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.
Technology Stocks : Advanced Micro Devices - Moderated (AMD)
AMD 246.76-0.5%Nov 14 9:30 AM EST

 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  
To: pgerassi who wrote (227044)3/1/2007 2:24:10 AM
From: Ali ChenRead Replies (3) of 275872
 
Dear Pete, you are really pushing me to take another 2-day vacation from SI. Your confusion is unbounded, and worse, you confuse other readers. Look:

"Using Base is artificial. Even SPEC does not say peak in their scores, just plain SPECfp_2006. For base they use the qualifier base. And with the rules SPECfp_2006 uses the base scores, if there aren't peak ones. So even SPEC uses the peak scores in comparisons."

Nothing is natural in benchmarking, but every aspect of SPEC has practical meaning. Who understands the spirit of SPEC, they realize that "peak" scores are theoretical maximums for self-compiled applications, and look even into individual scores that resembles their own computational problems. If you want to evaluate potential performance of a platform on pre-compiled applications from a shelf, you better look at base scores. Of course, nothing is perfect. However, I would look at a compiler that unifies its performance potential in few base flags as an achievement, not a flaw. If you want to find a ground for criticism, better look at feedback guided optimization concept, which is clearly mis-applicable for off-shelf applications.

"Using the same type OS is also not artificial. When comparing two things one tries to keep as much similar as possible."

Wrong again. We are not here for an academic discussion which IPC are better or which branch predictor is smarter. PC buyers are for plain performance result, how fast a PC can compute their application of run a CAD program, regardless of whether ithe memory is PC8500 or RDDR or how shine is the case.

"Using Linux and 3rd party compilers, Opteron with RDDR PC-3200 got the same percentage of Woodcrest with FBDIMM 667MHz in both SPECint_2006 and SPECfp_2006,"

Again, irrelevant. When I buy a PC to run AUTOCAD or compile FPGAs, and I have a licence only for WinDos platform, your Linux "same percentage" does not buy me any performance. As I said to Petz, you are misguiding yourself by looking at a narrow field of applications. Sure, servers are a nobel application, but the bulk of money is in mass production of media PCs, general PCs, and workstations.

Get real,

- Ali
Report TOU ViolationShare This Post
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext