A CPU that runs entirely on GPU
github.com
March 4, 2026
8 min read
Summary
nCPU is a CPU architecture that operates entirely on GPU, utilizing tensors for registers, memory, flags, and the program counter. All arithmetic operations, including addition, multiplication, bitwise operations, and shifts, are performed through trained neural networks, with specific methods like Kogge-Stone carry-lookahead for addition and learned byte-pair lookup tables for multiplication.
Key Takeaways
- The nCPU operates entirely on GPU, utilizing tensors for registers, memory, flags, and the program counter, with all arithmetic performed through trained neural networks.
- The nCPU achieves 100% accuracy on integer arithmetic, verified by 347 automated tests, and includes 23 models totaling approximately 135 MB.
- Multiplication in the nCPU is 12 times faster than addition, contrasting with conventional CPUs where multiplication is typically slower than addition.
- The nCPU benchmarks show execution times ranging from 136 to 262 microseconds per cycle, depending on the instruction mix, with models loading in 60 milliseconds.
Community Sentiment
MixedPositives
- The exploration of running a CPU entirely on a GPU opens up intriguing possibilities for future computing architectures, potentially revolutionizing how we approach processing tasks.
Concerns
- The skepticism about completely replacing CPUs with GPUs highlights fundamental differences in how each handles latency and processing, suggesting that a full transition may not be feasible.
Related Articles
Flash-MoE: Running a 397B Parameter Model on a Laptop
Mar 22, 2026
Parakeet.cpp – Parakeet ASR inference in pure C++ with Metal GPU acceleration
Feb 27, 2026
BarraCUDA Open-source CUDA compiler targeting AMD GPUs
Feb 17, 2026
Run a 1T parameter model on a 32gb Mac by streaming tensors from NVMe
Mar 24, 2026
Source
github.com
Published
March 4, 2026
Reading Time
8 minutes
Relevance Score
61/100
Why It Matters
This page is optimized for focused reading: quick context up top, a clean summary block, and a direct path to the original source when you want the full story.