HPC cloud specialist Rescale is partnering with Intel and HPC resource provider R Systems to offer first-ever cloud access to Xeon Phi “Knights Landing” processors. The infrastructure is based on the 68-core Intel Knights Landing processor with integrated Omni-Path fabric (the 7250F Xeon Phi).

“It is a three-way joint initiative,” Tyler Smith, head of partnerships at Rescale told HPCwire. R Systems is hosting and managing the Intel technology at its HPC datacenters in Champaign, Illinois, with access to the cluster available through Rescale’s “big compute” ScaleX Pro platform.

R Systems Principal Brian Kucic said his company is committed to offering advanced, bare metal technologies to the HPC research community through its partnership with Intel and Rescale. At Intel, General Manager of the Accelerated Workgroup Barry Davis emphasized that the partnership is intended to support increased adoption of HPC applications to drive discovery and innovation.

Starting today, the Xeon Phi cloud is being made available at no cost for four weeks on a first-come-first-serve basis. Via a single queue, end users can submit one job at a time for up to six single-socket Knights Landing nodes. Running follow-up jobs will entail getting back in the queue. Rescale has plans to offer a premium, priority queue that will allow users to run multiple jobs simultaneously. Initially, however, Smith said the effort should be viewed as a test lab that brings next-generation computation power to the cloud.

After the complimentary period expires, the cost will be $0.03 / core / hour, so for the 68-core platform, the price will be about $2 per CPU.

While the new Intel gear is initially targeting dev and test environments, at least two Rescale customers have expressed interest in running larger workloads, according to the company. “We can reserve a certain number of nodes for a specific customer to do a POC if they’d like but I think that’s all going to come with time,” said Smith.

The KNL cloud offering is part of new effort called ScaleX Labs, and Rescale expects this one to be the first of many, anticipating future opportunities both with Intel and with other hardware providers. “The [larger] initiative is reflective of what we’ve done [here] with Intel, to provide customers with access to pre-release or early-release hardware or be the first to offer a product in the cloud,” Smith explained. “Depending on the hardware partner, the lab can be used in a few different ways: it can hopefully create demand from software partners, to optimize their code, or create demand from the public cloud provider. And it allows Rescale to provide a platform in which our customers can test the latest and greatest hardware.”

Rescale offers the cloud labs through its ScaleX Pro platform, which features an intuitive GUI for job management and collaboration and secure SSL data transfer and encryption at rest, according to the company. ScaleX Labs users will also receive beta access to ScaleX Developer, Rescale’s product that allows software developers to integrate and deploy their own software on the ScaleX platform. ScaleX Developer is currently on track for general availability in Q3.

“I think one thing that attracted Intel to partner with Rescale on this initiative was our ability to enable developers to have beta access to our ScaleX Developer product so they can deploy their applications directly to the ScaleX platform and then end users could bring proprietary code in a bring-your-own-software model and run it on the KNL,” said Smith.

A single-socket 68-core Knights Landing CPU running at 1.60 GHz, with two AVX-512 vector units (512-bit) per core, provides around 3 teraflops of double-precision (peak) performance. The bootable Knights Landing chips, based on the second generation of Intel’s Many Integrated Core (MIC) architecture, are largely competing against Nvidia’s Pascal-generation P100 GPUs released last year.

Rescale doesn’t own or operate any of its infrastructure. Instead the San Francisco company partners with cloud providers, both major IaaS purveyors and others that are more regional in nature. And with ScaleX Labs, Rescale also has a path to work directly with hardware suppliers.

“Once we do the API integration, that’s where our secret sauce is, where we can quickly add our software application portfolio to that cloud provider and within a very short period of time make those software applications available in that cloud provider,” said Smith.

He also emphasized the advantages of a multi-cloud strategy. “[The only way to have] access to an “infinite” amount of resources, [is to] partner with multiple cloud providers. Google had access to the Skylake processor first; AWS came out with a bigger GPU offering. If you’re locked into one public cloud provider, you’re limited. It’s a competitive world out there and folks are worried about time-to-market and performance is highly sensitive. It’s also about scalability. If you’re ever bound or constrained by computing you’re not really a cloud provider; it kind of goes against the ethos of it, at least in my opinion.”

Although Rescale supports 220 software applications, the ScaleX Labs with Intel Knights Landing will launch with only a few software packages at the outset. The aim is to have it grow organically through customer demand. “We want to offer it within the parameters that Intel has set out to target applications that have been optimized for KNLs and will see a performance boost or will see a performance boost from the Omni-Path fabric,” said Smith. “There’s also a lot of proprietary code out there and this will be a way for end users to test out how it performs on KNL. We’ll expect to grow the software as more commercial software vendors support KNL, which is coming, and secondly, as other customers request it, which our sales guys are already starting to see happen.”

AI and deep learning frameworks won’t be there at launch, but according to Smith, a proactive effort is underway to build up this support over the next couple of months. Given Intel’s AI ambitions and proclamations for Knights Landing’s machine learning chops, it makes sense that this would be a priority.