9 ms·
I ran into this issue on one of my builds. Aida64 has a benchmark (floating point photo or something?) that uses AVX instructions. Pressing the "run benchmark
by jrockway 7y ago
I ran into this issue on one of my builds. Aida64 has a benchmark (floating point photo or something?) that uses AVX instructions. Pressing the "run benchmark" button would instantly black-screen crash my machine with 100% certainty.
I debugged this problem over a number of years... I replaced the RAM, I replaced the motherboard, I eventually replaced the CPU... still happened no matter what I did. Even if I underclocked the machine and kept the voltages the same, instant crash. It was maddening.
Exasperated, I eventually busted out an oscilloscope and looked at the waveform on the 12V supply to the CPU. When starting the AVX benchmark, there was a huge brownout. That basically explained everything; my power supply essentially turned off when the CPU started drawing a ton of power. I replaced the power supply and got lucky -- it handled it fine and I could run the benchmark. I even got some overclock out of it.
After this whole experience, I've never looked at computers the same way again. You can buy high-spec brand-name components, and it's all just a crapshoot. Maybe your computer won't crash in the middle of an important task. Maybe it will. There isn't much you can do but cross your fingers.
- wtallis 7y agoYou can do a little bit to help by ensuring you shop for the right kind of high-spec: workstation rather than gaming, so that the extra money goes toward useful engineering and QA rather than RGB LEDs. But even then, it often ends up that the best you can hope for is a long warranty with a quick and easy replacement process.
- CoolGuySteve 7y agoI've had enough failed server and workstation components to believe this is snake oil. It seems like a lot of "Enterprise" hardware (particularly hard drives) is mostly the same chipset/components with a longer warranty and a more boring PCB color. Or more annoyingly, differentiated by feature lockout in some firmware. Gaming hardware has larger economies of scale which leads to more people running the hardware and submitting RMAs. Gamers are also more likely to overvolt their hardware, leading to most consumer components being overspec'ed for heat and power. As such, the second revision of a gaming board imo is more refined than a workstation/server board that almost never gets a new revision. On a more practical level, consumer CPUs usually have higher clockrates and newer architectures and gaming NVidia GPUs are so much cheaper than the enterprise product lines that you can buy multiple. My workflow is faster with the cheap stuff, sometimes twice as fast.
- wtallis 7y agoTo some extent, this depends on what kind of component you're talking about. I agree with what you're saying about consumer vs workstation GPUs because the price disparity is so large, but consumer GPUs don't usually get the full RGB treatment anyways. I think you're also overestimating how big the gaming desktop components market is, and underestimating how fragmented it is. For motherboards, the economies of scale apply to the OEM custom motherboards used by Dell, HP, etc., which have far more volume than any one consumer retail motherboard model. You're probably also vastly overestimating how many gaming-oriented motherboards make it past revision 1.0. I just checked Gigabyte's top gaming product line, and out of 75 models listed, all but 8 were revision 1.0. Browsing through a similar number of their most recent mainstream consumer motherboards didn't turn up any that went beyond rev 1.0. Some gamer-oriented motherboards are over-specced for power delivery and cooling. It's definitely common to find VRM heatsinks that are oversized gaudy sculptures, but that doesn't help product reliability. There are also tons of examples of gamer-oriented boards that advertise a large number of VRM phases, but have obvious shortcomings in their power delivery design once the heatsinks are removed and an EE takes a close look at things. And then there's the motherboard at issue here. When gamer boards are overbuilt, it's not thorough, it's usually just the most visible aspects that get beefed up (and the price). RAM is a really obvious area where paying a bit extra for workstation-class ECC DIMMs gets you a real reliability advantage, but paying extra for high-speed consumer modules with RGB LEDs and big heatsinks does not. RGB LEDs are still thankfully pretty uncommon on SSDs, but the handful of examples show that the impact varies from compromised performance to catastrophic instability. Most gamer-oriented SSDs with no LEDs but with flashy heatsinks also fail to outperform or outlive the best pedestrian-looking models. Enterprise and NAS hard drives are definitely diverging from mainstream consumer hard drives in more ways than simple firmware feature gating. Some of the technologies they're using to drive density beyond consumer HDD needs have a questionable impact on reliability, but filling them with helium does help reliability, and you don't find that on a WD Black or Seagate Barracuda.
