3rd 2GB Kit, we have a winner!

Dirk

Dirk

Dirk

Associate
Joined
20 Dec 2005
Posts
1,932
The first kit was Corsair 2x1GB XMS3500LLPro. Infineon chips on this RAM meant good timings but didn't benefit from increased voltage. Maxed out @ 245 2.5,3,2,6 2.7v. Not going to get close to going 1:1 to 300 HTT :p

The second kit was Corsair 2x1GB XMS4000pt. Samsung UCCC chips have looser timings than Infineon but i saw reviews with this RAM running over 290Mz so i wanted to try it. This RAM didn't want to budge off it's stock speeds. Would bench at 255Mhz but would only be Prime stable on stock 250Mhz. Voltage all the way to 3.0v didn't help tighten the timings and achieve Prime stablity even at stock. Maybe a hint of that whole DFI/Corsair incompatability question mark?

All the previous o/clocks had benched ok. The 3500 pair couldn't get me to 3Ghz without running them near stock speed on a divider. The 4000 pair got me to 3Ghz but only at 250Mhz 3,4,4,8. To get great benches on air cooling i wanted RAM that would go 300Mhz at good timings. So.....

Crucial Ballistix. This article decided it for me....
http://www.anandtech.com/memory/showdoc.aspx?i=2676&p=5

300x10Crucial_Benches.jpg



300x10Crucial_3DMark06.jpg






...and one of the previous best with the Corsair memory..




254x12Corsair_Benches.jpg



Would have like to have tried Mushkin Redline 4000 2x1GB but availability is poor. The Crucial Ballistix performs better but has that reliability question. According to the AnandTech article...
..."some users have been killing their Crucial Ballistix 2GB kits. It appears that the culprit is high memory voltages. We did a survey of a number of Forums and found that, in most cases, the memory was dying at voltages of 2.85V and higher. Our advice, which we also followed in benchmarking for this review, is to keep voltages below 2.85V with the 2GB Crucial Ballistix kits."

Haven't pushed my luck so far with a one shot benchmarking run. I need better cooling to go beyond 3.06Ghz. Needed to loosen timings a bit once i reached 290, to be able to boot, from 3,4,4,8,6 1T to 3,4,4,10,8 1T 2.8v. Good as gold on 300Mhx now. :D
 
Wow, i'm really surprised at the poor results from the 3500LLs. I've got mine stably to 280 3-4-4-8 at 2.8v and 270 3-3-3-7. I still think it's better to go for the lower FSB and tighter timings with a divider, but there are still some 1:1 options there (though not for my opty 144).
 
Well, as ever, mileage varies between identical parts. The cheaper option would have been to try a few BIOS Revisions to see if there is one that would have broken the stranglehold.

I've always been an advocate of Corsair memory as it's good quality, reliable memory. There is truth however in saying that they are not the best overclockers.

http://www.anandtech.com/memory/showdoc.aspx?i=2676
 
You're quite right, it's got a good reputation for reliability and stability. But they have come on leaps and bounds in the overclocking department too since the AMD64. On an Intel platform they will probably lag behind Mushkin, OCZ, Patriot and quite possibly ballistix and G Skill too, but on a A64, they right up there at the moment. At least if my stick are anything to go by.
 
On reflection it does occur to me that maybe there is as much an affinity between Corsair and Asus board (they certainly perport to suit onanother well) as there is a lack of affinity between Corsair and DFI boards.

Just a thought.
 
Ah, screw it. I don't know what the rules for these Corsair are. I just know they kick ass in my rig and I still think i've got more to squeeze out of them.

Street, you should try dropping your timings to 2.5-3-3-5. So far i've found 5 to be just as attainable as 6. But i am using 2.9v at the moment.
 
I've found mine dont respond well to more voltage... If i put more through them, they start throwing up errors in memtest:(

Ill give 2.5-3-3-5 a go... back in a sec:)
 
i haven't experimented much with voltages, but i did find that 236 at 2-3-2-5 was stable at 2.9v but not at 2.8v.

