Newb Ram latency questions

[Bambini]

Alright, so I know very little about overclocking and latencies, but here goes.

I have 2 corsair xms 3200ctcp's (http://www.corsairmemory.com/corsair/products/specs/twinx1024-3200c2.pdf) (total 1024 mb matched pair), and the corsair site says they're rated at 2-3-3-6. Right now im running them in a soc 754 amd 64 3200+ (clawhammer) in a Asus k8v se (http://usa.asus.com/products/mb/socket754/k8vse-d/overview.htm). Both Cpu-z and my bios are reporting that im running at 3-3-3-6 @ 200mhz fsb (2000 mhz/10).

My questions are:
1) Would bringing the cas latency down to 2 result in any noticable performance increase?
2) How do I go about doing this?
3) Should I consider changing anything else (FSB multiplier, or latencies?)

[CandyKid]

Bambini,

Corsair TwinX1024-3200C2PT modules are rated to CAS 2.5-3-3-6 on AMD systems, 2-3-3-6 on Intel systems.

However, if you bump the DDR Voltage (VDimm) to about 2.85 volts, you may be able to run them at CAS 2 on your system. The reason behind this was a change in the IC's used to make the modules and they would not reliably run CAS 2 on AMD systems. This only applies to revision 3.x and later. Revision 2.x's are mostly based on the Winbond -C Rev 6 IC (I think I remembered that correctly) which did run CAS 2 quite well on AMD systems.

As for overclocking, that's a whole new story. If you're overlcocking, 3-3-3-8 is fine until you get the OC you want. That usually requires an HTT of 3x and then you can increase the FSB until it starts to fail.

However, I'd need a full listing of all the hardware currently running in your system in order to detail a proper overclocking outline. The only bad news is that the Clawhammers weren't amazing overclockers. So you might also add in what you hope to achieve from an OC.

PLUR
CK

p.s. 2.85 volts on the VDimm will NOT void your warranty on Corsair XMS.

[Bambini]

Alright heres my rundown:
Clawhammer 64 3200+ (thermaltake venus 12 heatsink)
2x512 Xms Twinx C2PT
Asus k8v se Deluxe
2x Seagate SATA (raid 0) (main-drive, os:win x64) || 160 gig WD pata || 120 gig Seagate pata
thermaltake silent purepower butterfly 480 watt psu
sapphire x850 xt
audigy 2 zs
Liteon dvd-rom
NEC Dvdrw
and of course a floppy drive
All in an antec lanboy super (has that ugly blue 120mm fan on the front)

Both the cpu and video card run at around 38 deg no load, and ~42+5 full load.

What am I looking to gain outa overclocking? Well
1) more fps outa games, being the gamer I am
2) having the knowledge of how to OC properly so that in the future I dont do something stupid :)

edit: The only times i've tried Ocing (read: twice) I didnt really know what I was doing, and all that resulted was BSoD @ boot.

[CandyKid]

Well, you're going to have two limiting factors here... the CPU and your memory modules.


The C2 modules aren't great overclockers, so here's what I'd do...

Put the VDimm on 2.85 volts (DO NOT exceed 2.9), the HTT on 4x and try the FSB at 210 MHz and the memory at 3-3-3-8.

See if it's stable, a good rule is to run Memtest86+ (www.memtest.org) to see if the system is still stable (FYI, look especially hard at test 5 thru 7 because errors here are not necessarily memory related!). If all is well (I'd also benchmark with 3DMark or such) then try 215 FSB, test again.

If you want, you can increase your VCore by 0.05 volts and that may help you get a little higher on the FSB as well. If you have to drop the HTT to 3x, then IMHO that'd be a performance loss and not really worth the OC you will get from this set of hardware.

After 215, just go up by 5 MHz at a time and see where you can get to, eventually it'll become unstable or too hot, then back it off 5 or 10 MHz and leave it be!

PLUR
CK

[Pug]

Good advice CK, nice one.

If you're right about memory beginning to be hit by heat with the voltage adjustments on some modern motherboards, I'm going to have to eat some words on active RAM cooling solutions. :o)
I've always been stopped by errors long before heat ever became a problem - even in a fanless box (all other components w/ced) and at 2.9v (iirc) too.

[zer0]

cool can i post my system specs here and get help overclocking too?

(i have no idea that HTT is or how to change it or where to change my cas settings in my bios so i might need more help)

[CandyKid]

Good advice CK, nice one.

Thank-you kind sir! Always happy to share!

If you're right about memory beginning to be hit by heat with the voltage adjustments on some modern motherboards, I'm going to have to eat some words on active RAM cooling solutions. :o)
I've always been stopped by errors long before heat ever became a problem - even in a fanless box (all other components w/ced) and at 2.9v (iirc) too.

