AMD unveils the world’s fastest HPC accelerator GPU
AMD has announced the development of the world’s fastest high-performance computing accelerator GPU.
The new AMD Instinct MI100 accelerator is set to revolutionise the industry because it excels in all capabilities, providing Artificial Intelligence (AI) and industry-leading compute performance. The GPU is also supported by new accelerated compute platforms from Dell, Gigabyte, HPE, and Supermicro.
AMD made the announcement following the recent Intel GPU of the Intel Iris Xe series for laptops in October, the company’s first foray back into the discreet GPU market since 1998. The first laptops to implement the Intel’s new GPU are the Acer Swift 3x, Asus VivoBook Flip TP470 and Dell Inspiron 15 7000 2-in-1. Intel has also hinted that it will launch a DGI for desktop video cards in 2021.
But while Intel concentrates on the PC market, AMD’s innovation looks set to transform computing for the global, business arena. Combined with AMD EPYCTM CPUs and the ROCm 4.0 open software platform, the MI100 is designed to drive forward new discoveries ahead of the exascale era. Its unprecedented performance will enable scientists and researchers worldwide to accelerate discoveries in industries including life sciences, energy, finance, academics, government, defence and more.
, assistant vice president in NCBU, GIGABYTE said the new computing power offered by the MI100 would be a big draw for customers from several industries. He explained, “AMD Instinct MI100 accelerators represent the next level of high-performance computing in the data centre. As a new accelerator in the GIGABYTE portfolio, our customers can look to benefit from improved performance across a range of scientific and industrial HPC workloads.”
The demand for high-performance computing (HPC) has risen considerably as industries adopt the technology to improve products, reduce production costs and decrease the time it takes to develop new products. As the demand for Big Data increases, so does the need for analysis, with HPC providing the solution.
AMD’s latest breakthrough will offer companies the chance to aggregate data at high speed, producing increasingly accurate, real-time analysis of performance.
Advocates of the technology say AMD’s innovation will also drive forward the development of exascale computing, which is currently defined as systems that can perform at least one exaflop―or one quintillion (a billion billion)—calculations per second.
The AMD Instinct MI100 GPU is built on AMD CDNA architecture, creating a new standard of accelerated systems for HPC and AI when paired with second-generation AMD EPYC processors. It also delivers seven times the FP16 theoretical peak floating-point performance for AI training workloads when compared to AMD’s previous generation accelerators.
, director of science, Oak Ridge Leadership Computing Facility has been trialling the technology. He said, “We’ve received early access to the MI100 accelerator, and the preliminary results are very encouraging. We’ve typically seen significant performance boosts, up to two to three times compared to other GPUs.”
Currently, the AMD ROCm developer software provides the foundation for exascale computing. The open-source toolset consists of compilers, programming APIs and libraries. ROCm is used by exascale software developers to create high-performance applications and has also been optimised to deliver performance at scale for MI100-based systems.
Messer commented, “What’s also important to recognise is the impact software has on performance. The fact that the ROCm open software platform and HIP developer tools are open source and work on a variety of platforms, it is something that we have been absolutely almost obsessed with since we fielded the very first hybrid CPU/GPU system.”
The performance of the GPU accelerator and its capacity for compatibility has made it popular with leading computer and technology manufacturers. , senior vice president, PowerEdge Servers, Dell Technologies said it was a step forward in data centre technology.
“Dell EMC PowerEdge servers will support the new AMD Instinct MI100, which will enable faster insights from data,” said Pendekanti. “This would help our customers achieve more robust and efficient HPC and AI results rapidly. The high-performance capabilities of AMD Instinct accelerators are a natural fit for our PowerEdge server AI and HPC portfolio.”
AMD’s corporate vice president, Data Centre GPU and Accelerated Processing, revealed the technology was created specifically with the scientific community in mind and he hoped the MI100 would benefit research work on a broad scale.
He said, “Today AMD takes a major step forward in the journey toward exascale computing as we unveil the AMD Instinct MI100 – the world’s fastest HPC GPU. Squarely targeted toward the workloads that matter in scientific computing, our latest accelerator, when combined with the AMD ROCm open software platform, is designed to provide scientists and researchers a superior foundation for their work in HPC.”
