UXL Foundation is an evolution of oneAPI initiative. oneAPI is an open model for programming across multiple architectures – and that is one reason it has become the starting point for the Linux Foundation’s Unified Acceleration (UXL) Foundation.
Fujitsu's involvement in the oneDAL (Data Analytics Library) project align with its vision for delivering open, flexible, and interoperable AI and HPC systems.
Fujitsu-MONAKA, targeted to be released in fiscal 2027, is a 2nm ARM CPU.
Dr. Priyanka Sharma, Director, Software Engineering, and Head of MONAKA Software R&D Unit, Fujitsu Research of India Pvt Ltd (FRIPL), tells us more. Excerpts from an interview:
DQ: How does Fujitsu's involvement in the oneDAL (Data Analytics Library) project align with its vision for delivering open, flexible, and interoperable AI and HPC systems?
Dr. Priyanka Sharma: Basically, UXL Foundation is an evolution of oneAPI initiative. oneAPI is an open model for programming across multiple architectures – and that is one reason it has become the starting point for the Linux Foundation’s Unified Acceleration (UXL) Foundation. UXL’s mission is to drive a vendor-neutral software ecosystem for multiarchitecture accelerated computing.
Here, it’s important to understand that demand for acceleration in HPC and AI requires multiple architectures – CPUs, GPUs, other accelerators. Developers typically do not want to have to program in different ways for different devices. So, when we say open standards for accelerators, it means that our software stack would be supporting CPU, GPU architectures and FPGAs and other accelerators as well. oneAPI provides a solution of one programming model across various hardware architectures, meaning we are not locked into one specific software stack.
This initiative of contributing towards an open ecosystem also aligns with Fujitsu’s Technology Vision towards driving human-centric sustainable digital transformation. And, now, further relating this with FUJITSU-MONAKA, our 2nm ARM based CPU, our major focus is towards supporting large number of open-source software ecosystem for solving some of the cutting-edge problems of the world.
The second important aspect is that Fujitsu is one of the biggest contributors to ARM ecosystem. Fujitsu’s previous chip A64FX Arm CPU was used to build FUGAKU, that had ranked as world’s fastest computer computer on Top500 for 4 consecutive terms till 2021 and even now as world’s number 1 on HPCG and Graph500 benchmarks.
FUJITSU-MONAKA, targeted to be released in fiscal year 2027, is a 2nm ARM CPU that adopts innovative 3D many-core architecture to realize more cores, low latency and high throughput. It will have 288 cores per node (i.e. 144 cores X 2 sockets). It will support SVE2 and Confidential Computing Architecture (CCA).
With FUJITSU-MONAKA, our new power efficient CPU for datacenters, we will be focusing on a wide range of usage in the data center including AI and HPC and contribute to the realization of carbon-neutral society. Hence, as a part of our software ecosystem, we will be working to support a large number of open-source software stack.
oneAPI being an important open model for programming across multiple architectures is an important software stack. We are collaborating to enable some of its stack like oneDAL (that was originally not enabled for Arm) for the Arm ecosystem.
DQ: What specific components or features of the OneAPI Stack has Fujitsu focused on optimizing for MONAKA? How do these optimizations benefit end-uses in terms of performance, energy efficiency, or ease of use?
Dr. Priyanka Sharma: Fujitsu’s vision is to develop an energy-efficient AI software stack to enable maximised Arm CPU performance and support green data centres for sustainable digital transformation. In data centre operations, 80% workloads are machine learning applications. Our focus is on AI software acceleration to contribute towards low-power green data centres with Arm-based HPC technologies.
In terms of our collaboration with oneAPI, we have recently worked towards enabling oneDAL stack for Arm. oneDAL was actually not enabled for Arm ecosystem. So, in order to accelerate some of the important ML workloads on Arm, our team worked towards replacing some of the MKL calls with open-source function calls, and this resulted in oneDAL enablement on Arm.
This development is now merged with the main branch of oneDAL and is also one of the first successful OSS Pull Request through UXL and Fujitsu partnership. We will be presenting details of our work in this reference during LinaroConnect 2024 being held in Spain in May 2024.
Even in the past, Fujitsu optimized and ported the oneDNN DL process library software for the Armv8-A instruction set so that it can be run at high speed on the Fugaku supercomputer. Fujitsu managed to make full use of Arm SVE architecture, resulting in improving performance 9.2 times in training and 7.8 times in inference.
The assembler we developed has been officially incorporated into the main branch of oneDNN along with the source code optimized for the Armv8-A instruction set.
Towards the end, I would like to say that Fujitsu will continue to work towards expanding MONAKA’s software stack with OSS communities, and our partners for easy-to-use software ecosystem. MONAKA Project supported by NEDO, i.e., “New Energy and Industrial Technology Development Organization”, a national research and development agency in Japan.
Our MONAKA 2nm Arm CPU is focused to meet future computing demand of perf, power, reliability and usability for wide range of usage in the data center, including AI and HPC. With this we look forward to contributing to the realization of carbon neutral society by our computing technologies and collaborations with users and partners.
DQ: Elaborate on the Unified Acceleration Foundation (UXL). What are its goals?
Dr. Priyanka Sharma: Unified Acceleration (UXL) Foundation is a consortium under Linux Foundation that is focused towards promoting open source, open standards, and cross-architecture software for accelerated computing. The foundation aims to build a multi-architecture and multi-vendor software ecosystem for all accelerators. It focuses on defining a standards-based, cross-architecture open specification for accelerated computing and fostering the open-source implementations of the specification.
The foundation hosts various working groups and special interest groups (SIGs) that meet regularly to organize community proposals and contributions to the oneAPI specification and open source projects.
The UXL Foundation also invites the community to join meetings, review the oneAPI Specification, and contribute to the development through working groups or by posting comments or questions as GitHub issues.
The foundation plays a crucial role in advancing the field of accelerated computing by promoting collaboration and open standards across different architectures and vendors.
Fujitsu is one of the founding steering members of the UXL Foundation. Fujitsu has a longstanding presence in the open-source community, contributing since 2005. This includes development in mission-critical systems, and in the supercomputer, Fugaku.
Fujitsu is committed to continuing this legacy through its FUJITSU-MONAKA project. We are collaborating with other industry leaders under the UXL Foundation to address the challenges of diverse hardware acceleration, and to make it easier for applications to perform optimally in heterogeneous environments.
Through its involvement with the UXL Foundation, Fujitsu is at the forefront of promoting and developing a vendor-neutral software ecosystem that supports accelerated computing across various architectures for sustainable digital transformation.