Follow

So my desktop has 32 CPUs and 64 gigs of memory, and 2x nvme 960 pro in RAID 1 configuration. Not to mention 4x Vega 64 water cooled frontier edition graphics card (for OpenCL work).

I'm not even sure I can call this thing a Desktop Computer anymore, its more like a low-end supercomputer. No matter what I throw at it it responds instantly. What a fucking beast!

ยท ยท 0 ยท 7 ยท 21

@freemo That is amazing! All I got is my laptop with an i7 8th Gen and 8GB of RAM (upgradable to 32).

What operating system do you got on it?

@MOTT Archlinux, I actually have a farm of similar last gen computers. My other one has 4 last gen watercooled gpu too, the others are more affordable air cooled but 6 cards per box instead of 4.. So have a pretty decent sized farm to work with. This one is my desktop too and the flagship of the lot.

@freemo Wow. That is impressive. I am currently running Antergos right now because I don't want to install vanilla Arch Linux right now and have little reason to.

@freemo WOW. That is pretty fantastic...

I've got an MSI gaming laptop that badly needs an internal cleaning. >_>

@freemo It works well enough for Stellaris and I certainly don't need to power you do for your work.

Still, I wouldn't mind an upgrade sometime in the future. For now, I'm still in the laptop world just because of space and whatnot. Another two years or so, and I'll probably upgrade at that point.

@Surasanji The irony is that gamers drool over my box and it neither runs windows now has any 3d games on it. Its kinda funny cause if i play any game it is a text based game like nethack and the only job the massively over powered gpus need to do is power the terminal transparency so the background is 50% transparent.

Talk about overkill :)

@freemo I wouldn't want 4 graphics cards, even in a dream build. It's just not required. The amount of power draw vs the actual gain in FPS just wouldn't be worth it. Diminishing returns and all.

Maybe two in my *dream* build, maybe.

But, you're doing machine learning and AI stuff- which makes a lot more sense for what you got there.

I consider myself a gamer, and while I'd love to see what your rig could do with my favorite games, it is as you say, overkill, for gaming. :)

@Surasanji For most people yea it would be a waste of power even for extreme games (one of those cards can max out any game). For what i do though I max out not just my cards on one box but across half a dozen similar boxes all at once.

@freemo For sure. But, it would be FUN to see what could happen. Just for shits and giggles.

@Surasanji gonna let my friends load windows and games on an external driver to try their favorite games maxed out :)

@freemo @Surasanji I don't know when this is happening, but this is sort of related. I'm actually thinking of getting myself Adele precision 3630 workstation. Probably next year, but the funny thing is, even though Dell doesn't advertise it. You can actually game using one of those things. I'm unfortunately limited as to which games I can play, but that's something totally different.

@freemo I'm as hopelessly addicted to as can be. I won't buy anything else if I can avoid it LOL. Then again, I'm used to how those things work. I'm sort of in the enterprise/monitoring mindset, so how does one monitor a custom build for hardware failure? For Dell it's called OpenManage, and for HP, it's called iLO. Do most of the custom build manufacturers have similar sensors built in?

@cambridgeport90 yI'd have to investigate how and what those sensors detect to give you an answer t that. Most server monitoring i use tends to be software based and really depends on what you monitor. Kubernetes is the goto commercially right now but that is mostly focusing on monitoring the software infrastructure not non-critical hardware failures, there is different stuff for that.

@freemo Dell and HP have systems that can monitor for hardware failures, temperature changes (not such a big deal, considering my apartment's already freaking cold anyways), and stuff like RAID rebuilds if needed. For instance, my current little server, he's adorable, but the only way I could get any insight into his RAID degradation (he has some, considering he's got two eight-year-old hard drives inside him), is through the actual BIOS interface, which would require a technical sighty to read for me. Hence why that sort of thing matters more so than it might for others. Those are actually coming from the board itself.

@cambridgeport90 All of those monitoring capabilities are availible in linux userland. mdadm monitors raid failures and rebuilds and can notify you when it happens, other software for other types of failure and even some suites to do it all in one. Lots of foss options as far as that goes.

@freemo I guess it's what you're used to. Even my very techie friend is now staying away from custom builds. A friend of ours just gave him a HP Z420, actually. Haven't gotten a chance to check it out, though. He'll probably sell it for a Precision, though, knowing him.

@cambridgeport90 Well less to do with what im used to and more to do with the fact that it is as good as the best you can buy, and at a fraction of the price. I will always pick the solution that is cheaper and functionally equivalent even if I need to get used to new software. In
fact I rarely "get used to" anything in my business as the software changes from day to day and client to client. So I just never get the luxury of getting used to anything. So if i get to pick i usually just pick what is practical.

@freemo Point. I always tend to stick to the secondhand markets or the outlets. there's always some poor sap who thinks that it isn't moronic to sell a fully loaded workstation or server for a steal. Idiots probably don't even realize what they end up with and just want to get rid of it. And on the odd paranormal side, me and computers. Only get me started on that if you're in for some stories, though ...

@cambridgeport90 I have very specific needs for the stuff i buy for my own R&D work for my company. The specs are high;y specific to the type of algos i run. Something prebuilt to handle it either doesnt exist or would easily cost 10x the price.

@freemo Ah. My day job, I run into that sort of thing all the time. Not so much for the stuff I do outside of it, though.

Show more

@freemo Sometimes I wish I had a machine like that... But it would be totally wasted on me! ๐Ÿ˜†

@hyperlinkyourheart The gpus alone were $8K, so unless your doing GPGPU work of some kind as I do it would be wasted on almost anyone

@Strabysme I have a farm of opencl bots mostly for R&D, when idle they can mine crypto to recover some of their costs but their primary purpose and most of the time what their running is ML based research algos.

@freemo I get the "R&D" and "crypto" part but not the "ML" one ^^"

It's really impressive :blobaww:

@nerthos About 10k - 12k if you bought it now. About 2k - 4k over that when I first bought it like a year or so bleeding edge

@nerthos First two i thought i answered the same person, when i realized i didnt i deleted one. Then i realized the one that i kept had a typo that changed the meaning significantly so i deleted and redrafted. Sorry for the triple post. Only one should remain though.

@freemo Should, but with federation and all that I end up seeing if not retaining the three.
Honestly the ideal way to deal with typos is to just make a second message correcting the first one rather than attempting to delete and redo. Skips a lot of trouble and is less confusing

@nerthos hmmmm, weird never saw that artifact here yet. You on mastodon?

@freemo Thankfully no. GNUsocial, soon to be migrated to plemora. It's a common thing though, and even if the messages correctly deleted themselves, were I looking at the timeline actively and tried to reply, I'd get errors when trying to, and in general it'd be very annoying.

@nerthos fair the artifacts probably dont persist as long on mastodon instances thought, just a guess. regardless your suggestion is valid.

@mono I acutally use them on the wires already for the power stuff. Might need some more though

@freemo you need better cable management and a modular power supply ;)

@Aisyk The power supply is modular, the cooling lines are very ridgid and not much i could do there. The cables are all pretty short and zip tied close to the case.

The issue isnt cable management the ussue is that this thing actively uses the full power capability and needs a lot of wires to carry that.

@arteteco Ill throw up a few more shots tomrrow of tha whole room (has my EE lab in it too). Onday i might get the other servers in the farm in some photos too.

Sign in to participate in the conversation
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.