3 ways crypto mining is impacting the data centre industry
Around the world - particularly in Russia, Eastern Europe and China - the global rise of crypto currency values has been driving an en masse industrialisation of the mining process. The trend has been bubbling away for several years, as the home mining rig has largely found itself edged out by hyperscale server farms comprising some of the largest data centres anywhere in the industry - all designed to mine crypto.
The demands placed on a facility built and run as a mining operation are somewhat different to those placed on a hyperscale cloud facility or enterprise data centre. Reliability isn’t so much of an issue; if a mine goes down for a few hours, money is lost, but your data centre won’t take half the websites in Western Europe down along with it.
On the flipside, density and cooling are much, much more important. To make a crypto mining operation profitable, you need to be harvesting more crypto currency (be it Ethereum, Dogecoin, or the perennial Bitcoin) than you’re paying for electricity by a significant margin. As a result, some of the most efficient cooling and hyper-dense rack architecture from the past few years - like two-stage liquid cooling - has originated as a crypto mining solution. Now, hyperscale cloud operators in particular are recognising the benefits of these innovations and applying them to other aspects of the data centre industry.
1. Liquid Cooling
Crypto data centres have always been as dense as possible, with their racks running at maximum capacity all day, all year round. By contrast, the average enterprise or cloud data centre isn’t necessarily running at peak capacity 24/7; workloads fluctuate with demand. However, as that demand has skyrocketed over the past year in particular, cloud and enterprise operators have looked to crypto’s preference for liquid cooling as a way to run data centres closer to the ragged edge of performance than ever before.
One example of this is LiquidStack. The Hong Kong startup makes a revolutionary two-phase liquid cooling solution for data centres, which was developed over a number of years inside Bitfury, one of the world’s leading crypto miners. “Bitfury is sharing our knowledge with the global data center community and we are excited that Microsoft and other internet giants can benefit from our years of experience and investment to best practice liquid cooling,” said Joe Capes, CEO of LiquidStack in an interview with Data Centre Magazine.
Now, LiquidStack is going mainstream, with substantiated rumours that Microsoft is looking to adopt their DataTank solutions across its ever-expanding portfolio of hyperscale cloud regions.
2. Denser HPC
One of the issues that liquid cooling solves is how to create ultra-dense server racks that can function at high temperatures. Crypto miners have been grappling with this problem for about a decade now, and the lessons they’ve learned are being happily adopted by the burgeoning data centre HPC market - which is swelling in response to greater AI adoption and increasingly-sizable data sets.
With the density that mining rigs can achieve, server architects are cramming hundreds of kilowatts into individual racks - although it should be noted that this is still relatively rare. A 2020 survey from the Uptime Institute still found that the average density of data centre racks was growing rapidly, however.
“We expect density to keep rising. Our research shows that the use of virtualization and software containers pushes IT utilization up, in turn requiring more power and cooling. With Moore’s law slowing down, improvements in IT can require more multi-core processors and, consequently, more power consumption per operation, especially if utilization is low. Even setting aside new workloads, increases in density can be regarded a long-term trend,” said the report.
In 2020, average rack densities of 20kW and higher became a reality for many data centre operator.
3. Sustainability Concerns
Now for the more worrying news. The industrial scale and massive power consumption inherent to the crypto mining business - and the negative attention that miners are now starting to receive from government - could point towards a concerning future for data centre operators in the wider industry.
Last week, the Chinese government announced that it would open an inquiry into the participation of Beijing’s largest data centre operators - which include the country’s three largest telecom firms - in crypto mining. At a time when the PRC government is attempting a significant reversal of its approach towards sustainability, the significant power draw of crypto mining activities may be one more hurdle than China cares to deal with.
The Indian government is mulling a blanket criminalisation of all crypto mining in the country and, in the US, the State of New York is also looking into tightening regulatory restrictions on the industry.
While crypto mining data centres are not the same as cloud or enterprise facilities, operators should be careful lest the ire of lawmakers be the latest trend to make its way from the crypto sector into the mainstream.