Fair - I used to work at a company that has VM in the name, and last I checked, some of my code is still in use, and many of the docs I wrote have been revised but still are. Now I work for one of the other companies in that group.I do love it when people assume. Maybe you are gods gift to virtualisation. Pretty big board though so it’s a bit presumptive. I know there is an AWS level 7 from ec2 product team for example. I’ve done VMware since 2005, had vExpert, have done kvm for years but I honestly don’t know what he’s talking about if he gets on that shit.
That's the one part I'm debating. I'm tearing down the home lab where I used to farm some of this stuff out to and hoping to consolidate everything but the serious gaming down to one machine. I'm thinking 4x NVMe drives (boot/games/video/VMs), but only one or two of those would be doing major IO at a time, I ~think~. The rest will be pretty idle - not going to play a game, run a demo, and compile video all at once.Anyway, don’t really know what you’re doing, it looks like you’ve got some pretty niche stuff. I’d have thought those would guide you anyway, demoing nvmeof I’d assume you use nvme pass through on a guest and just map it as a target through a vswitch. Point being drive for that, drive for your vms (make sure you need nvme) If you had graphics in there too you’re gonna be pushing it on lanes so maybe headroom does make sense
Less criticality, more I have a crap ton of it I technically can't resell, but could use myself. Like, 768G of the stuffECC works fine with threadripper btw, just depends on degree of critically whether it’s worth it. Farm machines we use it as a job can take more than 24 hours and one app in particular is a single context, non-distributable monstrosity. Bit slower vs losing a day is an easy calculation
Trying to justify it? Sure. Yeah, I am. But I'm also not an idiot, and I'll admit that trying to justify it doesn't necessarily mean it really is justifiable, and sometimes having someone call you out on your bullshit is useful to really narrow down what you need. I NEED 4 NVMe drives and a mass store of some kind - that is sane. But any of the good x570 boards will do that dance. I need 8 USB ports - same. I don't NEED 24 cores - and even the demo really can be fudged without it - I need 12-16. I need 64G of ram, but I don't really need 128 or 256 - and if I do, I still have access to a monster farm that could do it, I just have to carry the source material in and let it rock on that for a bit (network transfer in and out is miserable). It's an annoyance if I ever got there, but not a show-stopper.Standard workstations we don’t. Just test them when they come in and if someone gets a crash they’ll lose nominal time unless they’ve been stupid.
It sounds like you want someone to say yes, (we’ve all been there), speaking for myself I went with it for home PC (am waiting for my water block)
Got a K8s cluster on actual hardware, and LOL at the JVMFor me I know I’m right on the borderline of it making sense and the stuff running is pretty...hefty
Host
Gaming VM
Multiple windows client vms (otherwise pita when you have logins for lots of companies)
K8s Cluster (utilities, maven cache, remote development containers etc)
Pytorch w/gpu
Database server, uses 90gb of ram. Fuck you very much jvm.
Yep. Hell, I'm debating the 10Gbe need myself - I've got it, but the storage devices hooked up don't anymore - got tired of the squirrel fans.I already know 16 cores is enough for that as I don’t use everything together, 4:1 ratios are fine as my workflow doesn’t drive contention and I just switch shit off if I need to run ‘big’ and don’t want the cpu queue issues that big vms cause. Plus big stuff can just take a bit longer if needed. I need 128gb, 32gb dimms get me there, I need 10gbe which is a factor but solvable on an x570.
In the end it just came down to the fact I’m buying this for a 5 year machine and want a single box to replace a couple, plus I’m pretty sure I’ll hit the lane limit on Ryzen by the time I get the updated optane etc. Supported by the fact that whilst I’m not one to spend needlessly, it’s not a problematic amount of money.
Of course my real preference would be to not have shit internet so I could just pay a few bucks an hour for the times I need most of that
We can rationalise anything to ourselves with time but Threadripper very much does not make sense for almost everyone. There’s no getting round that. Its basically “do you earn money from it”, and then “does it make you more money”, if answer is no, you don’t need it. You just want it, which is good and what [H] is about but it’s not rational.
So if you want it, go for it. If you’re still not sure, flip a coin. If you’re disappointed with the answer then you’ll know.
The shit internet is a valid point. I used to run 10 ESXi servers and two storage arrays locally; now, with the net being as slow as it is, and them getting older - wasn't worth the power cost. Still have the work farm, but I have to hand-carry the raw data in if I need to do something there. Stopped hosting my own shit because it just wasn't worth it.
To quote you:
“do you earn money from it” - Nope, not really.
“does it make you more money” - ehhhh..... schmaybe?
I don't think I can justify it right now. And it's not like a 3950 or 3900 will suck that much for the moment. If I do need TR in reality, next year I can figure that out and build a loop for the first time in 15 years. Which I'm really not looking forward to - I freaking hate the idea of water cooling.