TR Forums

TLDR, have you ever had one of those days?

I had one yesterday, nothing wanted to work.

A cheap CPU upgrade from ebay arrived in the morning and I though I'd just pop it in... 5 minutes work right?

No.

First of all when I unscrewed the mounting for the heatsink (a coolermaster evo something or other) the mount unscrewed from the motherboard and fell off the back rather than the heatsink unscrewing from the mount leading to me having to pull the whole board and refix the mount.

With the new CPU installed I turned it on for the first time and got random MCEs from the kernel and it wouldn't boot. At this point I though I'd gone too far in trusting ebay but I figured I'd try a new bios before shouting at the seller. That installed OK going straight from F4 to F8 but it didn't help getting the machine booting so back to the original CPU.

Which now also refused to boot, **** **** **** **** ****. I wasn't getting the MCEs but it seemed to lock up part way through the boot.

To my horror I discovered I wasn't allowed to go back to the F4 bios I knew worked and could only go back to F7 which still wouldn't let the old CPU boot the system. Bollox.

All the other unnecessary hardware now came out and it still refused to start. At this point I noticed my CPU fan wasn't turning, odd. Cable connected OK and no obviously loose or damaged wires, tried it on another fan header and still nothing, tried another fan on the same header and it worked. Somehow I'd killed the fan taking it out the case. Still the bios wasn't exactly reporting silly temperatures so this didn't look like the cause of the MCEs and while a new fan brought them down a from around 55C to 35C problems remained.

As a final throw of the dice I decided to see what happened with a clean install on a spare drive. The bios worked fine with the new CPU installed and I'd have thought I'd see some weirdness if the CPU really was a dud (crashes in the bios, spontaneous reboots, random failed POSTs etc) . The new installed worked flawlessly with the new CPU, no MCEs, so it looked like the ebay seller was off the hook and the CPU was fine. Going back to the original install the MCEs had disappeared there too but it still wouldn't boot completely behaving like the original CPU and just locking up part way through.

I wish I knew more about how UEFI works but I'm guessing there was some kind microcode patching being done by the Linux UEFI boot entry and doing a clean install with the new CPU installed fixed it.

Anyway it still wasn't booting but it appeared to be graphics driver related rather than the CPU which was confirmed by going back to the new test installation and adding the nvidia driver. Now that install behaved like the old one. Poking around the logs I noticed some references to Intel graphics and IIRC Prime. It looked like the new bios allowed the nvidia driver to works as hybrid graphics and part way through the boot process I was being switched the Intel which looked like a lockup. Disabling the Intel graphics in the bios sorted that and I finally had a booting machine with my original install.

I tried a couple of games and everything seemed lovely, until I rebooted. Lots of reboots later I discovered that it worked properly maybe 1 in every 10 reboots. The rest of the time things were a slideshow, games were at maybe 5-10 FPS and even the desktop seemed pretty sluggish. Clock speeds were the first port of call but they looked OK which left the PCIe bus. The only control my mobo gave me for that was to choose gen1, 2, 3 or auto. So I gave gen 2 a go rather than "auto".

Bingo, everything now works.

Obviously I'm relived but why can't life be simple?

Oh and I had to do some actual paid work after this and neither of those jobs really worked properly either.

EDIT for anyone interest I was upgrading an i5 4460 to an i7 4790 on a Gigabyte Z97P-D3 with a 1080 graphics card.

At the office we call days like this "going down the rabbit hole". By the time the dust settles you've learned a bunch of stuff you didn't know you needed to know!

And some things yet to be learned, perhaps: What's with the PCIe mode forcing all of the sudden? Did the CPU change the equation somehow?

meerkt wrote:

And some things yet to be learned, perhaps: What's with the PCIe mode forcing all of the sudden? Did the CPU change the equation somehow?

Yeah it's bizarre. The problem is that so many things changed and some of them, like the bios update, can't be undone. My guess is that either something to do with the new CPU, like its extra power draw just pushing something else out of spec or that firmware update caused it. I would put the old CPU back in to see if it was still a problem but I'm a bit reluctant to fiddle with it now

Since I have no idea what you bought (and the OP is light on such detail), I'm just guessing you replaced a Sandy Bridge CPU with an Ivy Bridge CPU and the board was somehow incapable of dealing with PCIe 3.0. Further guessing you're using a Z68 board that had early PCIe 3.0 support and it's probably buggy as hell.