This was rather cryptic, IMHO, but just for reference, the CPU die of an AMD Barton-core Athlon XP can reasonably function up to 85 centigrade. Mind you, this is HOT, but it is still covered by warranty and should not cause the system to fail. Regardless of how dangerous it is.

Your RAM IC's (memory chips on the PCB) can also run HOT without a problem. I mean really hot. They're not working as fast as a CPU for one, they're not even running on a full voltage scale for that matter, but there's a whole lot of tech spec to explain that. Basically, if heat was causing soft errors, then you'd have to have a blow-torch inside your case or set the system right beside the hottest radiator in your house. If that wasn't the case, then something would be seriously wrong with either voltages, the physical components, or your installation (I got to hear about a guy who used a hack-saw to make the DDR-I notch on some SDR memory to make it fit. Go boom.).


zer0, feel free. I'll do my best to walk you through it, but I also expect you to RTFM a bit. Deal?

PLUR
CK

[fivecheebs]

(I got to hear about a guy who used a hack-saw to make the DDR-I notch on some SDR memory to make it fit. Go boom.).

:shock: Are people really that stupid?

[CandyKid]

No, most people are not blundering idiots, but there are a few.

I've seen sombody break their "drink holder" for real. No joke.

My favorite is the, "It says to enter disk 3, but only disk 1 and disk 2 will fit in there."

The most retarded was, "I can't get it to turn on or anything." "Well, sir, can you go check to see if the circuit breaker has been hit?" "No, the power is out in the whole building and I can't see."

Favorite one I heard lately, "Yes ma'am, you only need to bring in the box for us to repar it." *Lady walks into store with just the monitor.*

You'd be suprised. I used to work 'miracles' by plugging things in and turning on the power strip. If asked, the ONLY appropriate answer to a member of management is, "I'm sorry, it's pretty technical."

PLUR
CK

[fivecheebs]

LOL!

Well i have been support and part time OEM for a few years now and i am happy to say i have never seen such incompetence as you describe! Most of the people i deal with are scared to touch too much (other than clicking on popups and stupid ads, and downloading misc crap progs) thank god!

[Bambini]

haha. those were some good ones.

Ok, so heres what I did (This is posted from work so some of this isn't 100% accurate):

I upped the dimm voltage to 2.8 (only other setting in bios), changed the FSB to 210, and I think I left the timings as 3-3-3-6. As for HTT, my bios doesn't report it in 4x (IIRC you can change it from 800 mhz and down).

Results: I only had windows x64 installed, and I didn't have iso burning software at the time so I couldn't run memtest. I have it now though, and I'll try it tonight.

I ran 3dmark03, and I gained around 150 points from just upping the vdimm, and the fsb by 10 mhz.

Windows was also a little bit more responsive.

However, since I only have one performance pc, I was still a little scared to play around, but im slowly coming over it, and will mess around more this weekend.


Quick question though: Why do you want to decrease the HTT multiplier, and what does HTT do exactly?

[CandyKid]

HTT is basically the FSB of an A64 system, but kind of backwards.

You should definately try to keep it upwards of 4 (800) if your system is a max 800, but for OC purposes that's not always possible. The idea is to increase the FSB enough to make up for (and supercede) any performance loss of the 3x HTT.

Now, if you could pull 10 or 15 over 200 and maintain a 4x, that'd rock and it's worth trying.

Nice work so far.

PLUR
CK

[CandyKid]

cheebs,

I'm mostly a hardware man and most people don't know anything about hardware... that's probably why we see some good ones.

PLUR
CK

[Da_Rude_Baboon]

My best one was a student who came running through to my office saying the monitors gone all funny and keeps grunting at me! I went through to the lab and the monitor looked fine. At a loss as to what it could be i checked a couple of things and couldn't seen anything wrong. I asked her to come and get me if it happened again so i could see it for myself. At that point her mobile phone, which was on silent and vibrate, went off. Sure enough the monitor went funny and her phone vibrating on the wooden desk made a nice grunting sound. :duh:

[zer0]

ok here is my info meaby you can get my system to overclock above 3.12ghz i shure as hell cant.

Intel Pentium 4 Presscott 3.0Ghz (not a J)
DFI Lanparty UT 915p-T12
1GB (2x512mb) Kingston ddr 400 memory
OCZ 520watt Modstream PSU

and of corse liquid cooling so heat isnt an issue.

(just a side not i was able to get my system up to 3.2ghz but i got bsods left and right while playing games)

i know my memory is probobly limiting the OC but it cant possobly be by that much.

