# Apple's M5 Ultra glues four dies into one chip

> Apple bolted two dual-die M5 Max chips together and called the result its most powerful silicon yet.
- Title: Apple's M5 Ultra glues four dies into one chip · MacRumors
- Summary: Apple bolted two dual-die M5 Max chips together and called the result its most powerful silicon yet. The M5 Ultra's real pitch is memory: 512GB of unified…
- Keywords: apple silicon, m5 ultra, mac studio, chips, on-device ai, technology, news, Apple's, Ultra, glues, four, dies
- Source: MacRumors — https://macrumors.com/2026/08/25/apple-debuts-m5-ultra
- Published: 2026-08-25T13:12:00+00:00
- Read time: 1 min
- Topics: apple silicon, m5 ultra, mac studio, chips, on-device ai, technology, news
## Apple's biggest chip yet
Apple announced the M5 Ultra on August 25, 2026, calling it the most powerful chip it has ever built.

> Apple today announced the M5 Ultra, which the company says is its most powerful chip ever.
## Four dies, one processor
Next-gen UltraFusion stitches two dual-die M5 Max chips together, Apple's first quad-die M-series design. Inter-die bandwidth tops 4.4TB/s with over 6x the connection density.
## 36 CPU cores
Twelve super cores and 24 performance cores push single-threaded speed up 1.25x and multithreaded up 1.3x against the M3 Ultra.
## AI compute jumps 4.3x
The GPU scales to 80 cores, each with a built-in Neural Accelerator for the first time on an Ultra chip. Graphics run up to 1.8x faster than the previous generation.
## 33 streams of 8K
A 32-core Neural Engine pairs with a Media Engine carrying twice the video encode and decode blocks of the M5 Max, enough for 33 simultaneous 8K ProRes 422 streams at 30 fps.
## 512GB of unified memory
Memory bandwidth hits 1.2TB/s, a 50% jump over M3 Ultra. Apple says that headroom runs large language models with hundreds of billions of parameters fully on device.

> run large language models with hundreds of billions of parameters entirely on device
## Key takeaway

The M5 Ultra's real pitch is memory: 512GB of unified RAM lets you run hundred-billion-parameter models entirely on your desk.