Stability issues running a quad core (775/750 SLI)

Discussion in 'Intel Processors' started by Mr. Bluntman, Apr 22, 2019.

  1. Mr. Bluntman

    Mr. Bluntman [H]ardness Supreme

    Messages:
    6,392
    Joined:
    Jun 25, 2007
    So, I've recently picked up a new retro system and I'm having an issue maintaining stability with a Core 2 Quad installed. System specs are listed in my signature as my secondary for brevity. With a Core 2 Duo E8600 installed the system is perfectly stable, but as soon as I install the Q9450 the system will crash within 10-15 minutes (BSOD with memory dump to disk, reboots). Bumping voltage on the memory to 2.1v (kit is model KHX6400D2LLK2/2GN for reference) and voltage on CPU to 1.3375/1.35v will help stability for up to a couple hours, but system will still BSOD, even idle at the desktop. It's not load dependent. Also of note is that in CPU-Z the CPU is detected as a Q9450S - which it is not AFAIK.

    Am I looking at a bad CPU? Are 750i SLI boards just that shitty? Or am I completely missing something?

    Any help would be appreciated, and can provide additional info as needed.
     
  2. Armenius

    Armenius I Drive Myself to the [H]ospital

    Messages:
    16,827
    Joined:
    Jan 28, 2014
    Do you have the latest BIOS for that board? According to the website support for the Q9450 was added in 0502.

    Do you have the 4-pin auxillary power for the CPU connected? I know with my board (P5N32-E SLI) it could boot without it using a quad core, but you would always run into issues.

    Have you tried using the default settings in the BIOS (in case you changed anything for the dual core)?

    For memory I remember needing to bump up the voltage to 2.2V to get it running at 1066 MHz, but it was just fine operating within JEDEC specs.
     
    auntjemima likes this.
  3. Mr. Bluntman

    Mr. Bluntman [H]ardness Supreme

    Messages:
    6,392
    Joined:
    Jun 25, 2007
    Yes, revision 1401. Was 1001 when I recieved the board and the system had even more issues before I updated.

    Oh yes. Before I power on any build I inspect everything to make sure nothing is unplugged that shouldn't be, and that everything that does need plugging in had been so already. Amazing that your board worked at all without the 4-pin connected!

    Yes. I've also tried swapping PSUs and RAM to rule it out. No effect.

    Memory settings being off other than a module being bad was what crossed my mind first. Restoring to JEDEC defaults didn't help. Bumping to 2.2v didn't either.
     
  4. Zareek

    Zareek Limp Gawd

    Messages:
    191
    Joined:
    Sep 5, 2011
    I had a C2Quad Q9400 it was solid as a rock overclocked to 3Ghz with 4 x 2GB sticks of RAM @ DDR 800 and then one of the memory channels went bad. Just out of no where, no mater what I did one of the memory channels would produce errors. I tried a different motherboard that had a fully functional E6600 in it. Obviously down clocking and upping voltages. I tried a bunch of ram, I even tried single sided and dual sided DIMMs. Nothing worked, I eventually gave up on it. Nothing like taking more memory bandwidth away from an already memory bandwidth starved quad core.
     
    Armenius likes this.
  5. Armenius

    Armenius I Drive Myself to the [H]ospital

    Messages:
    16,827
    Joined:
    Jan 28, 2014
    If the socket looks fine and the pads on the CPU look fine I'm going to say it's a bad CPU.
     
  6. GotNoRice

    GotNoRice [H]ardForum Junkie

    Messages:
    9,110
    Joined:
    Jul 11, 2001
    Socket 775 Nvidia chipsets were notoriously unreliable compared to Socket 775 Intel chipsets of the era such as the Intel P35 and P45. The 750i was not bad as older chipsets such as the 680i but still nothing compared to the stability of the Intel chipsets. IIRC the nvidia chipsets tended to run very hot, so maybe you could try to increase cooling around the chipset area. Either way if you are basing your overclocking expectations around how that CPU might perform in an intel chipset motherboard, you are likely to be disappointed.
     
