Quadric

Learn about the latest AI/ML breakthrough - KAN enables dramatically smaller model sizes with orders of magnitude lower power consumption. https://lnkd.in/gkY4dQYH

The fallacy of operator fallback and the future of AL/ML accelerators https://lnkd.in/gF4rTGx5

How big is failback failure? NPU plus DSP products are 32x, 64x, or even 128X slower than Quadric's Chimera. See blog with details at https://lnkd.in/gZVBQRVN

Going to Design & Reuse IP SoC Day? Be sure to see Steve Roddy, CMO of Quadric, at 1 pm. See https://lnkd.in/gmS-2CVx for details

New ML networks far outperform old ones. What can you do with old hardware? Nothing. Your hardware needs to be programmable to use the new networks. https://lnkd.in/g8KeaZnB

Quadric is getting noticed! Today we were happy to see investment firm #Activant Capital include Quadric in their coverage of new disruptive AI/ML technologies. https://lnkd.in/eXpyDqY8

Right now, companies from the mid-cap sector to the enterprise are racing to acquire AI hardware so they can deploy next-gen applications faster than their competitors and create defensible “moats” that might be able to last for years. Yet deploying AI hardware/software and developing ML algorithms for the next phase of AI - inference - presents challenges for CTOs and leaders, many of which are rooted in the process of evaluating what hardware to invest in. One of the challenges is that the shortage in hardware over the past few years limited options. Another is that hardware often comes with proprietary software, so developers may have to learn new languages to program models. On top of that, porting models from one vendor’s chip to another can be impractical, especially if the models were optimized for the original hardware. The greatest challenge might be the murkiness surrounding benchmark results. Companies use benchmark tests to show what their HW can do, but the HW is often tuned to achieve optimal performance, which doesn’t necessarily reflect real-world scenarios. To effectively democratize AI/ML, we need HW companies to address the pain points while also increasing the transparency of performance benchmarks and empowering potential customers to make informed choices. Ease of use, model portability, and transparency can be part of a strong business model that fosters customer loyalty rather than creating customers that become flight risks. This was demonstrated clearly to me when I took an afternoon from the Chiplet Summit to visit the headquarters of DeGirum Corp. and interview their CTO Shashi Kiran Chilappagari. I have no doubts that transparency will be vital to DeGirum’s success, and I’ll dive deeper into my conversation with Shashi another time. Meanwhile, here are two other startups we see having a substantial impact on the democratization of AI/ML due to creating holistic platforms where ease of use and transparency are virtues: Quadric - Not only does Quadric offer something radically different with their General Purpose NPU, but they also offer a different approach to benchmarking. CEO Veerbhan K. and his team decided early that Quadric would push their Chimera toolchain out in the open for customers and their DevStudio includes the source code for all the benchmark nodes, giving engineers and developers unprecedented visibility without the complexity. SiMa.ai - They do create great hardware, yet CEO Krishna Rangasayee and the SiMa.ai team view themselves as “a software company that builds its own silicon”. This is because they’ve co-created HW/SW in parallel to build their highly-regarded developer platform called Pallette which offers full functionality for developing ML models at the edge and empowers customers to use their own proprietary algorithm development and testing practices. #semiconductorindustry #ai #machinelearning

Great discussion of Chimera at approximately 7 minutes into this video

Hasn't this always been the case. That's why it's great to work at a start-up!

Start watching at 7 minutes to find out more about Chimera.