Newsflash: QVLs are Meaningless

[gordan]

You'd think that memory listed by the motherboard manufacturer as supported would in fact be supported at the speed/timings the memory is rated at, right? Wrong!

DFI list F2-8500CL5D-4GBPK as supported on the UT 790FX M2R board. So, you'd reasonably expect it to work at it's rated 5-5-5-15 timings, right? Except there's a minor problem - the motherboard BIOS doesn't offer the option to set tRAS to less than 16, so the nearest it can be set to is 5-5-5-16. Which would be fine, as loosening the timings won't cause instability. Except there's another problem:

DIMM EPP SPD also lists tRC: 48, tWR: 13. But the motherboard doesn't allow tRC to be set higher than 41, and tWR to be set higher than 8! So these two timings cannot be loosened enough to get the settings within the memory spec! And yet the memory is listed, as DDR2-1066, and supported.

It's official - DFI suck!

EDIT: Oh, and no, 5-5-5-16-41-8 isn't stable.

 

[decto]

Not really a newsflash.... the memory controller is built into your CPU so it is dependant on the quality of the CPU, perhaps you have a dud. Assume you have phenom as X2 64 only supports 800FSB official

Other factors that can affect include the voltage of the CPU and DDR (2.0V - 2.1V) as set in BIOS. Memory timings being set correctly, it should work with Auto but can still need some tweaking. There should be no issue setting tRas to 4 upwards, are you sure you're changing the correct setting? tRC should be in the 20's ... the SPD data is usually very slack.... set everything to Auto except the 5-5-5-15.

If it's not stable ... how does this show .. have you used memtest ... does it not boot....


If you're looking for help with your issue then post a photo of your memory settings and detail your CPU , OC if any and system details.

 

[gordan]

It's a Phenom 9950, supposedly AM2+, which _should_ support up to DDR2-1066.
The problem is that the EPP profile lists 5-5-5-15, tRC=48, tWR=13. BIOS only allows a max tRC of 41, and max tWR of 8.

Yes, I did set the vDIMM manually to the lowest setting >= 2.10V (Actually 2.12V, but should be close enough). There is a thread that someone on the DFI forum linked me that says that OCZ Reaper modules that are supposed to be 5-5-5-18 need to be run as 5-6-6-24 on the Phenom, and tRC should be 30 (which seems odd considering the other settings are being relaxed). So the "auto" setting of 5-7-7-20 was actually pretty close, and probably should have worked. The key problem appears to be linked to the tRFC for each bank being auto-set wrong (way too short). Supposedly it should be 195ns on all banks, but since that didn't appear to be in the SPD/EPP dump, I'm at a loss as to why the BIOS didn't actually set it correctly (since it is implied that "it's a Phenom thing").

It boots (always did), but it errors out under OCCT RAM test after 2-12 hours. I don't use memtest because it is completely useless for stability testing. So far it seems stable for 5 hours of OCCT, but I won't trust it unless it manages to be error free for at least a few days under OCCT test. Only time will tell.

 

[Senture]

Could very well be a duff board. My P35 DS4 gives memory errors in memtest no matter what sticks I use and it fails Prime 95 blend in a few seconds. Hardly ever crashes (always been thanks tothe useless Netgear drivers) and the sticks check out fine in another machine at the same settings.

 

[decto]

You seem to have all the bases covered.

If you're getting 2+ hours of OCCT before a failure I'd suspect it may not be a true memory issue. Unstable memory would see you with a stream of errors especially if this was due to the tWR. A lot of performance ram uses a value of 6 so 14 seems to be a programming error or fail-safe setting.

Have you adjusted any of the CPU settings, eg a small increment in voltage? Are the tempertures of the CPU and case well controlled?

AD

 

[gordan]

I'm not OC-ing, and since 1066 is supposed to be in spec of both CPU and mobo, I see no reason why anything should be over-volted, either.

I'm pretty sure it is a memory problem. OCCT is a better tool than others, but 2 hours isn't enough to test properly, even with that. As I said, I don't accept it's stable until it lasts a week with OCCT. I like my systems really stable for 24/7 operation - especially when I'm not actually overclocking!