At SC19, QCT showed off its top-end 4U accelerator server with something unique: a FPGA plus Intel Xeon accelerator card. QCT is showing this solution along with a face detection and feature recognition demo. STH grabbed a few photos of the demo to share.
QCT QuantaGrid D52G-4U Server
The QCT QuantaGrid D52G-4U server is a 4U GPU server with some interesting features. First, it has front power supplies located under the drive bays. These PSUs are then cabled to the rear of the chassis so the PSUs can be serviced from the front, just like with drives, but the PDUs can still be located in the rear. One can also see the array of 2.5″ hot-swap bays.
Moving to the rear of the chassis, one can see 20x expansion I/O slots for accelerators plus additional slots below for networking. By moving the PSUs to the front of the chassis, QCT gets more flexibility in how it configures the entire system.
A fun picture we took is of this service panel. One can see the large fans in the system. One can also see near the top of the photo the latching system for this panel. Instead of using a screw, or something similar that can be difficult to service, this area opens for service by simply pinching two large plastic bars. This s a great feature.
QCT sells this server with fairly standard accelerators such as FPGAs, NVIDIA Tesla V100, and Tesla T4 GPUs. At SC19, it was showing off something else, the QCT E3S10 accelerator.
QCT E3S10 Accelerator
This is perhaps one of the most interesting product names. The QCT E3S10 sounds like a very reasonable name for combining an Intel Xeon E3-1500 V5 CPU (for the “E3”) and a Stratix 10 FPGA (for the “S10”.) The one challenge with this name is that it is extremely hard to search for given that some people use E3 as “Episode 3” and S10 as “Season 10” generating a ton of search results for E3S10 that are unrelated to the computer industry.
The card is intriguing because the Intel Xeon E3-1500 V5 series included Intel’s Iris Pro graphics. You can read about that more in our Intel Xeon E3-1585 V5 review. Choosing this GPU means that QCT has the ability to have a full system on the module using Xeon E3 similar to the Intel Visual Compute Accelerator. A big difference is that QCT has the higher-end GPU solution along with a FPGA.
The demo itself was running Intel OpenVINO and said it was looking for people and features across 48 different video streams with the solution. The system needed to stay closed to continue running the demo, so the only shots we got of the QCT E3S10 were from the rear of the chassis.
Video analytics and feature detection have become an enormous area for hardware companies. Seeing novel solutions like the QCT E3S10 in action is always exciting. QCT showing a custom-designed product for this market shows that the company is getting demand from customers on how to solve real-world problems today with video analytics.