One marketing tag line often heard from Arm is “From Sensors to Supercomputers”, highlighting our company’s ubiquitous nature across so many of today’s technology sectors. A great example of this is found in the vital Oil and Gas sector where proficient seismic exploration is a key to success. Here we see millions of connected sensors being deployed to harvest waveform data and some of the largest supercomputers in the world being deployed to interpret that data into actionable high-fidelity maps of the earth’s crust.
While Arm partners have been supplying Cortex M class processors into the sensor space for many years, Arm is relatively new to the decades old supercomputing sector. In just the past 12 months we have seen announcements from our silicon partners reaching or nearing production: Marvell with its production ThunderX2, Fujitsu touting its advanced A64FX design, and quite recently, Huawei with its efficient Kunpeng 920 SoC. A deeper look at the Arm partner landscape will note that stalwarts HPE, CRAY, ATOS-Bull, Fujitsu, and Huawei are significant server OEM’s in the HPC market, all with Arm based offerings that align well to this space. Two notables from the above, we saw HPE deploy our first production Top 500 supercomputer cluster at Sandia National Labs last year and heard great news from Fujitsu at SC18 where they highlighted the expected performance and compute density of their upcoming A64FX chip which should supply an unheard of 1 PFLOPS of compute per rack.
On the software side of things, Arm is a major player in HPC with our earlier acquisition of Allinea and their software tools. The Arm Forge tools enable debug and analysis of complex HPC codes across all the prevalent platforms and languages (including python). For heterogenous Oil and Gas datacenters which have significant amounts of “owned codes”, these software tools are key.
Anyway, enough of the market segment recap. Let’s delve a bit deeper into the Oil and Gas landscape as it pertains to HPC and Arm.
The Oil and Gas industry leverages HPC for several key applications including:
The seismic and reservoir codes are normally quite data and compute intensive. Extreme amounts of data stem from the large field sizes (many miles deep and wide), which can be 2D, 3D, or 4D in scope, and have a huge number of sensors and factors involved. Upwards of a million different sensors may be used with to harvest seismic wave data. High numbers of sensors and high sampling rates are key in order to meet fidelity requirements and generate actionable models. The seismic data input files can easily be on the order of several terabytes in size. A major trend from Arm silicon partners in this space is to step-up the memory and IO throughput of their designs on a per socket basis in order to more efficiently handle these big data applications.
At the heart of popular O&G workflows such as the Reverse time Migration (RTM) or the Full Wave inversion (FWI), seismic wave modelling solvers need to take advantage of both post-Petascale computing facilities and the latest features at the chip level. This often corresponds to hundreds of thousands of high-end computing cores routinely used for seismic inversion in the largest computing facilities. Consequently, these applications make the most of reliable software stacks and ecosystems, including optimized compilers toolchains and math libraries.
To tackle the significant amount of data and compute required, parallelism is heavily leveraged by dividing the data sets and problem spaces and then distributing these across a large # of compute nodes and cores in a cluster. Well-tuned connectivity stacks and low-latency interconnects such as Mellanox InfiniBand support the efficient distribution of data across available nodes. Furthermore, hardware platforms need to have the right balance of CPU, memory and IO performance. Moving vast amounts of data on and off each compute node can take as much time and energy as the arithmetic computation cycles themselves and leave the processors stalled a disproportionate amount of time.
As Arm entered the HPC arena, we initially engaged with a strategic set of HPC end-users with a research and co-design mentality in order to understand how we should best evolve and extend the Arm architecture for future large-scale systems.
Below we note some of the features in the Armv8 (64-bit) architecture and related design IP that go a long way in addressing many of the challenges noted above for HPC and Oil and Gas applications:
Today’s heterogenous datacenters and solutions demand cross-platform software tools. Oil and Gas companies will deploy a mix of x86, Power, Arm, and GPGPU compute solutions. With Arm’s acquisition of Allinea two years ago, we have become a growing software presence in the Oil and Gas sector by offering Arm FORGE tools for software analysis and debug of complex HPC codes across a variety of prevalent architectures, including those mentioned above. We also offer the Arm Performance Libraries with prevalent optimized functions such as BLAS, LAPACK, and FFT for a variety of partner micro-architectures.
2019 is a great time to evaluate Arm for Oil and Gas related applications. A strong set of OEM and silicon suppliers are bringing a new balance in terms of performance, efficiency, and platform choices to the HPC market.
Connect with Arm HPC Tools experts and partners at the following upcoming event:
Please visit Arm.com/HPC for more info on the Arm HPC ecosystem.