  7. Mr. Bluntman

    Mr. Bluntman [H]ardness Supreme

    Messages:
    6,392
    Joined:
    Jun 25, 2007
    I'm not overclocking at all, for the record. I know these chipsets suck at FSB overclocking and really are only any good for multi-only OCing with the X and QX series Core 2. Any 32-bit version of Vista leads to BSODs at random intervals. I'm burned out of trying any new versions immediately though as I've been troubleshooting and reinstalling different versions of Windows on this thing for nearly a week. I have stumbled onto something though; switching to 64-bit seems to fix the crashing. I've been running 16h without a BSOD so far. With 32-bit it would BSOD right on first boot. Strangely it only does this with the quad core installed, it's well behaved with the dual core CPUs I have. Could it still be a bad CPU (seems unlikely with the 64-bit stability)? Or am I looking at board and BIOS problems instead of a bad Q9450?
     
    Last edited: Apr 24, 2019
    Armenius likes this.
  8. Armenius

    Armenius I Drive Myself to the [H]ospital

    Messages:
    16,827
    Joined:
    Jan 28, 2014
    Hmm... I never tried the 32-bit version of Vista, but would be interesting if that is the cause of your woes. I'm actually interested in researching this. I know the release version of Vista would fail installation if you had more than 2GB of memory installed, which is the only issue I had with installing the 64-bit version. That and other than the networking issue I mentioned in your other thread I've never had an issue running Vista 64-bit on my 680i system (QX6700, 4GB DDR2-1066, 8800GTX SLI).
     
    Araxie and Mr. Bluntman like this.
  9. CAD4466HK

    CAD4466HK [H]ard|Gawd

    Messages:
    1,219
    Joined:
    Jul 24, 2008
    Vista32 was a shitshow for me after I upgraded from XP to see what Crysis looked like in DX10. Everything felt stutterly and laggy, not just games either. Only took around 3 days to get V64 after that. Major difference. And that was on a A64 X2 4200+ , 2GB 533MHz, 9800GTX and RAID 0 on a pair of 80GB Caviars.

    Not to long afterwards, I had a 9550 on a 780i w/ 4GB of 1066 and 9800GTX SLI and V64. IMO Vista w/SP 2 is really no different than W7 other than the memory management.
    32bit Vista is without a doubt no where as stable as 64bit.

    I wish I could say the same for 64bit XP.
     
    Armenius and Mr. Bluntman like this.
  10. Mr. Bluntman

    Mr. Bluntman [H]ardness Supreme

    Messages:
    6,392
    Joined:
    Jun 25, 2007
    I don't have any blank disks at the moment to burn another version of Windows, but so far it seems that it could be a 32-bit compatibility issue with Yorkfield CPUs installed. It just flat out won't run stable, but 64-bit Vista works without a hitch. My head simply explodes when trying to think of a reason why other than a buggy motherboard or BIOS implementation that is causing the Windows kernel to promptly shit kittens in a fine spackle everywhere. I wish I had a Q/QX6700 to validate the theory that it is a Yorkfield incompatibility issue.

    Too bad discovering this issue is 10 years to late to make any difference.
     
  11. GiGaBiTe

    GiGaBiTe Gawd

    Messages:
    689
    Joined:
    Apr 26, 2013
    The sticks the OP listed the P/N for are 800 MHz (PC2-6400), not 1066 MHz (PC2-8500). and require 1.8v.

    CPUs can partially fail and result in weird crashes that only happen when performing certain workloads. Multi-core CPUs can have one or multiple bad cores and still behave normally until some bit of code upsets them and causes a crash. Heavily used and abused CPUs also can be unstable due to electron migration degrading the die(s). I've seen some server CPUs which had been run at pretty much max load 24/7 for years start to misbehave, but they could be made to work again by bumping up the core voltage or lowering the clock speed.
     
