Memory efficiency....in SISoft Sandra Memory Benchmark

Hi,

I don't see many people here concerning about the memory efficiency that reports in SiSoft Sandra Memory Benchmark. Most people here seek for high FSB but low efficiency, around 80%.

So why doesn't this % mean anything to us? Or do we just ignore it and go for high score. What is really count?

I dunno what its all about but i know i only get 85%

i think it's actual vs potential bandwidth, which is determined by chipset speeds or something...i just ignore it, mine's in the low 90's anyway

The memory's rating (eg. PC3200) is an estimate of the theoretical bandwidth, which doesn't concern latencies. It is purely an estimate based on the clock frequency and the number of bits (which is 64):


DDR400 * 64 bits / 8 = 3200 MB/s


The percentage-rating in Sandra indicates the actual bandwidth that is measured during the test. So you have to subtract the bandwidth loss that is caused by the delays that occur when a memory is performing a read or write. You can see that the bars always show lower numbers than the estimate, which is also reported by the program. The bars show the actual bandwidth (integer/float).

The main rig in my sig usually gets 95% integer / 87% float. If loosening the timings, the difference is quite clear.

the efficiency is just sandra deviding the theoretical with the current settings (just the bandwidth times bitrate or what not). It's a useless figure really.

nicspits said:

the efficiency is just sandra deviding the theoretical with the current settings (just the bandwidth times bitrate or what not). It's a useless figure really.

Click to expand...

What? Which current settings are you referring to? If this is the case, why does the program actually run a test? It could as well have done the estimate like I posted above (in a few nanoseconds), instead of actually testing the RAM. The memory is stressed quite a bit by this test.

However I don't know how relevant these test results are to real life-performance.