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: j3pflynn who wrote (235602)	7/3/2007 4:08:48 PM
From: graphicsguru	Read Replies (1) \| Respond to of 275872

It's from a document dated Feb. It comes from this vd.verysell.ru realworldtech.com Hans' interpretation (below) is probably correct. The slides are talking about throughput improvements from doubling the number of cores. When you derate for the 15% lower clock speed of actual Barcelon vs. 2.3Ghz, the throughput improvement is rather unimpressive. Let's see. 1.4 * 2.0 / 2.3 = 1.21. So they're claiming that you'll get an average of 21% throughput increase for those server tasks by doubling your cores. Wow. I'm floored. I don't know how to break it to everyone, but if you replace a Woodcrest with Clovertown, you get a better throughput improvement than 21%. And Woodcrest outperformed Opteron to start with . . . . (yes, I've assumed linear scaling of performance with frequency to derate the numbers to 2.0Ghz just to get in the ballpark. But if Barcelona can't scale pretty well from 2.0Ghz to 2.3, it has even more issues than previously revealed). David Kanter (dkanter@realworldtech.com) on 7/2/07 wrote: --------------------------- > >I'm not sure about the scale on those relative barcelona performance slides. > >DK I would say they give the percentage of improvement of a two socket system (slide 71) and a four socket system (slide 72), when going from 3 GHz 2222's to 2.3 GHz Barcelona's This would correspond to a >40% median performance increase when 3 GHz SE2222's are replaced by 2.0 GHz Barcelona's (with a maximum of ~70%)

To: j3pflynn who wrote (235602)	7/3/2007 4:46:54 PM
From: dougSF30	Respond to of 275872

To what do you refer, Paul? Theo made quite a mistake in that article, referring to a Barcelona 2.6GHz claim as a 2.3GHz claim. Perhaps you should read closer?

To: j3pflynn who wrote (235602)	7/3/2007 6:11:59 PM
From: Petz	Respond to of 275872

re: benchmark charts in xbitlabs xbitlabs.com WOW, Intel only got 42% TPC-C performance improvement from quad core, that's X5355 relative to the 3 GHz DC Xeon 5160, and Opteron 3 GHz was faster than 5160. The 2P graph shows 2.3 GHz Barcelona using SQL or Oracle to be 55 to 70% faster than 2P Opteron 3 GHz 8222SE. So, in short, database performance going from DC to QC gains a LOT more performance for AMD vs. Intel, even though, initially, you have to reduce the clock speed. This means 2 GHz Barcelona will be 40% to 60% faster than 8222SE, depending on database used. (I am assuming benchmark/)frequency scaling is between 60 and 100%.) OK, so take a look at TPC.org There are no scores using 8222, the closest is a 2.8GHz 8220. It gets 139693. So the 8222 "baseline" score is about 150,000. According to the 2P graph, add about 56% to get Barcelona 2.3 GHz performance using SQL. That's 234,000. This is better than one 2.66 GHz Clovertown result, and below the other one. So it takes 2.3 GHz Barcelona to match 2.66 GHz Clovertown. 2.5 GHz will easily match 3 GHz, since Clovertown scaling is pretty bad until they bump up the FSB. So, for 2P TPC-C using SQL, AMD needs 2.5 GHz to match a 3 GHz Clovertown. 4S servers are another story. 4S 2GHz (not 2.3) Barcelona should get about 390,000, eclipsing all quad-core systems from Intel, Itanium or Xeon and doing it at a lower cost per tpmC than any 2S or 4S Intel system in the >200K tpmC range. Petz