Dividers: hurt performance alot??

[FSH42NA]

I'm new to OC'ing and am still getting acquainted to the terminology. There a couple of issues that I keep wondering about:

(1). Is there a significant drop in performance when you have to use a divider to slow down your memory in order to allow a better overclock of your cpu? If cpu speed is the most important, at what point, does your underclocking of memory significantly negate any gains you receive from overclocking your cpu?

(2). What is the preferred order of adjustments to your rig as you overclock? In other words, when you hit the wall and can't oc any further, what do you tweak 1st (2nd, 3rd, etc.) to overcome this obstacle without affecting overall performance?


Thanks for your input

 

[peleejosh]

There is no decrease in performace if you use a divider on an Athlon 64 system
Always go for the highest cpu speed you can get, even if it mean running at a relatively low ram speed.

 

[FSH42NA]

Thanks, that's just what I needed to hear!

 

[AppaYipYip]

I'm overclocking my 175 Toledo Opteron, and its at 2.6 with my DDR 400 memory reading 175MHz in CPU-Z, is this acceptable you think? I try to keep it at 200 but can't seem to figure out how to get it there. I'm running a pretty heavy "divider" with my Asus board, it's set at DDR 266 in BIOS.

 

[GuitarDaddy]

175mhz if fine! You may be able to tighten your times tighter than spec at that speed, low latency seems to make more difference than mhz on A64's anyway.

 

[CP5670]

Yeah, I keep my memory at 183mhz, at which 2-3-2 is possible, instead of the stock 200mhz 2.5-3-3 and it's slightly faster.

 

[EXEEMLITE]

So your saying that there is no ill efect if I am running my ram at say 100mhz as long as I have my opty as high as it will go I will not see any sluggishness? please explain thanks for your replies

 

[malvare1]

you want to stay above 200mhz

 

[Doctorweir]

Originally posted by: EXEEMLITE
So your saying that there is no ill efect if I am running my ram at say 100mhz as long as I have my opty as high as it will go I will not see any sluggishness? please explain thanks for your replies

No. 100 is of course way too low and will severely hurt performance. The thing is to maximize CPU clock and also maximize the RAM clock as far as possible (depending on RAM specs).
E.g. with DDR400 rams you want to stay as close as possible to 200. It does not hurt however if you have to go down to 190, if your CPU maximizes at that value and getting the ram to 200 would cost you 100MHz CPU speed...
As Guitardaddy said: If you go lower than the rated speed of your mem, try to tighten timings (if not already 2-2-2-5 ram ).
However there is a limit in going down...175 is quite low already...

 

[CP5670]

The original Opterons only supported 166mhz (DDR333) memory and didn't gain a whole lot (about 5-6% on average IIRC) going to 200mhz memory. 100 is extreme, but I don't think you'll lose much down to 175 or so at least, especially since you can compensate for it with lower timings.

 

[Indred]

I know SysTool is a good program and it was able to let me have more options on memory deviders. My board only has like 133, 166, &200. With sys tool I have increased my OC. I'm in the process of finding "the sweet spot" ballancing the max ram and max cpu. Like most of the How-To's in this forum say find your max cpu, htt, ram and then you will have a better understanding of what your over all max set up could be. will help you to find what the best mem divider would be also.

 

[GuitarDaddy]

I was recently amased by how well old slow ram does in A64's. I built my son a system with a 3200+, epox 9NPA ultra nf4, 2x512 PC3200 2-2-2-5, and it had a problem that required RMA'ing the ram. So I stuck 2x256 of old PC2100 2-2-2-5 from my old Tbird 1.4 system(cirica 1999), and it runs 100mhz. I couldn't even tell a difference in general usage the old ram at 100mhz is very snappy. All the games except HL2 and Doom3 showed very little difference, HL2 and Doom3 lag significantly though because of the drop from 1gb to 512mb.

 

[Falloutboy]

its really amazing the athlon can do so well with slowish ram compared to intel, I'm really woundering if DDR2 will actually hurt performance a bit with the increased latency

 

[Regalk]

"There is no decrease in performace if you use a divider on an Athlon 64 system
Always go for the highest cpu speed you can get, even if it mean running at a relatively low ram speed.. etc etc"

I have ran the following:
303 X 9 DDR at 5/6 2.5 4 4 10 1T - faster mem speed = 246
320 X 8.5 DDR at 5/6 2.5 4 4 10 1T - same as above mem speed = 248
337 X 8 DDR at 2/3 2.5 3 3 8 1T - slower.

So I am not so sure about the running RAM at lower speeds theory doesn't hurt performance. 3Dmark 2001/3/5 all show significant increases as well as UT2003/4 when I run RAM speed up to as close as possible to 250 at 1T as opposed to running below say 225.

 

[Indred]

Is it possible to make a table (I know it would take some work) for memory timmings for us that know nothing about it. For example
XXXX would be the best timmings
XXXX would be the next best all the way down to the worst.
I don't know if this is possible (I don't see why not) but I know it would help me GREATLY and probly most of the other N00B's here(like me).

