GPUs, CPUs, builds and the silicon arms race · page 5
both give you ~128GB of unified memory for local AI, and their memory bandwidth is almost the same — 273 vs 256 GB/s. so the $2,700 gap buys something else entirely. here's what, and who it's actually for.
Ravi Malhotra · Jul 19 · 6 min
the Framework Desktop and its Ryzen AI Max+ 395 twins hand you ~96GB of GPU memory for two grand — enough to load models no single consumer GPU can touch. whether they actually run fast comes down to one word: bandwidth.
Ravi Malhotra · Jul 19 · 7 min
the cards are done. they're sitting at board partners right now. what's stopping them from shipping? a single memory chip that costs about $50 more than the one it replaces.
Ravi Malhotra · Jul 18 · 4 min
a Medusa Point engineering sample smoked AMD's current flagship in a leaked Geekbench run. it was also quietly clocked 300MHz higher than the chip it 'beat.' let's read a leak properly.
Ravi Malhotra · Jul 17 · 5 min
a 360mm server cooler on a five-year-old RTX 3080 dropped VRAM temps by 54°C. The frame-rate gain was under 9%, and that gap is the most useful cooling lesson I've seen all year.
Ravi Malhotra · Jul 14 · 7 min