Who says 4GB can't be fast?

[Deathspawner]

Last month, I took a look at some 4GB DDR2 action, but compared to 2GB it just wasn't that interesting since they both ran in 2T. I figured it was time to pay some attention to DDR1, since I've been neglecting it for a while.

/enter 4:00am overclocking session

The modules I used for testing were Corsair 3500LL PRO. Some oddities during overclocking arose, but I will tackle those shortly.

These are obviously DDR438 modules, so overclocking was going to prove fun. They use Infineon BE-5, and after recent Infineon CE-5 hassles, I was almost worried. However, I was impressed to see that these clock just as well, if not better than the 4400PRO kit currently out.

My first goal was to see if the max OC on 2 * 1GB could hit the same max overclock of my 4400PRO, and they did. I didn't bother running benchmarks at the time, because I wanted to see if 4 * 1GB could equal that OC. Which it did..

4 * 1GB 275MHz 3-3-2-0 2T 2.8V (Valid)


After a few hours of testing and tweaking, I found the max stable overclock to be:

4 * 1GB 284MHz 3-3-2-0 2T 2.8V (Valid)


I consider that setting to be stable, after running a 5 hour Prime95 test, a one hour loop of 3D Mark '06 and a one hour real time play of Half-Life 2 and Battlefield II.

Here's where things get strange though. For some reason, at those max settings, 2 * 1GB will cause 3D Mark '01 and Half-Life 2 to freeze at various points, but only for a few seconds. Once the extra sticks are popped back in though, there are no stability problems at -all-. Even changing the 2 * 1GB config to use 2T timings did not help anything, so I have a little bit more of tweaking to do

MemTest Stable:


Ignore A64 Tweaker in that picture. The second picture above has the correct values for each setting used. I am going to tweak a little more to see if I can boost the settings higher for either config, but I am very pleased with the results. I should also mention, during all this testing, -no- sort of active cooling was used. I added a 120mm fan this afternoon, but it didn't seem to do anything for performance.

Here are some quick 2GB vs 4GB results.

Everest 2.80 219MHz 2-3-2-6: 6756/6363 vs 6613/4533
Sandra Unbuffered 219MHz 2-3-2-6: 2986/2988 vs 2605/2605
Sciencemark 2 219MHz 2-3-2-6: 5728.01MB/s vs 4716.73MB/s

Everest 2.80 250MHz 2.5-3-3-6: 7740/7262 vs 7494/5144
Sandra Unbuffered 250MHz 2.5-3-3-6: 3310/3392 vs 2874/2891
Sciencemark 2 250MHz 2.5-3-3-6: 6488.28MB/s vs 5328.94MB/s

Everest 2.80 284MHz 3-3-2-0: 8720/8256 vs 8550/5860
Sandra Unbuffered 284MHz 3-3-2-0: 3310/3392 vs 3249/3291
Sciencemark 2 284MHz 3-3-2-0: 6488.28MB/s vs 6054.58MB/s

Hope you found this post interesting. I am back to tweaking...

[s7e9h3n]

Last month, I took a look at some 4GB DDR2 action, but compared to 2GB it just wasn't that interesting since they both ran in 2T. I figured it was time to pay some attention to DDR1, since I've been neglecting it for a while.

/enter 4:00am overclocking session

The modules I used for testing were Corsair 3500LL PRO. Some oddities during overclocking arose, but I will tackle those shortly.

These are obviously DDR438 modules, so overclocking was going to prove fun. They use Infineon BE-5, and after recent Infineon CE-5 hassles, I was almost worried. However, I was impressed to see that these clock just as well, if not better than the 4400PRO kit currently out.

My first goal was to see if the max OC on 2 * 1GB could hit the same max overclock of my 4400PRO, and they did. I didn't bother running benchmarks at the time, because I wanted to see if 4 * 1GB could equal that OC. Which it did..

4 * 1GB 275MHz 3-3-2-0 2T 2.8V (Valid)


After a few hours of testing and tweaking, I found the max stable overclock to be:

4 * 1GB 284MHz 3-3-2-0 2T 2.8V (Valid)


I consider that setting to be stable, after running a 5 hour Prime95 test, a one hour loop of 3D Mark '06 and a one hour real time play of Half-Life 2 and Battlefield II.

