March 28th, 2018 by Adam Armstrong
GTC 2108 News 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. At the GPU Technology Conference (GTC) 2018 in Silicon Valley, StorageReview was onsite, there were several announcements that either small or would potentially get wiped out by some of the larger announcements so we have collected them here. Announcements come from companies such as OSS, NVIDIA, Nyriad, Advanced HPC, H2O.ai, WekaIO, TYAN, Fastdata.io, and Pure Storage.
OSS Rack Scale GPU Accelerator Doubles HPC Efficiency
One Stop Systems, Inc. (OSS) introduced its new rack scale NVIDIA GPU accelerator, GPUltima-CI. The GPUltima-CI can have a maximum configuration of 32 dual Intel Xeon Scalable Architecture compute nodes, 64 network adapters, 48 NVIDIA Volta GPUs, and 32 NVMe drives on a 128Gb PCIe switched fabric, while supporting tens of thousands of composable server configurations per rack. The composable nature of the product allows unused GPU, storage, and network resources to be rerouted to other applications that need the performance.
NVIDIA Releases Boosts To Several VR Use Cases
NVIDIA is one of the bigger players in the GPU space and many of these announcements deal with the company and its enterprise use cases. NVIDIA also release a few “fun” things at GTC including a Holodeck that allows players the chance to enter the dystopian world of Ready Player One. Several industry leaders are adopting NVIDIA VRWorks that help them create rich 360-Degree Video stories. These companies include Z-Cam, Pixvana, and STRIVR. NVIDIA announced the relase of VRWorks 360 video SDK version 1.5 with support for Linux.
Nyriad & Advanced HPC Partner On GPU-Accelerated Data Protection
Nyriad and Advanced HPC announced that they are partnering on a solution that the two companies claim will elevate data protection beyond any RAID solution out there. The combined solution is Advanced HPC’s Orion System bundled with Nyriad’s NSULATE software and an NVIDIA Tesla P4 and 64GB of Netlist NVDIMMs. This new solution is aimed at HPC, big data, and hyperscale storage.
H2O.ia Released H2O4GPU
H2O.ia releases its open source GPU-accelerated machine learning package, H2O4GPU as well as announcing that it and the company’s automated machine learning platform, Driverless AI, are fully optimized for the NVIDIA Tesla V100 and CUDA 9 software. New benefits and capabilities include:
- Truncated SVD (singular value decomposition) & PCA (principal component analysis) for dimensionality reduction and feature engineering
- A new R API brings the benefits of GPU-accelerated machine learning to the R user community
- Improved K-means and XGBoost performance on multi-GPU workloads for faster performance and ability to handle larger datasets
WekaIO Matrix Software Hits 5GB/s Per Client
WekaIO demonstrated that its Matrix software, running NVIDIA TensorRT inference optimizer over Mellanox EDR 100Gb/s InfiniBand on 8 NVIDIA Tesla V100 GPUs, was able to hit 5GB/s throughput per client. This level of performance is ideal for customers looking to leverage AI applications.
TYAN Showcases Server Platforms With Support For NVIDIA’s Latest Technology
TYAN had a wide range of server platforms that support the latest NVIDIA technology include Tesla V100, V100 32GB, P40, P4 PCIe, and V100 SXM2 GPU accelerators. These new server platforms will help customers working in fields such as Machine Learning, AI, Deep Neural Networks, and Inference applications. The platforms include:
- Thunder HX TA88-B7107 – Takes full advantage of the NVIDIA NVLink technology, offering eight NVIDIA Tesla V100 SXM2 GPU accelerators packed within a 2U server enclosure. With four PCIe x16 slots available for high-speed networking and 24 DIMM slots supporting up to 3TB of system RAM
- Thunder HX FT77D-B7109 – Supports up to 8 GPUs including the NVIDIA Tesla V100 32 GB with double the memory capacity, P40, and P4 PCIe GPU accelerators
- Thunder HX FA77-B7119 – Supports up to 10 GPUs including the NVIDIA Tesla V100 32 GB with double the memory capacity, P40, and P4 PCIe GPU accelerators
- The Intel Xeon Scalable Processor-based Thunder HX GA88-B5631 and AMD EPYC processor-based Transport HX GA88-B8021 both feature support for up to four NVIDIA Tesla V100 32 GB GPUs within a 1U server
Fastdata.io Launches GPU-Native Stream Processing Engine
Fastdata.io launch its GPU-native software, FDIO Engine. The software is said to exploit the massively parallel processing capability of NVIDIA GPUs to provide real-time stream processing of big data in motion. Other benefits include:
- Accelerate stream processing up to 1,000 times faster than any CPU-based software.
- Cut total real-time data processing costs more than 70%.
- Run Apache Spark Structured Streaming workloads without any recoding required.
- Reduce power and space requirements more than 98% in the data center whether running in the public cloud or on private servers. This new level of efficiency will save billions of dollars in power costs while tremendously reducing greenhouse emissions.
Pure Storage & NVIDIA Announce AIRI
Pure Storage announced what it is referring as the industry’s first comprehensive AI-Ready Infrastructure, AIRI. Powered by NVIDIA, AIRI extends the power of DGX-1 and operationalizes AI-at-scale for every enterprise. AIRI is a combination of Pure’s FlashBlade, four NVIDIA DGX-1’s, interconnected with Arista 100GbE switches, and NVIDIA Tesla V100 GPUs. AIRI is set to deliver insights at an unprecedented rate.