Best RAM make for NF7-S?

[Knoxville]

Yeah but i can't go higher with my fsb speed as the FN41 board loves to complain about it So i just tightened the timings.

[TooNice]

Thats what I've been wondering.

I know that prior DDR, CAS2 PC133 RAM was all the rage. People wouldn't even consider CAS3 if they wanted to OC.

PC2100 came and the first ones were all CAS2.5.

Now I see some CAS2s around, but with ~£10+ premium over CAS2.5. Is there anything about Dual Channel which would make the 0.5 difference less significant?

[KAIN]

Erm Austin, do you know anything about speed binning and QA testing on CPU's GPU's?

The 9500 NP's that were released that would mod to 9700 Pro's were the lower quality card's. If you look none of them reach the same overclock's as the 9700 Pro's. They are different cards... They may be the same physical layout, but the actual quality of the silicon on the cheaper one is much lower.

Same applies to 4200/4400, expcept not quite cause NVidia's binning at that time went a little AWOL.

I know what you are saying, often in times of shortage higher specced parts are sold as lower specced parts. 2.26B's, XP1700+'s, Celeron 667mhz etc. But normally the case is the lower models are not capable of running the higher speed's at the defined tolerances, for example running their rated speed at a temperature of TMax-5% etc. There are many ways to test the capability of a CPU, if it doesn't pass the stringent test's it means it will most likely fail earlier than a CPU that does at that speed.

But enough off topic and very boring subject

I'm not disagreeing completely, just generally parts that exceed rated speeds it is more through luck than judgement. And will do immeasurable damage even at stock voltage, I know I've killed many Graphics cards using stock voltages. CPU's do have much higher tolerances due to their manufacture, but they still are reduced significantly

[Austin]

Last time I looked into latency was back in the XP2000+ type days and it showed in real usage that the perf diff between 2.0-2-2-5 and 2.5-3-3-7 was only around 1%, hardly worth the effort as RAM can be hard to certify 100% stable when o/c'ed. Biggest reason for lower latency RAM being popular is so you can raise the latency and often run at a faster speed (in mhz).

Dual Channel lessens the effects of latency as that is a large part of how Dual Channel speeds things up, by managing the RAM channels intelligently in can largely remove the time taken for RAM to refresh for example. It is especially so for SktA mobos as the extra bandwidth Dual Channel offers is almost totally useless, much of the speed gain comes from eliminating latency. I hope I explained that okay ... it's late (well early).

KAIN we can agree to disagree, at least we understand each others points and have stated our cases.

[Agent]

Originally posted by TooNice
Is there anything about Dual Channel which would make the 0.5 difference less significant?

Nope.
Dual DDR has a default CAS of 2.5 regardless of what your RAM's timings are. You can change it to a CAS of 2, but this is overclocking the Dual DDR bus. The performace gain of this would be small, although of course if its free, have it

[TooNice]

So, how overclockable are those CAS2.5 Twinmos? Is it quite common for people to run it at CAS2?

Also, what are all the other values?
(e.g. 2.0-2-2-5 and 2.5-3-3-7)

[Knoxville]

I had a gig of TwinMos PC2700 CAS 2.5 running at 2.0-2-2-6 100% stable for ages before i traded those sticks in for my mushkin 3500

[Austin]

TwinMOS is well known to be very o/c'able both in latency and pure speed (mhz). As for latency ...

The CL rating (Cas Latency) is usually quoted with the designed speed of the RAM (eg 333mhz for PC2700) but this and the numbers which (should) follow can be daunting.

RAM is effectively arranged in rows and columns (much like a spreadsheet) so then CAS is Column Address Strobe, RAS is Row Address Strobe. The RAS, CAS and the RAS-to-CAS latency are basically the time spent waiting to locate, read or write a given area of memory but if a consecutive block of memory is needed the significance of any latency is pretty minimal, so latency is of greatest importance when using memory randomly.

The problem is all that's often quoted about latency or RAM timings is the CL number, it's the most important but less than 1/5th of the story. If you line the timings all up you can state a more complete picture. You should get quoted CL2.0 as being 2-2-2-5 (but could also be 2-3-3-7) and CL2.5 as being 2.5-3-3-7. The 1st number is Cas Latency, 2nd RAS Precharge, 3rd RAS-CAS, 4th Act Precharge (IIRC). BIOS' may use diff notations but that's basically the principle and lower is always faster as it means less waiting. Other things to look out for are Command: T1 (faster) or T2 (normal) and Bank Mode: off, 2 (fast) or 4 (faster). Anything to do with Memory holes, caches (not L1 or L2), shadowing and the like should be disabled. AGP Aperture is best at 256MB but there isn't (technically) really a perf hit in going lower, it's only assigned if the system has it free and your gfx card's own RAM 'runs out'.

If adding a new DIMM to an existing DIMM rem that the limit and timings are dictated by the slower DIMM. 256MB DDR333 CL2.0 coupled with 256MB DDR266 CL2.5 will both effectively be limited to DDR266 CL2.5. Cas Latency should be selectable in the BIOS, SPD is automatic, if you don't find many settings for latency then try the latest BIOS or CTRL+F1 in the BIOS on Gigabyte mobos.

I'm not saying any of the above is 100% definitive by any means, just what I've gathered from notes I've made coupled with experience, should be somewhere near the mark. Please note again that precise notation varies.

[Agent]

Dammit, i need to mod this board soon, 225mhz @ 7-3-3-2.0 isnt good enough
Why cant this board only supply 2.9 volts to the RAM ? ARHGG
Same for the chipset - 1.7 volts ?. Bleh, i could fart out 1.7 volts

Wish i had the cash to go upto a gig - i had a gig running at 200mhz and that made more of a difference than i thought. Games loaded quicker, lot less swaping in / after games, and those few annoying jerkys in Ut2k3 went away.

Maybe after chrimbo..

[Knoxville]

A gig of 3700, you greedy buggeh

btw, i'd love to see you fart out 1.7v

[Agent]

Thing is, im looking at £125 for a 512meg stick thats branded PC3700, minimum

[KAIN]

Originally posted by Austin
Last time I looked into latency was back in the XP2000+ type days and it showed in real usage that the perf diff between 2.0-2-2-5 and 2.5-3-3-7 was only around 1%, hardly worth the effort as RAM can be hard to certify 100% stable when o/c'ed. Biggest reason for lower latency RAM being popular is so you can raise the latency and often run at a faster speed (in mhz).

Erm, nope 1% isn't true. Try SETI bench is the best proof in the pudding for this although pifast is quite good at it as well

Will just run a quick test

[KAIN]

Ok ran the test

2500 (12.5*200 2-2-2-5)



2500 (12.5*200 2.5-3-3-7)



That's only on a short test, I could do 3dmark and SETI but I can't be bothered this second

Looks around 7% to me, which is nothing to complain about merely from changing some timings.

[Agent]

Remember, pi-test is mainly CPU based, not FSB

[KAIN]

Not sure what you mean by that But yeah it is not hugely affected by memory speed so a 7% difference on something like that is actually quite significant

[Austin]

Exactly why I find synthetic tests interesting but pointless. 3Dmark can at least offer some kind of real world perf evaluation but the likes of SETI and esp PIfast are pointless IMHO. Games are what's important to most people, that the main reason for sinking extra cash into a PC and the gains there are very small. Anyway this is all I could find from an independent source (I'd expect the diffs to be even smaller on Dual Channel setups) ...

http://www.tomshardware.com/mainboar...507/index.html