BSOD "hardware error: consult hardware vendor"

[BonzaiDuck]

I'm looking for insights while I troubleshoot this difficulty.

Here are the specifics of the system:

E8400 CPU
680i Striker Extreme [BIOS 1603 August, 2008]
G.SKILL 4GB 2x2GB DDR2-1000
Seagate 7200.10 320GB x 2 in RAID0
BFG [nvidia] 9600 GT 512MB x 2 in SLI
Cooling: Noctua NH-U12P
Seasonic 650HT PSU
VISTA-64

First, cooling is great.

I ran up the FSB unlinked to RAM to 1600 -- almost immediately 1:1 with DDR2-800 and stock 5,5,5,15, 2T latencies. After a few iterations, I found the stable voltage for 3.6Ghz and a 9-hour test run with PRIME95 small-FFT test. I'd tested the RAM with MEMTEST86+ -- but only for the minimum recommended test-bank iterations and about 5 runs at test #5.

VCORE setting was 1.32V -- rounded up from the 5-digit set value. This, of course, was showing idle monitor readings around 1.28V, and the mobo has a 0.02V Vdroop.

Over the last few days, I've been pushing to see if I can't get up close to 4Ghz. So I'd moved up the VCORE and settings incrementally, until I finally had a 5 hour, 45 min preliminary small-FFTs run -- which I manually stopped with 0 err, 0 warns. At that point, VCORE setting was just below 1.37V showing monitor readings at idle around 1.34V and load of about 1.32V. This last setting follows the pattern set by the details above, at 3.82 Ghz. Coretemp readings, both cores, never broke 59C with room-ambient of a carefully-measured 80F degrees.

So I decided to call it quits for a while on this "OC exploration," and dropped back to 3.6Ghz -- same settings and lower VCORE as shown above.

Here, I decided I'd better submit the G.SKILLs to rigorous "Blend-Test" at the DDR2-800 setting and stock latencies.

At this point, I'd get through about 20 minutes of testing, whereupon a BSOD would pop up saying "hardware error: consult hardware vendor."

My first thoughts about this: I was stressing the memory -- it might be the memory. That's OK -- easy to RMA, and I've got plenty of spare kits. moreover, the G.SKILLs are now selling for about $90.

My WORST thoughts: some sort of damage to the graphics card, maybe the chipset. I just don't see how it would be either GFX card, though, because the PCI_E buses were fixed at stock values, and they're not OC'd. The motherboard has hardly been used more than a couple weeks total. It was a spare board I picked up last year building my other rig. The Graphics cards are as old as this system -- about 3 weeks.

I've swapped in a set of Crucial Tracer DDR2-800's for now after clearing CMOS. I'm testing at the stock defaults (3.0 Ghz) at the moment. Plan to test at the 3.6 Ghz setting.

Anyone ever had symptoms like this? OBviously I'm trying to save myself time and trouble by asking. Or perhaps you have some insights as to the source and cause of these symptoms. I'd be happy as a pig in poop if it just proves to be RAM.

I just don't see the processor going south at that voltage and the brief time it was running at that speed.

 

[Zenoth]

If your spare RAM kits don't show errors, and if you re-try tests on your G-SKILL kit and it still fails, then it's highly likely to be a memory issue indeed, especially if you run CPU-focused test runs and leave the RAM tests behind. But if you really want to test RAM I suggest you to test it under Windows conditions (I.E Windows running, simply).

For that you should get HCI MemTest instead of the regular BIOS one, although the regular version is good for detecting issues at default values, but when over-clocked it's better to stress-test RAM only while Windows run, to make sure that it can sustain that condition under OC'ed frequencies especially for gaming, since obviously no one play games from the BIOS and outside of Windows OS.

 

[Idontcare]

Hey BonzaiDuck my 680i Striker Extreme killed off not one, not two, not three, but all four sticks of my Mushkin redline. That was with active cooling on the DIMMs and not even running the Vdimm as high (I thought) as they were spec'ed by Mushkin to operate.

