Maximum Intelligence. Zero Waste.

What's new

From 32-bit floats to 1-bit weights

PNN V4.3 runs on NVIDIA GPUs and carries 4-bit and 1-bit weights end to end — model file, training, CPU kernels and chip, bit-exact on an FPGA — and 1-bit PNN puts a vision transformer into 8.9 MB. INT8 stays the default.

• V4.3

  Faster on NVIDIA GPUs

  PNN kernels for NVIDIA GPUs: a 7B-shaped decoder runs 2.1–3.4× faster than the same model dense at one token on T4, L4 and H100. 1-bit weights end to end — model file, training, CPU kernels and chip, bit-exact on an Artix-7 FPGA.

• 4-bit

  4-bit weights, end to end

  INT4 in the model file, the CPU runtime and the chip. A dual-mode multiplier computes one 8-bit or two 4-bit products per cycle, formally proven.

• 8.9 MB

  1-bit PNN

  ViT-B/16 at 71.19 % ImageNet top-1 with sign-only weights — 39× smaller than dense, 2.0× faster than dense int8 on a 4-core CPU.

• 4 CPUs

  CPU runtime

  Pre-built for x86-64 and ARM64. Faster than tuned dense int8 on Intel, AMD, Arm Neoverse and Apple M1, bit-exact — SLNG: single load, no gather.

• INT8

  Quantised without retraining

  PNN ViT-B/16 goes to INT8 for 0.32 points of top-1, 49 MB against 346 MB dense FP32. Published on Hugging Face.

• FPGA

  Bit-exact on hardware

  The generated 4-bit and 1-bit inference cores run on a Xilinx Artix-7: 52,000 outputs equal to the bit-exact model.

• 0.17×

  Smaller silicon

  Through sky130 place and route, a 1-bit processing element takes 0.17× the area and 0.07× the power of the 8-bit one; 4-bit takes 0.50× and 0.37×.

Key markets

Where the PNN chip goes

• Datacenter

  Double your AI capacity

  Running customers with PNN's smaller memory footprint on PNN kernels with faster execution will virtually double your resources. Spend 10 % on PNN to get 100 % more capacity.

• Automotive

  Perception at a fraction of the parameters

  On real Waymo LiDAR, PNN matches the dense model to every printed digit with 6.7× fewer parameters — headroom for radar, LiDAR and camera stacks on automotive chips.

• Robotics

  Vision on board, within a robot's power budget

  A 1-bit PNN vision transformer runs in 8.9 MB; a chip for it is modelled at 0.6 W — perception that stays on the robot (est).

• Edge & IoT

  Cameras, sensor heads, battery-powered devices

  Whole models in on-chip memory, no DRAM for weights: a 12 mm² chip modelled to run ViT-B/16 ten times faster than a Jetson module (est).

• Smartphones & PCs

  Smaller models on the processors you already ship

  PNN inference runs faster than dense int8 on Arm and x86 CPUs, and 4-bit and 1-bit weights cut the memory a model needs.

• Industrial

  Fixed models, long lifetimes, low power

  A model fixed in silicon — the 1-bit processing element needs 0.17× the area and 0.07× the power of an 8-bit one, in place and route.

(est) = engineering estimate; the other figures are measured.