[Greeny]

If you believe the memory is limiting you then the simple answer is to take the memory speed out of the equation by using a lower multiple for the FSB to Memory frequency ratio. Drop the memory down to known stable frequency or less with a divider and keeping thrashing on the CPU's FSB.

For myself I test individually how high the memory frequency and settings can go whilst keeping the CPU at or near a proven stable value, then move onto overclocking the CPU only while keeping the memory within known good parameters.

I just find this technique easier and quicker for finding the perfect sweet spot.

[zer0]

If you believe the memory is limiting you then the simple answer is to take the memory speed out of the equation by using a lower multiple for the FSB to Memory frequency ratio. Drop the memory down to known stable frequency or less with a divider and keeping thrashing on the CPU's FSB.

For myself I test individually how high the memory frequency and settings can go whilst keeping the CPU at or near a proven stable value, then move onto overclocking the CPU only while keeping the memory within known good parameters.

I just find this technique easier and quicker for finding the perfect sweet spot.

i understood about 5 words out of all of that.

[Bambini]

i understood about 5 words out of all of that.

double goes for me.

[fivecheebs]

Greeny is suggesting that you run the RAM asyncronously with the CPU.

You will find in your BIOS somewhere a ram divider. Its like a ratio of how fast the momory bus runs in relation to the FSB. There could eb a setting that says auto then youll probably have numbers like 5:6 etc. Syncronous woulda be something like 6:6. If you set it to 5:6 and the CPU is on 200MHz FSB the ram would be running at 166 (200/6=33 , 33*5=166).

So now you have the ram running slower than the CPU, if you know how fast the RAM will reach you can figure out what you can reach without the RAM being a problem by using the same calculations i did earlier.

Greeny also says he pushes one part of the system at a time to find its, or the motherboards, maximum overclock. So he would reduce the RAM divider loads so he know it wont be a problem and pushes his CPU until it squeels. Then he will lower the CPU multiplyer and push the RAM. Now he knows the limits of his componants he can find a happy ballence wetween all of the individual overclocks.

Is that any clearer?

[zer0]

ok i feel like a real n00b but here is all the options in the section that looked most like it was for overclocking
Ext. Clock (CPU/PCI) (200mhz ->)
PCI-Express Frequency (Auto ->)
N/B Strap CPU as ( By CPU, FSB533, FSB800)
Minimize FMB15 CPU Ratio (Enable, Disable)
DRAM Type (DDR333, DDR400)

CPU Voltage
DRAM Voltage
N/B Voltage

Auto Detect PCI Clk. (Enable, Disable)
Spread Spectrum ( 0.35%, 0.50%, 0.75%, 1.00%)

[fivecheebs]

TBH im not familiar with intels, I just know roughly the way it works.

Im guessing here but does the "DRAM Type" force the memory bus speed? If so then my guess is its doing it by dividers, it just called it something else. Im sure you can appreciate all BIOSes are different in the terms they use.

Assuming the above is correct, and you believe the RAM to be the limiting factor, turn the DRAM type down a bit then boot up and check the speeds with CPUz. Then make a small adjustment to the FSB and see if the memory bus changes accoringly. If it does then my post above explains what is happening.

[zer0]

ok i screwed with my DRAM type setting and i did see a drop in ratio of the fsb and the memory clock (5:4) however i also tryed upping my fsb up with my memory set to DDR333 and i still cant get any higher than without changing that.

[fivecheebs]

As you increased the FSB again did you see the memory bus increase too? What speeds do you get too?

Did you tweak any voltages?

What i would do is try to find the limit of your RAM, then you will know when thats causing the errors or not. Get Memtest (http://www.memtest86.com/), i usually use the iso but use what you want. Set your FSB to the point where it causes errors, then reboot with memtest (it runs before windows ever comes into play so you dont have to wory about stability problems). Let it run through the tests a couple of times, the last few are the hardest for it to pass. If memtest passes you know the ram is ok at that point so up the FSB a bit more and run it again. etc...

[Greeny]

Well explained Fivecheebs, my apologies for not being around to clarify I forget that not everyone knows the basics. You've hit the nail bang on the head.

A more modern variant of Memtest which has readouts of frequencies for most boards/chipsets and wider compatibility can be found: here (http://www.memtest.org/) . I use the older version where possible as it has proved its worth to me many times, the new one I cannot be so certain about as I have yet to use it extensively and get to trust it. Sometimes it's the only version that works though.

