New HPC and Deep Learning POWER8-Pascal GPU Cloud Available Now
I am excited to tell you that Nimbix, a high-performance computing (HPC) cloud provider, deployed a new cloud capability this week that features the extremely high-performance IBM servers that include the new POWER8 with NVIDIA NVLink processor and the new NVIDIA Tesla P100 Pascal GPU accelerator. These NVIDIA NVLink-equipped servers, the IBM POWER S822LC for HPC servers, have two POWER8 CPUs connected to four NVIDIA Pascal GPUs via the high-speed NVIDIA NVLink interface and half a terabyte of system memory.
I believe this new Nimbix POWER8-Pascal cloud provides the best infrastructure for running HPC applications and workloads, along with new advanced analytics workloads like deep learning and GPU-accelerated databases. This is because not only does it have the fastest processor-GPU combination, but also because it uses a Mellanox Infiniband network to interconnect the servers. Since deep learning and HPC applications often use multiple servers, up to hundreds at a time, this combination of a high performance “fat” server node with a high-speed network backbone enables large scale clusters to be used for high performance work flows and applications.
The Nimbix POWER8-Pascal cloud service features both baremetal POWER8 servers along with an image with the popular deep learning packages, CAFFE, Torch, and Theano, pre-installed to enable quickly launching these applications. We have already demonstrated amazing results on the deep learning frameworks with the new IBM S822LC for HPC.
Enabling Deep Learning and GPU-Accelerated Databases as a Service
This cloud enables startups, research teams and enterprises developing or using deep learning applications and GPU-accelerated databases to easily offer or run their applications as a service. In fact both Minds.ai, which provides a high-speed deep learning service, and Kinetica, a GPU-accelerated database company will soon be available on this new POWER8-Pascal Nimbix cloud.
JARVICE, a unique platform capability of the Nimbix Cloud, enables easy deployment of jobs and scaling of on-prem applications to burst into the cloud.Developers and application providers can leverage Nimbix PushToCompute™ technology to rapidly build and deploy new applications targeting the POWER8-NVLink-Pascal server at bare-metal speed.
Easy Developer Access to new NVLink-based P100 GPU
I am excited about this new cloud because it enables easy access to developers who want to try out the new NVIDIA Tesla P100 Pascal GPU and take advantage of the high-speed NVIDIA NVLink interface that connects the POWER8 CPU and NVIDIA Tesla P100 Pascal GPU together. NVLink enables applications to move data between the CPU and GPU 2.5 times faster than over PCIe, which helps most applications that have large datasets.
Access to the Power8 capability of Nimbix Cloud is available at: https://power.jarvice.com.
Please do post how your experiences are as you use this new cloud; I would love to hear how CPU-GPU NVLink helps your application run faster and makes programming faster.
Here are some performance results on this Power8-NVLink "Minsky" server from Stone Ridge Technologies <a href="https://www.garudax.id/pulse/echelon-nvidia-p100-vincent-natoli?trk=prof-post">https://www.garudax.id/pulse/echelon-nvidia-p100-vincent-natoli</a>. This is a reservoir simulation application that is used in oil and gas exploration and drilling.
Interesting. The questions arise are - a) Price performance comparison with AWS P2 instances? b) Where to store the learning sets ? c) What with scaling out DL ?
What he said
Would love to see this benchmarked.