I've been having a number of problems with a system I recently built, and I am beginning to think it might be the motherboard itself. Here are the specs:
Gigabyte MA770T motherboard
AMD Athlon II X4 630
OCZ Gold 4GB (2 x 2GB) DDR3 1333
POWERCOLOR Radeon HD 5770
Seventeam ST-750PAF 750W ATX 12V V2.2 power supply
Western Digital 500GB Green hard drive
Out of the box, I wasn't able to install windows 7 64, as it kept blue screening. I eventually pulled out a stick of RAM and got it to work. Seemed to be fine for a couple of days, then started freezing with vertical gray stripes filling the screen. It would always crash about 30 seconds into an avi file. This seems to be a common issue with the Radeon 5770, but there were other issues as well. I put in both RAMs, made sure the CAS settings were correct (9-9-9) and ran memtest. Both sticks came back with errors, always on test 8. I RMAed the RAM. After putting the new RAM in, everything seemed to be fine for a day or so - I tested out an avi file and it made a few minutes at least. I didn't have time to go all the way through, but it got past where it used to.
I then ran another memtest for 25 hours, with both sticks in, and it comes back with errors on test 5. It seems highly unlikely that I got two faulty memory kits, so now I am suspecting the mobo itself. Additionally, I have run Prime95, both with the old memory and the new, and it had errors.
Any suggestions on how to narrow this down, or should I RMA the mobo at this point?
Assuming that you're doing the (obligatory) "Load Optimized" from the BIOS to initialize your RAM parameters, you might take a look at the procedure here: http://www.tomshardware.com/forum/269711-30-ma790x-ud4p...
which will let you compare the theoretically correct sub-timings (the SPD values) with what you're actually getting fron the init...
Could you please post a full MOBO p/n? I'd like to pull down a manual copy (my normal collection is inaccessible - my workstation has been down now for eighteen days waiting for a PSU cross-ship from zalman ) to look at the documented memory settings; I went half-blind before I thought to go to imageshack and look at the originals, and I think I'm about fried for tonight - but - several odd things jump out... First off, the EPP table lists 609MHz, which, besides being a really peculiar value, should be 1218, not 1333 (which should show 666 in the table); second, the Column Address Strobe values, beside being 'greyed out' seem, in both examples, to be showing a value two less than desired - and I'm wondering if this is a program peculiarity, or if the BIOS is showing the same values?
In addition, as I can't access my 'table of interesting pointers', I've got to search for two things - a reference page that shows the structural differences between DDR2 & DDR3 (maths to get some of the sub-timings), as well as an article that relates specific MemTest errors to likely causes...
Ok, I've d/l'd the manual (actually, manuals, as I found there is a different one for the rev1.1 MOBO, but examination of the memory settings listed shows no variation in the memory settings...), and can maybe give a couple hints.
First, the 'peculiarities': I still find it odd that the SPD lists its fourth slot as 609MHZ (DDR3-9745), rather than the claimed 1333 (DDR3-10666); tRFC being set in nanoseconds, rather than counts, as well as being set on a 'per-DIMM' basis; and the fact that the 'extended' settings apper in the fourth 'slot' of the SPD table, rather than the fifth as I've usually seen it...
My overall advice would be to set the RAM up as 1066, for a couple of reasons.
First, the available 'range' of tRC; tRC is typically tRAS + tRP, with tRAS being tCL (otherwise simply known as CAS) + tRCD + 2; while the tRC of all of your shown settings (the first SPD for 1066, the fourth for a nominal 1218, and the settings from OCZ) aligns with this, the tRC calculates for the 1066 setting to be 23 - however, for the nominal 1218 of the fourth SPD, I get 29 - but your MOBO's BIOS appears to limit this setting to 24 or lower - that, alone, tells me it's never going to be 'happy' at the highest SPD, much less the stated 1333... That said, your 'current settings' page on both shown speeds lists 27 and 33 - and I can't figure out how they got 'set'?
