Running memory with divider

[Speedo]

As many others on these forums, I'm awaiting the nForce4 mobos to come out, so I can get it and slap on a winchester and overclock my hearts out.

But I have one unanswered question while figuring out my system setup. In my current AXP setup I have some 2x 512MB Geil PC3200 Ultra Dual Channel memory. I can't decide whether its a good choice to keep this memory in my upcoming setup.

My goal will be to reach 2.5-2.6GHz, with either the winchester 3000+ or the 3200+. I'm trying to figure out which memory divider I will have to run to reach these CPU speeds. I haven't pushed this memory yet (my A7N8X-Deluxe doesn't like high FSBs), but judging by reviews I found on the web it seems like this memory will do around 225Mhz (450DDR) with its speced 2-6-3-3 timings.

So what would be my best choice? Aim for 250-260FSB with memory (divider) set to 166 in bios (which should make the mem run just above spec) coupled with the 3200+ which has a 10x multiplier to reach my target speed, or should I get the cheaper 3000+ with a 9x multiplier and aim for FSB 280-290MHz, with memory set to 133 in bios? In the later case I think my memory would run a little below spec?

So basically wondering if I can save a lot of cash and keep my current memory, and if so, should I put up some extra cash to get the 10x multiplier of the 3200+ instead of 9x with the 3000+. I'm also very curious about how the asyncronous memory settings suggested above would affect my overall system performance... Would I get a big hit in performance if I ended up with async memory running running a bit below 400MHz?

 

[Zebo]

All mem on a64's run async.

get the 3200. Reason being is 2500Mhz is likly and 10X multi allows for 250Mhz Bus keeping your ram in spec at 207 Mhz using the 166 memsetting.

If you get the 3000 and hit 2500Mhz (9x277 FSB) your ram will probably not run at 230Mhz using the 166 divider or will be signifiganty underclocked using the 133 mem setting (184Mhz)

Edit theres simply much more options with the 3200 as you can lower multiplier but still have the 10 allowing for much more flexibility with your ram or any ram for that matter.

 

[Speedo]

All mem on a64's run async.

Even if you set it to 200/1:1 in bios ??

get the 3200. Reason being is 2500Mhz is likly and 10X multi allows for 250Mhz Bus keeping your ram in spec at 207 Mhz using the 166 memsetting.

Yeah, that seems logical. I was originally only considering the 3000+, but you are definitely right that I should go for the 3200+ instead. Funny thing is it seems as the majority of users in these forums have choosen the 3000+? I guess they might have better/faster memory though...

Are there any benches on the web showing the difference between running at 133/166/200(1:1)?

 

[Zebo]

Yes at the 200 setting the memory in an Athlon 64 system is running async still. The memory speed is arrived at by using a divisor off the Athlon 64 CPU speed (i.e. a /12 divisor.)

http://www.ocforums.com/showth...?p=2902632#post2902632

 

[StriderGT]

DEFINATELY try async...
Your memory is A ok, and has nothing to do with the CPU OC potential check the calculator at the bottom on how to reach those freqs with the 3000+ and the 3200+
Important: Use manual memory settings (eg 3-3-8-3, not Auto) because the BIOS is miscalculating/misinterpreting SPD of the memory when you run async...

 

[Speedo]

Originally posted by: Zebo
Yes at the 200 setting the memory in an Athlon 64 system is running async still. The memory speed is arrived at by using a divisor off the Athlon 64 CPU speed (i.e. a /12 divisor.)

So you mean the memory speed changes when changing the CPU multiplier as well? (Since CPU speed is affected by multiplier, and you say mem speed is arrived by a divisor of CPU speed) Definitely new to me...

StriderGT: Thanks about the "manual setting" tip. I actually noticed this when building a S454 for a friend. I noted the spd settings at default speed, 1:1, and locked it at that.

Still I wonder though, what the performance hit will be by running the memory with a divider, instead of buying new memory which can run at high FSB speeds 1:1 ...

 

[StriderGT]

Originally posted by: Speedo

Originally posted by: Zebo
Yes at the 200 setting the memory in an Athlon 64 system is running async still. The memory speed is arrived at by using a divisor off the Athlon 64 CPU speed (i.e. a /12 divisor.)

So you mean the memory speed changes when changing the CPU multiplier as well? (Since CPU speed is affected by multiplier, and you say mem speed is arrived by a divisor of CPU speed) Definitely new to me...

Check OCA64, there is a math type and yes mem speed changes when you change cpu multiplier!!!
There is negligible/no performance hit, as Zebo said A64 is always Async

 

[Speedo]

Ok... no performance hit by using a divider itself, but what about the different/lower memory speeds? I know the A64 isn't as bandwidth limited as the P4 systems, but I guess it does affect performance, just not as much...

 

[StriderGT]

You can blow a donkey as we say in my parts...
There are so many multiplier combinations that can be used even with the 3000+, so that:

1. You run your memory at its highest stable rate
2. You run your CPU at its highest stable rate
3. You run your FSB high enough to accomodate the previous two (270-300+ doable)
4. You run your HyperTransport at something below 1000-1050 (this is the only real showstopper)

Use the calculator to find out how to do all four...

 

[Speedo]

StriderGT: I have looked at your excel dokument. This made me realize even more that I should get the 3200+. Then mem and HT would be just above spec with CPU at 2550MHz. With a 3000+ though, I'd had to go down in memory divider and HT, which will result in them running below spec.

I still have a hard time believing that changing only the CPU multiplier would affect the memory speed. If that is the case, what's the formula to get the resulting frequency with a given FSB, divider and CPU multiplier? Now, please don't tell me the HT speed also affects memory speed.....

 

[StriderGT]

Originally posted by: Speedo
StriderGT: I have looked at your excel dokument. This made me realize even more that I should get the 3200+. Then mem and HT would be just above spec with CPU at 2550MHz. With a 3000+ though, I'd had to go down in memory divider and HT, which will result in them running below spec.

Yes 3200+ is better since it has the holy 10x multiplier, but 1020 HT can not yield noticable increase in performance over 956 (even 850 only if the benchmark or application is HT dependant - very rare).
I personally would also choose the 3200+ but the point is that even with the 3000+ you will get your job done!

Originally posted by: Speedo
I still have a hard time believing that changing only the CPU multiplier would affect the memory speed. If that is the case, what's the formula to get the resulting frequency with a given FSB, divider and CPU multiplier? Now, please don't tell me the HT speed also affects memory speed.....

No HT speed does not affect memory speed.

OCA64 Calculator type:

Div. = CPU Frequency / Memory Frequency
= CPU Multiplier * ( FSB / Memory Frequency )
= Rounded [ CPU Multiplier * ( 1 / Div. Ratio )]

=> Memory Frequency = (CPU Multiplier * FSB) / Rounded^ (CPU Multiplier / Div. Ratio)

where Div. Ratio = 1, 5/6, 2/3, 1/2 (Max Mem Clock = 200,166,133,100)

I don't think it can get any clearer than that