Archived

This topic is now archived and is closed to further replies.

simba_nl

Optimized DDR2-1066 settings

Recommended Posts

Is there anyone who can give me a good timing hint for my new purchased Corsair Dominator DDR2-1066 memory?I'm running a E8500 on a Gigabyte P35-DS3 board and 9,5 x 400 MhzTHX

Share this post


Link to post
Share on other sites
Help AVSIM continue to serve you!
Please donate today!

Isn't the default 5-5-5-15? If memory serves me correctly, you get to the timings by pressing and holding Ctrl + F1 at the main menu of the BIOS. Set the voltage to the Mfgr's suggested value which I believe is 2.1v. Regards,Jim Karn

Share this post


Link to post
Share on other sites

I have mine running at 1104MHz with 5-5-3-13 timings. Makes quite a difference with the stutters. I've had it up to 1130MHz at these timings (see pic below), but that's about as high as it's going at those timings. I had some stability issues because I didn't really want to push the vDIMM. So...http://forums.avsim.net/user_files/192252.jpg ...My settings now are just a bit more modest. In this second pic, note the subtimings. The most important one here is the tRRD, set at 1. http://forums.avsim.net/user_files/192253.jpgThis is a bit better shot. The pic above didn't show the memory boxes filled in correctly.http://forums.avsim.net/user_files/192256.jpgThese work for me. I don't make any promises for your system. As always, you assume the risk for changes to your system.Hope this helps.Best Regards,Jeff

Share this post


Link to post
Share on other sites

Jeffhew,Thx for your settings, my stutter is slightly reduced now but I'm still not 100% happy. Don't know why but my former cheap DDR-800 was running better. No stutter at all... (?!)

Share this post


Link to post
Share on other sites

Nick_N who frequents this forum is the local expert on RAM so hopefully he will chip in. In case he doesn't, I suspect he would emphasize the importance of timing as much as/ more than the frequency of the RAM. The higher frequency of your new RAM may imply higher performance, but your 800MHz DDR2 may actually have been operating at a more optimal combination of frequency and timings than can be achieved by concentrating just on getting a higher frequency. He has recommended using fast DDR3 for best results.Tim

Share this post


Link to post
Share on other sites

Yes, Tim is exactly right! It's been my experience that with many games and apps, timings aren't that big of an issue but, with FS, timings can be more important than speed. I know in my case, if I had to choose between fast RAM or tighter timings, I'd take tighter timings every time for FS. I agree with pulling back on the speed, 100%. The tighter timings are responsible for reducing the stutters, not the speed. Several people told me I could get more speed out of my RAM, and they are right. I have run it faster but, only with the looser timings. The extra speed isn't worth it. This is as tight as I can get these sticks on my system. The Corsair is good RAM but, my one regret is that I let an expert talk me out of trying the Mushkin CAS4 memory.Tim is also right about Nick. He is THE expert. He's written a number of useful articles. You can take a look at some of them here:http://www.simviation.com/cgi-bin/yabb2/Ya...m=1208959973/30If you need more in depth help, I would definitely seek him out!There was an article floating around a few years ago comparing performance of FS9 at faster RAM speeds vs tighter timings. Tighter timings resulted in much better graphics performance compared to faster speed. I don't recall the author but it certainly could have been Nick. Best Regards,Jeff

