I just finished a webinar by Dustin Franklin, but there wasn’t enough time for my questions. My company is currently using TX2 as an computation accelerating module of our proprietary algorithms. We want to understand Jetson AGX Xavier to align our future projects.
At the same power consumption as TX2, how much faster would it be?
The module has GigE only, any plans for faster nBaseT?
How long will the product be supported (life cycle)? Is this a consumer / industrial / automotive market?
Thanks for tuning into the webinar and following-up with your questions, Will!
For deep learning inference currently up to 10X performance and efficiency with early-access sw and both devices in 15W mode, which we expect to continue improving over time in the future releases with support for DLA INT8 and additional optimizations for the Xavier GPU.
You can attach Ethernet adapters to the devkit today via the PCIe slot, and I would also recommend checking with Ecosystem partners about their plans for adding more Ethernet to their carrier boards. You should see some new products from that area announced soon.
It would be nice if NVidia can release some sample code (i.e. updated DeepStream SDK) that can be used on both Xavier and TX2, so we can run it side-by-side to verify 10x speed improvement on Xavier. Without seeing it I’ll take this claim with a grain of salt.
Hi albertr, we’ll be releasing DeepStream SDK v3 later this year, and we plan to make the 30-stream demo for Xavier available sooner.
The TensorRT samples are also available now and include GPU INT8 and DLA support. TensorRT is what was used for the results in the webinar slides.