Hello everyone,
I am buying a new workstation for my University research group. We mainly intend to use it for training deep reinforcement learning networks to control a robot. The model of the robot is programmed in a single-threaded C++ environment. Currently, we use python's "miltiprocessing" library to run 8 instances (I have a 4-core, 8-thread processor) of the simulator so we can speedup training with multiple experiments at the same time. The workload of training the network is marginal when compared to the workload of running the model, so it seemed obvious to us that the more models we can run in parallel, or the faster they run, the better. Given that, we are currently considering 2 options: a workstation with 2 Intel Xeon Silver 4110 (or similar), or a workstation with an Intel Core i7-9800X (or similar). So my questions are:
- Which option do you believe its the best? I have seen in the Internet (and some Linus videos) that dual-processor systems not always perform better than single-processor ones, but I believe that my particular case (little amount of shared resources, intense CPU usage) the speedup can be considerable. If you have other processor suggestions within a similar price range, please let me know.
- In the future we may want to use the workstation for a more GPU-intense use, like CNN training. For that case, we will equip the station with a GTX or RTX GPU (we have yet to decide which one). Do you believe that for that purpose the dual Xeon system can be a bottleneck for the GPU? I have seen some discussions about the performance of dual-cpu systems in CNN training, but it would be just OK for us if in this case the system works nearly as fine as the core-i7.
Thank you all for your answers!