  12. Mr. Bluntman

    Mr. Bluntman [H]ardness Supreme

    Messages:
    6,392
    Joined:
    Jun 25, 2007
    Thanks for the information. I have seriously been wondering if this CPU has been mistreated as a line of pads have a darker patina on them even after cleaning, as if they were exposed to a lot of heat over time. It is both unusual and baffling that it only crashes while running x86 Vista, as 64-bit runs completely stable AFAIK, but there is a reason it's not assembled now and being stability tested as we speak. I would prefer to run x86 as it seems better suited to run on 2GB machines and has better compatibility than x64 with a few programs I wish to run on it. Background processes running on a fully updated SP2 install of Vista definitely needs more than 2GB of RAM, as disk swapping from background processes is so bad the computer is almost completely unresponsive. But that's a whole different problem entirely.

    Upping the voltage all the way up to 1.35v didn't help stability, which leads me to believe that it might be a BIOS or chipset bug on this particular motherboard, but I'm not an EE. The problem only goes away when I install a dual core CPU, but I have no other quads to test. If I had access to one, that would be the next step to rule out the motherboard. For now, it's been disassembled and will be shipped back to the seller for him to further troubleshoot and possibly exchange for a different CPU like a Q6700 or different Q9450.
     
  13. GiGaBiTe

    GiGaBiTe Gawd

    Messages:
    689
    Joined:
    Apr 26, 2013
    Darker pads would indicate they got really hot, which is either indicative of heavy current draw (past the design limit) or a bad connection to the socket caused resistance and again got it far too hot. I would suspect the former is more likely than the latter.

    Yeah.. Windows Vista isn't going to run well unless you have at least 4 GB of RAM. 2 GB will work if you only plan on running one application at a time and not do web browsing, anything else will grind it to a halt. Going 32 bit won't help that much, except on CPUs that run slower in 64 bit mode (eg. the Pentium 4 6xx series.)

    If a dual core works fine, it's most certainly the quad that's bad. If you can look at the sSpec printed on the top of the CPU lid, you can figure out what stepping it is, hopefully you don't have an engineering sample or something.
     
    Mr. Bluntman likes this.
  14. Mr. Bluntman

    Mr. Bluntman [H]ardness Supreme

    Messages:
    6,392
    Joined:
    Jun 25, 2007
    They are only slightly darker, but you can definitely tell they are in the light. I was afraid of as much.

    It does work fine. I've tried two different CPUs (an E8600 and a E7500) without issue. It's only the Q9450 that crashes but only in x86. x64 is 24h stable so far.

    You are right about RAM usage, loading a web browser is sluggish. Loading a memory intensive program is rough. During heavy background processes and disk activity Youtube videos occasionally freeze outright. ~120MB sequential disk read/write isn't much, but it's what I've got.

    Thanks for the reply, very helpful.
     
  15. Mr. Bluntman

    Mr. Bluntman [H]ardness Supreme

    Messages:
    6,392
    Joined:
    Jun 25, 2007
    A 12-day-later update and x64 Vista and 7 are stable. x86 flavors of XP, Vista, and 7 are not. So with the Q9450 only x64 OS are stable in this particular motherboard (ASUS P5N-D). I'm being sent a Q6600 and Q9300 to eliminate both my CPU being bad, problems running quad cores, and problems just running Yorkfield based quads before I get a replacement anything.
     
  16. Furious_Styles

    Furious_Styles [H]ard|Gawd

    Messages:
    1,191
    Joined:
    Jan 16, 2013
    An odd issue. Unrelated bluntman but what cooler are you using? I've recently bought a few 775 mobos and I'm having a difficult time locking down the cpu HSF. I have 2 different stock intel ones, a low profile one and one with a bigger metal portion. They both don't seem to fit nearly as good as the 115X ones.
     
  17. Mr. Bluntman

    Mr. Bluntman [H]ardness Supreme

    Messages:
    6,392
    Joined:
    Jun 25, 2007
    I'm using a copper core thicker style Intel stock radial heatsink with Arctic Silver Ceramique 2, Temps rarely go above 55c and never above 68-70c under a Prime95 max heat/power consumption load.

    Protip: The trick with the pushpin heatsinks is to apply firm pressure at opposite ends diagonally until you hear a soft click. Rinse, repeat. Go over every corner again to nake sure things have truly snapped in place securely and there is no give in the push pin struts.

