when o/c an amd cpu, is it harder to do it w/ dual ddr?

how high can I set my FSB if I am using dual ddr vs a single stick of ddr?

Hi all, I am planning on o/c an amd cpu. Is it harder to do it w/ dual ddr than a single stick of ram? The reason is that I am on a tight budget and I think it's cheaper to get two slower sticks than one fast stick.

Is that true that for a stick of PC3200, I can set the FSB to
3200 / 8 / 2 = 200

How do I calculate the max FSB I can set when I am using dual DDR? Like how much FSB I can go for two sticks of PC2700?

Thanks in advance.

Re: how high can I set my FSB if I am using dual ddr vs a single stick of ddr?

smallghost said:

Is it harder to do it w/ dual ddr than a single stick of ram?

Click to expand...

No, it isn't harder but you'll get different results.

smallghost said:

Is that true that for a stick of PC3200, I can set the FSB to
3200 / 8 / 2 = 200

How do I calculate the max FSB I can set when I am using dual DDR? Like how much FSB I can go for two sticks of PC2700?

Click to expand...

3200 is suppose to run at a 200fsb. But it'll depend on your mobo too. Buying 3200 doesn't mean the mobo will be able to reach 200. A number of factors involved.

You can't calculate the overclock you'll get with 2700. It'll depend on the particular sticks you get. Some brands overclock better than others and even for the same brand one stick will differ from another.

actually, it is harder to get 200fsb with 2 sticks, ime not all the dimms on an nf2 board can go 200fsb, especially dimm1 (furthest one from the cpu) on the nf2 boards, just like you can be limited by your worst stick of ram in the sys you can be limited by your worst dimm

even though i'm using one stick, i have it in dimm3 now because at dimm1 with the same settings it wasn't stable, now in dimm3 it's rock solid, i've also read about this on this forum and others for other peoples boards

you can't calculate what you can get, but you can test for which is the most you can get with components remaining stable

I would like to see this test done on the Nforce2 platform to show that in principal Dual channel would have real world benifits if the FSB was quad pumped like Intels P4 bus

Run the FSB at 200mhz

Run memory at 133mhz single channel and bench

Run memory at 133mhz Dual channel and bench

The 200FSB will give the memory at 133mhz Dual channel a chance to stretch it's legs and show that given some headroom it does work. If thereis no difference in this extremme technology test than Nvidia Dual channel really BS.

All other memory comparisons are pointless to show Dual channel
advantages with DDR and AMD double pipeline bus when the limiting factor is always the AMD BUS.

Seraphic said:

I would like to see this test done on the Nforce2 platform to show that in principal Dual channel would have real world benifits if the FSB was quad pumped like Intels P4 bus

Run the FSB at 200mhz

Run memory at 133mhz single channel and bench

Run memory at 133mhz Dual channel and bench

The 200FSB will give the memory at 133mhz Dual channel a chance to stretch it's legs and show that given some headroom it does work. If thereis no difference in this extremme technology test than Nvidia Dual channel really BS.

All other memory comparisons are pointless to show Dual channel
advantages with DDR and AMD double pipeline bus when the limiting factor is always the AMD BUS.

Click to expand...

I don't understand. The AMD bus has been and remains a limiting factor. 200 is faster than 133. Dual channel is faster than single but not by much. What's your point? You picking the age old Intel is better than AMD fight?

let me first say I have both Intel and AMD systems so I am not trying to start an AMD vs Intel.

AMD BUS is double pumped and DDR is double data rate. Running
both in sync uses up the maximum available bandwidth. So when I see reviews of the Nforce2 there is very little difference between running memeory dual channel and single channel in sync with bus speed.

Do you agree? K

Now what I want to know is does the Nvidia implementation of Dual channel memory bus really work. How can one show this?

Well you give the dual channel room in the Bus. So you run the bus speed faster than the memory speed and then run in single channel then dual channel. This would prove that Nvidia's dual channel when given room in the systems BUS does indeed offer tangible performance benefits.

The reason I threw Intel into the mix is because of their Quad pumped Bus, DDR in Dual mode has the chance to offer higher performance than using DDR in single channel becuase Itels quad pumped bus is not saturated.

In short: I want to see Nvidia's Dual channel really work and this test I proposed would prove that it does.

Seraphic, thanks for the clearification.