Share this post


Link to post
Share on other sites

Its a trial and error situationI would go through it with you however I just do not have timePerformance is gained by using FSB/STRAP - set the strap to 333 if available and with that system most likely 400-410 will be your FSB limits however I did manage to get Anthony Vos who is on DDR2 running 8x450 Q9450 @ 5-5-5-18 2x2GB DDR2 1086... these sticks: http://www.newegg.com/Product/Product.aspx...N82E16820145197your best stable FSB depends on the motherboard and processorMemory SPEED = Higher is better but within the limits of the lowest timings possible at the highest speed.. Take a lower CAS over 100MHz more in speedMemory timing = Lowest stable CAS possible at the highest speed @ 333MHz CPU strap... and try to maintain a TRAS of CAS + TRP + 2 (or 4 max) However with DDR2 you may have to accept a TRAS of 18 or 20 even if it calc's to be 14 or 16the rest other than the 4 primary timing values, just set to AUTO is usually fine. BIOS .. you may need to work both Vcore and VDD to get stable memory clocks established. Do not assume the memory voltage is the only thing that influences memory stability. In a CPU/Mem clock a touch more Vcore can also help, sometimes not. With corsair and clocking you typically have to run their maximum warrantee memory voltage to stabilize a high speed, low latency clock. As long as its within that Corsair spec, your fine. Outside that you are on your own but I do sometimes bump it up a touch (.05v) over 3PD memory tool = MEMSPEED = LOWEST 'performance level' or tRD MCH Read Delay value at the lowest CAS @ the highest speed = best performance and always set the TREF to the highest T value in the dropdown. Memteat allows you to save the settings for boot but dont do that until you are sure you have a stable clock. Once established, then set memtest to make any trimmed timing changes as Windows boots up. If you do and you cant get boot into Windows use safe mode to delete the saved memory profile and then get back into Windows to make changes.a memtest PL setting of 6 is pretty good.. 5 is a holy grail and rare to hit,, 7 is the max I would allow a system to run and that would be on DDR3 bandwidth.I cant just throw numbers at you.. I can suggest but then you have to work it out and stabilize it with stress tests based on your CPU clock. Start with the factory numbers on the highest memory voltage and start working your way up and downlast... 1T CMD ... that is one I always go for however it is another rare one to get with many DDR2 clocks. I will usually accept 100MHz loss for a stable 1T Memtest Performance Level 6 clock. I will usually set 1T CMD FIRST and try and work the rest out with 1T enabled.. but as I said, with DDR2 which is not specified to run 1T, thats a tough one.Your major performance is in timing, not bandwidth, but the key is to hit the lowest timing values @ the fastest speed possible, safe.Use the Everest benchmark to check your latency.. its crude as a suggestion in use however getting it down to 50ns and under is the real goal, with the fastest memory speed possible.With Nehalem, its possible you may see memory lantency in the 20's and possibly the teens with clocks.. and without raising FSB past 166, and, with memory read speed nearly double the 10-11 and 12K marks DDR3 hits today.. but thats 'just' a guess :-hah

Share this post


Link to post
Share on other sites

I did a DDR2/DDR3 test here at AVSIM using Gary's benchmark as the baseDDR3 on the right CAS always wins :-hah same with DDR2.. the timing is the key, faster speed is secondary but still importantThe images are gone now but it starts here and you can follow it to the end discussion through the links in the 6 partshttp://forums.avsim.net/dcboard.php?az=sho..._id=37203&page=here is the last thread with the evaluation of the results.. that thread also includes my discussion of memory timing and bandwidth and how that can be tuned better http://forums.avsim.net/dcboard.php?az=sho...37210&mode=full

Share this post


Link to post
Share on other sites

Keep in mind that the Everst "Latency" number is the final result of any timing and/or speed adjustments you might make. Even the Everest Read/Write MB/s are just measures of internal ram chip processes that assist this final "Latency" result. Latency represents how long it takes to transfer a data bit between the memory buss and the Front Side Buss. This is the job that needs to be done. Everything else is about measuring the processes that cause this outcome to occur. (For the airplane guys, Latency is "N1" or "EPR" The rest are N2, EGT, FF, Etc.) A definitive measure of the result of a lower (quicker) latency number has been illusive. Even huge, 50% decreases in this Latency number (for instance the difference between a 70ns latency and a 45ns latency) has produced performance changes that have been subjective, at best.

