@mmu_man I've had that before, but it was mainly because I was needing to run it on 386s and the like. This is a modern machine so there should be no problems.
yanked the odd hardware like GPU and hard drives, as well as the common stuff like DVI video capture device, both floppy drives, optical drive, and serial port
I had to go find another keyboard though. my spare keyboard on hand is an AT keyboard, and while of course I have a (hand made!) AT-to-PS/2 adapter, this computer for some reason only has USB ports.
I can't understand these companies just expecting us to have obscure hardware on hand to diagnose their failures.
WE'VE GOT ERRORS! hitting a bunch of RAM problems around 31-39gb.
nastily that's divided between two banks. So... it's probably not just one pair of DIMMS going bad.
(and of course it's also possible it's some other problem that's causing RAM issues)
with half ram, I did two successful passes of memtest86+, no errors. time to check THE OTHER HALF (to confirm if it was just a needs-reseating/too much ram for the mobo problem)
and it passes too.
So, either my motherboard has stopped being able to support 64gb, or it was just a reseating problem. Let's stick it back in and see.
I really don't like that it corrupted outside of the text window.
That shouldn't be happening if this is just a RAM problem. That's like... a GPU corruption problem.
Which is very bad because I'm currently using the CPU's on-die GPU
NOPE! it ran for 3 hours, threw a ton of memory errors, then crashed in the same matrixy way.
So, back to just one pair of sticks, and I'm gonna let memtest run for a while and see if that seems stable.
OK so ran the half-ram test for nearly 8 hours with zero issues.
Now I've reattached the GPU, usb ports, serial port, floppy drives, and rust drives and optical drives (but not the m.2 ssds yet because they're boot drives) and I'm running it again, to confirm it's not a PSU problem
Memtest didn't complete, but for a very silly reason: I have my computer up on the desk and plugged into a different outlet for this testing, which means there's a power cable crossing the doorway.
When I left the room, moving the cable so I could go under it slid it out of the power supply and it shut off
memtest completed and now I'm going full on stress testing: playing fullscreen video, 3D games, processing a bunch of shit in the background, some VMs. you know, the usual stuff I do on an average day
So, it's probably one of motherboard, cpu, or PSU. At a stretch, it could be the GPU.
I have another spare GPU I could swap in. I have a near-identical CPU that I could swap in (it's in use, but I can temporarily borrow it).
PSU and mobo are trickier.
So, I'll have to try the easy ones first. Swap the GPU and see if windows still hard crashes like that, then the cpu, then start working on the others.
@baishen I've got one somewhere but this seems too subtle a problem for it to detect. It's not failing to boot, it's just falling over after running for 20-30 minutes
If you'd like to help me get back online (and gay cats, of course), donations would help. I'm kinda broke and not having a working computer is not going to help.
@foone default uncore/ring ratio I assume. What is vccsa and vccio read (voltages, system agent and IO)? also how much memory, dual rank sticks, or single rank? 2 dimm or 4 dimm, what mobo?
@foone the only reason I ask is because 8th gen Intel memory controller likes to fall over when dealing with 4 sticks of dual rank memory aside from micron e die, and z1/4xx boards (6th-9th) boards were notorious for shoving a ton of sa and io voltage when all dimms were populated or xmp was enabled, and by a ton. I mean voltage that will degrade or kill the imc fairly quickly, like 6 months to a year.
okay today's first test: I yanked out my GPU and I'm running on just the internal GPU. I'm gonna load up some videos, VMs, 3D games, and a bunch of browser tabs. See if this falls over too
somehow I got my youtube video playing over my actual speakers but one of the games playing out the HDMI and the little speakers on the monitor. that's weird.
okay I've made it an hour running sans-GPU. That doesn't mean it's the GPU though. This machine is using way less power without the GPU... so it could still be a PSU related problem.
so after running fine for about 3 hours with no GPU, I've gone out and bought a new... power supply.
yeah I don't think it's the GPU. And a flakey PSU could easily fail with the GPU and not without, since the power usage is way lower without a GPU in there
okay new PSU is in. That took way longer than it should.
