Every benchmark I post comes from one machine. That machine is also my dev box, my test rig and my daily driver.
I'm Sebastian. I push local AI on Macs as far as it goes and share everything openly.
More hardware means more models and more numbers for you:
https://t.co/q3z0dgArNl
Two AI launches today, both pitched against a meter.
Apple's 27.3 betas put a daily cap on cloud Siri.
OpenAI's GPT-6.2 pitch: fewer output tokens, "so you can get more done within your usage limits."
The only limit on the model running on my Mac is how much RAM it has.
Your Mac Studio was never the expensive way to get 512GB.
NVIDIA's new DGX Spark HPC kit vs one M5 Ultra Mac Studio:
4x DGX Spark: 512GB across four boxes in a ring, $27,800 at $6,950 each before cables
Mac Studio M5 Ultra: 512GB in one box, 1.2TB/s, $22,299
Same memory. The Mac is $5,500 cheaper and never splits your model across a network.
The kit is for CUDA and MPI research. For one big model at home, it's the Mac.
Apple just put a daily meter on cloud Siri.
The 27.3 betas cap Apple Intelligence per device. More quota means more iCloud+ storage: up to 2x on 200GB, up to 20x on 12TB at $59.99 a month. Apple hasn't said what 1x is.
The model running locally on your Mac has no meter.
@samaschke Agreed. Once several agents hit it with long contexts at the same time, prefill becomes the bottleneck, and four Sparks have a lot more compute to throw at it.
Did you know that running your Mac for more than 50 days without a reboot will fuck with your internet connection?
Yes, really. There is a 32-bit integer timing your TCP connections and if you let it overflow MacOS will run into all kinds of fucky issues with its TCP stack.
For example I was unable to upload things bigger than 4MB. I was wondering why my Claude sessions started to hang seemingly at random.
The fix? Literally just reboot your machine. That's it.
What the fuck
https://t.co/B7kd0PsO5g