Share this post


Link to post
Share on other sites

===========================================CONSIDERATION OF RESULTS:I would probably pay more attention to those +3% to +5% TRUE average increases in benchmark frame rates in conjunction with the lower timing, lower latency memory products

Share this post


Link to post
Share on other sites

Correction>a memtest PL setting of 6 is pretty good.. 5 is a holy grail and rare to hit,, 7 is the max I would allow a system to run and that would be on DDR3 bandwidth.PL (tRD) 7 is fine on a good CAS 5 DDR2 1066 and above clock although 6 would be better, just harder to hit on low CAS without a voltage increase. Some sticks will do it and some won't...and that too is called: MEMSET not MEMTEST or MEMSPEED... senior momentLOL

Share this post


Link to post
Share on other sites

Hi Nick,It is pretty new stuff for me and my English is not my native language so I have to study your info in depth but correct me if I'm wrong but the goal is to get the Latency around the 50 ns...I have done a quick test and getting the follow default results in Everest build 460.1500 with the Memory Benchmark with tRRD=11066 5-5-5-15 = 71,4ns800 4-4-4-12 = 79,1ns667 3-3-3-9 = 72,4ns (raised from 2.1V to 2.4V)I have set the CPU at standard 333x9,5It's all pretty slow I think and quick tests in FSX give me stutter in all different settings.ThxPS: I'm running at 1920x1200 with a 8800GTX/768 and have 1 Raptor for OS in multiboot (XP and Vista64 for fs9 and FSX) and 1 raptor dedicated to FSX, defragmented with disktrix 2008 file/folder high performance FSX outerside of disk

Share this post


Link to post
Share on other sites

The quickest and easiest way to play with latency is to use Memset's "Performance Level" setting. This the tRD setting. It sets how may FSB cycles are skipped between data bit transfers between the memory and the Front Side busses. The smaller the tRD number, the more often (or more quickly : aka "Latency") data bits are transferred between the memory and the FSB. This is the whole point of the memory tweaking drill. (Note: all that "Strap" does is change this tRD number.)You can adjust tRD with Memset in real time. The change is immediate. Each notch of Performance Level (tRD) is worth ~ 2ns of memory latency, depending on your FSB speed. Play with it. Make an adjustment, then run Everest and see the Latency difference. Keep turning it down until your system locks up. That was one notch too far! Reboot and remember that no-go setting.http://forums.avsim.net/user_files/192324.jpgSet a very high performance level setting (tRD) and note the (higher) memory latency with Everest. Now run the sim. Note the sim's performance. Now set a low performance level setting (tRD) and note the (lower memory )latency with Everest. Now run the sim. Note the sim's performance. How much of a latency change were you able to accomplish? How much of a difference did this latency change make in the sim's performance? Now you know.

Share this post


Link to post
Share on other sites

Yup, and at those settings, PL6 locks up my system instantly. After reading Nick's posts, I started playing around again. I actually got down to 4-4-3-10 but, at only 881MHz and a 1:1 divider. This, of course, forced my FSB/CPU speed down and basically negated the advantages of the tighter timings. Bandwidth and latency were a bit slower and higher respectively but, the 1:1 divider wasn't as smooth. Stability also became a factor. The 5:6 ratio is just smoother on my system. My PL stayed at 7. The biggest thing to understand is that when overclocking, as Nick alluded, it is more than just setting one variable. Changing one setting may have an impact on several others. The only way to do it is trial and error, and watching how your system responds to those changes. You have to be sitting in front of the system and know what it does under normal settings to be able to determine what is causing any problems. I can tell just by the way my system posts (or doesn't), and loads the OS(or doesn't), what the problem might be. Only experience with the specific system will give you this knowledge. No two systems, however identical they may be, will work exactly the same.There's just no way around it. To be successful, you have to put the time in researching and studying. If you skip this step, you can do alot of damage to your systems components without even knowing it. Overclocking is as much art as it is science.Regards,Jeff

Share this post


Link to post
Share on other sites

Good words Jeffhowever as the demo above pointing out the use of memtest did not stress strong enough and correctly.. you must work the CPU FSB into and past the 400Mhz range with a CPU strap of 333MHz (5:6 ratio) before you ever start playing with Memset.The higher you get that FSB up into the 450Mhz range, the better you will do on that STRAP setting. It has to do with how that FSB and STRAP affects changes in the northbridge through the BIOS programming, which is why Jeff sees better performance at 5:6 than 1:1

Share this post


Link to post
Share on other sites

This drill is Entirely about decreasing Memory <> FSB latency (or Everest's "Latency" number). 1) Each 100Mhz of FSB is worth ~ 1ns of Memory <> FSB latency. 2) The "Strap" simply changes the tRD number (Memset's preformance level adjustment). This is certainly not about tweaking a single setting, however the Goal IS singular. Tweak "this" and adjust "that," but remember the - Single Goal - is to reduce Memory <> FSB latency (Everest's latency number).