Apparently between the RM650x and the RM850x, Corsair redesigned their modular cables, so I couldn't just swap the PSU and reuse the cables. So now I have a cable management nightmare, but it's running. Let's put the stress-testing pants on
the worst part is that I forgot to double-check that the new PSU would come with the right cables to let me hook up my floppy drives. Thankfully, it did.
So it turns out I can't get it to do 16x in the other slot. My motherboard has 4 16x slots, but it does them in sequential order: if the first one is full, it gets 16x. if the first and second are full, they get 16x. if the second one is full and the first isn't, the second gets 8x.
I'm gonna run my stress test with some performance logging on to see if it's overheating.
I did realize my card has a physical switch for "high fan" vs "quiet fan" and switched it to "high fan".
I can't imagine that'd be why it was crashing but maybe.
over an hour with the GPU back in the Crashy Slot and no crashes.
huh. Maybe it was temperature based?
My GPU isn't getting THAT hot, my fans aren't even maxing out.
GPU temp hit a max of 65C with a hot spot of 76C.
Those aren't out of range for a GPU under load, and they're not trending upward at all, it's stable.
@foone your story sounds like my ampere gpu which also just crashes randomly (sometimes after hours, sometimes minutes), but that's a quadro a6000 I got from nvidia through work, other GPUs work just fine on the identical setup/slot.
So that's, like, 1200$ if I wanted to replace it all. That's definitely not going to happen. So for now I think I'm just gonna have to wait to see if more shit breaks, but ordering more RAM is next on the list, since I'd like to be back up to full RAM and it'd be useful to know if adding RAM back in causes it to crash again
New RAM installed. I had a fun moment where the system seemed to be completely dead... Turns out one of the modular cables slipped out while I was installing the RAM.
But no crashes so far!
I also updated the BIOS, to a version that has "dram stability improvements" listed on on the changelog.
@foone Yeah, like I was saying, a lot of the boards up until 10th gen were very not memory focused, so they didnt have great shielding or trace layout, you often even got really bad t topology layouts or extremely long daisy chain that would fall over at anything above 2133 with 4 dimms.
Have you looked at the bios on the board, there may be a few updates that include "improved memory stability" and usually those are extremely helpful for running dimms made well after the board.
@foone
The only thing more insane than doing the same thing again and again expecting a different outcome is doing the same thing again and again and GETTING a different outcome... 🤷
@OtterMatic it's possible, but I've pulled and re-inserted it like 5-6 times through this whole saga, and it was still crashing up until this last time
@foone@foone what are your core temps and clocks like in hwinfo? Heatsinks all snug? All internals unobstructed? Checked the board for any spicy caps? When you tested RAM, was it a matched pair? Anything overclocked or undervolted? Tried another PCIe slot for the GPU?
Worst case the motherboard is just dying, I had an X99 Asus board which was notorious for eating CPUs (which mine did, while simultaneously FUBARing itself). That's probably a niche edge case though 😅
@cw I don't have the temps on the crashy setup, but I've not seen any temps outside of reasonable ranges. No cooling problems. RAM was matched pair, nothing should be overclocked or undervolted. I've not tried a new PCIe slot yet, that's definitely something I should try out. I think it'd cause my GPU to run at lower speed but at least I can confirm it works/doesn't work there
@foone for a moment I thought about how I need to make sure I still have adapters in my collection to power 3.5" drives with the 0.1" center 4 pin JST header off of a full size power connector. aaahhhh!
@flameeyes@foone Even if you keep the same model (this happened to me years ago with 600W Enermax – replacement had 5 pin connectors on PSU side instead of 6).
@foone I had the same thing happen years ago when a 600W Enermax died – got it replaced under warranty, but the new one had different connectors on the PSU side (despite being the same model), so I had to redo all my cables…
@foone this is sounding like a memory clocking problem. XMPP speeds are only (usually) for a single stick per channel, and down-clocking to the stock speed is (usually) required for multiple sticks per channel
@foone I recently dealt with an Intel i5 with dying GPU. As long as the Intel driver was in use, the screen was full of artifacts, and moving windows around created random polygons all over the screen.
@foone
I Had errors like this on an AMD Radeon card after a year, and it turned out to be shitty software. It cost me a heap (new RAM, power) to find out that a stupid driver update was the cause and by editing some files I could have the kernel overcome that.
If the RAM tests say it's fine, it might be crap software.
Add comment