    Cooler installs don't get easier, at least for me. There are many I am aware that detest if not outright loathe the push--pin, but I guess I'm lucky in not having any problems (other than having to replace a plastic tab once which is super easy if you have a donor cooler and a standard screwdriver).
     
  18. Furious_Styles

    Furious_Styles [H]ard|Gawd

    Messages:
    1,191
    Joined:
    Jan 16, 2013
    Yeah I just get way more resistance and even popping out with 775. All the 115X ones are extremely easy and never give me a problem. The push-pin is hard to beat for convenience (test bench for me) but long-term I use the bottom bracket like on the noctua nh-d14/15.
     
  19. mda

    mda [H]ard|Gawd

    Messages:
    1,446
    Joined:
    Mar 23, 2011
    Bump up the VTT a little. I do remember (I could be wrong) the Nvidia chipset boards not being the best for Quads.
     
  20. crazycrave

    crazycrave Limp Gawd

    Messages:
    346
    Joined:
    Mar 31, 2016
    Never stated if you wiped the cmos after the cpu change out .. my X58 I sometime have wipe it and start over just for Windows to get stable as it needs to rebalance the change out .
     
  21. KATEKATEKATE

    KATEKATEKATE [H]Lite

    Messages:
    85
    Joined:
    Jan 20, 2019
    i had an nForce 780i board once (MSI P7N Diamond) and it was nothing but trouble. It kind of worked with an E6400 but wasn't stable with a Q6600 and eventually just stopped booting for unknown reasons.

    Would it be an option to just keep the dual-core in the nForce board and find a cheap P35/45 or x38/48 board for the Q9450? Intel chipsets should be more stable/robust and offer better OC'ing
     
  22. Dan_D

    Dan_D [H]ardOCP Motherboard Editor

    Messages:
    53,244
    Joined:
    Feb 9, 2002
    If you aren't running SLI video cards, get yourself an Intel chipset based board. That's the short version of a very long post on the subject.

    Statistically, it's unlikely you are looking at a bad CPU. I've serviced thousand and thousands of systems. That's not even remotely an exaggeration. I've seen maybe a dozen or so CPU's die over two decades and very few failed without an external cause. Less than half of them in case you were wondering. CPU's rarely die. Are 750i SLI boards just that shitty? Absofuckinglutely, they are indeed a soggy shit sandwich. They are only slightly better than 780i SLI boards, but not by much. We will go over the reasons they are shitty shortly. I reviewed probably a half dozen or so NVIDIA SLI chipset based boards, probably more. I have personally owned 13 of them. All of them died within 18 months. Most died in two to three months or were DOA. One lasted the 18 months, but it wouldn't clock RAM or FSB for shit. Even by 680i SLI standards.

    You are correct on most of your points. I will add that cooling won't help unless the thermal paste needs to be applied or your fixing an active cooling solution that's shit the bed on you. I didn't catch anything about overclocking, but if he's using a Yorkfield, he should give up on that idea entirely.

    I'll start off by saying, you are indeed correct on the point about overclocking. FSB overclocking on the NVIDIA chipset boards was a shit show. Extreme Edition CPU's were the only game in town for overclocking on NVIDIA boards. Next, I'll say 32bit versions of Windows post-Windows XP Professional x64 Edition are the bane of the computing industry and arguably had no right to exist. Sure, you can make a case for Vista being transitional, but by Windows 7, 32bit OSes should have largely died off. The continued existence of shitty 32bit OSes you can blame on the OEMs like HP and Dell who were too fucking lazy to transition their garbage-ware (pre-installed bloatware) to 64bit. They also used Intel CPU's, and during the Pentium 4 days with EM64T, these chips were slower in long mode (64bit) than they were in legacy mode. Naturally, the move to 64bit computing wasn't embraced. There was also a perception that running through the WOW emulator in Windows x64 would bring performance penalties and that there would be various compatibility issues with older software. They did exist, but were rarer than you'd imagine.

