21:39:40 <nikolam> If using VT-d , do to you think that PCI-passthrough could be safe ? And not to give the whole box/compute node to that GPU user ?
21:41:41 <nikolam> Also I am seeing a video using unprivileged LXC on Proxmox sharing GPU (with cgroups on Linux host). Wonder how secure that actually is (With Intel A380 GPU) but it seems it makes GPU accelerated light virtualization possible . 
21:42:14 <nikolam> Or I am just not taking into account actual GPU drivers for illumso for Intel and AMD GPUs in the first place. 
22:04:19 <danmcd> Unless vgatext is attached to the device, illumos probably doesn't have the GPU driver (esp. not SmartOS... OI might, but not SmartOS).
22:08:50 <nikolam> I hear people mentioning new wave of all types of (desktop, other) apps would ask for GPU to function properly 
22:09:15 <nikolam> Like AI parts in it doing it's job locally 
23:06:59 <sommerfeld> Possibly dumb question: is there way to programatticaly determine your cpu cap inside a smartos zone?  
23:12:51 <sommerfeld> ("getconf NPROCESSORS_ONLN" gives a number larger than what I can use)
23:19:14 <jbk> sommerfeld: i think it might be in the metadata
23:20:32 <jbk> though depends i guess
23:21:16 <bahamat> sommerfeld: You can do `mdata-get sdc:cpu_cap`
23:22:02 <bahamat> sommerfeld: These are the committed fields available: https://eng.tritondatacenter.com/mdata/datadict.html 
23:22:21 <bahamat> sommerfeld: There are others available, but they're considered internal.
23:25:15 <bahamat> The others available but not documented are basically because they're OS dependent.
23:46:23 <sommerfeld> bahamat: thanks.   units are centi-cpus?  (n% of a cpu?)
23:48:57 <bahamat> sommerfeld: yes
23:54:21 <bahamat> sommerfeld: those are properties mapped directly in from vmadm, so thats your auxiliary documentation.