silicon & Math
AI Is Math.
Breakthroughs in AI can only come from Math.
The Zeroth Layer
Purpose-built logarithmic math, hard-wired in 3 nm silicon, ready for the largest models and real-time 4K video.
Log MAth
Cut Chip Compute Area with Native Log Math
By replacing every multiply with lightweight log-math adders, we free chip compute area versus today’s FP8/INT8 GPUs. Fewer transistors means chips run cooler and more energy-efficient.
FP16 Multiply
Tensordyne Log Math 16bit Multiply
Silicon
Silicon for Beast-Mode Performance and Economics
Fewer transistors frees up chip real estate. This allows us to pack in extra tensor engines, significantly more high-bandwidth SRAM & HBM3e, plus any-to-any fabric that feeds data at full throttle so the beast never starves. All of this in a smaller chip means tighter yield nodes, resulting in better hardware economics.
Quality
Uncompromised Accuracy, Full Dynamic Range
Tensordyne Log Math delivers the ideal number system for AI - achieving lower error compared to floating-point while preserving the dynamic range. Accuracy greater than 99.9% relative to any trained language, vision, or video model comes standard. With Tensordyne Log Math, you gain uncompromised precision and efficiency. Say good-bye to tradeoffs and model degradation that come from using ultra-low-bit formats just to meet power budgets.
Language Models
| Model | Relative Accuracy |
|---|---|
| OPT -66B | 99.92% |
| Llama2-13B | 99.98% |
| Llama2-70B | 99.97% |
| Llama3-8B | 99.97% |
| Llama3-70B | 99.93% |
| Llama3.1-405B | 99.91% |
| Falcon-180B | 99.90% |
| Mistral 7B | 99.98% |
| Mixtral 8x7B | 99.98% |
Image Models
| Model | Relative Accuracy |
|---|---|
| SD 1.5 | 99.97% |
| SD XL | 99.92% |
| SD 3 | 99.91% |
Efficiency
Unlocking High-Res Real-Time Video
Higher resolution real-time video thrives on log math’s high dynamic range and in some cases even better than native resolution. Use cases previously out of reach are now technically possible and economically feasible.
Future outlook
Raising the Roof for AI
Shrinking process nodes and pruning weights help—but they’re incremental. At some point, the industry needs a revolution. Logarithmic compute not only raises the roof for AI, but also reshapes the scaling curve.