Will downclocking memory to run in sync with CPU improve performance?

[Live]

Hi,

I can?t hit 166 Mhz FSB with my locked 1900XP. Since I don?t wont to unlock it please don?t suggest it. Not hitting 166 MHz FSB means that my memory (166Mhz PC 2700) runs asynchronous with the CPU. Would it improve performance to lower the memory to 133 MHz and then overclock the memory and the CPU to run synchronous? I can reach a FSB of about 142-145.

When does the higher Mhz of the memory start to pay of? I understand that the performance increase of PC 2700 over PC 2100 is about 2%. But what if I overclock the FSB from 133 to lets say half eg 149,5 FSB. Would I then have gained more because of the system running synchronous? Of course the memory will be more overclocked if I keep the PC2700 rating with 166 MHz.

I hope you understand the question, and hopefully some of you can shed some light on the issue.

 

[PliotronX]

Running the memory asyncronously certainly doesn't hurt performance, so what happens when the memory is +33 when the FSB is at 142? If it's an absolute choice between sync mem at 142 FSB or async +33 mem with 133 FSB I'd go with the sync mem at 142 FSB because the FSB overclock takes the whole system up, and widens the memory bus to the CPU for an overall efficiency increase. But something strange is going on here.

What happens when you run at 166MHz FSB? It should run properly given that you lowered the multiplier where your CPU can handle the speed (i.e., you can hit 142 FSB which would be 1704 with a 12x multi so try a 10x multi with 166 FSB).

 

[Smilin]

142 sync -vs- 166 async eh?

I don't know dude, I think it's going to take an experiment to find out. I think the 166 may have the edge but not by much.

 

[Live]

Sorry for not stating it clear the first time. We are working with a locked CPU here. So the usual way to solve this by lowering the multiplier, will not work.

It is stated that the VIA KT 333 and 400 is optimized for asynchronous operation. Yet in many reviews synchronous operation is recommended. My low self esteem makes it almost impossible to experiment for myself since I have a hard time believing my results

 

[CrazySaint]

Running memory asynchronously with CPU on Athlon systems increases memory latencies and doesn't add any bandwidth that the CPU can use, thus performance goes down. Running the memory synchronously with the CPU at 142FSB vs. running the CPU at 133 and the memory at 166, synched at 142 will win hands down. However, at 142FSB you're starting to run your PCI/AGP busses pretty well out of spec and begin to run the risk of damaging some of your components. Since you're unwilling to unlock your CPU, the obvious solution to your problem is to sell your XP 1900+ for ~$75 or whatever and buy a 1600+ from newegg for $53 - you'll end up with a faster (OC'd) CPU, MUCH faster RAM, and PCI/AGP busses in spec

 

[Live]

Originally posted by: CrazySaint
the obvious solution to your problem is to sell your XP 1900+ for ~$75 or whatever and buy a 1600+ from newegg for $53 - you'll end up with a faster (OC'd) CPU, MUCH faster RAM, and PCI/AGP busses in spec

Thanks for your reply

Well I have thought of buying a 1600+ since they are so dirt cheap just to have some fun with.

But this is for me not only about getting the better performance, silly me wants to understand why so any link that could enlighten me would be great.

 

[PliotronX]

There is good article about this here that explains how the Athlon XP's memory bus is starved and the future dual channel DDR setups for P4s won't offer much improvement when run asyncronously like the Athlon XP. Overall I would go with the 142 FSB, but does the RAM crap out asyncronously at 142 FSB? You could try running the mem with higher voltages (2.8v max) and/or more relaxed timings (but depending on the timings this could negate most performance gains).

edit- fixed linky

 

[Smilin]

Originally posted by: Live
My low self esteem makes it almost impossible to experiment for myself since I have a hard time believing my results

You can do it man!!! You are a good lookin fella and I have the utmost confidence in you.

Keep it simple...run quake3 in a really low res to be sure it will be cpu dependent..two passes at each memory setting ought to do the trick.

Remember you are a dashing, handsome person!

 

[Budmantom]

You can do it man!!! You are a good lookin fella and I have the utmost confidence in you.

Sometimes this place cracks me up (and this is one of those times) lol.

 

[CrazySaint]

Originally posted by: Smilin

Originally posted by: Live
My low self esteem makes it almost impossible to experiment for myself since I have a hard time believing my results

You can do it man!!! You are a good lookin fella and I have the utmost confidence in you.