Share this post


Link to post
Share on other sites

>>The higher you get that FSB up into the 450Mhz range, the>better you will do on that STRAP setting. It has to do with>how that FSB and STRAP affects changes in the northbridge>through the BIOS programming, which is why Jeff sees better>performance at 5:6 than 1:1>>Thanks for saving me the trouble of actually askng the question Nick. I couldn't quite put my finger on it. The northbridge is one of those more elusive settings to me in that it's hard to point to any one symptom and say "Ahh, that's the northbridge". My northbridge is set to 1.61v in the BIOS for an actual of 1.63v. This puts my NB temp at about 55c on a normal day and up to about 63c on a real hot day; as much as 67c (hot day) under load. Would you agree that this is about as far as I should push it? If I leave it in AUTO, I've seen it as high as 1.74v which makes me a bit uncomfortable.Best Regards,Jeff

Share this post


Link to post
Share on other sites

>>This is certainly not about tweaking a single setting, however>the Goal IS singular. Tweak "this" and adjust "that," but>remember the - Single Goal - is to reduce Memory <> FSB>latency (Everest's latency number). Hi Sam,I wasn't responding to you with my comments about single vs many settings. I meant no insult. I hope you didn't take it that way. I was addressing the OP with that. He said he didn't know a whole lot about OCing and I just wanted him to understand that any change can affect alot more than he might realize. It's important to understand the correlation between settings. This is exactly why I always hesitate to give specific settings because I can't be sure of the persons knowledge or experience. I just don't want to be the cause of someone frying their system.Best Regards,Jeff

Share this post


Link to post
Share on other sites

No worrys. The real point is that this is a lot easier than it might other wise seem . . . . and tweaking 'puters is great fun. That tRD excercise is a good one cuz it's just impossible to hurt anything. If you go too far, the system will just lock. A simple reboot resets it all. Using this drill, a user can really see the difference that BIG latency changes make in game. Heck, you can even pause the sim, reset the performance Level and continue the flight. For instance, if 30ns makes "this much" difference, an additional 10ns decrease will be "a bit more." That'll give the user an idea of how much good all this memory tweaking can do.

Share this post


Link to post
Share on other sites

>My northbridge is set to 1.61v in the BIOS for an actual of>1.63v. This puts my NB temp at about 55c on a normal day and>up to about 63c on a real hot day; as much as 67c (hot day)>under load. Would you agree that this is about as far as I>should push it? If I leave it in AUTO, I've seen it as high>as 1.74v which makes me a bit uncomfortable.>>Best Regards,>Jeff>OuchWarning.. off the top of my head I do think the termination temp of that NB is 80c and if I am not mistaken there is no safety for that like the system in a processor unless that BIOS is programmed to shut the system down before point of no return. Do you find it necessary to run 1.63 to maintain stability?My rule of thumb on that is 1.52 without any special cooling. 1.55 and above requires either a cool environment or special cooling for the NB. Auto mode sets that to 1.74? I assume you are not running any AI Overclocking settings. How are you verifying its running 1.74v in Auto? If it is, that would be a first for me to see. Not to say it is not possible however I find it quite surprising auto mode would run the NB on anything other than the specified default for safe operation.

Share this post


Link to post
Share on other sites

>2) The "Strap" simply changes the tRD number (Memset's preformance level adjustment). I have really had about enough of this backyard electronic engineering with an english majorThat one goes hand in hand with "A Raptor only gives you 7 seconds better FSX loading time and no difference after"http://forums.avsim.net/dcboard.php?az=sho..._id=43364&page=What, after FSX is loaded the hard drive is never used????And

Share this post


Link to post
Share on other sites