Memory divider, noticable difference??

[nitro28]

When a memory divider must be used to be stable, do you in real terms notice a difference while doing various things on your computer? If you have a athlon 64 3200 running at 2.5GHz with memory at 250Mhz is it noticably faster than the same computer at 2.5 Ghz with memory running with a divider at less than 200Mhz? Would the lower latencies most likely being run on the computer with the divider make up for some of the difference?

 

[beatle]

Bump, as I have the same question. My PC3200 memory runs @ 189, 2-2-2-10-1T but won't overclock for squat (maybe 10mhz), even with the timings @ 3-4-4-8-2T. I'm wondering if it'd be worth my while to pick up some high bandwidth stuff or not.

 

[nitro28]

anyone???

 

[bmillerd]

Yes, in fact the real world performance is where you are going to be hurt most. Dividers will up your benchmark score but in real world applications the divider actually hurts performance more than it helps. If you're gaming, sync is the way to go.

 

[miketheidiot]

Originally posted by: bmillerd
Yes, in fact the real world performance is where you are going to be hurt most. Dividers will up your benchmark score but in real world applications the divider actually hurts performance more than it helps. If you're gaming, sync is the way to go.

Wrong. I've done this with both athlon xp's and 64's. 2500 @ 2.3 with a 220 fsb divided 4/3 runs better in my opinion than the same set up with @ 2.3 and a 17(?) FSB and 1:1 with a higher multipler. HL2 was noticeably smoother anyways.

Of course if the ram could do 220 fsb that would be better than having a mem divider, but IMO assycronous bus is better than a signifigantly smaller bus and higher mult. for athlonxp as the bus is used for other things besides accessing ram.

as for the a64 question, the memory never runs asyncronous. When you change is to 166/200 or 133/200, your really just changing the multipler. CPU-z shows my 3200 (currently 265x10 with a 166/200 "divider") as 221x12. the actual change difference between 265 mem and 220 or even 178 is negligable at best. High bandwidth memory and expensive memory in general is a waste of money for a64.

 

[ssvegeta1010]

Yes, higher clocked RAM can help in many games and real world apps. Besides, PC-3500 up is getting cheaper by the day.

 

[beatle]

Does anyone have any benchmarks? Example: CPU @ 2.5GHz, mem @ 200 vs 2.5 @ 250. I've never seen benchmarks like this, so I'm unsure as to where people are getting their info as to how much something affects performance other than guessing.

 

[rogue1979]

Using a memory divider to run faster memory speeds will help performance on a Pentium 4 or an Athlon 64.

It will not help on a standard socket A. You might get a better Sandra Memory Bandwidth score, but you will lose performance in gaming as well as other applications.

This is at the same cpu speed of course. If the only way to overclock the cpu faster is to use a memory divider to slow down or speed up the memory, then you will get more performance.

But again, at the same cpu speed a 1/1 memory ratio will give best performance on socket A.

 

[nitro28]

So would setting the memory at a lower latency or not running a memory divider make a bigger "real world" difference?

 

[gobucks]

i think the more important question is this - is there a penalty between running a slower processor to a faster speed with a divider, as opposed to a faster processor running stock. It's obvious that the extra memory bandwidth helps, but what's not obvious is whether it hurts performance to run memory at the same speed out of sych. For example, if you run a 5/6 divider on a A64 3200+, and clock the memory back to 200MHz, then you'll end up with a 2.4GHz A64 running DDR400, equivalent to a 3800+. I'd like to see the 2 directly compared, to see if there is any penalty resulting solely from the divider and not related to the memory speed difference.

 

[shira]

Originally posted by: gobucks
i think the more important question is this - is there a penalty between running a slower processor to a faster speed with a divider, as opposed to a faster processor running stock. It's obvious that the extra memory bandwidth helps, but what's not obvious is whether it hurts performance to run memory at the same speed out of sych. For example, if you run a 5/6 divider on a A64 3200+, and clock the memory back to 200MHz, then you'll end up with a 2.4GHz A64 running DDR400, equivalent to a 3800+. I'd like to see the 2 directly compared, to see if there is any penalty resulting solely from the divider and not related to the memory speed difference.

I've seen posts that say that using memory dividers adds inefficiencies on older MBs, but that Socket 939 MBs don't have that issue. Like you, I'd like to see some actual tests.

Frankly, it shouldn't be too difficult:

Using an NF4 Ultra or SLi MB:

Case one: Athlon 64 3200 (2.0GHz). Run at stock.
Case two: Same CPU, with multiplier set to 5, bus speed set to 400Mhz, and memory divider set to 100Mhz, HT ratio of 2.5.

With this setup, if there are no inherent inefficiencies in using memory dividers, the performance should be virturally identical for the two setups.

My system is old, so I can't run this experiment. Anyone else care to give it a whirl?

 

[richardrds]

For A64 owners read Zebo's Memory Sticky at the top of this forum. It has all the answers you are looking for and you will be surprised by the results!!!!

 

[shira]

Originally posted by: richardrds
For A64 owners read Zebo's Memory Sticky at the top of this forum. It has all the answers you are looking for and you will be surprised by the results!!!!

Actually, it doesn't. The sticky compares performance based on various memories at different settings.

What's we're wondering here is if the SAME memory is set up two different ways on the same system, with the effect that the memory is clocked at exactly the same frequency and with the same timings, will there by any difference in system performance.