Could you elaborate on your experience with local models on your card? I've been thinking of upgrading to 9070 XT, and was thinking the 16GB would be okay-ish to at least run something usable locally, no?

Usable certainly. But my impression is that useful models still need a bit more than 16GB. Something like Qwen 3.8 27B is useful but squeezing it into 16GB requires fairly aggressive quantisation which will make it unreliable (e.g it'll get stuck in loops) and won't leave enough space for a long context (which qwen 3.8 really likes)

I’m the parent of this thread, the person with the with the RX 9070.

My understanding would be that if you’re interested in this sort of card for AI that you should go with the AI PRO R9700, which is basically the professional version of the RX 9070XT but with 32GB of memory.

It’s significantly more money but not crazy like a 5090.

I just happen to have the 9070XT primarily for gaming purposes.

I’m not quite sure how to describe my experience using it other than “rudimentary,” and a lot of that is on me for not really understanding the best way to set it up.

If you have been using cloud hosted models, you will be severely disappointed with what you’d be able to run on 16GB VRAM. You will spend most of your time fighting with the model to fix its mistakes.

What issues did you have? I used Dino and Conversations for six months or so a year ago, and enjoyed the experience. Our conversations were all OMEMO encrypted.

That sounds super cool, I'd love to see some pictures if you have any!

I play Overwatch and CS:GO and in competitive matches it is the norm to use in-game voice, if you don't - you'll get flamed about it.

Casual game modes - that's different.

Yeah but in fact almost no one solo Qs.

Sysadmin tool suggests installing itself with sh & curl combo? This makes me suspicious of the tool itself.

I was thinking the same thing. As a grey haired sysadmin it's disconcerting to see this blasé attitude toward executing local commands from a remote host.

In a practical sense, I'm not sure it's as bad as it looks. It is https. And if you weren't going to read the source code, it's not really different from downloading the program before executing it. In this case, its a small shell script, so it's easy to glance through the code to make sure it isn't doing anything sketchy. For large code bases, that isn't always feasible, so at some point you have to simply trust the source, whether you install from a tarball or (ugh) let a shell execute commands from a remote host.

Also, if you're the kind of sysadmin that fires up a VM or a container any time you want to experiment with a new piece of software, you can afford to take risks. At worst, they'll steal your public key.

It does seem like a bad habit. If you get used to sh+curl install legitimate projects, it isn't such a stretch to sh+curl miscellaneous suggestions on forums.

I saw this link elsewhere on HN today: "is curl|bash insecure?" [1]

[1] https://sandstorm.io/news/2015-09-24-is-curl-bash-insecure-p...

Mine is waiting for ElasticSearch reindex after changing one number to see actual difference in fuzzy-matched search results. 7 minutes is a long time when you need to do it multiple times a day. (While I am typing this, ElasticSearch is reindexing - again)

Also, coffee machine is too far away.

Yes, reindexing ES is a pain in the ass. While it is reindexing, I constantly stare at the screen asking myself "Why the f* do you have to reindex, I just added a field to the mapping you f* piece of sh* moronic database".