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.

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

To: Amy J who wrote (53264)	3/23/1999 8:13:00 AM
From: Scumbria	Respond to of 1583403

Amy, I find Winstone to be a very informative benchmark. It has been observed for some time now that larger MHz increases are producing smaller Winstone score improvements. intel.co.il Winstone* 98 Business Performance Comparison Pentium II processor MHz 233 266 300 333 350 400 450 Winstone 98 Business Scores 20.3 22.1 23.4 24.6 25.6 27.1 28.6 Going from 233 to 266 Mhz produced a performance gain of 0.055 Winstones/MHz, but going from 400 MHz to 450 MHz only showed a gain of 0.03 Winstones/MHz. It becomes clear that the MHz vs. system performance curve is on an asymptote. The problem is mostly tied to DRAM latency. At 1GHz, an L1/L2 miss has a latency of well over 100 CPU clocks. It is of no use having 20 execution units on a chip, if they are starved for data. Intel's approach to fixing the problem has been: 1. Larger L2 caches 2. DRDRAM #1 is a very good idea, but as cache sizes increase it is producing diminishing returns. #2 will be a disappointment because DRDRAM latencies are no better than SDRAM latencies. CPU architects tend to have this mentality that they can fix the problem by throwing more execution units at it, but it has been proven repeatedly that it doesn't work. The PPC 604 was a great example of this phenomenon. Scumbria

To: Amy J who wrote (53264)	3/24/1999 2:33:00 AM
From: Scumbria	Read Replies (3) \| Respond to of 1583403

Amy, I calculated some numbers for PII price/performance. Prices are the lowest price at: pricewatch.com Performance numbers are from the Intel website. MHz 333 350 400 450 Winstone 98 24.6 25.6 27.1 28.6 Cost $130 $147 $243 $425 Winstones/MHz 0.074 0.073 0.068 0.064 $/Winstone 5.28 5.74 8.97 14.86 $/MHz 0.390 0.420 0.608 0.944 Note that the $/Winstone of a 450 MHz PII is almost 300% that of a 333 MHz, yet the performance gain is less than 15%. Scumbria