I have been told though that these Asus boards require you to set the voltage on the ram a tad higher than with other boards. At least that's what the tech guy ot OCUK told me when I bought it.
 
Last edited:
Hey Dirk,

Could you do a 3DMark '06 bench with the timings tighter and a harsher divider please?

I'd be intrigued to see how it responds to the decreased latency, but decreased bandwidth.

Can it do 2-2-2 at all?

300 Mhz @ 3-4-4 has roughly the same latency as 2-2-2 does @ 164 Mhz :eek:
 
Last edited:
Ive just memtested my ram at 2.5-3-3-5, it all works fine at 250MHz with 2.6V, no errors what so ever:)

Ive found that knocking the voltage up to 2.8+ makes my ram not as stable, 2.6 and 2.7 seem fine though.
 
I'm glad dropping to 5 worked, it seems to be the surprising little trick of this ram.

I may have to give lower voltages a go when I get home too. I'll also have to start memtesting again. I've tended recently to use only superpi and prime to test stability. Maybe these great times i can get won't stand up to memtest.
 
Jimbo Mahoney said:
Hey Dirk,

Could you do a 3DMark '06 bench with the timings tighter and a harsher divider please?

I'd be intrigued to see how it responds to the decreased latency, but decreased bandwidth.

Can it do 2-2-2 at all?

300 Mhz @ 3-4-4 has roughly the same latency as 2-2-2 does @ 164 Mhz :eek:
Click to expand...


I'll test it when i get time but i can tell you already that it scales in the same way as......

Clipboard01.jpg




3DMark06 only gives you a general CPU score and that scales mostly with CPU speed. But as it does count toward the score, does improve with RAM bandwidth and with every point counting it does matter.

Previous o/c 254x12 Corsair scores-5205 (1185CPU)
Now-5215 (1192CPU)

http://service.futuremark.com/orb/projectdetails.jsp?projectType=14&projectId=35152


At 292x10 5/6 243 @ 2.5,3,2,6-51643DMarks(1160)

http://service.futuremark.com/orb/projectdetails.jsp?projectType=14&projectId=2311
 
Dirk said:
I'll test it when i get time but i can tell you already that it scales in the same way as......

Clipboard01.jpg
Click to expand...

OK, thanks.

I am still convinced that for most applications, memory bandwidth has very little effect on performance. However, I am NOT saying that it is pointless to try to get the highest bandwidth possible. What I AM saying is that, in my experience (perhaps it's because I only clock low-end CPUs like the Venice 3000 and Opteron 144), running the RAM at a very high clock stresses the on-die memory controller and reduces the highest stable CPU clock. Under this circumstance, sacrificing CPU clock speed for memory bandwidth is a very bad thing to do (in almost all applications).

Another point is that just increasing the CPU clockspeed increases the memory bandwidth...

I'd be intrigued to know how the higher-end CPU's memory controllers fare.

I also DO NOT want to seem as if I am arguing with you (I am not) but I don't want everyone to go off and clock their memory like crazy, sacrificing the CPU clock speed because it will reduce overall performance.

e.g. Keeping the clock speed @ 2.4 Ghz in your table, when the memory speed is increased 50% (from 200 Mhz to 300 Mhz), the Quake 3 FPS only increase some 3%, the Wolfenstein only 2%. The memory bandwidth has increased some 15%.

I guess my experience has been this:

high memory clocks = reduced CPU clock

large amount of memory = reduced CPU clock

Maybe with the higher-end processors, another part of the CPU is the weak point?

A good read here:

http://www.neoseeker.com/Articles/Hardware/Guides/athlon64oc/

My findings have been the same.
 
According to their website they are not in production any more. That's not to say they won't be again or that you can't find any back stcok, but with their record it's just not worth it to keep producing them.
 
man_from_uncle said:
That's an interesting and useful theory Jimbo. I'll try dropping back to a 133 divider and see if i can get stable at 2.8ghz (which so far has just illuded me).
Click to expand...

Drop your LDT multi too.

I always use a 2x.

Even using 1x has less than 1% performance impact (check the link I gave), but I've found lower multis help stability.
 
You must log in or register to reply here.
Back
Top Bottom