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: fyodor_ who wrote (74934)	3/19/2002 8:59:37 AM
From: Dan3	Read Replies (1) \| Respond to of 275872

Re: Because they were conducted with overclocked FSBs and, as the few datasets I pulled from AnandTech show, 3 datapoints are not a lot to guess a trend from, considering the error margins on the benchmarks (or benchmarkers) in question. Hmmmm. You cheerfully posted some other results with 3 or fewer datapoints. (but I do think that it was only because you felt it added a few otherwise missing, but key, datapoints. The thing is, you compared a linear scale P4, with a non-linear scale XP's 1.5X-500. Then you induced a more linear curve from P4's data and a less-linear curve from XP's data. This is not a revelation. The fact that you used this analysis to refute an assertion that Athlon's slower, but much more sophisticated cache would let it provide higher performance on a given memory technology wasn't particularly relevant either way. Then you responded that system performance was all that mattered so, since Intel offered dual channel chipsets and AMD didn't, AMD's better performance on a given memory technology didn't matter. I see your point there, but would respond that memory technology has always changed much more slowly than CPU speed, and that it's usually much easier to just adopt a current, known, technology than create a new one. I see Athlon's ability to extract more performance from a given memory technology as significant, because I think Intel may be stuck between a rock and a hard place due to their basic CPU architecture. Their very long pipeline fills up with bubbles unles they use a very fast cache, but very fast caches must be kept simple and fairly small. AMD's shorter pipeline can handle bubbles with less performance loss, so AMD's architecture can afford a more sophisticated, if somewhat slower, cache design. A more sophisticated cache lets AMD extract more performance from a given memory technology, which could mean that P4's core architecture cannot ever compete with Athlon's (and Hammer is supposed to follow the basic design architecture of Athlon). If chip speeds continue to increase faster than main memory speeds (which has been just about as consistent as Moore's law), memory perfomrance will become the major performance bottleneck, and Athlon/Hammer's architecture should substantially outperform Intel's. That was the point I was making when you induced AMD's model numbering system from posted benchmarks. Regards, Dan

To: fyodor_ who wrote (74934)	3/19/2002 10:50:18 AM
From: Ali Chen	Read Replies (1) \| Respond to of 275872

"The conclusion that quantiherz (using the current forumla)" And exactly what the current formula is? "The change in fps from each successive addition of 100 quantiherz is, and should be, a monotonically decreasing function" That's right. However: Each "successive addition" changes the CPU core frequency in less and less diminished steps (in %%-wise); therefore the difference between consecutive scores should not drop faster than it started at 1500+, which was 0.1/step. Yet the Dp dropped by 0.3 and 0.2 during steps to 1900+ and 2000+. Therefore we can safely conclude that those 1900 and 2000 scores are at fault, and not the 2100. Actually, no matter how funny it sounds, but the ability of P4 to "scale better" is a direct consequence of it's lower IPC. Just think of a processor that can do zillions of Instructions per Clock. Until it hits a cache miss and stalls waiting for data from memory. Then it runs whatever available instructions are in almost no time (remember, it can do zillions!), until next stall. In this case (assuming the memory/FSB still runs at the same frequency), the change in df will make almost no difference in performance, so your dP/df will be zero. So, "low dP/df" = "high IPC". Take care, - Ali