On Monday I had a chance to go to NVIDIA's GPU conference in San Jose, CA. I was surprised how well this conference was organized in comparison with some other conferences I had a chance to visit in past. Plenty of seats to sit at any talk despite high attendance and plenty of good talks to choose from.
My two favorite ones were S7172 - AUTONOMOUS DRONE NAVIGATION WITH DEEP LEARNING (cool potential applications) and S7634 - BUILD A NEURAL TRANSLATION SYSTEM FROM SCRATCH WITH PYTORCH (amazing what one person can do by himself), but I was visiting only for one day.
Now that I read news about the rest of the week at conference, it is quite exciting that NVIDIA finally hopped over 5k GPU cores with Tesla V100 (they have been stuck with 4k GPU cores for a while now). Still a way to go to match 8k cores of AMD Radeon Pro Duo. I know it's not only about count of GPU cores, but also how fast can these cores access and work with memory and I also understand CUDA language is further than C++ OpenCL, but still it's nice to see the upgrade of GPU core counts as well (and yes, tensor cores for matrix operations are nice extra - it's not like matrices and operating with them is going to go away). More details about NVIDIA Tesla V100 can be read at AnandTech.
The cutest surprise of conference was the gadget (bag) I received. I usually do not keep conference gadgets for long, but this one might be exception.