[fivecheebs]

No problems mate. We all rush posts every now and again. I think im pobrobably more guilty that anyone of that crime. Im just glad you agreed :lol:

[Greeny]

Not having much experience with Intel systems until recently I don't know what current convention is as regards what the FSB to DRAM ratio is called. On both my native Dothan setup and my new hybrid setup it is controlled by changing the DRAM Frequency in the BIOS. As the FSB scales higher different options for DRAM Frequency appear as more dividers become available at higher bus speeds.

Let's say I select my FSB of 200Mhz for the CPU and I set my DRAM Frequency to 200Mhz, as I step up the FSB of the machine gradually I would expect the DRAM frequency to also go up 1Mhz for every 1Mhz step onto the FSB. Thats becasue rather than being a static value which I have input it selects a ratio based on what I selected in the BIOS. In this case the ratio would be 1:1 as both values are identical.

For the next example I select the same 200FSB for the CPU, but this time I set my DRAM Frequency to 166Mhz. This is a ratio of 6:5 now (divide both numbers by 33Mhz to derive) and not the standard 1:1 so for every 6Mhz I add onto the FSB I will only add 5Mhz onto the memory frequency.

The mere fact that the DRAM Frequency scales at the same time as the FSB frequency scales shows that the DRAM Frequency is not fixed, but rather it is derived from a divider ratio of the Front Side Bus Speed as is the case for most motherboards and most overclocking endeavours.

[Greeny]

Not really a rush job I just failed to accurately gauge my target audience. :P

Everyone likes to overclock these days and it's hard to determine just how much they understand the mechanics of the operation. I spent weeks/months reading up on the classic techniques and learning the maths etc. before I began (not that theirs much too it, but I'm a perfectionist).

[zer0]

Not having much experience with Intel systems until recently I don't know what current convention is as regards what the FSB to DRAM ratio is called. On both my native Dothan setup and my new hybrid setup it is controlled by changing the DRAM Frequency in the BIOS. As the FSB scales higher different options for DRAM Frequency appear as more dividers become available at higher bus speeds. so that might explain there only being 2 options at the stock FSB clock.

ill go check this out.

do i need to turn of hyperthreading while overclocking (i do anyways but it would be nice to know)

[edit]
i changed the ddr type down to ddr333 and pushed the fsb as high as i could be before the system quit posting.

here is everything i know.
cpu speed: 3.240Ghz
FSB: 216Mhz
Memory frequency 172.8Mhz
CAS settings: 2.5-4-4-8

NEW RECORD LOW ON CPU TEMP! 26C

i have not tryed changing the voltage settings in any way

i havnet ran memtest yet ill have to dig up one of my linux disks and boot up memtest (if im not mistaken memtest 86 is linux based?)

[fivecheebs]

You can d/l a bootable iso from the website i linked to earlier mate. No need for anything else.

[zer0]

ok a very small update but its kinda nice

by changing my dram voltage to 2.7V i was able to squeeze a few more Mhz out of my system.

3.270Mhz
FSB: 218Mhz
Memory: 174.4Mhz

further change in dram voltage has not helped incrice speed.

[walkitiki]

i havnet ran memtest yet ill have to dig up one of my linux disks and boot up memtest (if im not mistaken memtest 86 is linux based?)
I always ran it from a DOS 6.0 floppy

[Greeny]

The bootable ISO is the best idea it will not require any linux discs and will not touch your hard drive contents, if you have Fedora Linux discs lying about you can boot to one of them and type memtest86 at the install prompt as well.

I'm not a ram expert but the fact that your ram is only rated for 2-3-3-6 at 200Mhz indicates that it's not quite as good as other DIMMS which are available now. I would expect the main limiting factor to be the CMDRate since 2-3-3-6 stuff usually runs 1T at 200Mhz but will likely not run 1T much above 210-215~ish at a rough guesstimate.

There are ways to slacken the 1T CMDRate, sometimes theyres a direct setting for it in the BIOS or with the P-4 boards I believe the performace mode (PAT/GAT) controls the CMDRate setting. I've read some say that it slackens to 2T above 200Mhz FSB+ anyway, but I have little experience in the matter as I'm still at comparitively low frequency to your machine.

Cas Latency is also likely to need to move to 2.5 as the frequency rises if you really want to wring them to high frequency. Common convention would be to move the TRas value to 7 cycles if you go to 2.5-3-3 so as to avoid TRas violation as well.

I wouldn't be surprised to see even those semi-mediocre looking modules hit 240~ish Mhz 2.5-3-3-7-2T, but the question will be whether this produces more bandwidth or not. Look at the the bandwidth value on Memtest86 I find it's results very reproducable.

That would be the most likely settings you would need to slacken to get the ram to higher frequency. It then becomes a case of testing whether lower and tighter is better than higher and slacker as the CMDRate is one of the ram timings which has greatest effect on performance as noted by the CandyKid elsewhere I believe.

