Sandra Bandwidth Test

I have an IT7 w/845E chipset. When I run the Sandra memory test I see an
efficiency of 42%. I have a:
p4 3.06GHz OC to 150MHz
1 Gig PC2700 DDR - 300 Mhz

Can anyone explain why the efficiency is so low? My other PC running is 85%
(ECS junk board with p4-2GHz/PC2700 Memory)

joe

 

you answered your own ? i think.if you say it is at 150 fsb,ddr
300,that means you are running a 1-1 divider.go into the bios,and set
the divider to 3-4.
but then your pc2700 ram will be running way out of spec at
ddr400-pc3200
150 fsb divided by 3 x4 =200fsb on the ram

 

What version of Sandra, and what are the actual figures?

Have you manually set the 1:1 FSB to memory divider, or have you allowed the
board to set the memory bus speed automatically?

We can't explain with 100% authority, no, as we don't have your PC in front
of us to reproduce the tests and check your settings. The best we can manage
in the circumstances will be educated guesses, and even there we really need
some more information.

What happens to the memory bandwidth figures if you clock the PC back down
to default speed? Does the bandwidth efficiency alter to any significant
degree?

Is it overclocked? Is it a single or dual channel board? What are the actual
bandwidth figures it reports?
--


Richard Hopkins
Cardiff, Wales, United Kingdom
(replace nospam with pipex in reply address)

The UK's leading technology reseller www.dabs.com

 

for "Real Life" memory performance efficiency to compare with others
see mx site under comp/benchmarks, link @ "updated" ...

 

Latest LITE Version
2180 MBITS/sec

YES manualy set to 1:1

Efficiency Improves very slightly at Ck=133MHz

Not overclocked with DDR266 frequency single channel memory
Reports 2040MBITS/Sec.

I thought that that it was posible that I was causing a degradation that
would be obvious to an experienced user. For instance, operating the memory
at a frequency out of sync with the CPU (If that makes any difference&).
Joe

 

My experance is that some chipsets don't give good bandwidth,

My last KT400 chipset was similar (single channel), although the NF4 Ultra I
have now gives 92% @ 6.02Gbits (dual channel), both with the same 512MB
PC3200 memory sticks.

Dazzer

 

thats true, but it is not all in the bandwidth ...
(DDR has double that than SDRAM, but using same real clock frequencies
& latencies, the second is faster in real life (& using WinRAR3.3
benchmark & hardware test under Tools tab, DDR2 is even slower than
DDR1 ...)

Sandra has some good subtests for "real life" benchmarking & comparing
(depends also on revision of it) like drive index for HD performance
and other useless ones like max.mem.bandwith test (but good for
marketing purposes). The last is almost useless for "real life" IMHO,
since practically almost never the transfers between mem. controller
reach its peak speeds ...

time to time I have access (repairing stuff too) to different
platforms so I checked also DDR based sistem with same real CPU
clocks/FSB/memClocks & same peripherals with SAME OS install (minimum)
between my setup and I can tell you that mine is faster in average
work (you can feel it & bench it with stop watch too). The major
difference between these platforms was just a WinRar mem Score !

Even thru Sandra mem bench showed much better performance for an AMD
platform of the time (AMD XP1500+ for example), my setup was faster -
noticeable (more than 10%); WinRar mem score was almost 50% higher on
mine! IMHO WinRar developers made a hell of a memory benchmark, but
is not accepted yet in the "benchmark" comunity since that uses widely
established Sandra results, but does not understand fully certain
stuff, because people have been bombarded with marketing blurb for
years & not tried to brainstorm with a bit of distance away from that
"widwly accepted thruths" ...