Keep it simple...run quake3 in a really low res to be sure it will be cpu dependent..two passes at each memory setting ought to do the trick.

Remember you are a dashing, handsome person!

But what if Live's a woman?

 

[SupermanCK]

LOL

 

[fluxquantum]

Originally posted by: CrazySaint

Originally posted by: Smilin

Originally posted by: Live
My low self esteem makes it almost impossible to experiment for myself since I have a hard time believing my results

You can do it man!!! You are a good lookin fella and I have the utmost confidence in you.

Keep it simple...run quake3 in a really low res to be sure it will be cpu dependent..two passes at each memory setting ought to do the trick.

Remember you are a dashing, handsome person!

But what if Live's a woman?

heh heh

 

[valhar2000]

I have much the same that Live has: I can only get to 145 MHz FSB, with my Athlon XP 2000+, even though at this speed my processor runs at a nice 33 degrees. I am running the momery asynchronously too, at 181 MHz.

I am going to make some measurements now different settings, and see what results I get.

And, by the way, Live: independent of your self esteem, your computer is an individual, and the setting at which it runs fastest is the setting at which it runs fastest. If your tests give you one result, that is the result that applies to your computer.

Any way, I'll keep you informed.

 

[Necrolezbeast]

I'm not sure, but I think the higher multiplier is going to limit your fsb with both the 1900+ and 2000+. No matter the temp, this chip just wan't made for that high of an overclock, whilke the 1600+ is the bottom of the barrel along with a 10.5x multiplier, so it naturally can handle the higher fsb. valhar, your memory might be crapping out at 181mhz, I don't know what kind you have or what not but that is pretty high.

 

[valhar2000]

I performed some tests with FSB at 133, 145, 145 and memory at 166, 145, and 181 respectively. I used Sisoft Sandra, 3Dmark2001, Povray 3.5, rendering the included chess2.pov scene, and I compressed a 135 meg wav file with ACE and with Lame in very high quality VBR.

Well, here are my results:

FSB: 145, MEM: 181
CPU arithmetic=5145/2525
multimedia=10000/11065
Memory=2236/2156 Mb/s
3Dmark= 8458
Povray= 3:49 s
Ace= 58 s
Lame= 2:54 s

FSB: 133, MEM: 166
CPU arithmetic=4805/2336
multimedia=9312/10323
Memory=2080/2011 Mb/s
3Dmark= 8357
Povray= 4:06 s
Ace= 1:03 s
Lame= 3:07 s

FSB: 145, MEM: 145
CPU arithmetic=5171/2527
multimedia=10000/11068
Memory=2194/2108 Mb/s
3Dmark= 8379
Povray= 3:47 s
Ace= 59 s
Lame= 2:54 s

In the light of these findings, it seems quite clear that it is better to keep your memory and FSB synchronous and overclock both, than to clock your memory at 166 with no further overclocking.
It also seems that keeping the 4:5 FSB:memory ratio at 145 MHz FSB gives an advantage, but it is not very large. I would recommmend that you attempt it, but if your memory cannot take it, I very strongly urge you to raise your FSB to 145 and use a synchrous memory clock frequency.

 

[valhar2000]

Does anyone know where I could find a good guide on how to unlock an ATHlon XP CPU? Given my results, I have become more interested than I used to be in synchronous operation.

My memory is a Corsair XMS PC2700 stick, 256 Mb worth of RAM. It is working very well even at an FSB of 181 MHz, and I suspect it is capable of going higher.

 

[valhar2000]

I meant a frequency of 181 MHz. My FSB is still 145 MHz.

 

[Live]

Who knew this would turn out to be such a entertaining thread. It got me laughing pretty hard

PliotronX just the kind of article we needed. Thanks!

valhar2000 great job with the tests. We need more of this.

I will, thanks to the great support and trust shown in me especially by what seems to be the Jedi master Smilin, do my own test and return here with the results. I can feel the force growing stronger now by the second. But is it the dark force I am being lured into? Only time will tell.

I do believe we are on to something here and tough not a secret it is not often stated in overclocking guides and the likes.

As for unlocking there are many, many guides to unlocking. Just do a google and you will find tons.

Here is one with lots of good pictures to get you started:

www.cluboverclocker.com/guides/amd_unlocking

And here is a video showing you how to:

tomshardware/unlockingXP/video