[TASOS]

Hi everybody

I've got a rev1.1 board and did the Vdd mod @1.82

I can run sync @213fsb with memory@Cas2-2-2-6 (Kingston Value ram PC3200 with Winbond BH5)@2.9

But when i use memory @120% my bench with Sandra 2003 Pro drops...why ?? (i havent tested any other mem bench)

The same thing happens with lower fsb (180..190..200)
and 125%..133%..150% mem

[dave4925]

I think your memory timings change when you run it that fast

[Medicated]

the memory timings only change if you have them set to auto, if you set them to expert they stay at what you set them to, regardless of the mem freq.

running asynch can cause loss of performance due to the time it takes to sync for data transfer, i.e. memory is ready to transfer data, has to wait to sync itself to cpu clock to send the data to the cpu. also, running the memory at a higher speed than the fsb won't show any performance increase because the cpu won't be able to use that "extra" memory bandwidth.

[EluSiOn]

sandra memory benchmark only measure memory bandwidth from the cpu... instead of entire system's memory bandwidth... so when you running async, it will not able to see any increase of performance at all.

[Yomama]

Originally posted by EluSiOn
sandra memory benchmark only measure memory bandwidth from the cpu... instead of entire system's memory bandwidth... so when you running async, it will not able to see any increase of performance at all.

I think there is more to it:

By running 120% you are running the CPU slower than the memory. Since in an AMD system the overall memory bandwidth is determined by the CPU throughput, it will thus drop to the level at which you are running the CPU (e.g. at 166/200 you will see 2700).

Ideally one would try to run the CPU faster than the memory (i.e. 75% or so), but at least the EPOX board has stability issues with 83% and 80% and at 75% the memory speed in general is too low(240/180 is worse than 200/200), and I could not get the board to run at 250 CPU FSB.

Yo

Yo

[EluSiOn]

dcddr got two memory bus/controller... so it can send memory data to cpu and agp card at the same time and that is why when people running in dcddr mode, their 3dmarks is higher...

AMD system is not totally relied on cpu throughput. Running Async and using sandra to measure the bandwidth will not able to find any gain 100% vs 120% because the path from CPU to RAM is only running at FSB...

[Yomama]

Originally posted by EluSiOn
dcddr got two memory bus/controller... so it can send memory data to cpu and agp card at the same time and that is why when people running in dcddr mode, their 3dmarks is higher...

AMD system is not totally relied on cpu throughput. Running Async and using sandra to measure the bandwidth will not able to find any gain 100% vs 120% because the path from CPU to RAM is only running at FSB...

Right on - but as far as program execution goes the path CPU to RAM is the determining factor. Actually even at 200/200 with 4.2.2. and CAS2.0 I am getting Sandra scores below 3200, so as opposed to a PIV the TBred is definitely limiting memory throughput below the capabilities of the RAM at a given speed.

Yo

[EluSiOn]

is there any other program able to test memory bandwidth besides sandra?

[muzz]

Sciencemark( I believe) pcwizard and passmark( I'm pretty sure they all do).

[EluSiOn]

it would be great to test the memory bandwidth using those apps i think...

I am very interested in this subject as well.. and also... would different brand running at the same fsb produce different memory bandwidth result (shouldn't be right)?

[Yomama]

Originally posted by EluSiOn
is there any other program able to test memory bandwidth besides sandra?

I am using SETI. It is a highly memory speed dependend program that also stresses the CPU very well.

Yo