match an 866 MHz Coppermine.
4. The basis of the comparison is the CPUMark99 benchmark, which currently favors the Intel Coppermine by a few percent at equivalent clock speeds.
5. The extrapolation is based on only three benchmark tests of an Athlon with various cache speeds, so there might be some measurement error in the predicted 11.3% increase in CPUMark 99.
6. Benchmarks with heavy floating point usage or heavy disk IO will benefit less than 11.3%. However, the Athlon is already ahead of CuMine in these benchmarks.

Petz