← Feed

Posts

• View post

  @decryption The hype about them is it's not a mac - with a 9B model my studio (M1 Ultra) is faster using ollama. The GB10 is about 5% faster on a 16-20B model and the Studio dies with a 30B model which eats about 70Gb of RAM on the GB10. An M5 Ultra Studio would shit all over these. Even a handful of the M3 Ultras chained together beat out a GB300 and for way less than half the cost. (still expensive - like $80k vs $180k) But the GB10/300 is not a mac if you're not into that and I think...

• View post

  @decryption I was playing around with some local models today on a mac studio with 64Gb and a dell GB10 with 128Gb. Managed to crash the studio because it didnt have enough ram. The models I could run on the GB10 pretty much suck compared to claude or whatever. Those are $6000-$7000 devices. Local LLM's are better than they used to be but are barely better than curiosities on anything within reasonable reach of mortals.

• View post

  swiftDialog has a (version) birthday today 3.0 is out now with new stuff and better old stuff https://github.com/swiftDialog/swiftDialog/releases/tag/v3.0.0 #macadmins