Voltage plays a big part in how some DIMMS overclock and what timings they can achieve and does sweet FA for others so I suggest going slowly with it and setting yourself a maximum which is safe. Stuff like Winbond BH-5 can take up to about 3.8-4V with active cooling (it's rated for about 3.8V IIRC), whereas the opposite is true for Samsung TCCD's in that for the most part they do not require large voltages and some of the dimms really don't like it.

While a board with a set of basic adjustments for ram timings is much easier for the novice (and even for the lazy knowledgable user) you have to remember that their are many other hidden timings in the BIOS which are most likely either hard coded and non-adjustable or vary as you change the other timings as appropriate. I've spent time with both types of board and I will take the former over the latter as the myriad of options are not well documented for the most part, finding out what some of them do is more complicated than just playing about with them. These unforseen values can also affect how well your ram overclocking session goes.

[zer0]

do you think i would get better results with ddr2 memory (newegg has a nice matched pair (pqi 1gb) for only $109 with good cas latancy) (http://www.newegg.com/Product/Product.asp?Item=N82E16820141200)

im also thinking about getting ddr2 memory because i can use it with the new socket A64/AX2s when they come out.

[Greeny]

I love the customer reviews. Hehe. :P

Think about whether the price justifies the extra performance difference is about all I can say. I have no idea what the difference wil be, but if you want a guess, not much.

[fivecheebs]

I dont think spending more money on RAM will help you much at this stage. Not until you have found out exactly how well the individual parts of your system. Have you underclocked the processor to push the ram further whilst keeping the CPU within its stable range? Seriously, you should do this with lots of memtests so you can say its your ram for sure. When you start to get errors, raise the voltage a little, and/or loosen the timings and try again Keep going until you reach a voltage you would consider to be a safe maximum. Keep a note of the timings at what voltages etc. it will come in handy after you discovered the CPU's max in a similar way except using prime95 as the stress test.

As for performance gain with DDR2, i think CK is the best person to aswer that right now.

[zer0]

some times reviews are a great way to see how far you can push it.

the big diffrences are the reviewer have better memory and a faster cpu (from the same family as mine) so i might be able to get a bit better

the settings im going to try are

CPU Voltage = 1.4375V
FSB Max = 250MHz (or as high as i can push it)
Northbridge Voltage = 1.8V


i think i might have found a big issue...

I was somewhat dismayed to find that the 3.2GHz system would barely reach 3.5GHz before failing to post. I tried a number of different settings, to no avail. Anything over 220FSB, or 3.52 GHz, resulted in the system not booting at all. On a whim, I swapped out the 6600GT video card we used for testing with an ATI X700 Pro, and lo and behold, the system suddenly had legs. After the swap, the system no longer complained when I started cranking up the clock, and it not only passed 3.5GHz, but 3.8 and 3.9 GHz with no fiddling required.
it seems my 6600GT is to blame for the awful overclocking results.

[Bambini]

they are teh crapzor. But an x850xt. great investment :thumbsup:

[zer0]

im sorry but i dissagree my 6600Gt is a great card. i dont plan on getting rid of it until i can afford a 6800GT/ulta or 7800GTX because i dont like ATI not only are the drivers unstable and vastly incompleate (in linux) but 30FPS on Doom 3 with a x800 (in linux) what the hell even my 6600GT gets better than that.

[Greeny]

The PCI-E bus is still a but of an unknown commoddity, I believe the frequencies on it are not fixed like they are with the AGP Bus these days, but again are a ratio like the PCI/AGP Bus used to be. This could well explain why the machine makes it further with a different card.

Before the locks it used to be a case of selecting the components which you know would operate at high frequency without error.

[CandyKid]

I'm not a ram expert but the fact that your ram is only rated for 2-3-3-6 at 200Mhz indicates that it's not quite as good as other DIMMS which are available now. I would expect the main limiting factor to be the CMDRate since 2-3-3-6 stuff usually runs 1T at 200Mhz but will likely not run 1T much above 210-215~ish at a rough guesstimate.

The command rate is actually hardware dependant on the chipset, MCH, and CPU. The command rate is very slightly dependant on the memory modules.

As for running higher with 2.5-3-3-7 @ 2T, that's a pretty safe bet, but try to keep the command rate at 1T because it's impact is pretty substantial, especially for gaming or progs that require a lot of random accesses... not so much for burst.

For testing purposes, you can also go 3-4-4-8 on the latencies.

FYI, if you see memory like this: 2-2-3-6, that means CAS-Row Precharge-RAS toCAS-Cycle Time (tRAS).

