Higher frame times when overclocked???

[CJ-5]

When I overclock my computer (sig rig) my frame times get longer by several minutes not shorter. :shrug: I use AI Booster to set the OC and I'm only folding one instance at a time and it is a QMD. I do leave the rig on for several frames to let the new frame times adjust to the computer having been restarted.

Could it be that AI is competing for CPU cycles enough to slow down folding?

Has anyone had any experience like this using AI Booster to overclock Asus boards that could confirm this, or maybe has another explaination as to why the frame times get longer when overclocked?

[Macaholic]

AI Booster can take up some processing power, but I would take a closer look at your RAM timings. Loose timings can easily affect output.

[muddocktor]

Also check for marginal stability. I've seen that before when overclocking a system hard to where it still looks to be stable at first glance, but then after running prime or F@H you find random errors. A lot of times it has to do with ram timings, like Macaholic said.

[ChasR]

You also have to be sure you're comparing frame times from the same length QMD. some have as few as 2040 steps. The longest I've seen so far had 2237 steps. Each step (after convergence) takes very close to the same amount of time. If you compare a 20 step frame to a 22 step frame, you'll see a 10% difference in the frame time.
Effective latency is a crucial factor in QMD folding. CAS 2 ram at 400MHz has a far lower effective latency than CAS 3 at 500 MHz. Since the QMD data set doesn't fit into the cache, which has a much lower latency than ram, ram latency makes a big difference. AI booster is likely loosening timings to attain a high FSB and cpu speed to the benefit of most every application except folding QMDs.

[CJ-5]

Thanks for such quick, and I might add helpful, responses.

I see the general consensus, from three most highly respected team members, is ram timings. I really don't know what my timings are set at and from what you say, ChasR, AI is probably changing them (looser) for me anyway.

My ram is Kingston pc2 4200 value ram with CL 4 latency and they're not even a matched pair, so yes my ram is probably the problem.

It looks like I'm going to have to get better ram and take control of my overclock settings instead of letting AI do it for me. I just didn't want to stop folding long enough to go through the learning curve (overclocking noob) which was why I was hoping AI would do it for me.

[Macaholic]

If you've got the funds, I would highly recommend dual channel low latency. You can check your timings easily by using CPU-Z (http://www.cpuid.org/cpuz.php). Yes, manual settings in the BIOS are really the best way to get a finely tuned overclock. Have fun!

[CJ-5]

Here's a follow up:

I unlocked the multiplier, with the bios, which provided me with an additional multiplier of 14 to work with in addition to the default multiplier of 17.

I manually set the FSB to 266 MHz with the newly aquired multiplier of 14 to get a CPU clock of 3.74 GHz. I set the ram frequency in the bios to 533 MHz, so I have a 1to1 FSB to Dram ratio with ram timings at 4,4,4,12. (Can't go any tighter on this value ram; haven't tried changing the ram voltage as my learning curve hasn't gotten that far yet.)

This combination has resulted in a stable overclock which has reduced my QMD frame time rate by about 1.5 minutes or around 12 minutes and 40 seconds per frame folding only one QMD.

Conclusion: Manually overclocking is a lot more fun than using AI Booster and much more educational. Sometime in the near future I will get better ram. Then the real fun begins. :)

[TollhouseFrank]

congrats on the findings dude!

i'm sure with a slight overvolt on the ram, you can do some tightening of those settings to like 3-3-3-8

[ChasR]

Tightening tRAS could prove counterproductive.

[TollhouseFrank]

it could? i thought on Intel Machines like his, that a lower tRAS would benefit him, whereas a higher tRAS benefits A64 guys.

Did i misunderstand this?

[CJ-5]

ChasR: Tightening tRAS could prove counterproductive.

You're reffering to counterproductive as far as folding QMD's and not benchmarks right?

[ChasR]

Actually my data on tRAS on DDR2 has been superceded. Anandtech benches on DDR2 memory show that you get higher bandwidth and better stability with tRAS in the 10 range on 4-4-4 memory. However there are some newer DDR2 memory modules that run @ 3-2-2 that will see a performance boost from lower tRAS. Anandtech found bandwidth to be the same from tRAS 2 to tRAS 9 at 3-2-2 timings. It was slightly lower (3%) at 10 to 12. Running 3-3-3-8 is probably OK and also probably provides the same bandwidth as tRAS 10.
Mushkin says tRAS shouldn't ever be less than CAS latency + tRCD +2 on DDR memory. Obviously some memory runs fine with less but probably has no greater bandwidth. Apparently on DDR2 memory tRAS can be set much lower but to no benefit. As I understand it, tRAS latency is sort of backwards from the rest of the timings. tRAS is the number of clock cycles tramsmission of data is allowed to occur. The others are how many clock cycles transmission is delayed. Set tRAS too low and the transmission window closes before the data can be sent thus reducing bandwidth and reducing stability.
You had it right on Frank.

[orionlion82]

wow, its amazing how much you can learn about your hobby when you fold as well.

[CJ-5]

I just changed the ram timings to 4,4,4,10 as ChasR suggested based on the Anandtech review of DDR2. Everything seems to be stable without an increase in ram voltage.

Still contemplating TollhouseFrank's suggestion of raising the ram voltage and going to 3,3,3,8 on the value ram. It's kind of like crossing a raging mountain river in a Jeep CJ-5. It looks worse than it probably is, you just have to get up your courage and try it. Of course the real thrill is getting your feet wet not sitting on the bank contemplating.

A little more comtemplating and research is needed as I don't want to fry my computer. Right now I know just enough to be dangerous and probably break something.

