A64 Memory / Motherboard / Voltage / Speed Questions.

[Lukano]

I know this thread really could be placed in the AMD or Motherboard specific forums - but as this pertains directly to overclocking but generalizes the way the parts operate - I figured I'd stick it in here (the Overclocking forum).

I've got a variety of questions that have been plaguing me in my attempt to overclock my new machine (specs as per my sig below). I'm trying to come to a finer understanding of how the guts behind the machine work, in an attempt to better quantify and tune my overclock.

I understand that with Athlon64's the memory controller is in chip. My current ram is rated 2-3-3-8 @ 2.6-2.8v and 2-2-2-8 @ 3.2v. I've tested and confirmed both speeds with memtest while running at default/stock/auto settings on the chip/motherboard - and am pretty confident that it's fully capable of those speeds.

Yet I've seemingly hit a threshold between 275 and 300 HTT - where either the memory or the memory controller are starting to flake out. This is where and why I need to become better educated in order to make some informed decisions about what I change or tweak in order to increase my stability (currently capable of 300 HTT x 9 multi @ 2-3-3-8 3.2v desktop and gaming stable, but it is not by any stretch Prime stable - which has me confused). I've been able to run the machine at the 300 HTT with relatively tight timings (2-3-3-8) and the machine is "day-to-day" useable (gaming, workload, etc). Lockups or crashes are nearly non-existent, or at least rare enough that further tweaking can fix that - yet Prime95 (using 24.11 beta which fixed a lot of A64 errors I was getting previously) still won't run for more than a few minutes stress-test without giving me hardware, sum, or rounding errors.

So, am I guessing correctly when I assume that as long as ram is running within spec no latency/speed adjustments are required? ie; if PC3200 is running in/around 200mhz spec and it's rated 2-2-2-8 @ 3.2v (and memtest confirms it'll do 2-2-2-8 @ stock 200 HTT with no errors) then if 300 HTT is applied, but the divider is dropped to 2:3 or to keep it at or below 200mhz it should still run 2-2-2-8? Or am I guessing wrong and infact the increase in HTT, no matter the memory speed - requires a bit more leniency in the ram (therefore requiring me to loosen the timing/latency on the ram to 2-4-4-10, or even 2.5/3.0 -x-x-x in order to maintain a stable 300 HTT overclock)? Or would loosening the timings to those mentioned simply be in order to overclock the ram past it's stock speeds?

Or is it just not as simple as that - and very subjective to both ram, chip, motherboard, voltage - etc - meaning I'm not going to be able to quantify it in such terms?

I am also correct in understanding that the LDT voltage is for the HTT controller? If so, bumping voltage on this would be in proportion to a higher clock HTT (and if the HTT multiplier is dropped, would the voltage still need to remain raised?). Or would it only need to remain raised if you were technically overclocking the HTT past 1000mhz?

The chipset on the other hand, controls what? Where and why would the increased voltage on the chipset be required, and what would it proportionately or inversely effect?

I'm trying to understand how each of these parts affects the stability of the machine as a whole - past the relatively simple HTT & CPU Multiplier portions of overclocking.

 

[Herulach]

What timings are as much a feature of the4 controller as the ram i think, but not knowing how A64 internals work, i dont know if the ram controller bit scales with Ram frequency, cpu frequency or both. I suppose one way to find out is run the ram at the same speed, but bump the HT and see if that screws up memtest, if it does drop the timings, if it fixes it youve got your answer there.

 

[Lukano]

What timings are as much a feature of the4 controller as the ram i think, but not knowing how A64 internals work, i dont know if the ram controller bit scales with Ram frequency, cpu frequency or both. I suppose one way to find out is run the ram at the same speed, but bump the HT and see if that screws up memtest, if it does drop the timings, if it fixes it youve got your answer there.

RAM speed does scale with HTT increases. For instance, if you were to leave ram divider at a 1:1 ratio and increase your HTT from 200mhz to 250mhz you would see a corresponding mhz increase on the ram speed. That's why the divider is so important in order to keep the ram within spec. I've tried exactly this, and found that if the ram gets even a bit over 200-210 at the same tight(ish) timings memtest starts getting errors. (note : this is why I want to learn more, as I'm not sure if this is simply that my ram max's out at 210mhz - or if it needs to mature/burn-in - or if it's something else affecting it. And this is also where/why my 300 HTT threshold and gaming stable but not prime stable is driving me nuts).

I guess to clarify my original question - if you are scaling HTT up but scaling ram divider down to keep the ram within spec speeds - is it a logical or untrue assumption that the timings themselves (CAS 2 2-2-8) can remain the same as long as the ram itself is still within/close-to/lower-than 200mhz after the divider is applied? Or - is there still some additional latency to account for (up or down) that would require loosening the timings on the ram (2-3-3-8 or lower) in order to account for the scaled HTT (again, even though the ram itself is still 200mhz or less as per our example)?