Voltage plays a big part in how some DIMMS overclock and what timings they can achieve ... Winbond BH-5 can take up to about 3.8-4V with active cooling (it's rated for about 3.8V IIRC), whereas the opposite is true for Samsung TCCD's...

Running 3.8v through your MCH/chipset and memory will void any warranty on the planet and for good reason. FYI, BH-5's I've seen usually require just over 3 volts for really good timings and overclocks while TCCD's often require 2.85 to pull 2-2-2-5 @ 200MHz. Often TCCD based modules won't run stable at anything less than 2.75 volts.

...These unforseen values can also affect how well your ram overclocking session goes.

Very true wich is the best argument for allowing access to these values. For persons not understanding them, then an 'auto' setting is sufficient, but somtimes it's those archaic settings that can make or break a great OC.

On another note about dividers and memory, take note that on some of the bigger dividers such as 2:3 (200MHz FSB turns into DDR600), the MCH can sometimes get overloaded and start tossing errors. This is a limitation to 'pushing' the limits of the memory without pushing the CPU and vice-versa, but the idea is still a good one.

PLUR
CK

p.s. I only use Memtest86+ (http://www.memtest.org/) these days. Always seems to run a little more smooth. Also, the last tests are the hardest, but tests 5-7 are the most important for the REST of your system and are often the target of errors originating from the CPU or chipset.

[Starbuck3733T]

Candykid, mr. memory, can you explain the GAT controls to me?

Specially, on an abit IC-7 I have the GAT set to auto-auto-auto-disabled-disabled where the last disabled is CPC.

If I set the first auto to Turbo, F1, of street racer, I get a non-posting machine.

memory is 4x512MB PC4000 micron chips (ballistix :rolleyes:)

[CandyKid]

I'd have to look up your BIOS to see the settings (Sorry, don't have 'em memorized!), but the game acceleration technology does bad arbitrary things for the most part.

The first one arbitrarily sets certain memory latencies for which a PC4000 IC will NOT be rated... actually, most PC4000 IC's are rated at CAS 3 so any setting on GAT will try to reduce these latencies to an unsupported value.

You're better off setting them manually, TBH!

The GAT settings try to automagically speed up certain aspects of your computer, usually the memory, but it does so without checking for compatibility.

PLUR
CK

[Starbuck3733T]

So the only good thing under them that I should try is turning on command per clock?

CPU-Z also shows that my "performance mode" is disabled... what's that mean?

I just downloaded my mobo manual, and it doesn't show the options I have. IO guess a screen digi-photo could be in order.

[CandyKid]

Go for the pic...

Performance mode should refer to PAT which is a good thing (but also messes with latencies). I'm not sure of it's functionality w/ 4 modules in your system though.

...command per clock should have a 1T or 2T setting, 1T is preferrable.

[Bambini]

Ok, I've been exploring a little bit, but I still dont know alot. So here is a list of all things I have ? on, or am just curious. These are all the options I found for overclocking/latancy tweaking I found in my bios:

Memclock Mode: Auto/Limit

This allows me to get to 1:1, 2:3, 2:1, 5:3, 4:1, and other fsb to memory speed ratios. I know you explained how they worked before, but Im still confused as to why we would want to have these options.

Timings with settings/default:
Bank Interleaving: Auto/disabled
Cas (Cl):..........2, 2.5, 3/Auto
TRC:................7 to 13/Auto
TRFC:..............9 to 15/Auto
TRCD:..............2 to 6/Auto
TWR:...............2 to 3/Auto
TRWT:.............1 to 6/Auto
TRAS:..............5 to 15/Auto
TRP:................2 to 6/Auto
DDR Clock Delay:1 to 8/Auto
2t Command:.....Auto/enabled/disabled

Can you tell me, or give me a link of what all these are? What settings should I change?

DDR Voltage: Auto/2.6/2.7/2.8

Should I leave it as auto? Or should I manually set it?

CPU Volt Adjust: None/+0.2 V
V-Link Voltage: 2.5/26

Should I be messing with those last two?

HT Data Width Up: 16/8
HT Data Width Down:16/8

What are these? Mess with them?


Right now, im running with a FSB of 215 mhz @ 37° c. I've attached a SS of my CPU-Z Report. All I did was change the FSB. Am I doing this right? It says I have a multiplier of 10, but my bios doesnt do it in x10, its a max of 800mhz.


CPU-Z SS: http://www.filefarmer.com/bambini/cpu-z.JPG
CPU-Z HTML REPORT: http://www.filefarmer.com/bambini/cpuz.htm

Thanks for all your help

[CandyKid]

Good stuff, I've got to get some reports done, but I'll try to get to it soon! You're making progress!

