Decatron.aiDecaGEMM · Efficient AI compute
Efficient AI infrastructure

9.55× faster execution
vs. NVIDIA cuBLAS*

DecaGEMM brings Decatron’s energy-based computation to the matrix operations at the core of AI. In our latest benchmark, DecaGEMM achieved 9.55× faster execution than NVIDIA cuBLAS.

* Measured on an RTX 5060 Ti with FP64, 192×192 GEMM, one million requests and 90% result reuse.
See measured performance Request an evaluation
9.55×faster execution vs. NVIDIA cuBLAS*
84.667 µsNVIDIA cuBLAS · time per request
8.864 µsDecaGEMM · time per request
Software-onlyGEMM acceleration path
Training + inferenceone deterministic engine
Bit-identicalverified mathematical output†
Drop-in directiondesigned for current AI stacks
The energy compute wall

AI does not only need more compute.
It needs more Compute Capacity per Watt

Conventional acceleration optimizes hardware, kernels, precision or scheduling — while the underlying matrix-multiplication work remains.

DecaGEMM brings Decatron’s energy-based computation to GEMM. It uses structure-aware energy convergence to reduce redundant arithmetic and data movement while preserving the required result.

DecaGEMM · The product

A new matrix-multiplication engine for current AI workloads.

Designed as a software-only acceleration path for enterprise, cloud and AI infrastructure.

AI workloadInput matrices
DecaGEMMEnergy convergence
AI workloadExact output

Structure-aware compute

Identify reusable mathematical structure and eliminate redundant work.

Less data movement

Reduce the matrix representation and the work that must move through memory.

Elastic depth

Compute only what is required and stop when the result is proven.

Deterministic correctness

Preserve exact output behavior without conventional FP accumulation inside GEMM.

Measured performance

9.55× faster execution than NVIDIA cuBLAS*

In our latest benchmark, DecaGEMM reduced execution time per request from 84.667 µs to 8.864 µs, achieving 9.55× faster execution.

Execution time per request

Microseconds (µs) · Lower is better

DecaGEMM8.864 µs
NVIDIA cuBLAS84.667 µs

Relative execution speed

cuBLAS = 1.00× · Higher is better

DecaGEMM is 9.55 times faster than NVIDIA cuBLAS in the measured benchmark.NVIDIA cuBLAS is the 1.00 times baseline. DecaGEMM achieves 9.55 times the execution speed. Both bars use a linear scale from zero to ten. 10×5×0× 1.00×9.55× cuBLASDecaGEMM
NVIDIA cuBLAS comparison9.55×

faster execution*

8.864 µs per request for DecaGEMM vs. 84.667 µs for NVIDIA cuBLAS.

Execution time per request89.53%

less execution time*

Calculated from the measured request times of 84.667 µs and 8.864 µs.

Evidence integrity0

mismatches†

Across 104,857,600 verified outputs in the August 2026 measured campaign.

Intel GEMV85.33×

fewer calculations†

6,291,264 vs. 536,854,528 scalar operations with bit-identical reconstruction.

* Measured on an RTX 5060 Ti with FP64, 192×192 GEMM, one million requests and 90% result reuse.

† Output verification and Intel GEMV figures are from the supplied August 2026 technical deck. Results are configuration- and workload-dependent and require external validation before generalized production claims.

Designed for adoption

Meet the current stack where it already runs.

DecaGEMM is being productized for evaluation without changing the model or adding specialized hardware.

01Designed for drop-in integration

Designed as a cuBLAS-compatible DLL path for familiar AI deployment workflows.

02Training and inference

One deterministic GEMM mechanism across both major AI compute modes.

03Measure, compare, verify

Evaluate throughput, energy per completed job and exact output under a fixed protocol.

Save energy now

Put DecaGEMM against your real workload.

Request an evaluation