Customized computer?

[Buffalobiian]

My brother's AMD 5600X CPU from 2021 had been throwing intermittent crashes since we had bought it. The more memory you run on it, or the quicker you clock it, the more frequently it crashes. He decided to RMA it when there weren't any games games coming out that he was interested in playing, and the shop ended up refunding him after AMD found a fault. I presume it's the memory controller on the chip.

So we bought a 5800X3D instead. It still threw the intermittent error when the RAM/infinity fabric was overclocked, with windows event viewer logging it as a CPU/memory channel? thing. So after that, I left the infinity fabric and memory on default.

A few days later, the PC resets and fails to POST. Motherboard diagnostic LED says "DRAM", but apparently that just means the PC encounters a fatal error as it tries to check the DRAM. Swapping out different kits of RAM didn't resolve the problem, and I can't reflash the BIOS without key components being operable.

So now I'm going through the RMA process again. Perhaps I got unlucky with AMD's memory controller lottery. Perhaps his board is frying CPUs (though the old chip didn't exactly get worse with time). We'll see.

I do have my own computer's mobo and CPU on the same platform that we could test, but mine is rock solid and I don't really feel like changing that.

[Kraco]

Why are you trying to overclock them in the first place? CPUs these days aren't mean to be overclocked, except for setting the memory to the specs the memory kit is meant for. Otherwise only overclock them if you want to compete in overclocking for benchmarking. Especially the AMD X3D CPUs are really bad choices for overclocking because the extra cache layer hampers heat transfer, making them run hot even on stock clocks. If you continue on your overclocking journey, you will sooner or later get a negative RMA request, with the explanation being overclocking. You don't even gain that much from it.

[Buffalobiian]

You overclock the infinity fabric commonly to match the XMP speeds on your RAM so their data cycles are in sync. Many chips will reach 1800 fine, though 5000 series chips only actually support up to 1600 I think.

The motherboard automatically does this when you set everything to auto but set your RAM to DOCP (equivalent to XMP)

XMP is technically overclocking since it's not a JEDEC spec.

[Kraco]

Sorry, I must have read your post poorly, since I was basically criticising you for what I suggested myself. I got an impression you were really trying to push the frequency up, beyond merely the manufacturer's stated specs. While using XMP indeed is overclocking (unless it's some terribly cheap memory meant to be used on the CPU default settings), I'm more used to hearing memory overclocking in the context of pushing it further than the XMP.

[Buffalobiian]

My brother's AMD 5600X CPU from 2021 had been throwing intermittent crashes since we had bought it. The more memory you run on it, or the quicker you clock it, the more frequently it crashes. He decided to RMA it when there weren't any games games coming out that he was interested in playing, and the shop ended up refunding him after AMD found a fault. I presume it's the memory controller on the chip.

So we bought a 5800X3D instead. It still threw the intermittent error when the RAM/infinity fabric was overclocked, with windows event viewer logging it as a CPU/memory channel? thing. So after that, I left the infinity fabric and memory on default.

A few days later, the PC resets and fails to POST. Motherboard diagnostic LED says "DRAM", but apparently that just means the PC encounters a fatal error as it tries to check the DRAM. Swapping out different kits of RAM didn't resolve the problem, and I can't reflash the BIOS without key components being operable.

So now I'm going through the RMA process again. Perhaps I got unlucky with AMD's memory controller lottery. Perhaps his board is frying CPUs (though the old chip didn't exactly get worse with time). We'll see.

I do have my own computer's mobo and CPU on the same platform that we could test, but mine is rock solid and I don't really feel like changing that.

Well the good news is the store accepted the CPU return without much scrutiny and offered a refund immediately.

The bad news is that the PC still won't boot after I bought yet another CPU for it. So now it's a hunt for other things.

I should unplug the sound card and other peripherals, then see if I can find a power supply somewhere to test. If all that is done and fruitless then I think I have a pretty good case for swapping out the mobo.

[shinta|hikari]

Just trash the whole thing and start anew.

