RAM Timings

[congo]

Hi Everyone,

I use a crude method to determine ram timings (other than spd settings) and then it takes me a long time (trial and error) to test variations to try and get the fastest stable settings.

Does anyone know if there is a universal, or even genre/chipset specific method to resolving the optimum performance memory timings for a particular system?

I am looking for a general method that can be applied at various and non standard clock speeds for overclocked systems as well as those with default settings.

[Nick N]

Hi Everyone,

I use a crude method to determine ram timings (other than spd settings) and then it takes me a long time (trial and error) to test variations to try and get the fastest stable settings.

Does anyone know if there is a universal, or even genre/chipset specific method to resolving the optimum performance memory timings for a particular system?

I am looking for a general method that can be applied at various and non standard clock speeds for overclocked systems as well as those with default settings.

Sounds to me like you are using the method that works the best. The problem is mainly in the diverse design specs for all the hardware involved in the system. Vdoop and other factors that a motherboard may place on CPU and memory stability.

One of the things I may do past finding SPD is to research the chips on the sticks themselves because in many cases you can find out if the SPD timings are being realistic, or if they are being conservative which would allow a much greater potential for clocking them up on tight numbers.

The ATi x800xt series with samsung memory chips were like that. The spec from Ati was 500MHz but the chips would easily do 600-625 under the same timing conditions. RAM modules are the same. If the manufacture designed a good stable circuit board and supporting components, finding out what chips are on the modules and matching the SPD numbers is a good way to know if they will clock like mad or crash shortly past spec.

[congo]

Thanks NickN,

The reason I asked is not only for myself, a lot of people with current intel chipsets are having instability problems with their overclocks. This is probably mainly due to inexperience, but there are some people who should be getting better results than they are.... ie. random instability issues. I think a lot of it might have to do with ram timings on cheap ram. There seems to be a mindset that thinks cheap generic ram is fine for high overclocks on core2duos as long as they use their dividers properly. However, people are having problems with supposedly good ram like high spec OCZ and Corsair modules with cas4 at 800mhz.

My theory is that the SPD chips, bios or a combination of both is causing a range of symptoms including shutdowns, blue screens and outright crashes.

Does anyone know if BIOS needs to somehow support spd settings or is spd completely independent of bios? I'm trying to rule out bios as a cause if the problem is memory error related if that makes any sense.

I have been advising some of these people to get spd timings at the ram's rated speed, and lock those timings for use with the optimum divider, when their final ram speed will be close to the rated speed. This is because I noticed on my particular board, that if the divider is changed, the timings on the auto setting seem to go haywire, I thought this might be occuring on their systems, but their bios does not allow them to see these effects, and their timings may be set inappropriately.

Any ideas?

[Nick N]

One thing about motherboards and BIOS

[Nick N]

There is another factor involved with PSU

[Nick N]

The MEMORY BUS will NEVER talk to the CPU at the same speed the CPU is able to chat, THEREFORE the best clock for a system is based on the speed of the FSB/Memory in relation to the CPU.

For AMD64, the architecture of the memory controller is different from that of Conroe. With AMD, it is usually best for a high clock to run a divider ratio and run the FSB up as high as possible, on the LOWEST memory timings WITH 1T CMD enabled. That is not always the case and with AMD it requires a much better defined memory product to produce a clock that will give the system its best PUNCH. Sometimes running a divider is not as efficient however with AMD I find that is usually not the case. AMD64 also responds better to lower FAB-4.

With Conroe, the architecture of the memory controller dictates the MEMORY/FSB want to stay in a 1:1 relationship with the CPU and NEVER divided. Because of that, memory timing OTHER THAN CMD RATE is not important and very high FAB-4 is fine with no significant loss to the bus. Lower stick timing is always better in any system but with Conroe the PUNCH comes from keeping the DDR2 and the FSB in this relation: FSB/4 (LTD 4x) = MEM x 2 (DDR2)

Example for DDR2 667 memory: set 1333FSB and Memory speed to 333 with an LTD of 4x and you get a 1:1 relation to what ever the CPU is running for a multiplier, no matter what. 333.5x2=1333/4x x2= DDR2 667

333.5x2 = DDR2 667 the formula is balanced on both sides of the equation and therefore 1:1

DDR2 800 memory: set 1600FSB and Memory speed to 400 with an LTD of 4x and you get a 1:1 relation to what ever the CPU is running for a multiplier, no matter what. 400x2 =1600/4x x2=DDR2 800

400x2 = DDR2 800 is balanced on both sides of the equation and therefore 1:1

In the case of the northbridge with DDR2 800, you may need to raise or drop the LTD because of the limits the chipset may apply. In that case the formula used is still the same. The FSB and memory numbers you are looking to plug into that formula must still balance on both sides of the equation with the LTD being raised or lowered to provide a FSB value the northbridge can handle. As long as MEMx2 = FSB (LTD) x2 = DDR you are always running 1:1 with the CPU.