 

[Herulach]

RAM speed does scale with HTT increases. For instance, if you were to leave ram divider at a 1:1 ratio and increase your HTT from 200mhz to 250mhz you would see a corresponding mhz increase on the ram speed. That's why the divider is so important in order to keep the ram within spec. I've tried exactly this, and found that if the ram gets even a bit over 200-210 at the same tight(ish) timings memtest starts getting errors. (note : this is why I want to learn more, as I'm not sure if this is simply that my ram max's out at 210mhz - or if it needs to mature/burn-in - or if it's something else affecting it. And this is also where/why my 300 HTT threshold and gaming stable but not prime stable is driving me nuts).

I know, thats my point, if your ram memtest ok with out a divider (say 210 2225) but as soon as you bump up to 300 HT (with a divider to keep the ram at 210 ish) does it throw errors at the same timings? If so then yes clearly, the A64 memory controller performance is dependant on HT, as your ram is capable of those speeds/timings but your memory controller isnt, given the only change in factor is HT speed, then there you have it.

Suppose that doesnt take into account the board, but it at least narrows it down a bit. ALthough having said that, i have zero issues going to 300 HT (keeping cpu speed down with multi) and theres no way my chip can do anything like 300x9 (should have stuck it out for venice damn it, gues theres always X2).

Just an aside, have you tried above 300HT? I know it sounds a bit daft, but i hit a similar at 230 on my board, it will not post set to 230, 231 up is fine though.

I guess to clarify my original question - if you are scaling HTT up but scaling ram divider down to keep the ram within spec speeds - is it a logical or untrue assumption that the timings themselves (CAS 2 2-2-8) can remain the same as long as the ram itself is still within/close-to/lower-than 200mhz after the divider is applied? Or - is there still some additional latency to account for (up or down) that would require loosening the timings on the ram (2-3-3-8 or lower) in order to account for the scaled HTT (again, even though the ram itself is still 200mhz or less as per our example)?

This does seem reasonable, the memory controller might struggle at high speed, it all depends on whether the controller speed scales with ram speed or not. I imagine eclipse will know.

This post not guaranteed to make any sense, havent had anything like enough sleep for around 3 weeks bloody exams.

 

[Lukano]

I know, thats my point, if your ram memtest ok with out a divider (say 210 2225) but as soon as you bump up to 300 HT (with a divider to keep the ram at 210 ish) does it throw errors at the same timings? If so then yes clearly, the A64 memory controller performance is dependant on HT, as your ram is capable of those speeds/timings but your memory controller isnt, given the only change in factor is HT speed, then there you have it.

Suppose that doesnt take into account the board, but it at least narrows it down a bit. ALthough having said that, i have zero issues going to 300 HT (keeping cpu speed down with multi) and theres no way my chip can do anything like 300x9 (should have stuck it out for venice damn it, gues theres always X2).

Just an aside, have you tried above 300HT? I know it sounds a bit daft, but i hit a similar at 230 on my board, it will not post set to 230, 231 up is fine though.

This does seem reasonable, the memory controller might struggle at high speed, it all depends on whether the controller speed scales with ram speed or not. I imagine eclipse will know.

This post not guaranteed to make any sense, havent had anything like enough sleep for around 3 weeks bloody exams.

It makes enough sense, don't worry!

I've tried 305, 310, 315 as stepping above 300 and each and every one doesn't even get to DMI verification without errors (or lovely colored ascii splashes). Even with looser ram timings or increased voltage (haven't gone higher than 1.79 for fear of combustion) it doesn't seem to help. Then again, I may be missing a step here.

I am going to try going up to or over 300 HTT but using the cpu multiplier to keep the CPU speed down around 1.8GHz and see what results I get. I may not be able to turn those results into real-world decision making info - but at least it'll be more info to share and get advice on.

 

[Postal Dude]

So long as you keep your memory within its own specifications it should do its job no matter what the HT is set at. Thats what the use of dividers are for. Above the 300HT are you lowering your HT multi down at all? HAve you considered that you are reaching your cpu's limit? IF memtest is passing with no problems, but u are getting cpu related issues, does that not stand to reason? Isnt there a wall with some motherboards that above 300 isnt doable?
From what youve said i'd say your memory is fine and not at fault. Check that your HT multi is fine. You may have reached your limit... 2.7 is still GREAT! Im doin 2.5 on a winnie atm.

 

[Lukano]

So long as you keep your memory within its own specifications it should do its job no matter what the HT is set at. Thats what the use of dividers are for. Above the 300HT are you lowering your HT multi down at all? HAve you considered that you are reaching your cpu's limit? IF memtest is passing with no problems, but u are getting cpu related issues, does that not stand to reason? Isnt there a wall with some motherboards that above 300 isnt doable?
From what youve said i'd say your memory is fine and not at fault. Check that your HT multi is fine. You may have reached your limit... 2.7 is still GREAT! Im doin 2.5 on a winnie atm.

It's hardly that I'm refusing to believe my eyes, so to speak - that I've hit a ceiling. Moreso that I want to make absolutely sure that it's a physical ceiling and not something solveable by a setting or a burn-in process.

280 x 9 1.61v 2-2-2-8 seems to be my Prime stable. I just want to keep pushing.