[Bambini]

Hrm, I was running stable and cool for well over 6 hours (games/apps/andall) till about 9:44 pm. I bsod'ed after alt-tabbing a game. Something about driver_srlq_not_lessthan.

It might be because im using windows x64, but it might be because of the oc'ing.

[Starbuck3733T]

I've officially lost my mind. P4 is running at 1.8V. Screw sudden northwood death syndrome, I'm not afraid! I've always got another chip to fall back on anyway! 3.60 GHz stable from 2.8GHz northwood C.

[fivecheebs]

LOL nice one star. :wiz:

Is this under your G5?

[CandyKid]

Star, go for it as long as the temps stay down! Sudden death be damned! :p

Bambini, run Memtest86+ (www.memtest.org) for 3 passes to double check. I'd also run the Sisoftware Sandra memory burn-in wizard, just as a stress test.

PLUR
CK

[Da_Rude_Baboon]

driver_srlq_not_lessthan.


That normally means you have a bad driver. It could be flaky support XP64. If you type your BSOD messages into google or the microsoft knowledge base search you can normally get enough information to put you in the right direction.

[Greeny]

I can assure you that my statement about the maximum voltage rating for the BH-5 is correct. Look it up if you like. I completely agree in that it would very likely damage the DIMM's if you were to do so (and stated active cooling would be required), regardless BH-5 is rated for an extreme voltage on the datasheet.

:)

[CandyKid]

I know you are correct, but nobody run's IC's in their computers, we run DIMM's.

Winbond rated their IC's to a much higher voltage than the industry standard for DDR DIMM's, but there are other uses for the IC's which might require that kind of power. I didn't mean to jump on you, but I get too many people touting that great OC they got with 3.3+ volts and I have to make sure the possible effects of that are known.

Sorry, it comes from helping in the tech-support arena for a certain manufacturer.

That's one of the reasons the OCZ VX memory really pisses me off... n00b's won't understand why their new VX modules rated at PC4000 and CAS 2 on 3.3 volts caused their chipset to crash and burn...

IMHO, never exceed 2.9 volts on the VDimm unless you really know what you're doing.

PLUR
CK

[Greeny]

I would totally agree with that assessment, your maximum voltage is about the maximum value I set myself.

Quite a lot of the time it seems like the boards with high voltage settings fail to deliver the voltage anyway, depends on how the manufacturer impliments the settings in hardware. Certainly I've been dissapointed before...

Do you work for OCZ?

[CandyKid]

No, I don't work for OCZ, but I've thought they made good products in the past, it's just that they're now releasing modules rated to great timings at amazing speeds...

...they just forget to add that the voltage required can damage your hardware.

BTW, I'd agree with you as well, many boards that would technically allow you to up the voltage past 3v usually don't do a very good job. Some, but not most.

PLUR
CK

[Bambini]

so what I had done yesterday:
I had run memtest twice
3dmark03 once
Had been playing cs:s for 3-4 hours
Played a lil bit of guildwars. Went into windowed mode (guildwars was designed for this, which I love), minimized the window, opened up Internet Explorer to check on this post as a matter of fact ;) . And boom, it crashed. I have a feeling it was my audio drivers, as sound usually gets garbled or keeps playing even when the game is minimized.

[CandyKid]

There may be some issues running games in windowed mode and then running something which is a multimedia resource hog like IExplore.exe.

I'd agree, driver issue... btw, give me a little while to list out those settings from earlier...

PLUR
CK

[Bambini]

you've got all the time in the world m8. Or until they change the standards for ram timings, and all the lingo associated with it... and I can afford said memory (=never).

[CandyKid]

Memclock Mode: Auto/Limit

Leave this on auto or you'll set a maximum memory clock.

This allows me to get to 1:1, 2:3, 2:1, 5:3, 4:1, and other fsb to memory speed ratios. I know you explained how they worked before, but Im still confused as to why we would want to have these options.

Alright, so let's say that by some chance you found DDR600 DDR-I modules, but your HTT frequency was 200MHz (200MHz x 4 = 800). If you used a 2:3 divider, that would mean that the memory runs at 3/2 of your frequency equaling 300MHz or DDR600. The dividers are for running your memory asynchronously while minimizing the performance impact of doing so. Because a divider is pre-set, the memory controller knows how to deal with the offset frequencies and has a much better time of doing it... something odd like freq. at 200MHz and memory at 217MHz would give it troubles.

Timings with settings/default:
Bank Interleaving: Auto/disabled

You want this to be turned on. Bank interleaving is a method by which each individual memory module can act like a dual-channel setup (to a point) and basically reduce down-time latency. Set this to 4 if possible.

