Sorry for asking stupid questions, but this is not something I've ever played with. What do I look for, and what should I change it to?
Printable View
First question is: what kind of memory are you running? A link would be good if you have one.
And I am curious as to your MB model number too.
I have this RAM x2. And this is my MoBo.
I'd love to hear what you did to fix your issue...
The stock timings are Ummm... well.... I don't know. The overview has it listed as 8-8-8-21. The details page has it listed as 8-8-8-24. Use CPU-Z and check the memory tab it will tell you the timings. Then check the SPD and it will tell the timings which are supported.
Ok, I too have a Gigabyte MB so I am hoping that we have pretty much the same BIOS. Boot up computer, get into BIOS. There should be something labled MB intelligent tweeker. Enter that. Then there should be something labeled DRAM configuration. Enter that. There should be something labeled Timings and it is probably set to Auto. Switch it to manual. Now below that should be a tab labeled CAS timings. If it shows a 7, change it to and 8. If is shows 8, change it to a 9........F10 and reboot. I hope this made sense to you. If not, ask more questions and we will continue to work on it. Good luck.
Memtest: Not sure what win7 uses to do that. A good memory test can be run from a CD, and will usually take a day or two depending on the amount you have.
http://www.memtest.org/
Ok - with 8 Gigs of DDR3, the Memtest embedded with windows took 4+ hours.
Something is telling me that joker is either right, or on the right track...
I went and changed the memory settings in BIOS, and I'm at the most stable system I've had in the last week. I have both 5970s in the machine in Crossfire mode, running at stock settings.
With memory clocked back to 1066, at 7-7-7-19, I've been running MW for the last couple hours without incident. I tried putting it on Collatz at these settings, and it crapped out - but MW is running well so far.
Any thoughts on the next settings I should try or anything like that?
My only last thought is to clock it back to 8 since that is what it is rated at and see what happens.
Holy crap - I woke up after going to bed, and found my computer... running! And crunching!
Did you mess with it again or did it just "fix" itself?
Haven't touched it again. After the computer has run without erroring for about 18 hours now, I'm so giddy I'm afraid to touch anything. I might get back to this tonight or tomorrow, depending on how things go.
But this clearly seems to have been some RAM timing issue. No clue how or why it happened.
If it is working then, as my doctor keeps telling me............DONT PLAY WITH IT!
The Silver Hammer passed all tests except one, and the resulting error is the name of the thread. It ran on MW for a day and a half without erroring out - that made me happy. I had backed the RAM speed down to 1066, with timings of 7, 7, 7, 19. When I put it on Collatz, crash. Put RAM back to stock settings. Crash.
So now I need help (again) - what should I try next? I'm a little scared of changing things willy-nilly in BIOS, and need some suggestions of where to go...
Again, I'm rocking this RAM. Current timings and SPD page from CPU-Z below.
http://i1194.photobucket.com/albums/.../SH-Memory.png
http://i1194.photobucket.com/albums/...n82/SH-SPD.jpg
This is the last thing I can think of as far as memory goes. Manually set your Command Rate down to 2T.
Done. Didn't fix it. Bah.
So is there a way to adjust the GPU RAM settings, aside from the RAM clock? I ask because I can get my machine stable on MW, but not on Collatz. MW loads heavier on the GPU than Collatz, so I don't think load is the issue. But since Collatz uses the GPU RAM so differently than the other ATI projects, is there some setting I should adjust there?
Note again, that I've got everything running at stock settings...
Errrr..... Hmmmmmmm, I have an idea.
You have 2x sticks of RAM. Pull one out and run Collatz. Your system will probably run a little slower. After the test, pass or fail, replace the installed stick with the one you pulled. Test again.
Reasoning: It is rare to have all the RAM sticks go bad. And since it is fairly easy and painless to install/remove RAM, this will be a fast way of physically testing for bad RAM. So one of the sticks should be able to run Collatz. That would be the "Good Stick".
The "Bad Stick" is therefore possessed. Call for a Priest and have the silicon exorcised. Then burn it at the stake just to be sure.......
And punch a few more children.......
You know, I had a not dissimilar thought, but didn't know particularly how to test it...
I actually have 4 sticks - 4x2Gb sticks. The Mobo has two sets of two "matching slots" (technical term ;) ). Can I just pull one stick at a time and run with it? And since all four sticks are identical, is the empty RAM slot an arbitrary decision, or is there one (or two) slots I need to make sure are filled?