[Buffalobiian]

I came home after a night shift to find that one of my hard drives just went offline. No prior warning, but now it clicks and won't be recognised by the OS. It's a 12TB Western Digital Gold drive. Mainly just had anime and movies on it, and was the backup drive for holding my main drive SSD's images etc. So not a big deal.

While dealing with this, my other 12TB WD Gold drive dies. I'm presuming this is from the same batch since I bought both from the same store at the same time. Now this one hurt a bit more because it had recent anime as well as music, and I like my lossless music collection.

So that's around 24TB gone now. I'll see how much data recovery costs, mainly for the music, but chances are it'll cost between several hundred to several thousand dollars, and I might as well hunt down all the music again - along with all the annoyances of dealing with Chinese file hosting mirrors.

Even if I had them in RAID1, in this scenario things wouldn't have been saved, unless I had the foresight of backing up the music onto a much smaller 4TB HDD - which I didn't.

Anyway, I'm using some spare HDDs now for storage while my SSDs still chug along and my OS remains untouched. Gonna shop for some HDDs in the next few days. Apparently they go up to the low 20s in TBs now. Maybe I'll do 22TB RAID1 or something next.

[Kraco]

I'd recommend having backups of anything you aren't ready or capable to easily redownload on external HDs that aren't running 24/7. I like to presume not being turned on all the time prolongs their life, seeing how I have a couple of usb2 externals that still work, as much as the usb2 interface can be considered to work in 2024, being as slow as romance in a shounen manga.

Still, lousy luck to lose both like that. Smells like either planned obsolescence or your PC (PSU) sending a power spike the HDDs' way to fry them.

[Buffalobiian]

My intel SSD from the early 2010s has finally reached the end of its endurance lifespan. For the same amount of money that I paid for this 240GB drive then, I could buy a 4TB NVMe drive today.

Installation was fine, though it took more coordination trying to place then screw this drive in place (without removing the CPU tower cooler or graphics card) than most procedures I've needed to do.

Data migration using Samsung's SSD management app and then converting it from MBR to GPT was easy enough. The drive runs anywhere from 56 to 66C right now, which is within spec (<70C operating temp).

Hopefully this drive lasts for a long time as well.

I bought a drive with a heatsink, and I'm intrigued that most heatsinks on SSDs really are heatsinks that increase surface area slightly, but mainly increase thermal mass to soak up spikes in temperature during bursts of operation, despite PCIe4 drives being known to throttle under operation at times. It's only the PCIe5 drives that really offer substantial fins for cooling dissipation (as OEM anyway).

[Kraco]

I used to have a 1TB system drive (PCIe 4) with a heatsink and a 500GB game drive (PCIe 3) without. I later replaced the 500GB drive and got one with a heatsink. My system drive actually came without a heatsink, so I got a third party one for it at the same time. Both are Samsung 980 Pro drives now. So, theoretically I could compare if the aftermarket passive cooler is better than the original Samsung one, but I won't bother. There would be the airflow difference as well, seeing how one is above the video card, one well below.

[Buffalobiian]

My intel SSD from the early 2010s has finally reached the end of its endurance lifespan. For the same amount of money that I paid for this 240GB drive then, I could buy a 4TB NVMe drive today.

Installation was fine, though it took more coordination trying to place then screw this drive in place (without removing the CPU tower cooler or graphics card) than most procedures I've needed to do.

Data migration using Samsung's SSD management app and then converting it from MBR to GPT was easy enough. The drive runs anywhere from 56 to 66C right now, which is within spec (<70C operating temp).

Hopefully this drive lasts for a long time as well.

I bought a drive with a heatsink, and I'm intrigued that most heatsinks on SSDs really are heatsinks that increase surface area slightly, but mainly increase thermal mass to soak up spikes in temperature during bursts of operation, despite PCIe4 drives being known to throttle under operation at times. It's only the PCIe5 drives that really offer substantial fins for cooling dissipation (as OEM anyway).

Update on my NVMe drive purchase.

