March 20th, 2019 by Adam Armstrong
GTC 2019 Bits
Our News Bits is a roundup that typically covers news pieces that are just small in content, not in impact. This gives content room to breath even if it comes in less than what we normally cover. This week the 10th annual GPU Technology Conference (GTC) took place in San Jose, California. While we covered the major announcements by the heavy hitters, mainly NVIDIA, there were several other smaller announcements that would have gotten washed away in the deluge. These announcements come from vendors such as Kinetica, Aetina, TYAN, DDN, Cognata, Inspur, SQream, and Samsung.
Kinetica Launches Active Analytics Platform
Kinetica announced the launch of what it calls the first complete active analytics platform. This GPU-accelerated data platform unites the key elements of active analytics: historical analytics, streaming analytics, graph analytics, location intelligence, and machine learning-powered analytics. Use cases include:
- Finance: Institutional investors continuously assess value at risk via models in the background as an ongoing process, shifting from a stale view of risk to a dynamic, responsive view triggered by market events.
- Automotive: Automakers recognize unique driving behavior and the conditions that led to it to make more responsive route recommendations, learn how people drive in various scenarios, and ultimately build a better vehicle.
- Retail: Retailers build evolving customer profiles that update as shopping takes place, reacting to customer location, time, price points, preferences, history, and immediate situation, among others, in order to tailor offers and interactions and glean more about the target audience.
- Telecommunications: Telco service providers analyze streaming mobile signals to understand network demand versus coverage, in order to better target the network to users and reduce massive infrastructure costs.
Aetina Presenting Vertical AI Recognition Solutions
Aetina Corporation presented the recognition solution by VID (video-image-detection). While VID can be used for many things, facial recognition is porbably the most popular. Based on Aetina Jetson edge AI computing platform, built by Nvidia Jetson module and Aetina Jetson carrier board AX710/AN310, and third party algorithm can bring a stunning total solution of a stand-alone access control system.
TYAN Announces Several Solutions Supporting NVIDIA T4
TYAN announced several server platforms that NVIDIA T4, NVIDIA V100 Tensor Core, and NVIDIA Quadro RTX 6000 GPU accelerators. These include:
- 4U/20-GPU Thunder HX FA77-B7119: 4U dual-socket Intel Xeon Scalable Processor-based platform with support for up to 20 NVIDIA T4 GPU accelerators, 24 DDR4 DIMM slots, 1 spare PCIe x16 slot, and 14 2.5" hot-swap SATA 6Gb/s devices, four of the bays can support NVMe U.2 drives
- 4U/20-GPU Thunder HX FT85-B7119: 4U dual-socket Intel Xeon Scalable Processor-based platform with support for up to 20 NVIDIA T4 GPU accelerators, 24 DDR4 DIMM slots, 1 spare PCIe x16 slot, and 12 3.5" hot-swap SATA 6Gb/s devices, four of the bays can support NVMe U.2 drives
- 4U/16-GPU Thunder HX FT77D-B7109: 4U dual-socket Intel Xeon Scalable Processor-based platform with support for up to 16 NVIDIA T4 GPU accelerators, 24 DDR4 DIMM slots, 1 spare PCIe x16 slot, and 14 2.5" hot-swap SATA 6Gb/s devices, four of the bays can support NVMe U.2 drives
- 4U/10-GPU Thunder HX FT48T-B7105: Pedestal dual-socket Intel Xeon Scalable Processor-based platform with support for up to 10 NVIDIA T4 GPU accelerators, 12 DDR4 DIMM slots, 1 spare PCIe x16 slot, and 4 3.5" hot-swap SAS 12Gb/s or SATA 6Gb/s drives
- 2U/8-GPU Thunder HX TA88-B7107: 2U dual-socket Intel Xeon Scalable Processor-based platform with support for up to 8 NVIDIA V100 SXM2 GPU accelerators, 24 DDR4 DIMM slots, 4 PCIe x16 slots, and two 2.5" hot-swap NVMe U.2 drives
- 1U/6-GPU Transport HX GA88-B8021: 1U single-socket AMD EPYC processor-based platform with support for up to 6 NVIDIA T4 or 4 Quadro RTX 6000 GPU accelerators, 16 DDR4 DIMM slots, 5 PCIe x16 slots, and two 2.5" hot-swap SATA 6Gb/s devices
- 1U/4-GPU Thunder HX GA88-B5631: 1U single-socket Intel Xeon Scalable processor-based platform with support for up to 4 NVIDIA T4 or 4 Quadro RTX 6000 GPU accelerators, 12 DDR4 DIMM slots, 1 spare PCIe x16 slot, and two 2.5" hot-swap SATA 6Gb/s devices
DDN Showcases Its A³I Reference Architectures
DataDirect Networks (DDN) showcased its next generation of A³I reference architectures, which include NVIDIA’s DGX POD, DGX-2, and DDN’s AI400 parallel storage appliance. DDN sets out to prove how it establishes higher levels of application acceleration when combining the power of GPUs with parallel storage systems. DDN also announced a partnership with Robovision, the self-service, deep learning platform.
Cognata Works With NVIDIA On AV Simulation
Cognata and NVIDIA are working together to accelerate testing and validation within the automated driving development process. The two companies can deliver a wide array of scenario and traffic models for validation using large-scale, hardware-in-the-loop simulation. Cognata provides AI-powered traffic agents and NVIDIA uses its DRIVE Constellation platform to generate millions of miles of AV testing in bit-accurate simulation.
Inspur Announces Server Supporting 8 NVIDIA V100s
Inspur released the NF5488M5 server, the first to support eight NVIDIA V100 Tensor Core GPUs interconnected with ultra-high bandwidth NVSwitch in a 4U form factor. Features include:
- Eight NVIDIA Tesla V100 Tensor Core 32GB GPUs with 5,120 Tensor Cores, providing up to 1 PFlops of AI computing performance
- The option of two 28-core CPUs to provide top-level, general-purpose computing performance, and 6 TB of persistent memory for high-speed data access
- A design that fits a broad range of data center power and a space-conserving environment, especially for power-constrained racks
- Flexible GPU cluster expansion over PCIe fabric
- Designed to operate on 54VDC, a more power-efficient voltage for GPUs
- A multi-layer heat dissipation design and intelligent PID adjustment that provide industry-leading thermal management and control
SQream Announces Support For NVIDIA DGX-2
SQReam announced it now supports NVIDIA DGX-2. This support will allow customers of SQream DB (the GPU-accelerated data warehouse designed for rapidly analyzing massive data stores) to take advantage of the performance jump from the DGX-1 to the DGX-2. The company states that SQream DB customers will be able to build on NVIDIA DGX-2 high-throughput analytical pipelines from ingest to query for historical data analysis, trend identification, fraud detection, outlier detection and more.
Samsung Unveils New HBM Technology
Samsung announced its new HBM technology, Flashbolt. The new HBM2E delivers 3.2Gbps per in, a 33% increase of the pervious HBM2. It also double the capacity per die with 16Gb. Samsung states it can deliver 410GBps data bandwidth and 16GB of memory.