Japan Plans the World’s Fastest AI Supercomputer: FugakuNEXT
Move over Fugaku — Japan just announced FugakuNEXT, a next-generation supercomputer that promises to be the Godzilla of AI machines. Built through a powerhouse alliance of Fujitsu + Nvidia + Riken, this beast is set to go online by 2030.
What Makes FugakuNEXT Special?
Unlike its predecessor Fugaku (which already clocked a respectable 0.44 exaflops), FugakuNEXT is designed to combine:
-
High-performance scientific simulations
-
Artificial intelligence tasks
And for the first time in Japan’s top-tier supercomputer history, GPUs will play the lead role as accelerators.
How Fast Are We Talking?
The ambitious goal:
-
600+ exaflops in AI FP8 precision.
-
FP8 = an 8-bit floating-point format, tailor-made for AI’s math-heavy workloads.
Translation: it will crunch numbers so fast, Excel might just cry in shame.
Who’s Building What?
-
Nvidia → designing the GPU infrastructure (future “Feynman” GPU, arriving 2028, will be the star).
-
Fujitsu → developing the new Monaka-X CPU, with more cores, SIMD extensions, and AI-friendly matrix engines.
-
Riken Institute → focusing on algorithms and software to make sure this monster doesn’t just sit there flexing its silicon muscles.
Why Does It Matter?
-
For science: complex simulations, medical research, climate modeling, you name it.
-
For AI: next-level training speeds that could leave today’s GPTs looking like pocket calculators.
-
For Japan: a serious boost in global tech competitiveness — basically a digital samurai sword for the 21st century.
The Vision
Makoto Gonokami, President of Riken, waxed philosophical:
“From ancient times, human civilization has advanced through computing. Today, AI and quantum technologies are transforming science itself.”
Nvidia’s Ian Buck added a bit more hype:
“FugakuNEXT will drive progress for Japan and the entire world.”
Until then, the rest of us will just have to keep rebooting our laptops when Chrome eats too much RAM.
Anhjila

.png)
Comments
Post a Comment