Second, memory speed really doesn't scale on processors with on-board memory controllers (actually, on any processors - I have long been an advocate of only using fast DDR2 on systems intended to run significantly higher that 1600 FSBs, where you run into a 'memory-limited' bus with DDR2-6400), as you're never ever doing what a 'synthetic' (that's why they're called 'synthetic'!) memory benchmark tests - long, sustained read from/writes to memory; in 'real life', you're always grabbing stuff in short bursts - latencies are all that 'counts' - the main contribution of faster RAM is advantages to latency in 'setting granularity': latencies are physical periods of time (not 'so many' counts of the clock - the way they're set); if your memory clock 'just happens' to hit a spot where a latency - say RAS to CAS turnaround time - calculates out to 6.1 'counts', you have to set it to seven, always the next highest integer... If your clock speed doubles, that tenth of a count which forced you to go to seven, calculates out to 12.2 counts of a 'double-the-speed' clock, allowing you to set 13, instead of the equivalent 14 that the slower clock required - and that's pretty much it! There is a tiny contribution from the higher memclk speed, but, unless you're running a rendering station, doing 3-D work or FEA (finite element analysis), or transcoding video a lot (and not using either CUDA or AVIVO CODECS), you'll never 'see' a subjective difference... One of the reasons you will hardly ever see me trying to help with P55/X58 problems is that no one appears willing to listen - they've been sold a 'bill of goods' by the memory manufacturers, refuse to look at tests plainly showing the lack of 'scalability' of excessive memory speeds, and wind up having major troubles with other parts of their overclocks, simply to accomodate uselessly high memory speeds! There's a reason why Intel only supports 1066/1333!
I'd leave the "LoadOpt" settings, with a couple minor changes: based on my calculations, tRFC comes out to 111 or 116 nanoseconds (assuming I'm calculating it correctly - I almost never do actual pico-nano-second calcs, just 'comparative' ones ), and I think OCZ knows something here, as I see they've instructed you to set one pair of them to 160 ns - my guess would be to set them all to 160; the maximum read latency is listed at 59 for the 1066 SPD, but LoadOpt seems to set it at 57 - I'd try to set it at 60 or 64, either in CPU-Tweaker, or AMD OverDrive; and tFAW, which is hardly ever used, should likely be at 20, as in the OCZ settings...
Memtest ran for over 24 hours with no errors on the Optimized settings now. A shorter memtest ran with no errors on the OCZ recommended settings, and I haven't had any problems for a couple of days with them now. Attached are the settings now, with your recommendations, if I've understood them correctly.
Additionally, I'd like to look into slightly overclocking everything I can here. Any recommendations, or should I move that discussion over to the overclocking board now?
I've mostly experienced that particular error when OCing, with the CPU voltage still a 'tad' too low - but, I only do Intels - for AMDs, may be indicative of other troubles...
Got most of an OC for you - at least enough to give a try:
First, the 'preliminaries' -
On the "Advanced BIOS Features" page:
"AMD K8 Cool&Quiet control" to "Disabled"
I see no support for, in the BIOS, or at AMD's product page for that processor, 'C-states'; they would appear in your BIOS as entries for "C1e", "C3/C5", or the like - if they show up, either on "Advanced BIOS Features", or "M.I.T.", disable 'em all; the point of 'Cool&Quiet' and 'C-state' controls is to reduce voltage and frequencies when the processor is lightly loaded, or idling; as, in OCing, our main intent is to increase voltages and frequencies (within reason, of course), the two aims are inimicable...
"Full Screen LOGO Show" to "Disabled" (doesn't have anything to do with OCing, just for convenience entering the BIOS faster & easier...)
On the "Integrated Peripherals" page:
"Legacy USB storage detect" to "Disabled" (for 'boot-loop' prevention in case of 'oopsed' BIOS, CMOS parameters...)
& on to the 'main event' - the "MB Intelligent Tweaker(M.I.T.)" page:
"Advanced Clock Calibration" - far as I can tell, this only works with BEs, or, sometimes, for 'unlocked' cores...
"CPU Clock Ratio" should already be "14.0"
...somewhat out of order...
"CPU Host Clock Control" to "Manual"
"CPU Frequency (MHz)" to "240" (20% - only way I know of to OC, as only BEs have unlocked multipliers...)
"CPU NorthBridge Freq." ...depending on how expressed, either 2160, if it increments by 240, or x9, if a multiplier...
"PCIE Clock (MHz)" to "100" (not 'auto')
"HT Link Frequency" to "Auto"
"Set Memory Clock" to "Manual"
"Memory Clock" to "x5.33" (this should give you 1280 or so - pretty close to the nominal 1218 from the SPD...)
"DCTs Mode" to "Unganged"
...remainder of memory timings/sub-timings per the OCZ settins, set all tRFCs to 160
"System Voltage Control" to "Manual"
"CPU NB VID Control" to "+0.10V"
"CPU Voltage Control" have no idea where this is normally; if below a volt, bump up to 1.10; if over a volt, bump to 1.25 - should not need to go above 1.325!
"DRAM Voltage Control" to "1.60V"
and I think that should do it... May reboot twice at the <F10> 'save&exit'...