Bandwidth Efficiency What is it?

[NiteSmoker17]

i have just tested my PNY PC2100 256Mb stick and in sandra it says i have a 75% efficiency.....what is it? and is there a way to increase it? My spec's are in my sig.

[q149]

You can tighten up your timings.. since it is pc2100 100% efficiency would be 2100 mb/s and that's how it gets it

[jonspd]

My one question is how do you get your memory efficiency up from 93% or so the highiest I have ever seen mine was 94%.

[q149]

That's pretty much what you can get on nForce dual channel.. on Intel you can get closer to 200% though:p

[NiteSmoker17]

well im running a Shuttle AK32L Mobo and all i have for settings is cas, bank, and interleave. so i have no idea how to tighten up my settings other than thoes. i am also running @ 154 on my FSB.

[Deathknight]

Those are exactly the settings you need to tighten up. List out what you have currently.

[NiteSmoker17]

ok here is what i got.

Dram Timing: SPD - Manual (Manual)
DRAM Clock 100 - 133 - SPD (133)
SDRAM Cycle Length:2 - 2.5 (2)
Bank Interleave:Disabled - 2 Bank - 4 Bank (4Bank)
DRAM Command Rate:1T - 2T (1T)

how can i run it any tighter?

[hitechjb1]

For a system with DDR FSB such as KT266, KT333, KT400, nforce2,
the max memory bandwidth = FSB x 2 x 8 = 16 FSB
e.g. FSB = 200 MHz, max memory bandwidth = 3200 MB/s
(Lower FSB for KT266, KT333, but max bandwidth is still 16 FSB).

For a system with QDR FSB (quad pump) such as for P4,
the max memory bandwidth = FSB x 4 x 8 = 32 FSB
e.g. FSB = 200 MHz, max memory bandwidth = 6400 MB/s

But in reality, the actual effective memory bandwidth is lower than the max bandwidth.

memory_efficiency = measured_memory_bandwidth / max_memory_bandwidth

So memory_efficiency is a measure of how effective the memory bandwidth is for a particular system setup, normalized to the FSB MHz, in terms of
- memory ras/cas timing
- dual channel/single channel setup
- CPU FSB interface setting
- fsb:memory ratio
...

It measures the combined performance and overhead of the FSB, system memory controller, CPU FSB interface, memory modules, ... One can evaluate a particular setup and its setting, independent of the actual FSB overclocking (over a certain range) and the actual bandwidth in MB/s.

I found that

1. for nforce2 system, dual channel w/ aggressive CPU Interfacce setting, integer_memory_efficiency would be around 95-96%
- single channel would reduce it by around 2-3%
- less aggressive CPU Interface setting would reduce it by 6-8% (at least for NF7-S rev 2.0)
- CAS 2.5/3 would reduce it by 1-2% over CAS 2
E.g.
- fsb = memory = 200 MHz
effective_integer_memory_bandwidth = 200 x 16 x 95% = 3040 MB/s

2. for P4 dual channel system, dual channel integer_memory_efficiency would be 75% or below (depending on fsb:memory ratio)
E.g.
- fsb = 200 MHz, memory = 200 MHz
effective_integer_memory_bandwidth = 200 x 32 x 75% = 4800 MB/s (75% efficiency)
- fsb = 250 MHz, memory = 200 MHz (fsb:memory = 5:4)
effective_integer_memory_bandwidth = 225 x 32 x 75% = 5400 MB/s (estimated)
memory_efficiency = 5400 / (250 x 32) = 67.5%

[JigPu]

The only thing I really see 'wrong' with your settings is the Bank Interleave. I was told that a 256MB stick normally has only 2 banks, so 4 bank interleave could actually be slowing down performance for you (if I remember right). Try 2 Bank and see what it does.

JigPu