To run the FSB @ 200 and the memory @ 133 then I assume you're suggeting setting the sync to 66%. How does the fact you're now running async affect the results? Should the test be done with the memory at 100 instead? Obviously I don't fully understand your line of thought.

perhaps the Nvidia Dual memory controller might have a bug when running Async slower than FSB but I just want to see it go faster than single channel mode.

I understand that running memory slower than FSB is slower than sync with FSB.

The point of the excercise is to prove that given the headroom
of a faster FSB bandwidth than a memory bandwidth the use of 2 64bit memory controls can offer more bandwidth than one 128bit controller.

So we are trying to artificially create the same environment that shows how the Granite Bay and Sis655 enjoy.

I have not seen evidence to date to show that's Nvidia's Dual memory controller really work. Just excuses from reviews and Nvidia about how the AMD FSB is limited.

Simply Remove the limitation to show that it does work. One day AMD bus will be quad pumped and we will all know that from the very beginning Nvidia's Dual channel memory controllers just need more bandwidth and that it will improve performance significantly

Peace

I see what your train of thought here is, but all it would show is what is know already, dual ddr is a bit faster but not by much.

To answer smallghost's original question, what mem are you looking for toal 512mb i assume, just get either 2x 256mb 3200 or 1x 512mb 3200. I personally use 2x 256mb crucial and i've got no complaints. If you want to get into the high fsb range then you'd be better off getting 3200 ranther than 2700 as it's one less thing to worry about.

"I see what your train of thought here is, but all it would show is what is know already, dual ddr is a bit faster but not by much."

Yes this is the case when there is no headroom in the FSB for the memory to run free. If this is still the case when the memory does have headroom than the dual memory controller is a sham.

Have you seen any P4 dual memory Channel board reviews?
There is a really BIG Difference between single and Dual Channel.

That is proof positive than given the the headroom the dual channel memory contollers implemented by Intel and SIS Work.

Someone point me to a review that can show this with Nforce2?

Please don't give methe old AMDbus excuse. I just wrote 3 pages explain how to overcome that limitation given the available bios otptions.

Eg. My car can do 175 miles per hour but I can't go that fast on public roads. Go to a race track then and prove it. Enough with the excuses.

OK, ran the following tests @ 200Mhz FSB
1 stick memory @ 100Mhz
2 sticks memory @ 100Mhz

Tests Used
1) Sandra memory bandwidth and cpu (multipler @ 11), did run multiple tests
2) Prime95 benchmark (256K FFT length) averaged for 5 separate runs

1 stick
Sandra --
memory bandwidth -> 1469/1404
arthematic cpu -> 6081/3034
Prime95 -> 22.725 ms


2 sticks (dual ddr)
Sandra --
memory bandwidth -> 2165/2037
arthematic cpu -> 6031/3033
Prime95 -> 19.171 ms

Thus dual ddr improvements
Sandra --
memory bandwidth -> 47%/45%
arthematic cpu -> ~unchanged as expected
Prime95 -> 15.6% improvement



At 196Mhz FSB (cpu mult = 12) and sync memory for 1 stick
Sandra
memory bandwidth -> 2949/2738
arthematic cpu -> 6492/3352
Prime95 -> 16.812 ms


At 196Mhz FSB (cpu mult = 12) and sync memory for 2 stick dual ddr
Sandra
memory bandwidth -> 3013/2858
arthematic cpu -> 6518/3283
Prime95 -> 15.621 ms


Thus dual ddr improvements at my OC
Sandra --
memory bandwidth -> ~2%/4%
arthematic cpu -> ~unchanged as expected
Prime95 -> 7.1% improvement

deeppow you da Man!

So dual channel does have what it takes if it only had some bandwidth. Dual channel in this case is only an advantage for those who use memory that cannot keep up with FSB speeds.

So if someone had two sticks of PC2100 would get far better out of there memory in Dual channel mode on an Nforce2 than on a Via chipset. If they can run their FSB significantly higher. Good stuff!

Thanks.

When you do memory tests it only tests 1 stick.....so there is no real way of actually seeing what DC gets you.

Seraphic said:

So if someone had two sticks of PC2100 would get far better out of there memory in Dual channel mode on an Nforce2 than on a Via chipset. If they can run their FSB significantly higher. Good stuff!

Click to expand...

Yes that conclusion has been reached before. You can get better performance out of poor memory, thus it produces optimum performance for minimum dollars to a certain extent. However, you still get the best performance out of the best memory and cpu. The cost is just more for decreasing returns.