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: AK2004 who wrote (74840)	3/18/2002 1:21:22 PM
From: fyodor_	Read Replies (2) \| Respond to of 275872

Albert: could you you run me through your numbers because I, frankly, do not see how you got to your conclusion. I'm assuming this is to me, although my name wasn't "Joe" last time I checked ;-). sandra 2002 yields no difference considering how much of a difference especially considering what a bios upgrade made You seem to quote SiSoft Sandra as your only example. I did say that this particular "benchmark" is totally useless&#133 Still, the trend is there. As I said "Northwood consistently scales slightly better than AthlonXP". For AthlonXP, dP/dQf is around the 4300 mark. For Northwood, dP/df is around the 4900 mark. That's a 20% difference&#133 in Northwoods favor. The higher the dP/df (or dP/dQf), the greater the performance increase achieved by the higher CPU frequency (or quantispeed). I.e. Northwood clearly benefitted more from the 2000MHz -> 2100MHz jump than AthlonXP did from 2000+ -> 2100+. That said, there certainly are serious error margins to deal with, as the Willamette scores demonstrate. Clearly, dP/df should me a monotonically decreasing function of f (even d²P/df² should behave this way). ps sandra overall yields even less conclusive results Actually, that's SysMark2002 - a completely different benchmark from SiSoft Sandra. I gave a detailed explanation of why I am disregarding the AthlonXP 2100+ numbers (the BIOS was apparently upgraded, with "no affect on performance"&#133 cough). Disregarding that one data point, the AthlonXP dP/dQf numbers are: 40, 20, 60, 10, 10 (for the different speed grades respectively, ordered lower to higher speed grades). The Northwood gets 45, 105, 55. Not enough data points to really want to say too much with any overwhelming certainty, but the trend is clear: the Northwood numbers are higher. The same goes for the Willamette numbers for the same benchmark. what you showed does create an impression that NW scaling is better but from what I see it is far less than conclusive I chose 7 benchmarks fairly arbitrarily - the others in the referenced THG article exhibit the same behavior, as do the others in the AnandTech article. In a few places, I could have chosen absolute numbers that looked worse for AthlonXP (e.g. using the Intel-optimized Lightwave 7b instead of 3D Studio Max 4.2, or using the regular Quake3 demo1 instead of the (much) more complex NV15 demo), but the frequency dependence did not change. Of my 7 examples, I included 2 semi-synthetic benchmarks. These are the two you chose to quote from and I would certainly agree that based on these 2 alone, one would be hard pressed to conclude anything with any significant degree of certainty. Since, however, the other 5 (a range of REAL programs and applications) completely confirm the weak trend noted with the synthetics: Northwood "real MHz" scales better than AthlonXP "quantiherz". Sure, this may not be extremely surprising, but it does show that AMD needs to either change its quantispeed formula or feature some significant improvements in Tbred (or break with the "True" in True Performance Initiative (TPI)). -fyo