Pocster

Buy you a drink if you pop over ! Of course if you stopped (expletive deleted)ing about and finish your build …

Clearly you don’t appreciate the spender of an iced latte !

Of course it’s real ! Cornwall that isn’t !

There ! That stopped posts ! 👊🏻

Every post will receive a “ (expletive deleted) off look where I am photo “ 😎😆

Yes ! I responded again so now need another rest !

Reviews are really good for it . Most people just email / web browse e.g waste time on here - it seems very capable for that . I only about 6 months ago bought an air and tbh Neo would have been fine as compiling/llm/heavy work always done on server type hardware ; currently the m3 ultra . But you know what ; I’m having a break 😎

Windows slut !

Ram shortage is real . Every single report suggests it of course does affect Apple . The question is how much …. I think as you can tell it’s going to be bad

We’re all guessing ; but there are indicators . You guess that m5 256gb were ordered last year - no evidence of that . There is a ram shortage . m3 ultra/ m4 mini at end of manufacture unified memory is to some extent is built to order wwdc when ultra m5 supposedly released we will see . No ram shortage effecting apple then no issues with delays . Limited large memory bins and delivery slips to months and then “ not available “ - clearly ram shortage having an impact . My guess is that will happen . So I’ll play pessimistic rather than ram shortage doesn’t really affect Apple .

I reckon even if I’m first off the block with an m5 ultra order of say 256gb it will still be a 10 week wait for delivery . High memory bins are a massive problem so anyone wishing more ( 512gb if they even offer it ) will be waiting months or indeed Apple jump back to “ currently unavailable “ rather than show delivery dates in excess of 6 months . They are clearing old stock ready for new but ram shortage is going to cause a launch where essentially it becomes unavailable very quickly .

M4 options also reduced ! Sure new models are coming but ram shortages clearly becoming a massive issue

Apple have now removed the option for buying 256gb option ! Clearly they have significant constraints on ram higher bins . Previously it was a 6month wait until delivery , now no option . Saving ram for m5

Yes . But also I want local for stt real time . Can’t exactly do that cloud based and integrating into a local home assistant. Or at least not in any sane way and speed . Also lots of other things I wish to do that can only be done locally .

Oh I see lol ! Well with free chat I’d run out of credits , so I tried free Claude and run out of credits . I then upgraded Claude and run out . went back to chat + For 20 quid a month it’s exceptional I think . Our technical conversations can hit 50k tokens . So I was trying to use chat like a company might pay a proper subscription and use Claude . But I ain’t paying 200 bucks a month

What do you mean ? To build my watering system ? just chat . A bit of work to get it reliable with multi file projects ( it doesn’t reliably support git ) . Now . Me and chat design / spec . Local llm will write code / test . Another 2 local llms will deal with reasoning from voice commands to control / anticipate home assistant stuff

lol I’m just testing it out. So did simple “ how does it work “ tasks building up each time . chat and I spec / architect the real thing then just get local llm to do the code . So yeah I’ll be doing it that way . Don’t really need multiple agents etc . But I’m so surprised at the speed and quality of code when spec’d properly . But before this I got chat to write an entire esp32s3 project for me ( automated watering system - it’s on the forum ) . This though - wow . Not only are my demands far more ; but its capabilities are also .

Ooooo ! Forget code complete …. Used to be anticipate variable etc . Now write function - but tbh a bit crap - “ simplistic “ Rephrase the prompt and let it write the entire program ( just a test example ) then refactor it . Wow ! . Local speed surprised me as did code quality . Tomorrow setup a test local git and see what happens . Awesome stuff !

Sure . Just trying to get visual studio to connect to anything . Will need multiple models anyway for full Avalon

On qwen coder 30b getting a very respectable 80 tokens / sec . Estimate on ultra m5 same model might touch 200 !

Go big or don’t go in 😉

Of course not ! The m3 ultra 96gb will be the prototype!

Thing is @-rick- I was just playing with you a bit . Clearly demand will out strip supply for m5 ultra for sure I.e delivery times will slip like a dog . My intention is to order within minutes of it going live in the App Store . My ambitions probably do not need m5 with 256gb or more but it bugs me to buy “old” m3 ultra at near full whack when we all know new and similar priced is coming .

Oi (expletive deleted)ers ! This is my local llm thread ! Not political or supply issues . Get a (expletive deleted)ing room 👊🏻

M3 96gb arrived ! Now the work begins !!