Mushkin replaced all four sticks even though we all agreed thru emails that the ram wasn't at fault (what are the odds of four sticks going bad?) and Mushkin had lots of experience with DDR2 dying at the hands of mobo's with the 680i chipset. Per Mushkin tech support the 680i, for whatever reason, actually overvolts the DIMMs quite a bit, setting 2V in the bios was putting something like 2.2 or 2.3V to the dimms.

At any rate I replaced the Striker on my own coin, Mushkin replaced all the ram on their coin (could not believe they did that, it was not my request but they insisted) and I haven't looked back since.

 

[BonzaiDuck]

Ah, Idontcare, it's good to hear from an older Striker veteran!

You weren't specific about what you'd done with yours. I had this as a spare board I bought when rolling in coin. I'd panicked in July 07 from a bad BIOS flash -- I must've told the story, but both the story and I are so old, I can't freakin' remember! Turns out -- all I needed to do was order a PLCC chip from BIOSMan. I actually thought it prudent to keep the spare board though.

I've posted extensively here about discoveries very similar to the ones we discussed last year, concerning the BIOS revision CPU optimizations. Remember that?

So the Striker 680i was supposed to be Wolfdale compatible. Remember the trouble with the v.1305 BIOS? They hadn't ironed out the bugs regarding Wolfdale until after BIOS 1504. The newest is 1603, released about two weeks ago! BIOS revision two years from the time the board first became available!

I think I lost both prudence and patience. I should've done a lot of extensive RAM-testing first.

But I'm also exploring some other possibilities, possibly saving some coin on another $80 4GB RAM kit. [And frankly, I should test these things carefully. Right now, I'm trying to replicate the problem with the Tracer 2GB kit -- running Blend Test at 3.6 Ghz FSB=1600.]

This is getting "kinda long-ey," as GHW Bush "41" used to say. Suffice to summarize -- I've got VISTA 64 SP1 loaded on this second system. What bothers me is this: I'm only speculating that the problem COULD come from interaction of PRIME95 and certain "power-saving" features. I've gone through the Power Options screens in VISTA, entering "Never," "Never," "Off," "Off" etc. for sleeping the monitor and HD, disabling sleep and hibernate. What mystifies me: VISTA STILL keeps sleeping the monitor while PRIME95 is running! In other words, I told it "Never," but it keeps putting the monitor to sleep after about 20 minutes. The 20-minute time-frame seems to roughly correspond to the length of time before Blend Test fails. About 5 more minutes, and the Tracers will have passed that threshold. VISTA is such a memory hog, though . . . . the 2GB kit apparently means a lot of paging operations . . .

I'm surprised if you replaced the Striker with another Striker, or that you stuck with the 680i chipset. I think you had some solder pins touching a backplate or something under the mobo? EDIT: Somebody on the ASUS forums -- probably "von Deutschland gekommen," said: "The Striker Extreme is like a German girlfriend. You love her, but she doesn't love you back!"

 

[BonzaiDuck]

Originally posted by: Zenoth
If your spare RAM kits don't show errors, and if you re-try tests on your G-SKILL kit and it still fails, then it's highly likely to be a memory issue indeed, especially if you run CPU-focused test runs and leave the RAM tests behind. But if you really want to test RAM I suggest you to test it under Windows conditions (I.E Windows running, simply).

For that you should get HCI MemTest instead of the regular BIOS one, although the regular version is good for detecting issues at default values, but when over-clocked it's better to stress-test RAM only while Windows run, to make sure that it can sustain that condition under OC'ed frequencies especially for gaming, since obviously no one play games from the BIOS and outside of Windows OS.

What vCORE setting and reported value (from sensors) did you get for the 3.74Ghz setting? I think that may have been a very manageable setting when I tried it -- system ran small-FFT for two hours before I stopped it (0 err, 0 warn.) I just wanted to go on and push it higher.

