Subj : Re: MSI H97 motherboard q To : Barry Martin From : Ky Moffet Date : Sun Oct 06 2024 14:11:00 BARRY MARTIN wrote: > Hi Ky! > > > When I installed the NVMe in this computer I didn't add any extra > KM> Silver has one for I don't know what (currently > KM> other-drive-is-busy let's-prevent-fragmentation storage) and one > KM> that does swapfile, temp and browser cache, and VMs (3 > KM> partitions). > > That unknown function is interesting: of course the obvious question is > how do we know the NVMe is doing its job if we don't know what its job > is?! ...Must have come from a government computer - DMV?! LOL. I just never got around to using it foranything in particular, tho at the time I probably had plans. As it is, it'll eventually accumulate enough temporary crap to fill it up. ("Only" 512GB.) > KM> Zombie (the "dead" gamer board) apparently can only support ONE > KM> NVMe, no matter where it is -- still haven't found one that the > KM> mainboard slot supports, but it'll boot off one in a PCIe card. > KM> But only ONE. Two on the card, or one on each of two cards, nope > KM> -- only one is seen. The joys of between-generation tech... > > Being sort of curious, especially when you the usual go-to guy doesn't > know, did a quickie Google search: Hadn't looked. > > 2 NVMe SSDs require 4 PCIe lanes to operate. PCIe x16: your motherboard > often has a primary x16 slot for GPUs and additional x16 slots. 16 > PCIe lanes can support an expansion card with 4 additional M. 2 NVMe > SSDs. > > https://www.sabrepc.com/blog/Computer-Hardware/how-to-add-m-2-nvme-ssd-to-your- > otherboard Yeah, see, according to the article, it should work with 4 NVMes (and the 4-holer card I have appears to be the same as the Asus-branded one, and probably is, except for the fancy cover). But turns out there's also "bifurcation" -- the board has to be able to support splitting the PCIe bus between two or more NVMes. And apparently Zombie does not, while somewhat-older Silver does. > So PCIe issue? I haven't read the article yet but based on that snippet > either something not turned on (or off) in the BIOS or something like > the video card - and maybe the other PCIe cards - using the lanes the > other NVMe wants. Does seem off the motherboard has the NVMe slots but > not the capabilities. That could be. But Silver's mainboard is two years older and has zero issues with NVMes on two different cards. Then again, in some ways Silver's (Asus P9X79 LE) is a more competent board; it was designed for more than just gamer appeal; for one thing it supports 64GB RAM, which at the time was found only in servers and workstations. (Fireball, same age, dedicated workstation board, supports 192GB, more than most do today.) Used Asus P9X79 boards (initially popular with cryptominers, probably because all the x16 slots [3 to 6 depending on model] will simultaneously support a GPU) are now commonly being repurposed as budget server and workstation mainboards. That's another thing Zombie doesn't support -- multiple GPU cards, which I expect is the same problem. However, it has pretty good onboard video (actually in the CPU), and allows up to 512mb shared RAM, so I haven't even bothered with a vidcard. > KM> Silver is older, and doesn't natively support NVMe. But a driver > KM> exists, and the limit seems to be "how many can you stuff in > KM> here?" > > That's my kind of motherboard!! Mine too!!! > > KM> But flat against a heat source doesn't strike me as optimal, no. > > Maybe the simple blowing the air around is sufficient. As I recall the > KM> I'd think so. I've developed the impression that NVMe heatsinks > KM> and other blather are mostly for gamers who are doing huge reads > KM> repeatedly, already have a system full of bling and overclocking > KM> heat, and that starts heat-tripping the drive. > > Could also be a mindset of 'Heat Sinks Are Good!' and the manufacturer > add a 2› part to establish a $5 additional cost. Well, heat-trips are real enough, but you can buy most NVMes either with or without a mfgr-attached heatsink. And... > KM> Had the 1TB NVMe > KM> (on a PCIe card, no heatsink) working for a while the other day > KM> and afterward it's barely warm enough to tell it was powered on. Couple days ago two hours of continuous