When is a benchmark NOT a REAL benchmark?

[bwanaaa]

So we crank the fsb, lower the multi, tighten the ram, use dividers - but why are my sandra scores not impressively improving. Here's an example using the following hardware-
AMD 4800, epox 9npa+ultra, 7800gtx card and various flavors of ram (gskill pc4400 2.5-3-3-7 at 290 fsb, the same gskill at 2-2-2-5 at 200 to 216 fsb)
FSB multi LDT RAM freq CPU ARITH MM MemBandwidth
200 12 5 2x200 2.41 27959 95065 11567 (STOCK)
210 12 5 AUTO(2x168) 2.52 29216 99254 10073
216(2x227) 29239 99379 12094
233 29226 99628 12613
250(2x140) 29364 99892 12964

215 12 4 2x107 2.57 29776 101252 6464

270 9 4 AUTO (220) 28161 95841 12192
285 9 4 AUTO (233) 29747 101243 12892
290 9 4 AUTO 30216 103212 13122


As you can see, I can crank the FSB up almost 30% but memory bandwidth only really scales like the total cpu speed. Does this mean that SANDRA is missing something?

 

[bwanaaa]

sorry about the above jumble-it looked nice when in the post but when i submitted, all the extra spaces got stripped out. The first row is the column headings.

Basically, FSB goes from 200 -> 290, CPU goes from 2.4 to 2.6 and SANDRA benchmarks go up 10% (like the cpu) not 30% (like the FSB)

 

[RichUK]

Sisoft Sandra + Intel & its HT = :heart:

 

[ElTorrente]

Originally posted by: RichUK
Sisoft Sandra + Intel & its HT = :heart:

I think they are "sitting in a tree"