But check the March 08 article here at Anandtech on the E8500 "The Best Gets Better" -- or something like that. Very good info about Intel's testing and binning perspective. I just felt uncomfortable running a VCORE exceeding 1.37V when the retail box maximum says "1.25V." Even with that -- my reported idle value was 1.34V, and this mobo has a 0.02V Vdroop -- approximately.

 

[BonzaiDuck]

Originally posted by: Idontcare
Per Mushkin tech support the 680i, for whatever reason, actually overvolts the DIMMs quite a bit, setting 2V in the bios was putting something like 2.2 or 2.3V to the dimms.

Didn't we visit that issue before? I personally killed a set of Ballistix DDR2-1000's. The reported RAM voltages -- per my experience, were about 0.05V higher than "set," but if there was bias in the sensors, too, then --- I coulda killed the G.SKILLs. They're rated upper limit is 2.1. But mostly, I never set them higher than 1.975V -- reported as about 2.01V.

Reported values could be "bad intelligence."

 

[BonzaiDuck]

It's beginning to look like defective G.SKILL RAM. I'm actually relieved.

I can't replicate the BSOD with a Tracer 2GB kit. And VISTA has it all choked up. I'd almost think PRIME95 would crash under these conditions just because VISTA is such a memory hog. It's now gone 39 minutes, no problem. As I said, it would BSOD after 15 to 20 minutes using the G.SKILLs.

 

[Idontcare]

Originally posted by: BonzaiDuck
I'm surprised if you replaced the Striker with another Striker, or that you stuck with the 680i chipset. I think you had some solder pins touching a backplate or something under the mobo? EDIT: Somebody on the ASUS forums -- probably "von Deutschland gekommen," said: "The Striker Extreme is like a German girlfriend. You love her, but she doesn't love you back!"

Oh no it wasn't another striker, it was another Asus the...the P5E WS Pro.

Your memory is spot on, can't be too old, or perhaps you are falling behind in eating your share of mad-cow tainted beef?

And you are right, I am just astounded that Asus is still putting out BIOS revs for a 2yr old mobo. That's kinda unheard of actually. Must be part of their internal committment to maintaining the ROG brand.

Hope you get your ram straightened out.

 

[Zenoth]

Originally posted by: BonzaiDuck

Originally posted by: Zenoth
If your spare RAM kits don't show errors, and if you re-try tests on your G-SKILL kit and it still fails, then it's highly likely to be a memory issue indeed, especially if you run CPU-focused test runs and leave the RAM tests behind. But if you really want to test RAM I suggest you to test it under Windows conditions (I.E Windows running, simply).

For that you should get HCI MemTest instead of the regular BIOS one, although the regular version is good for detecting issues at default values, but when over-clocked it's better to stress-test RAM only while Windows run, to make sure that it can sustain that condition under OC'ed frequencies especially for gaming, since obviously no one play games from the BIOS and outside of Windows OS.

What vCORE setting and reported value (from sensors) did you get for the 3.74Ghz setting? I think that may have been a very manageable setting when I tried it -- system ran small-FFT for two hours before I stopped it (0 err, 0 warn.) I just wanted to go on and push it higher.

But check the March 08 article here at Anandtech on the E8500 "The Best Gets Better" -- or something like that. Very good info about Intel's testing and binning perspective. I just felt uncomfortable running a VCORE exceeding 1.37V when the retail box maximum says "1.25V." Even with that -- my reported idle value was 1.34V, and this mobo has a 0.02V Vdroop -- approximately.

My own E8400 is a poor over-clocker, has always been, a very bad batch, although I also suspect that my Power Supply isn't good enough, but it would honestly surprise me. As soon as I go above 3.78Ghz exactly I can get to Windows Desktop but just starting up a game, any, will result in an immediate crash, not even a BSOD, just a system instant hard reboot.