Here's where things get strange though. For some reason, at those max settings, 2 * 1GB will cause 3D Mark '01 and Half-Life 2 to freeze at various points, but only for a few seconds. Once the extra sticks are popped back in though, there are no stability problems at -all-. Even changing the 2 * 1GB config to use 2T timings did not help anything, so I have a little bit more of tweaking to do

MemTest Stable:


Ignore A64 Tweaker in that picture. The second picture above has the correct values for each setting used. I am going to tweak a little more to see if I can boost the settings higher for either config, but I am very pleased with the results. I should also mention, during all this testing, -no- sort of active cooling was used. I added a 120mm fan this afternoon, but it didn't seem to do anything for performance.

Here are some quick 2GB vs 4GB results.

Everest 2.80 219MHz 2-3-2-6: 6756/6363 vs 6613/4533
Sandra Unbuffered 219MHz 2-3-2-6: 2986/2988 vs 2605/2605
Sciencemark 2 219MHz 2-3-2-6: 5728.01MB/s vs 4716.73MB/s

Everest 2.80 250MHz 2.5-3-3-6: 7740/7262 vs 7494/5144
Sandra Unbuffered 250MHz 2.5-3-3-6: 3310/3392 vs 2874/2891
Sciencemark 2 250MHz 2.5-3-3-6: 6488.28MB/s vs 5328.94MB/s

Everest 2.80 284MHz 3-3-2-0: 8720/8256 vs 8550/5860
Sandra Unbuffered 284MHz 3-3-2-0: 3310/3392 vs 3249/3291
Sciencemark 2 284MHz 3-3-2-0: 6488.28MB/s vs 6054.58MB/s

Hope you found this post interesting. I am back to tweaking...

Nice work man

[Vassili]

Hey s7e9h3n, do you have to quote his post? Reading the results once is enough for me...

But indeed nice results, but 284MHz 3-3-2-0 3D(mark) stable is very, very nice!

What OS are you using? Win 2k3?

[Punisher!]

LOL man! Really really good!

[±Pink=FloYd±]

hmm... don't think 4GB of ram will look good on the cc bill lol
very nice job tho!

[Getttosmurf]

me wants 2 more GB Great oc Deathspawner

[Deathspawner]

Hey s7e9h3n, do you have to quote his post? Reading the results once is enough for me...

But indeed nice results, but 284MHz 3-3-3-2 3D(mark) stable is very, very nice!

What OS are you using? Win 2k3?

3-3-2-0

I am using XP 64-Bit (x64). It doesn't have the 3GB bug that the 32-Bit version of Windows has.

Thanks for the comments guys

[CHaynes112]

Was it 3DMark stable at 275MHz and 284MHz? I thought a lot of people were having trouble getting OCZ Plat EBs 3D stable above 265-270?! Are these Corsair modules just that much better or do you just have a set of monster overclockers?!

[Hystrix^]

Amazing sticks you got there

[bachus_anonym]

Whoa, nice 284MHz Rob Game stable too... What volts did you have to go with for that? More compared to 2x1GB?

[Deathspawner]

Was it 3DMark stable at 275MHz and 284MHz? I thought a lot of people were having trouble getting OCZ Plat EBs 3D stable above 265-270?! Are these Corsair modules just that much better or do you just have a set of monster overclockers?!

Yes, 3D Mark was stable at both settings. Though, at 284MHz, 3D Mark was -only- stable with 4GB, not 2GB. I know it makes no sense, but I am trying to figure out why At 275MHz though, 2GB worked fine throughout everything.

Whoa, nice 284MHz Rob Game stable too... What volts did you have to go with for that? More compared to 2x1GB?

Thanks man No more voltage was needed at all... the entire time the modules were at 2.8v. They may actually work with -less- than that, but I never moved it anywhere.

[celemine1Gig]

Yes, 3D Mark was stable at both settings. Though, at 284MHz, 3D Mark was -only- stable with 4GB, not 2GB. I know it makes no sense, but I am trying to figure out why At 275MHz though, 2GB worked fine throughout everything.
...

Well, perhaps the chips just need the full two clock cycles (2T) at those high fequencies, regardless, if they're working in 2GB or 4GB configuration. Could be due to the specific chip design. Anyone else tested his set at 2T when it was memtest stable at a certain speed, but not 3d-stable?