- garaetjjte 7y agoOh, WD Purple surveillance HDD brochures are just marketing bullshit™: "Using AllFrame™ technology, WD Purple™ drives improve video capturing and help to reduce errors, pixelation, and video interruptions that can occur in a video recorder system." "WD Purple 8TB, 10TB, 12TB & 14TB capacities feature AllFrame AI technology that enables not only recording up to 64 cameras, but also supports an additional 32 streams for Deep Learning analytics within the system." "WD Purple drives with AllFrame AI technology feature a workload rating up to 360TB/year to support the Deep Learning analytics that is featured in AI capable NVRs." https://documents.westerndigital.com/content/dam/doc-library/en_us/assets/public/western-digital/product/internal-drives/wd-purple-hdd/data-sheet-wd-purple-series-2879-800012.pdf https://documents.westerndigital.com/content/dam/doc-library...
- dman 7y agoIf there is one thing in a computer you should not cheap out on, it is the power supply. Buy from Seasonic and buy more power supply than you currently need.
- zymhan 7y agoYou say not to cheap out, and then recommend Seasonic. I'd say avoid a brand that was notoriously the cheapest option in recent history.
- Scramblejams 7y agoA lot of brands, Seasonic among them, vary from model to model. The last time I felt like I could just blindly stick with a brand was PC Power & Cooling, and they've since been absorbed by OCZ, so... ?
- kalenx 7y agoSome Seasonic are probably among the best consumer power supplies you can find (e.g. https://www.anandtech.com/show/11252/the-seasonic-prime-titanium-power-supply-review https://www.anandtech.com/show/11252/the-seasonic-prime-tita...)
- ricketycricket 7y agoAt the very least, Anandtech seems to rate Seasonic very highly: https://www.anandtech.com/show/11252/the-seasonic-prime-titanium-power-supply-review https://www.anandtech.com/show/11252/the-seasonic-prime-tita...
- kllrnohj 7y agoIt's not just Anandtech. All reviewers have nothing but praise for Seasonic. They have the best reputation in the industry at the moment and all their PSUs have a 10 year warranty to back it up. Jonnyguru's last Seasonic review: https://www.jonnyguru.com/blog/2018/07/03/seasonic-prime-ultra-750-titanium-power-supply/6/ https://www.jonnyguru.com/blog/2018/07/03/seasonic-prime-ult... Literally nothing in the bad & mediocre summary sections, and with nothing but recommendations for the entire series: > Our look at the PRIME Ultra Titanium series has now at last come to a conclusion. We can now definitively say there’s not a bad performer in the bunch, and while all do have some very minor drawbacks I can’t see any reason not to recommend any of them.
- Scramblejams 7y agoGood call. After learning the hard way in my youth, now when I build a machine I don't try to save money on the power supply. Instead I do a bunch of research, stick to the premium models, and I haven't regretted it.
- kuon 7y agoI had weird issues a few times, and it was always the power supply. Before I would never think of it as the responsible for issue, but now I always do. Each time a friend has a weird unexplainable problem with a gaming PC, I suggest to try another power supply and it always fixed the issue.
- seabrookmx 7y agoFor me it's the same, but with one additional step. Test the RAM in a known stable machine. If it passes an overnight memtest, then replace the flaky machine's PSU.
- jrockway 7y agoRAM has always been my "go to" "I bet this is it" problem, but I've debugged my own machines and other people's machines and I haven't seen bad RAM in ages. I avoid XMP, however, and typically manually specify the voltage that is listed in the RAM's spec sheet and run at the speed that the CPU manufacturer recommends. (So the RAM might be sold as 3200 but Intel says 2400 is the max, so I run it at 2400.)
- paol 7y agoI get your desire to be conservative here, but this is not good practice with modern CPUs. For example AMD desktop CPUs are designed with DDR4 speeds well north of 3000MHz in mind[1]. Problem is that JDEC profiles simply don't go that hight for some reason, so you you have to run with XMP profiles if you want the chip to perform as the manufacturer intended. [1] https://www.anandtech.com/show/14525/amd-zen-2-microarchitecture-analysis-ryzen-3000-and-epyc-rome/11 https://www.anandtech.com/show/14525/amd-zen-2-microarchitec...
- jrockway 7y agoI look at Intel ARK and use what they say the CPU is rated for. I am a couple generations behind, so not at the bleeding edge. (Haven't used an AMD CPU, but I would find their equivalent if I did.)
- arminiusreturns 7y agoI would say this is why burn-in has been ingrained into me as extremely important, especially target workload specific burn-in. I see way too many cots things cobbled together and then cursed for failing in some scenario that should have already been tested for (and specifically built for).