Random Reboots with 290/CPU at full load

[ccityinstaller]

Hey guys,

I swapped my old AsRock Z77 Extreme4 out for a Z77MPower since I am going to go Tri-Xfire (paid for with mining) and I am having some issues with the build. I am posting it in the subforum since the issue(s) seem to be more related to the GPU then the CPU. I am running the latest CAT 14.2 BETA drivers (also had the old 13.12, upgraded hoping it was a funky driver issue).

I have the CPU running @ 5.026Ghz (50x100 plus the little MSI roundup on the BLK) @ 1.4V (she is a golden chip)..I have tested the CPU through 25+ Passes of IBT on Extreme Mode with Max Memory, and everything is rock solid. I have verified the CPU Vcore via Multimeter and it stays right at 1.407V. Never flickers a single digit when loaded 100%. CPU core temps are 59, 60, 63, and 61C.

I have my first 290 running @ 1075Mhz/1325Mhz (also tested at stock) @ it's default Voltage (1.19V and 50% board power increase via TRIXX) and have it run through 3+ Hours of FurMark without a single issue. GPU Core peaks @ 35C and VRM 1 peaks @ 52C.

Now that I explained each component is working quite well under stress testing, we get to the issue I am having: running them both balls to wall. I am a firm believer that a component isn't 100% stable (if in a single loop) unless every device in the loop can produce it's maximum heat load without any issues.

My problem is that whenever I load both the CPU with Prime95 (small FFTS) and the GPU with Furmark, everything runs fine for 10~20 minutes, and then I get a random reboot. Nothing crashes, no blackscreens, just a reboot. I have noticed that when both the CPU and GPU are loaded, the GPU core clock wants to drop from 1075Mhz down to ~1036Mhz. I have noticed this ONLY when both are loaded 100%.

Anyone have any ideas? The things I have tried so far include:

1) Bumping the cpu Vcore from 1.39 (actual) to 1.407..I did since a tiny amount since the CPU alone was 100% stable @ the 1.39V setting.

2) Bumped CPU I/O from default (1.05) to 1.2V.

3) Tested CPU Vcore and 12V rail of PSU (reads 12.07V under load) with a higher end Fluke multimeter.

4) Upgraded the GPU drivers from CAT 13.12 to the latest 14.2 BETA.

I am ready to tear my hair out with this..I have had to tear the loop down twice already to due one stubborn sneaky ass leak that ended up coming from a tiny crack (yes a fucking crack) in one of my Koolance QDCs. I just want the damn thing to pass the stability tests so I cam resume mining/gaming and know that everything is tuned properly.

 

[Optik]

psu

/thread

 

[ccityinstaller]

psu

/thread

The PSU is a Seasonic X850W Gold that is less then 10 months old and has been rock solid in my last build which had a 7950 @ 1.325Ghz running 100% 24/7 along with the cpu. Do you honestly think that it just randomly started acting flakey? (I am not being smart, I am truly curious)..

 

[Vielgus-Kutas]

run it all without overclock and see if it reboots, if it doesn't it's likely the psu is too weak, if it still reboots it's likely the psu is too weak, it might be some other problem but it looks like psu

 

[ccityinstaller]

run it all without overclock and see if it reboots, if it doesn't it's likely the psu is too weak, if it still reboots it's likely the psu is too weak, it might be some other problem but it looks like psu

I am not sure If I was clear in my first post, since I did mention "going" to Tri-Fire, but at the moment I only have a SINGLE 290. The rest of the system is in my signature. I will give them both a shot at stock and report back.

EDIT: Just brought the system over from the test bench to where it will normally resides behind a UPS..Right now I am testing them both together @ the O/C speeds and I am seeing a ~475W draw, which is 5W more then my 7950 was drawing when it was in place of the 290.

 

[Ultima99]

I am not sure If I was clear in my first post, since I did mention "going" to Tri-Fire, but at the moment I only have a SINGLE 290. The rest of the system is in my signature. I will give them both a shot at stock and report back.

EDIT: Just brought the system over from the test bench to where it will normally resides behind a UPS..Right now I am testing them both together @ the O/C speeds and I am seeing a ~475W draw, which is 5W more then my 7950 was drawing when it was in place of the 290.

Its possible the PSU is acting up, but its also quite likely its fine. Too many people love to shout PSU when theres an issue. I've gone off this advice before (from a certain video card makers tech support no less) and replaced a perfectly good PSU and got what was then a brand new Corsair AX850 and seen no change when in fact they were just sending me iffy GTX280 cards that ran into issues on X58.

 

[ccityinstaller]

Just wanted to post an update to my situation..After massive amounts of time (50+ hours) of trying different BIOS releases and settings, I have finally ironed everything out.

My MB does not like to run my G.SKILL memory at anything tighter then 1600 11-11-11-1T (its a 4x4GB Quad Channel kit rated @ 2133 11-11-11-28 via XMP)..After spending some time over at the MSI forums, it appears I am far from the only one having issues running G.SKILL memory at its rated speed via manual or XMP timings. These issues range from BSoDs/BIOS update failures/Random Reboots (HEY!) etc... MSI is collecting information from users that are reporting the issue, and it seems to apply to all of the current Intel lineup, from X58 onward..There is speculation that G.SKILL was paid by Asus to tune their ram, but I have no confirmation of that.

I also am using the Sapphire Tri-X BIOS on my reference card, which is now 100% stable @ 1.25Ghz core/1.5Ghz memory (Elpida at that!) with 50% board power and a VDDC offset of 50mV..If I push the core to 1250Mhz, The card will perform without artifacts, but will randomly clock itself up and down, much like the whole "retail 290 variance" issue that some review sites ran with a while back.


I am curious if this throttling up/down is normal when you reach the max core clocks the card can support, and if anyone else has noticed it in their testing?