 

[Cheesetogo]

Here's one:Text
(scroll down)

 

[blurp]

Originally posted by: Cheesetogo
Here's one:Text
(scroll down)

Aewsome. Thanks.

 

[Zebo]

Originally posted by: malvare1
you want to stay above 200mhz

Not Really PC2700 or even 2500 at 2-2-2 will beat this corsiar value stuff people are buying and running PC3200 3-3-3.. They should underclock it and tighten timings.

 

[uOpt]

Benchmarks, most real-world.
http://cracauer-forum.cons.org/forum/crabench-memory.html

For high-clocked Opteron links see end of page.

 

[Mucker]

Nice testwork Martin :thumbsup:

 

[Zebo]

Originally posted by: MartinCracauer
Benchmarks, most real-world.
http://cracauer-forum.cons.org/forum/crabench-memory.html

For high-clocked Opteron links see end of page.

Great work martin and very much what I've found.. A64's not bandwitdh staved and benefits most from Low latency- and of course no effect of divider since all memory on A64's use a divider in the first place.

I think 1.5 CAS AMD upconverts to 2 though so I'm not sure you have any benefit there. I tried A64 tweaker and a couple benches which are really mem intense like winrar/farcry and 1.5 and 2 same. You may want to verify.

 

[jenny9580]

Originally posted by: Zebo

Originally posted by: malvare1
you want to stay above 200mhz

Not Really PC2700 or even 2500 at 2-2-2 will beat this corsiar value stuff people are buying and running PC3200 3-3-3.. They should underclock it and tighten timings.

Hmm.. so I'm actually better off running 2-2-2-5 @185MHz instead of the 2.5-4-3-7 @227 I'm doing now? If it only works at 2-3-2-5 (stock speed for the OCZ plat), is it still worth it to run @185? Could I keep stock ram voltage if it's underclocked, despite trying the lower latency?

These numbers are based on using either the 266/400 or 333/400 dividers with my current OC (in sig).

 

[Zebo]

Originally posted by: jenny9580

Originally posted by: Zebo

Originally posted by: malvare1
you want to stay above 200mhz

Not Really PC2700 or even 2500 at 2-2-2 will beat this corsiar value stuff people are buying and running PC3200 3-3-3.. They should underclock it and tighten timings.

Hmm.. so I'm actually better off running 2-2-2-5 @185MHz instead of the 2.5-4-3-7 @227 I'm doing now? If it only works at 2-3-2-5 (stock speed for the OCZ plat), is it still worth it to run @185? Could I keep stock ram voltage if it's underclocked, despite trying the lower latency?

These numbers are based on using either the 266/400 or 333/400 dividers with my current OC (in sig).

That's a little tougher but you can try it easy... the most demostrative test, and correlates best with games and other real world scenarios, is winrar's internal benchmark. (highest number wins) But even with a 41 Mhz deficiency, yes, I think you'll find the underclocked choice is faster.

I assume your 2-2-2-5 @185MHz is with 133 memsetting and your 2.5-4-3-7 @227 is 166?

That means you're 133 will actually be 178.71 Mhz or (DDR357) so it's almost 50Mhz difference.

 

[jenny9580]

Here are the results of my highly unscientific tests using the Winrar bench:

620 - 2.5-3-3-7 @227, 166 div
589 - 2.5-4-3-7 @227, 166 div
586 - 2-2-2-5 (errors) @185, 133 div
586 - 2-2-2-7 @185, 133 div
552 - 2-3-2-5 @185, 133 div

I didn't get Winrar errors at 620, but Prime is now yelling at me.

 

[uOpt]

Originally posted by: Zebo
That's a little tougher but you can try it easy... the most demostrative test, and correlates best with games and other real world scenarios, is winrar's internal benchmark. (highest number wins) But even with a 41 Mhz deficiency, yes, I think you'll find the underclocked choice is faster.

I assume your 2-2-2-5 @185MHz is with 133 memsetting and your 2.5-4-3-7 @227 is 166?

That means you're 133 will actually be 178.71 Mhz or (DDR357) so it's almost 50Mhz difference.

I disagree on the note that winrar resembles other real-world applications, even games.

While it happens to correlate with machine parameters when comparing it to some gaming benchmarks, these gaming benchmarks are usually those where people erronously try to "get the GPU out of the equotation" by running an older game at low video quality - at FPS number well above 150, even up to 300.

While such a high-FPS game benchmark is still a game run, it has no resemblance to another a modern game which is fps-challenged on today's CPUs will do.

Because at these low CPU loads any game basically turns into a useless memory bandwidth benchmark from shuffling its textures and other large data around. More modern games are entirely different as they do more actual computation, and on the grphics side they now push shaders, they don't suffle much more data around that the older games.

What would be excessively useful would be take the Quake 2 engine and make a variant which actually omits everything to do with the graphics cards, so that you can benchmark the code that makes up game mechanics, physics and AI. Or lets say me still move the graphics data round by to a specific place resembling a sink.