all major nation state intelligence services have an incentive to spread negative sentiment and reduce developer adoption of ai technology as they race to catch up with the united states.
idk why i am the only person that dislikes ghostty, i just have a tendency to put my machines under heavy load and keep using the cli and ghostty chugs super super badly when there's any gpu or cpu load, especially compared to "slow" terminals. zero issues with iterm2 on macos or gnome terminal on linux under load but that plus zero bells or whistles makes ghostty seem like a purposeless toy to me.
One of my favorite demos. First time seeing physics and interactive 3d combined. SGI had to change the name to 'newton' because 'jello' is a trademark.
FYI it also supports pre-training, reward model training and RL, not just fine tuning (sft). My team built a managed solution for training that runs on top of llama factory and it's quite excellent and well supported. You will need pretty serious equipment to get good results out of it, think 8xh200. For people at home i would look at doing an sft of gemma3 270m or maybe a 1.6b qwen3, but keep in mind you have to have the dataset in memory as well as the model and kv-cache. cheers
depends ln your goals of course. but worth mentioning there are plenty of narrowish tasks (think text-to-sql, and other less general language tasks) where llama8b or phi-4 (14b) or even up to 30b with quantization can be trained on 8xa100 with great results. plus these smaller models benefit from being able to be served on a single a100 or even L4 with post training quantization, with wicked fast generation thanks to the lighter model.
on a related note, at what point are people going to get tired of waiting 20s for an llm to answer their questions? i wish it were more common for smaller models to be used when sufficient.
This is oddly enough around the time I canceled my 10+ year premium subscription and moved to freetube. I cannot support the anti privacy crusade of asking for users government identification and i spoke with my wallet. What are the chances that enough people cancelled premium that it caused youtube to increase the amount of advertisements served? What are the odds that just drove away a bunch more people?
Doing things that way (virtualization rather than containerization) fell out of vogue specifically because of how bad vmware was to work with. CPU quotas were probably what pushed serious people away from the product instead of machine licenses. I was early in my career but working with vmware products was the bane of our existence because if we wanted to make any sort of configuration change or spin up a test machine or really do anything at all it had to run through accounting which was just an instant non starter. we all started fiddling with alternatives and docker swiftly became reliable at least for spinning up a new web server or testing the latest and greatest whatever. vmware did this to themselves.
Exactly. You end up having to burn licenses for stuff like that, although I will concede that VMware always gave us more licenses than we paid for and they always included extra products. It was weird though. There was no rhyme or reason to it. One time they gave us 1000 licenses for VMware Fusion even though we didn't have any Macs. Microsoft, on the other hand, let us use whatever we wanted. If it was still around when it came time to true-up, we paid for it.
But not on vmware, just zen and/or kvm with various management front ends.
VMware has always been a PITA, even in the late 2000's, we pivoted and bought several thousand physical machines for a new datacenter after they started to play tricks just weeks before we were going to turn up the DC.
They have always aspired to be Oracle like, where customers are hostages. Most people I knew who weren't stuck in the "Enterprise" trap moved to kvm/zen ASAP especially after the Westmere dramatically reduced the vm_exit() latency allowing for databases etc...
That was over 15 years ago, and outside of a very small number of niche use cases, tehre was no real argument to run container hosts on Vmware outside of a (IMHO) mistaken risk appetite.
It is really the fruit that ate itself, as had IT departments had a more data based risk assessment process, we would probably be heavily hybrid-cloud now. But the same Enterprise gravy train that VMware grew under killed them.
Shifting blame at great expense in licensing and agility to an _Enterprise_ solution was their jam...now Broadcom owns them an it is even worse.
Not in the F100. They're all VMs, all of the time, all on vSphere. Nutanix was the next best solution, with Hyper-V as a distant third. Hence why Broadcom ate them.
From what I hear and have seen, pre-Broadcom head-on VMware takeouts didn't go much of anywhere. But kubernetes-based (Kubevirt) products do seem to be having a degree of success.
100%. it was very hard (but not impossible) to compete against vmware in the virtualization space before the Hockening™; vSAN, vSphere and NSX were very very nice and super easy to use. nutanix succeeded more often than not.
orgs now are kind-of scrambling for alternatives. great for nutanix. better for the CSPs. also great for red hat (my employer) since we have openshift virt, which is based on kubevirt but is much easier to use and works nicely with the rest of the ecosystem.
Yeah, vsphere has a mile-long list of enterprise checkbox features that the sales managers can overwhelm the CIO's with on the golf course.
Kubernetes might have success, but AFAICS Kubernetes also sort of involves a new way of architecting applications (cloud native applications, 12-factor apps, microservices, etc.; whatever the buzzword du jour is). The idea with vmware was always to virtualize all those zillions of more or less idling physical servers, and get some snazzy management GUI to handle them all etc. etc.
Rearchitecting for containers is indeed a lot more effort and, indeed, one of the reasons for VMware's success was that it provided more efficiency without (at least initially) much in the way of operational changes.
But kuvevirt with Kubernetes does much of the same, especially for companies that are--or know they will--move to containerized workloads anyway.
kubernetes is where vsphere was circa 2010 or so. wide adoption with loads of room for growth and refining.
the vmware implosion + kubevirt gaining maturity is what will hyper-accelerate this.
nonetheless, it's a bit nuts that a vsphere alternative doesn't really exist for kubernetes. openshift, harvester and foss projects like portainer are close but not really the same thing. vsphere made it stupid easy to orchestrate vms at any scale with any level of experience. your cat could deploy a simple two-node vDC, and you could theoretically click your way through doing very very complicated things with your cluster (though you probably should).
esxi was silly easy to install most of the time also. provisioning kubernetes is finally (as of maybe three years ago?) pretty easy to do, but day two ops are still very command line. great for nerds like you and me; not good for former VIadmins who were suddenly thrown into the k8siverse.
this is even worse on the networking front. anyone who's messed with cilium or calico vs administrating NSX will know what i'm talking about. you can click your way through setting up a whole logical enterprise-grade networking fabric with vSANs, BGP, stateful firewall, the whole nine, all with near-linespeed performance throughout your vDC. cilium and calico, on the other hand, are powerful but have (to my knowledge) no real GUI equivalents.
storage is even worse. there's no vSAN frontend equivalent in the k8s landscape afaik.
This really seems only obviously true if you're counting docker/podman-desktop and similar dev tools which work via stashing containers in a VM. There are a ton of large scale kubernetes deployments made directly on baremetal.
how much money did they get out of your insurance? paradoxically people who have less coverage are safer in that circumstance (they let you put in 72 hours vs 72 days)
Around $1,500/day, totaling $550,000. I reported to Medicaid CMS that it's discriminatory False Claims Act fraud because I never hallucinated, self harmed, nor was violent. I never signed anything saying I would pay.