    Allot of that pre-installed garbageware fit the category of crap that didn't work and caused problems on 64bit OSes. The only reason the lazy OEM's moved to 64bit Windows is to avoid the warranty claims about failing or missing RAM as 32bit OSes showed anywhere between 3.9GB and 3.5GB of RAM instead of the 4GB customers believed they had. Those lazy bastards would probably have stuck with 32bit Windows longer than they did were it not for that fact. You should stop trying to use a 32bit OS. There are many reasons why. First and foremost, there are memory limitations with 32bit OSes which aren't generally understood fully by the public. You can technically run 4GB of RAM, but for direct memory access, RAM from graphics cards, option ROMs for RAID controllers, etc. have to be mapped somewhere. This is why you would see 3.5GB or so of usable RAM. Not only that, but the OS took 2GB for itself, and left 2GB allocated to user programs before talking about what was taken out for memory mapping. This is where I think the crux of your issues with 32bit Windows probably lies.

    BIOS implementations are very likely the reason for the problems. OROM compatibility and other things were often messed with in the NVIDIA reference BIOSes. Later BIOSes were sometimes tuned for overclocking while sacrificing things like PCIe RAID controller compatibility. I believe the feature they killed was PCIe spread spectrum of something like that. Its been a number of years so I'm fuzzy on some details.

    Yes, this is absolutely true. Having said that, I have rarely seen such a thing. I've only personally seen three examples of CPU degradation in more than 20 years of dealing with PC hardware.

    I've seen discoloration on the pads of processors before and I've never seen any signs that this was actually a problem. That's not to say that it doesn't indicate a problem in some cases, but your system running fine with 64bit Windows tells me that your CPU is good. If it wasn't, it wouldn't work like that. The reason your issues disappear with a dual core CPU has to do with the VRM design of your motherboard. NVIDIA actually designed and spec'ed all the NVIDIA reference designs. Even when a company like ASUS did their own thing, they were still forced to deal with NVIDIA's chipset silicon and reference BIOS. This isn't unlike when Intel fucked up the Z68 chipsets (SATA port issue) or AMD's early X370 AGESA code issues. NVIDIA had what I believe to have been QC issues with the silicon as well. This was either a fault in the design of the silicon or a problem with whomever produced the actual chipsets for NVIDIA. I am not sure which.

    As I said, your using a quad-core on a motherboard with a VRM that's really insufficient for use with quad-core CPUs. That's the biggest reason why overclocking was so much worse on quad cores and why quad-cores were so problematic with those boards. Users of quad cores often ended up with dead boards. Many dual-core guys came out of the 680i SLI era believing the EVGA 680i SLI reference board to be one of the best motherboards they've ever owned. If you stuck with dual cores and didn't go for high RAM speeds, this may be true. Pop in a quad core, RAM beyond 800MHz, use four DIMMs or any number of things and your experiences will likely end up being similar to mine.

    Do not do that if you do not have to. There are several problems with those boards relating to voltage. The VRM implementation is a joke. It barely met Intel's specifications to start with. The other problem is NVIDIA wrote the reference BIOS. It's an absolute cluster fuck. Any voltages set on AUTO are going to be way too aggressive for the chipset and CPU. Upping voltages past stock is going to generate more heat which is the last thing any component on that board needs. Unfortunately, you often need a little extra voltage for stability. Over time, this degrades the silicon of the chipset. I've actually seen these chipsets separate layer, by layer and almost peel apart. You can actually tell if your south bridge sits at anything close to 50c rather than 35c, its done. Throw it away. Several of the boards I had ran super hot no matter what you did. While I have no evidence beyond the sample size of around 20 (my 13, combined with extra review boards and friend's systems) leads me to conclude that boards with chipsets at 35c would work for at least a short time. Any board showing 50c south bridge temps was a time bomb. It wasn't likely to work for long if at all. Boards with those higher temperatures wouldn't show much of a reduction on water cooling. My EVGA Black Pearl had water cooled everything. The loop included the south bridge and eve the memory in my system.