Yeah dude! Mobos and BIOS/firmware updates suck.

I had similar **** happen to me when I got a bunch of DDR3-2400 for cheap off a friend who upgraded to Ryzen. So I popped it into my system (Gigabyte X79-UD7 on the F10 BIOS) and found it didn't play well. I thought it might be because my CPU is a locked Xeon, so I popped my old i7-3820 back in for some testing.

Still didn't play very well, so I update the BIOS to the latest F12f. Now everything is running great! Ran at 2133 right away, and I was able to get the full 2400 after upping the voltage on the memory controller a bit.

Great, right? So time to put the Xeon back in and as soon as I do so boot goes dark. Wut? Timings I guess right? So I cleared the CMOS so the RAM would start up in SPD instead of XMP, and this time it does boot, but it gets stuck at some weird dialogue before the post screen, unable to even go into the F8/BIOS setup menu.

WTF?!, I was thinking.

Last known good config was the i7-3820, so I put that back in and **** is working.

At this point I figured the F12f BIOS added better compatibility for faster RAM, but removed compatibility with v1 Sandy Bridge Xeons! What a head scratcher!

So I try to downgrade back to the F10 BIOS, but now Gigabyte's utilities wouldn't let me! A few hours and lots of wading though old forums posts later I learn that there's an Intel utility that'll flash any BIOS you want.

More than a whole day later I was back right where I started, and it was then I found the sense to take a look at ark.intel and found out the Xeon would've been locked at 1600 anyways, so the best I could have done was tweak the RAM timings.

I guess it was about time to re-paste CPU anyways lol

just brew it! wrote:

At the office we call days like this "going down the rabbit hole". By the time the dust settles you've learned a bunch of stuff you didn't know you needed to know!

+500 EXP.

Roll 2d6 for SAN loss.

bthylafh wrote:

just brew it! wrote:

At the office we call days like this "going down the rabbit hole". By the time the dust settles you've learned a bunch of stuff you didn't know you needed to know!

+500 EXP.

Roll 2d6 for SAN loss.

LOL... the product I help maintain is a large data storage system, so I'm quite familiar with scenarios like that!

How about some old buggy HBA firmware which mangles the model names of the HDDs it reports to the system? And a system which identifies individual drives using a hash of each drive's model name and serial number? (As expected, hilarity ensues... this was one of our recent dumpster fires.)

I believe I may have griped before about how horrid the firmware and drivers often are, even for "enterprise grade" gear.

just brew it! wrote:

At the office we call days like this "going down the rabbit hole". By the time the dust settles you've learned a bunch of stuff you didn't know you needed to know!

An iconic version of this immortalized by a young Walter White.

derFunkenstein wrote:

Since I have no idea what you bought (and the OP is light on such detail), I'm just guessing you replaced a Sandy Bridge CPU with an Ivy Bridge CPU and the board was somehow incapable of dealing with PCIe 3.0. Further guessing you're using a Z68 board that had early PCIe 3.0 support and it's probably buggy as hell.

I updated the OP with some basic specs, maybe there is something wrong with the CPU and it doesn't handle PCIe 3 properly and that's why it was being sold. It's not even like I was going from Haswell to Haswell refresh.

Anyway I left the computer on this morning and its been stable all day and that's good enough for me.

I went through a similar situation recently trying to secure erase my PNY SSD which doesn't have any manufacturer software for doing so. I went down a deep rabbit hole of programs and a Linux boot drive trying to get the thing secure erased. It was one roadblock after another, just like that. It's a true test of one's character and resolve, I'll say that. Though I eventually caved and took the easy way out by ultimately spending some $$ to solve the problem. Looks like you didn't have that option so I salute you.

Oh, that's really wild. Haswell to Haswell.

My guess is that the i7-4790's clock was pushed too much and for too long and that impacted its I/O handling capabilities. Noticed the deteriorating effect of overclocking on I/O years ago on my Sempron that ran at 3GHz for over a year and then developed random I/O related errors and crashes and general wonkiness.

Currently taking a break from a similar gremlin fight with an i5-750 and a GA-P55-USB3 motherboard. It fired up just fine at first build and even finished a couple different F@h work units, but the memory was running single channel. After that effort it all went sideways. Going to think long and hard before buying another Gigabyte motherboard.