The final Conroe GHz speed is determined by the processor being used and its internal multiplier. At that point, when instability is seen during FSB/MEM test checks, raising northbridge, memory and Vcore voltages is the next step, then applying cooling to the northbridge, memory clock gen IC and the CPU (if a higher Vcore is needed) until the highest stable clock is found.

The advantage to having control of the multiplier, a motherboard that supports that control and not buying a CHEAP Conroe has it advantages well over AMD using the formula I provided. Once the formula is applied to Conroe, you can then start to work with the multiplier and further gain as long as you DO NOT exceed the heat/performance ratio ANY processor will exhibit once it hits a certain point, EVEN IF the CPU is well within the visually displayed temp spec. With AMD having multiplier control is CRITICAL. Without it, you are stuck.

When it comes to the FAB-4 and other timings I will use a combination of the SPD data and information gathered from the engineering support for the sticks (I would never buy sticks without that support) and if I had to resort to looking it up myself I would remove the heatsinks, get the numbers off the chips and look up the JEDEC and manufacture specs for them, making my own assessments as to how to approach clocking them up or raising the timings. Without electronics training that can be difficult for someone to accomplish.

With AMD, it

[Nick N]

Because I had access to all the engineering data which included the Corsair information and analyzed all the components in his system, I was able to assess the problem with memory stability for this person, which is in fact, the PSU:

Post#3 on this page:
http://www.simviation.com/cgi-bin/yabb2 ... 157609/315


Corsair stated for the sticks in his system, the advertised voltage of 1.9v was too low and to run them at 2.0v-2.1v with Asus motherboards


If you do not have access to a factory rep with the better memory products, go to the HOUSE OF HELP, the official Corsair forum and do searches for the EXACT memory product having the issues, same with OCZ at bleedingedgesupport. You can usually find out if they are aware of issues in which they will provide updated settings to combat.



That

[congo]

OMG, I didn't expect that tome of info NickN. Thankyou very much for sharing that most interesting insight.

[Nick N]

[quote]OMG, I didn't expect that tome of info NickN. Thankyou very much for sharing that most interesting insight.

[Nick N]

As you will see in that BIOS list, that is the "science experiment' phase, where I apply the divider and the SPD information to the BIOS manually in a mild clock test, then test the system to see if it is indeed stable. If it is, and if the PSU has been confirmed, then I have a basis in theory to continue the clock and timing changes.

If it is not stable, then I have to figure out why, be it the memory product, the timings being applied to it, the motherboard or the BIOS being used. Once that is sorted out and I fix the problem, then and only then can I start the clock-up and timing changes.

The rest is pure speculation because every system has its own personality, so to speak. The best clock is accomplished by learning the personality of the system. Strange anaolgy, but very true. If I may make a change in the future, be it a BIOS upgrade or the memory/CPU product, since I know the system it stands to reason I will know why something may fail or succeed after such a change.

Its important after a major hardware upgrade or BIOS update to start at the 'science experiment' level again. By doing so you can see where changes have taken place be them good or bad. In some cases poor stability with similar settings may not be bad. It may simply require a change to memory timing which then allows an even higher FSB/MEM clock to be succesful.

[Nick N]

As a follow up to what I said about clockers and Conroe, this is a perfect example of what I said about how they make the #1 mistake:

http://madonion.com/forum/showthread.php?t=39859

CLUELESS

Is not about GHz until the best memory speed, timing and FSB have been maxed at a communication rate of 1:1, THEN, and ONLY THEN does the multiplier come into the mix and if it can be raised for a higher GHz it will provide another level of boost.


With AMD it is exactly the same EXCEPT because of the memory controller of the A64 a divider INCREASES the ability to raise the communication rate between the memory and CPU without degrading the real world performance, it has the opposite affect. Unlike Conroe thought you do want the highest CPU GHz you can get out of it.

Once the processor reaches its critical internal point of breakdown, higher GHz RETARDS performance depending on the CPU technology being used. Just because the CPU is running at just under the thermal breakdown barrier and is not overheating does not mean it is doing its job the best it can. Typically, that needs to stay at the very least 15-20c under the max temp level. AMD can run a touch hotter but still must remain well below its breakdown point.

[congo]

Awesome NickN!

Invaluable info there that will save a lot of people time and headaches.

Thankyou very much for taking the time to respond in such detail.

[Nick N]

Awesome NickN!

Invaluable info there that will save a lot of people time and headaches.

Thankyou very much for taking the time to respond in such detail.

I got the email and I see where you are having difficulty. Without access to the higher level settings in the BIOS and knowing they exact manufacture of memory with defined specs on it, what you are dealing with is tough.

I don't have time to post back in detail but I will think it over and re-read what you sent.

Do you have model # for the memory you are running or trying to get locked down? EDIT: PDC2G3500LLK

[Nick N]

I just looked over the motherboard manual, as a clocker you are screwed with that BIOS. Limited voltage controls, no control of the bridges, no multiplier, .. I am amazed you have not gone insane trying to work with it.

[congo]

Muahaha, this is gonna be fun!