Cas (Cl):..........2, 2.5, 3/Auto

The CAS latency is the column address strobe. It's giving the system time to locate the proper column within the selected IC and to precharge the circuits. 2 is faster than 2.5 by a lot. 2.5 is faster than 3 by a bit, but that's a long story... basically, half-clocks don't work so well, so a half clock is closer to the rounded-up value in terms of performance. You can play with this one.

TRC:................7 to 13/Auto

This defines the number of clocks between each row access. Unless you like doing math and looking up the nanosecond access times of your IC's and MCH, I'd leave it on auto. Don't mess with this one.

TRFC:..............9 to 15/Auto

This is the row refresh cycle time and is a 'wait so you can do it again properly' kind of thing. Don't mess with it.

TRCD:..............2 to 6/Auto

This is the RAS to CAS delay. This is the number of clock cycles between the row address being determined and the column address being output. Most performance modules can manage a 3 here, the TCCD+BP based modules will run a 2 at DDR400. You can play with this one.

TWR:...............2 to 3/Auto

This is a write latency allowing a certain number of clicks for the memory to properly write information into the IC. Don't mess with it.

TRWT:.............1 to 6/Auto

This is the read to write delay, defining a 'recoup' period between the actions. Don't mess with it.

TRAS:..............5 to 15/Auto

This is the row active time... if reading a byte of memory was like reading a single word on the page out loud (I.E. your friend is the CPU), then this defines how long you have to find the word and get it in your mind, but only start to read it. More technically, the system has to output half of the information back to the MCH and have the rest in the output registers.

There's lots of conventions here, but a tRAS of 11 or 12 seems to work really well on AMD systems. You can play with this one.

TRP:................2 to 6/Auto

Row precharge... that's it, just charging circuits, but it takes time. Good luck getting it below 2. :p You can play with this one.

DDR Clock Delay:1 to 8/Auto

Leave this on auto, if it needs to be done, let the system do it. Basically, there can be a difference in signal between the PLL and the module, but most systems are based on the 0ns 'perfect' ideal. This setting is in all of our systems, just most don't let you see it. Don't mess with it.

2t Command:.....Auto/enabled/disabled

I think we've been over this, but the DRAM command rate defines the amount of time the MCH has to latch a CS (chip select) signal onto the DRAM... err... how long it has to decide which IC has the information in it. With two modules, running a 1T shouldn't be a problem with an nF4 system. Try to keep this at 1T as the performance hit at 2T can be substantial.

Can you tell me, or give me a link of what all these are? What settings should I change?

Just did my friend!

DDR Voltage: Auto/2.6/2.7/2.8

Tight latencies and high clocks means less time to reach full voltage and a bigger requirement... blah blah blah... For most CAS 2 DDR modules at DDR400 2.7 to 2.8 volts is sufficient. Performance modules rarely run well on 2.5 volts, but value-based modules usually run well at 2.5 to 2.6 volts, albeit at loose latencies (hence the lower price). Don't exceed 2.9 volts on DDR-I DIMMs.

CPU Volt Adjust: None/+0.2 V
V-Link Voltage: 2.5/2.6

You may need to overvolt the CPU by 0.2v to keep an OC stable, but just watch your temperatures as the A64's can be finicky. The V-Link voltage probably doesn't need to be messed with except as a last resort for stability reasons.

HT Data Width Up: 16/8
HT Data Width Down:16/8

These define the bit-width of your bus. 16-bit is by far faster and better than 8-bit. 'nuff said.

Right now, im running with a FSB of 215 mhz @ 37° c. I've attached a SS of my CPU-Z Report. All I did was change the FSB. Am I doing this right? It says I have a multiplier of 10, but my bios doesnt do it in x10, its a max of 800mhz.

You're now talking about two different things. CPU-Z is looking at the CPU's core speed, not the HTT frequency, which is actually 4x (200MHz x 4 = 800, but with your OC 215 x 4 = 860).

Hope that helps some! In the end, the CAS, Row Precharge (tRP), RAS to CAS (tRCD), and Cycle Time (tRAS) are the big memory players.

PLUR
CK

[fillip]

http://wizdforums.co.uk/images/smilies/ph34r.gif Thats a lot of info to take in, but extremely helpful to the tweakers amongst us. http://wizdforums.co.uk/images/smilies/thumb.gif

[Bambini]

:worship: HOly bejeebuz! You are a god amongst men my friend! Absolutely amazing. Time to start messin around with this info. Thank you sooo much!

[CandyKid]

Nay, I am a novice compared to some that I know!

I hope it helps though! Perhaps that's another one to go into the DDR guide sticky thingy... hrm...