And even though my 3.74Ghz setting is 100% stable (Prime, Orthos, F@H, gaming, anything I throw at it it takes it no probs) I still need a ridiculously high 1.392v on the vCore along with a horribly high +0.25v on the FSB and an unbelievable extra +0.200v on the Northbridge (MCH). The memory is stable, I've ran it at higher frequencies and tighter timings before but I backed them up a bit to more reasonable settings. I've tried many, many things which I haven't mentioned and I could spend a day typing about my tests and so on. But I'm just one unlucky owner of an otherwise superb chip. The thing is the difference in performance from 3.74Ghz to 4.00Ghz would be negligible at the very best.

And as far as temperatures go, at the current OC'ed frequencies it idles at around 34ºC or so, depending on the room temperature. The idle temps are taken from the BIOS Hardware Monitor. And at load, taking readings via EVEREST I get an average (depending on the room temperature, as always) of 45ºC to 50ºC (around 50 during hot summer days and gaming for hours non-stop). I haven't seen higher than exactly 53ºC as of today and that came from Prime95 after 17 hours of Small FFT's. All that is with AS5 + OCZ Vendetta cooler. Just keep in mind that the mentioned temperatures are considering the voltages mentioned too, because at stock frequencies and voltages I haven't seen load temps going above 49ºC if I recall correctly. Although the environment isn't static, during a hot and humid summer day it could have been higher, but "usually" it gets around that and no more even during very long gaming sessions.

 

[BonzaiDuck]

Well -- to answer Idontcare's question -- It looks certain now. The hardware fault is the RAM. VISTA is a memory-hog. Not that the system runs like molasses on a cold day, but under load, its behavior makes you think it's gonna die any minute. And for all that, with the replacement using the paltry 2GB Crucial kit -- it just "takes a licking and keeps on ticking." It was the G.SKILLs that died -- and nothing else.

And I think I'll order something other than the G.SKILLs, even if I send them under RMA for replacement. It bothers me that these things could fail so quickly running UNDER their spec and at voltages just below their spec'd range. Maybe the SE board overvolts them more than I can imagine, but with a setting at 1.975V? That was as high as I'd ever set them in this . . . . obsessive . . . grueling fiasco of a project.

Zenoth, your remarks about the Wolfdales are revealing. I thought I saw -- at forums here and there -- that the OC potential of these CPUs varied quite a bit. It's becoming more certain the more I hear from other E8400 users.

Too bad I lost this memory kit. I could almost . . . . spring for an E8600 E0-stepping . . . .

By the way. The Reapers? I'm guessing the X38 board uses DDR3? I can get DDR2 4GB Reapers for under a Franklin-note. I just don't know how "tweakable" they might be, and these memory makers bob up and down in the ratings as each year and memory standard pass and change. . . .

 

[Zenoth]

The Reapers are awesome, they were highly recommended by a guy I know who's working at my local PC store's technical support department (he repairs, checks, analyze and even tests oc's from clients in his spare time during the week-end), he's been tweaking and playing around with memory since the 486 days, he knows quite a bit, not that it means much to you or anyone else here, but he often recommended me those amongst a few other selects. And from my own experiences so far he was damn right, with that kit I've been able to reach 1060Mhz with 2.2v at 4-5-5-4-12 (default values of 4-4-4-12, since it's the C4 / 4GB edition, there's a cheaper C5 / 2GB edition around).

I bought them retail at the store but here's a link I found of them on Newegg: http://www.newegg.com/Product/...x?Item=N82E16820227267

And their heat spreaders, as my guy said, aren't just for aesthetics, it does help from what he observed and I think he's right, I've never seen that kit fail on me, ever. In fact I could probably push it more with 5-5-5-15 settings like I currently set them at (for stability's sake only, but I know I'm doing this out of a slight paranoia about my E8400's OC which might not be stable with other settings than that even though I haven't fully tested under 4-4-4-12 settings). So because I decided to stay with a 1:1 ratio OC I let them run at 998Mhz on a specific memory strap in the BIOS and voila, I'm stable and it's fast enough, no over-heating, no crashing, nothing bad, but I do know it could be better given a better batch chip and, who knows, perhaps another PSU too.