[The Byter]

great oc there

what mobo is it ?

[Deathspawner]

Well, perhaps the chips just need the full two clock cycles (2T) at those high fequencies, regardless, if they're working in 2GB or 4GB configuration. Could be due to the specific chip design. Anyone else tested his set at 2T when it was memtest stable at a certain speed, but not 3d-stable?

As said in my original post, "Even changing the 2 * 1GB config to use 2T timings did not help anything, so I have a little bit more of tweaking to do." It still makes no real sense to me.

The Byter, it's the Ultra-D using the 704BTA BIOS

[Makubex_GB]

Very nice results!

[The Byter]

As said in my original post, "Even changing the 2 * 1GB config to use 2T timings did not help anything, so I have a little bit more of tweaking to do." It still makes no real sense to me.

The Byter, it's the Ultra-D using the 704BTA BIOS

I didn't know the ultra-d can run 4 dimms... and that high... stable

great job
10x for the answer

[Deathspawner]

I didn't know the ultra-d can run 4 dimms... and that high... stable

Yes, I was quite impressed really. I figured if anything would get bottlenecked, it would be my CPU, but it was rock stable. I think what is most interesting, is that lack of any cooling on the modules. That's a massive overclock to be THAT stable without any fan on them.

[Punisher!]

Yes, I was quite impressed really. I figured if anything would get bottlenecked, it would be my CPU, but it was rock stable. I think what is most interesting, is that lack of any cooling on the modules. That's a massive overclock to be THAT stable without any fan on them.

Yeah, impressive is how it can be stable! Ultra-D is really one of the best board ever.

[dnottis]

3-3-2-0

I am using XP 64-Bit (x64). It doesn't have the 3GB bug that the 32-Bit version of Windows has.

Thanks for the comments guys

That 3GB bug is only a SP2 bug, SP1 will address all 4GB - but SP1 wont see all 300GB of my Maxtor HD lol

Win some, you lose some.

[Deathspawner]

Dean,

Yeah, that 3GB bug is rediculous. I have a friend who had 4GB in his computer for a while, and he didn't know -forever- that he was only actually getting to use 3GB of it, hah.

"SP1 wont see all 300GB of my Maxtor HD"

Now, THAT is wierd.

[elec999]

Very nice
Thanks

[pcmoddingmy]

Nice one there, really..
But don't these 3500LLPROs perform worst at higher VDIMM?
My highest stable (25hour Prime) with just 2.4V at 250MHz (2.5-3-2-5-1T):

Maybe I'll get back to my 3500LLPRO once I'm done with the PC4400PRO.
Btw, my 3500LLPRO was v1.2 with batch code of 0542.

[Deathspawner]

Nice one there, really..
But don't these 3500LLPROs perform worst at higher VDIMM?
My highest stable (25hour Prime) with just 2.4V at 250MHz (2.5-3-2-5-1T):

Maybe I'll get back to my 3500LLPRO once I'm done with the PC4400PRO.
Btw, my 3500LLPRO was v1.2 with batch code of 0542.

2.4Vdimm? I don't even think my Ultra-D goes lower than 2.5v. I keep it usually at 2.8v all the time. They may work lower, I just haven't bothered. May give it another shot in the coming days.

As mentioned though, my OC was not THAT stable with 2GB, but was rock solid with 4GB. It's a mystery I still haven't figured out.

[celemine1Gig]

2.4Vdimm? I don't even think my Ultra-D goes lower than 2.5v. I keep it usually at 2.8v all the time. They may work lower, I just haven't bothered. May give it another shot in the coming days.

As mentioned though, my OC was not THAT stable with 2GB, but was rock solid with 4GB. It's a mystery I still haven't figured out.

If you did Vmods before, it shouldn't be really hard to get that Ultra-D under the 2.5V mark. If I'm not mistaking there already is a vdimm-mod how to out there for your board. Changing the mod slightly should do the trick. Even my ECS Nforce4-A939 was easily modded to go as low as 2V or lower.

[Deathspawner]

I've done the VMod on my board before, it allows it to go up to 4.0v. It may have allowed lower also, I'm not sure. Using the Vmod on the Ultra-D turned very ugly for the 4400PRO kit in my experience though. The computer would not even POST. Undoing the mod, it worked fine.