@Harley - My BIOS is telling me it's running at 1333, but CPU-Z is picking it up at 1339.2. What do I change to fix that? What is a RAM ratio?
And yeah, kicking something is not far off. Maybe a child petting a dog... :rolleyes:;)
OK, here is the deal.....
I decided I needed to read the manual for a change :)
My idea was to have only one stick of RAM on the MB which your MB does support. The caveat is that the memory will run a LOT slower because its dual channel memory.
Since you have 4 matching sticks you can use a modified version of the test.
It would be best to put a small piece of masking tape on each stick to label them as 1 through 4 or A through D.
Now, using only memory slots #1 and #2, test the sticks in pairs. 1+2, 3+4, 1+3, 2+4, 1+4 and 2+3.
Seems weird that whatever did this messed with so much. RAM timings, gpu clocks, etc. But since everything else is so out of whack, might want to check your voltage settings for the RAM too.
Actually, do a binary test. Theoretically one pair should work OK. So test 1+2, then 3+4. record if either pair works OK. If either pair fails, you now swap in one from the other pair to determine which one is at fault. So three tests total rather than doing all 6 combinations. :)
Time for a 1000W PSU!
He says to someone with a 1000W Gold Rated PSU already in there...
And the PSU is running MW right now, so I don't think it's a power issue. ;) Good suggestion, though...
You never know when your PCI bus has gone bad.
I have a similar machine. Similar, in that it a PITA. It's an i7 with dual 5870s, and 6x1gb RAM. It's about a year old now. For the first 6 months, it did like yours is doing. Constantly crashing with some projects, not others. I tried changing GPUs around, tried swapping DIMMs around, replacing DIMMS, less, more, different combos. Nothing made any difference. Then one day, it was stable, so long as I didn't put it on a heavy RAM project. 6 months later, I have been able to slowly add heavy RAM projects, and no problems so far. The lesson learned? There is no god damn rhyme or reason to this stuff sometimes. God hates you, and you have to put up with it for as long as he has his eye on you. Either that or life is completely random, which is even more unnerving. ~X(
Yes, Sauron is after your a$$!
And, I'm back to this. Woohoo.
I think I have narrowed it down to GPU RAM (yes, I've said this before...). The reason I say this is that my system is now rock solid stable, except when attempting to run Collatz and/or upping my RAM clocks. I'm currently running at 825/600 on MW (stock being 725/1000). Stable as hell. Temps great, no hiccups or lags or anything. The second I crunch a Collatz WU, or just a few seconds after upping the GPU RAM clocks, I get this:
http://i1194.photobucket.com/albums/...3-08223851.jpg
http://i1194.photobucket.com/albums/...3-08223732.jpg
When I run the "Test Custom Clocks" function from CCC (even with stock settings this happens), I get this:
http://i1194.photobucket.com/albums/...3-08224428.jpg
I have used ATI OverVolt to adjust RAM voltages, but that has a minimal effect, and resets itself after a computer restart anyway.
Any ideas, or should I continue punching myself in the nuts?
I hate to be a naysayer but it could be a PCI-E bus problem. Your MB might be at fault here.
Joker has a good point there. The bottom line is you need another system for testing or at least an equivalent GPU and PSU (Known to be good) to swap into your system for comparison. What about taking the funky system or one of the GPUs to work with you and use one of the lab boxes for a test bed?
Good news: It's not the motherboard/PCIe slots.
Bad news: It's the video cards.
I took the cards in to work, and tried them out there. Same error, same artifacts, same issues. Happened in combination and when I only put one card in at a time.
Harley has been a champ, and has been PMing me a ton of ideas on how to fix the cards, but I've had no luck so far. Am I at RMA stage, or are there other things I can try? It's weird to me, because these cards work great on MW, but freak out on Collatz.
This is now out of league. Good luck. %%-
I'd take the RMA route if the option is there. I hope they're not Sapphire's, I had an awful time with them recently on an RMA. I think you would have the same problem with MW if their work units ran longer.
Oh yeah, I forgot. You've punched puppies, kicked children petting puppies, and punched yourself in the nuts, but have you kicked a kitten yet?
You might have to go to your local petting zoo and kick some goats and sheep in the nuts. X_X
I think its time to grab the Bull by the horns, the Tiger by the tail and kick them both in the nuts.