    Once you start seeing USB problems, SATA issues and the like, your board is ready for the trash can.

    This was common and the reasons weren't unknown. NVIDIA cut corners while designing the VRM. It was fine for dual-core CPUs but it barely made the grade for quad-core CPUs. Stability couldn't always be maintained and it would often cause premature component failure. 600 and 700 series NVIDIA chipset based boards often took out memory modules when they'd die. This was due to bad VRM designs for the DIMM slots as well. QC was so bad that NVIDIA's reference BIOS over-volted the crap out of everything just to make it work. NVIDIA's efforts to improve overclocking on the quad-cores was like polishing a turd. They would literally break device compatibility to make them overclock more. I know this from experience. They removed the PCIe spread spectrum setting as disabling it would sometimes give you more OC headroom. Unfortunately, storage controllers like my LSI 8308 required that feature to be enabled. To win benchmarks back in the day, manufacturers would tune memory settings to be more aggressive than your modules wanted to operate at. So if you had low latency RAM, it would actually set the latency one point lower to speed up the system for benchmarking. Unfortunately, this created issues with reliability.

    It was a common practice back then. However, Intel boards were fine doing this. NVIDIA's, not so much. After 13 of those piece of shit boards, I would have paid my own money for one of these just to be able to game without worrying about my system breaking every few days to weeks: Intel Skulltrail I actually kept that board and processor combo. I ran that for years. I still have it and it still works.
     
    Araxie, Mr. Bluntman and KATEKATEKATE like this.
  23. Mr. Bluntman

    Mr. Bluntman [H]ardness Supreme

    Messages:
    6,392
    Joined:
    Jun 25, 2007
    What a long and detailed post filled with useful info. Unfortunately I am running 9800 GTs in SLI in this so an nVIDIA chipset board is a requirement. I would have got an X38 or P35 board if GeForce 9800 GX2 cards didn't go for almost as much as a GTX 970 used. I should note that the motherboard is an ASUS P5N-D and doesn't appear to be reference since it appears to be different than other 750i SLi based motherboards in the VRM department. I have also obtained a Q6600 and Q9300 - and other than high temps on the Q9300 (which lends credence to your statement about overvolting) both CPUs are fully stable, passing 24h Prime 95 runs. The E7500 and E8600 CPUs in my possession have never exhibited stability issues (again, confirming what you said regarding that).

    This motherboard is entirely passively cooled, so I have no idea what southbridge temps are sitting at, but I will keep an eye on the aforementioned USB/SATA issues you mentioned.

    And just like AMD's Quad FX, Skulltrail was something I drooled over but had no hope to afford back in the day. I bet it still holds its own today in day-to-day non-gaming usage.

    Honestly I am not sure if I have or not. The BSOD issue usually is at its worst after a fresh install of Windows. It will blue screen seconds after first boot begins. None of this occurs on four other CPUs I have, two dual and two quad. Just the Q9450.
     
    Last edited: May 25, 2019 at 5:52 PM
  24. Dan_D

    Dan_D [H]ardOCP Motherboard Editor

    Messages:
    53,244
    Joined:
    Feb 9, 2002
    I think I actually have a 9800GX2 sitting in a box. I should test it and see if it works. If it does, I will sell it if the thing is worth money. For me, its one of the most disappointing cards of all time. Its sad you are using SLI as those boards are absolute shit and they aren't getting any younger. The way they were designed doesn't lend themselves towards longevity. They are all time bombs to an extent.
     
  25. Mr. Bluntman

    Mr. Bluntman [H]ardness Supreme

    Messages:
    6,392
    Joined:
    Jun 25, 2007
    While I am not chained to SLI, per se, I did build the system with using two 9800 GTs in mind. I am getting the distinct impression that while my board may be better than the norm, it's still a far cry from the stability I am accustomed to while running Intel processors. eBay scalper prices are nuts. Too bad I don't have a larger monthly budget, I'd take your 9800 off your hands for you and go get a decent P35 motherboard. I appreciate your insights as I was mostly in the dark about how bad nVIDIA chipsets really are for Core 2.