use, barely warm to the touch. > KM> Case hanging open, but no case fan, just a CPU fan barely running > KM> and whatever the PSU has, also on low. Probably won't bother with > KM> a secondary case fan, doesn't seem very necessary. (Also won't > KM> have a bunch of hot spinning rust internal drives.... doors hang > KM> open on Silver's spinnies in hotswap bays). > > I read somewhere too many fans is unnecessary and sometimes even work > against each other. Depends. Normally the PSU has an exhaust fan (sometimes its own intake fan, too) and that really isn't enough. When the case is full of Hot Stuff, I like to add an intake fan to the case, which also keeps a lot of the dust and particularly lint out of the case. Otherwise that PSU fan creates enough vacuum to suck crud into every orifice (can be enough to totally clog up a floppy or optical drive). So give it a dedicated air source instead, and a filter mesh on the outside of that, if need be. Most of the higher-end cases of the past had an intake fan, so I'm not alone in this. (Silver, Paladin, and Bullet all do.) Dell's "engineered" cooling on what was in 2003 their top-of-the-line $4000 system was probably the worst I've ever seen. Tinker came to me as a two year old because it wouldn't stay running even with then-newfangled water cooling, but still using the dedicated air funnel... I threw out all Dell's crap, gave it a normal HSF and a case-intake fan; temp dropped 40F and it ran stable until the capacitors went a few years later. (Probably the most overpriced motherboard I've seen, too.) > To me it would seem more efficient to be able to move some of the heat > away from the source and redistribute to other parts of the heat sink -- > give additional removal spots. Sort of like the old stove/furnace in The heatsink is supposed to remove heat from the chip, not just move it around. It does no good to reheat another fin, and may harm heat flow. It needs to flow out the fins as directly as possible, that's all there is to it. Of course a larger slug can absorb more heat, but may not release it efficiently. Best seems to be a middling slug (big enough that it doesn't overheat, small enough that it doesn't also store a lot of heat) and lots of very conductive fins, with or without their own fan. (many server CPUs are passively-cooled, but use copper heatsinks.) Here is the real trick: copper everywhere you can have it. HSF foot, contact slug or heatpipes, and if possible the fins too. On a middling-hot AMD (well, all AMDs run relatively hot) I swapped the default aluminum HSF for an otherwise nearly-identical all-copper HSF (fins too), and its running temperature dropped 40F degrees. And even the most basic heatpipe outperforms the best of the slug type. > the middle of the house: hot next to the stove, cooler in the corners of > that central room, but the cold in the next room. Add ductwork to > distribute the stove heat into the adjoining rooms -- ahhh! That's more like moving air around inside the case, which is basically a lot of leaky ductwork. > KM> The rather-slower i5 that came on Zombie had been heat-tripped > KM> (HSF wasn't even touching for the most part) and firmly believes > KM> it runs at 99C all the time, even when it's barely warm. BUT... > KM> otherwise it still works. > > The sensor could either be damaged/stuck or the software reading the > sensor is incorrect. I've seen oddball readings on my hardware > indicating the device is at freezinf (0øC) or thinks it has liquid > nitrogen cooling (super-cold reading). No, it had definitely been heat-tripped (remember it came to me because it was at best flaky and had been deemed dead), and probably that set a flag in the CPU so now it believes if it's powered on, it's that hot. It takes seconds to get there, even if it's barely warm to the touch. Likely it's meant to indicate it ought to be replaced, but it behaves fine with better cooling. But the "new" Xeon is so much faster, I don't care what the i5 thinks. :D > .. Some Haiku express > depths of insight and beauty > but this one does not. That's the one I couldn't quite remember! þ RNET 2.10U: ILink: Techware BBS þ Hollywood, Ca þ www.techware2k.com --- QScan/PCB v1.20a / 01-0462 * Origin: ILink: CFBBS | cfbbs.no-ip.com (454:1/1) .