Maurice van der Aa (Advantech Europe B.V. / EMVA)
Jerome Gigot (Ambarella, Inc.)
Bram Verhoef (Axelera AI)
Christian Parr (MVTec GmbH)
If you want to do multimodal deployment, a GPU is more capable. For edge or near-edge applications, AI SOCs are definitely taking over.
Maurice van der Aa, EMVA
The vision team needs the same kind of skill level, not very additional skills, to work with AI accelerators, mostly because of unified interfaces like Halcon
Christian Parr, MVTec
Memory is incredibly important. You have to make sure that your memory architecture can handle both CNNs and transformer-based networks efficiently.
Bram Verhoef, Axelera
If you’re looking at GPUs as pure accelerators, GPUs definitely have a plus for retrofit installations where you don’t have AI.
Jerome Gigot, Ambarella
Compared to an AI SOC, the AI SOC is more integrated, usually has a smaller form factor, and is very relevant for applications like IoT or AR/VR.
Die komplette Podiumsdiskussion können Sie unter folgendem Link anschauen:
<a href=“http://tedo.link/BSHaxM