Performance-wise, I'm not noticing anything drastically different compared to 6Gbps SATA SSDs. I don't use it for anything outside of general browsing, office-stuff and gaming. Even games that stream textures on the fly should be fine with SATA SSDs I suspect, but I can't say I've actually put anything to the test.

Temperature-wise, it runs hot. Sometimes if I'm gaming and it's doing a backup, I think the controller can run up to 91C. The rest of the NAND can run between 65-high 70s.

I've had 2 crashes that I attribute entirely to the SSD, and both of those were when I open a browser and it tries to load 2000 tabs. After the crash, the PC actually goes to BIOS because it couldn't detect the NVMe drive on boot up. Turning it off and then back on again resumes normal function.

It works more often than not (it works several times a day whenever I close and re-open my browser, and has only crashed twice in this manner over the past year). There has been maybe one or two freezes other than that during the year, but I haven't attributed those to anything in particular.

It has a 5 year warranty so I'm not worried much about cost, but if the controller goes bad and slowly corrupts data instead of (or before) completely dying, then it could ruin my more recent backups.

I think I'm going to hunt for specials, and just get another SATA SSD at the same capacity, and just do a mirror backup onto it every few months in case this dies, then just use the SATA drive. I don't think I'll miss any performance loss and would be much happier knowing that the drive temp is well within range (usually 30C or so in my PC) and thus reliability is likely not affected.

If I were to get a NVMe drive again, I'd probably use an actual aftermarket heatsink that looks like a mini-CPU cooler. The heatsinks that come from SSD manufacturers themselves look nearly useless for any meaningful heat dissipation.




In other news, dad's computer started dying so I gave him my old CPU/RAM/Mobo set that I upgraded from. It'd be >10 years old now, but still worked. The combo didn't have integrated graphics though, so I whipped out an old Geforce 8500GT for him. I was both amazed that it worked, and that it had W10 drivers (last updated 2016).

It all worked out fine, but a few weeks later I helped him with something and found his web browser was lagging like hell. Turns out, the default setting of "Use Hardware Acceleration to improve performance" was maxxing out his meager 256MB of VRAM. Unticking that option did the trick. His 2013 CPU was with websites than his 2007 GPU was.

[shinta|hikari]

My real wife and I have had 0 trouble with our 2TB NVMes. Does not run hot either. Could you have gotten a faulty one?

[Buffalobiian]

It could be faulty.

My computer does run hot normally though, in that I only use one exhaust fan at 700rpm which only ramps up when CPU load is high, so the ambient air of my case probably runs at 60deg or something when gaming. My RAM also runs at 55-65C in there. The location of the NVMe doesn't help since it's right between the CPU and GPU.

All the temperatures drop readily when I open the case door (including the NVMe temperatures), so temp is just a function of my PC case and general practice of running things as quietly as possible.

My friend who has a >3yr old NVMe but from a different brand ran into issues where during hard writes it would also crash, and then boot to BIOS with no boot drive being visible to the PC. Eventually it started crashing more frequently and he's currently not able to complete any system file checks or reinstall windows successfully - so he's getting a new drive. I'm waiting to hear if this fixes his problems.

[Kraco]

I recycled the old 1TB Samsung PCIe4 NVMe from my old PC to the new one, again as the system drive. I replaced the other 1TB Samsung PCIe4 NVMe with a 2TB Samsung PCIe4 NVMe. I didn't feel like PCIe5 would bring anything but heat, seeing how I don't do massive video editing. I don't believe the SSDs have been giving me any trouble, but now that you mentioned it, who knows. I do believe I damaged my old CPU by running four DIMMs at the specs meant for two DIMMs, while according to the manual the speed should have been kept much slower. Every time I dropped the speed, the system was more stable... until it wasn't anymore. Win11 on this new PC hasn't been crashing, but I haven't yet checked the OS integrity on the system disk. If there are errors, again like on the old PC, then maybe my system drive is pining for the fjords as well.

I can't say I'd have anything bad to say about NVMe drives. They do load faster than SATA SSDs. But the PCIe5 drives with active cooling elements are just ridiculous.