[pscout]

a small bump in ddr voltage won't hurt anything IMO. Not sure what defaults are on that board . my p5wd2 are 1.8 ... bumping to 1.9 would not cause a problem.

But before bumping volts, bump the timing first ... you may not need any voltage bump to increase the speed with stability ... every stick of ram is a little different.

[CJ-5]

ChasR: Apparently on DDR2 memory tRAS can be set much lower but to no benefit. (this according to Anandtech)

Tightening the ram timings to 4,4,4,10 did take maybe 15 to 20 seconds off the frame times. So now I'm encouraged or maybe it was just a fluke. Only time will tell.

pscout: But before bumping volts, bump the timing first ... you may not need any voltage bump to increase the speed with stability

I tried 3,4,4,12; 4,3,4,12; 4,4,3,12 the other night and the computer wouldn't boot. I'll try it again using tRAS=10 instead of 12, although I wouldn't think it will make any difference. Maybe I'll be suprised. If it doesn't work I'll get my feet wet and up the ram voltage by 0.1V and try again.

[pscout]

not sure how you ensure stability but i like to run prime95 for a while before i put it to the folding test.

If you want to test on folding, backup the WORK subdir, queue.dat and unitinfo.txt before you run the console and disable your network interface ... that way if you get an early_unit_end on the WU you can just delete the work sudir and the 2 files from the fah directory and restore the ones you backed up. If you leave the network enabled it will report back the failed unit and you will need to start a new one ... makes more work for stanford.

I would run prime95 for a while first before trying the folding test ... if the folding test gets to completion ok you can just reconnect/enable your network so it can submit the WU.

[CJ-5]

pscout: I would run prime95 for a while first before trying the folding test

Too late, but I'll keep that in mind for next time. It does seem to be the safe thing to do.

I really never got far enough for folding to begin after start up except once. I was looking at CPUz and the computer restarted on its own. I believe the restart probably coinsided with FAH starting up. The FAH files seem to be ok since they never got a chance to run. All other attempts never made it into windows. There was just a message about bad Bios checksum.

I did not raise the voltage higher than the 1.9 volts you suggested since this is value ram and I don't know what it's upper limit voltage is.

I think I'll wait to continue until I get some performance ram. Most probably I am at or very near the limit on this ram at least for this setup and there is not much performance gain (for folding) to be gained. I would have to try raising the overclock on the CPU to get any additional gain. Something for me to think about.

This whole experience has added a new dimension to folding for me. Thanks for everyones help.

[ChasR]

If you're having stability problems, lowering tRAS isn't the thing to do. Since benches show tRAS makes very little to no difference in performance, it should be left high until stability is acheived with the other timings, that do make a difference. Higher tRAS will result in better stability.

[pscout]

If you're having stability problems, lowering tRAS isn't the thing to do. Since benches show tRAS makes very little to no difference in performance, it should be left high until stability is acheived with the other timings, that do make a difference. Higher tRAS will result in better stability.

Good to know ... i am just starting into the ddr2 tuning game ... once i get some more of it ... have been in the process of building a couple of rigs over the last several weeks including an rma of a p5wd2-p, and have a daul channel kit split between the 2 mobo's so am only running single channel atm.

There certainly are lots of dividers on the p5wd2! And looking for compatible memory is challenging ... seems many buyers assume any ddr2 mem will work on the board but this does not appear to be the case!

Not sure if the same issues exist on the p5ad2-e p but research before you buy.

It is a dimension of configuration i never worried about with ddr mem.

[CJ-5]

pscout: There certainly are lots of dividers on the p5wd2! And looking for compatible memory is challenging ... seems many buyers assume any ddr2 mem will work on the board but this does not appear to be the case!

I certainly have less experience than you or ChasR in choosing memory. I've been looking for performance ram for several weeks now and don't really know how to make a decision as to what is good for best all around performance. I would think that relative to my inexperience in general I would feel more comfortable staying with manufactures such as Crucial who promise a particular module will work on a particular MB.

As you point out there are lots of dividers available on these Asus boards with native support for several FSB settings and ddr choices. This makes the process of selection even more difficult for me. With lower latencys being desirable but increasing as ddr values go up, I get really confused as to what to choose. Right now I'm leaning toward Crucial PC2-6400 with 4,4,4,12 timings at 2.1 volts. Tomorrow I may learn something that changes my mind. This is my quandry. Probably what I'll do is give it a little more time for some test reports to come out that will solidify my decision.

[pscout]

I am in the same boat as you ... i have some corsair 5400c4 split between 2 machines atm. Just swapped sticks betwen mobos to prove my rma'ed p5wd2-p worked with dual channel (it did thank god!) ... in the process i put the supposedly other matched stick back into my p5wd2 with an 830 oc'ed to 3.9 and it trapped/rebooted twice before turning in an early unit end!

... just proved that matched for DC is only a relative statement based on the test equip the manufacturers use!

lowered the oc a bit and all seems fine again.

I have some pqi turbo 667 should arrive this week and am contemplating some ocz 5400 EB ... once they arrive i will have something to play around with in dual channel mode.

ddr2 seems very much an adventure being so new ... and a bit of fun too!

BTW ... i have been researching the forums ... the guys that supply the official ocz support don't seem to have the latest mobos for testing so search the web well ... not just the manufacturers links!

[ChasR]

I haven't been paying attention to DDR2 ram. Now that OCZ has come out with DDR2 that runs @ 3-2-2 (PC2-4200EB) and has tremendous headroom for overclocking, it's time to start. Maximum memory bandwidth is 7000+ on Anandtechs memory bench machine.