SiMa.ai secures best-in-class MLPerf benchmark results for three consecutive submissions, sustaining its leadership in the AI/ML Edge category
Key Takeaways
- SiMa.ai’s leadership in the AI/ML edge category was made possible by the Palette software that optimizes end-to-end model execution and performance.
- In the 2024 MLPerf submission, SiMa.ai performance as measured by FPS/W increased across all workloads between 7% and 16% .1
- Three consecutive best-in-class results demonstrates SiMa’s continued improvements and commitment as the industry standard for AI/ML at the embedded edge.
SiMa.ai achieved best-in-class results in the MLPerf benchmark for Edge category for the third consecutive time. In the MLPerf 4.0 submission, SiMa.ai maintained dominance in FPS/W, improving performance across all evaluated workloads with increases in FPS ranging between 7% and 16% in the Closed Edge Power category over its 3.1 submission last August.1 As perennial forerunners in MLPerf benchmarking, SiMa.ai continues to deliver on its ongoing commitment to industry leadership by enabling customers to realize real-world value of performance that does not sacrifice power efficiency.
SiMa.ai’s Palette software plays a crucial role in the execution of ML models with significant improvements, such as streamlined memory movements, just-in-time input and output image setup, and data structures aligned to the accelerator format. The result? More efficient use of the runtime resources propelled our SingleStream performance by 16%, MultiStream2 performance by 7% with a 2x lead against the nearest competitor and Offline performance by 14%. 1
[caption id="attachment_4783" align="aligncenter" width="750"]
comparing MLPerf(R) closed edge power SingleStream (3.1-0131,4.0-0077), Multistream (3.1-0131,4.0-0077), and Offline (3.1-0131,4.0-0077), results.[/caption]
We’ve made our position known in the competitive landscape of edge AI, even surpassing industry giants such as Dell and Qualcomm, a testament to our purpose-built ML architecture that drives edge AI workloads with native power efficiency. Our system encompasses the entire ML stack, from tool chain to compiler to software platform, and ML accelerator, all fine-tuned to drive power with maximum efficiency. With each submission, SiMa.ai continues to build on its expertise, not resting on its laurels. These accomplishments reflect a consistent pattern and unwavering commitment: to become the first company that can handle computer vision, vision transformers, and Generative AI, all in one platform. Sustaining innovation and growth where it matters for companies looking to infuse products and processes with AI/ML at the edge.
[caption id="attachment_4785" align="aligncenter" width="750"]
comparing MLPerf(R) closed edge power Multistream (4.0-0044, 4.0-0091, 4.0-0015, 4.0-0070, 4.0-0028, 4.0-0077) results. FPS/W calculated as 8000 / system energy millijoules[/caption]
Customers will experience significantly enhanced performance of machine learning models, unlocking a new level of value across a wide range of edge AI applications from industrial manufacturing to aerospace and defense. This is not simply a technical upgrade, but a strategic leap forward that deepens our leadership in edge AI performance, efficiency, and innovation.
We invite you to explore our detailed MLPerf results for a deeper understanding of our capabilities. If you would like to experience the SiMa.ai difference for yourself, our MLSoC Development Kit is available for purchase online (US only): https://devkit.sima.ai/.
1 comparing MLPerf(R) closed edge power SingleStream (3.1-0131,4.0-0077), Multistream (3.1-0131,4.0-0077), and Offline (3.1-0131,4.0-0077), results.
2 comparing MLPerf(R) closed edge power Multistream (4.0-0028, 4.0-0077) results