Reinders: “AVX-512 May Be a Hidden Gem” in Intel Xeon Scalable Processors

By James Reinders

June 29, 2017

In this contributed feature, James Reinders explores how AVX-512 vector processing will add flexibility to Intel’s forthcoming “Scalable” processors.

Imagine if we could use vector processing on something other than just floating point problems.  Today, GPUs and CPUs work tirelessly to accelerate algorithms based on floating point (FP) numbers. Algorithms can definitely benefit from basing their mathematics on bits and integers (bytes, words) if we could just accelerate them too. FPGAs can do this, but the hardware and software costs remain very high. GPUs aren’t designed to operate on non-FP data. Intel AVX introduced some support, and now Intel AVX-512 is bringing a great deal of flexibility to processors. I will share why I’m convinced that the “AVX512VL” capability in particular is a hidden gem that will let AVX-512 be much more useful for compilers and developers alike.

Fortunately for software developers, Intel has done a poor job keeping the “secret” that AVX-512 is coming to Intel’s recently announced Xeon Scalable processor line very soon. Amazon Web Services has publically touted AVX-512 on Skylake as coming soon!

It is timely to examine the new AVX-512 capabilities and their ability to impact beyond the more regular HPC needs for floating point only workloads. The hidden gem in all this, which enables shifting to AVX-512 more easily, is the “VL” (vector length) extensions which allow AVX-512 instructions to behave like SSE or AVX/AVX2 instructions when that suits us. This is a clever and powerful addition to enable its adoption in a wider assortment of software more quickly. The VL extensions mean that programmers (and compilers) do not need to shift immediately from 256-bits (AVX/AVX2) to 512-bits to use the new bit/byte/word manipulations. This transitional benefit is useful not only for an interim, but also for applications which find 256-bits more natural (perhaps a small, but important, subset of problems).

The “future Xeon processor” extensions for AVX-512 were first announced in mid-2014. Intel has done the right things to make software ready for it (for instance we can google for “gcc SKX” – where SKX is widely speculated to be a contraction of “Skylake” and “Xeon”).  For several years now, the gcc, Intel, Microsoft, clang and ispc compilers have supported these instructions. Also, some open source software such as embree and mkl-dnn include support. We can create programs today that use these new instructions, and run them easily using Intel’s Software Development Emulator (SDE). Finally, after years of “leaking via software patches,” in May Intel officially announced that AVX-512 will be in the highly anticipated Skylake/Purley platform as the first processor in the new Intel Xeon Processor Scalable Family (successor to the Xeon E5 and E7 product lines).

Vectorizing more than just Floating Point
The net effect: when Intel Xeon processors support AVX-512 we will have exciting new capabilities that extend the obvious use of AVX-512 for HPC and AI/ML/HPDA workloads to offer flexibility perfect for vectorization needs that include integer and bit-oriented data types as well as the strong floating-point support that first appeared with AVX-512 on Intel Xeon Phi processors. While most HPC and AI/ML/HPDA workloads lean on floating point today, there is plenty of reason to believe that algorithm innovations can benefit from integer and bit-oriented data types when hardware acceleration is available. This makes these AVX-512 developments very exciting!

AVX introduced some bit manipulation and byte/word capabilities. AVX-512 expands greatly on these capabilities. It’s the “VL” (vector length) extensions that ties SSE/AVX/AVX2/AVX_512 together in a clever way that makes these new capabilities easier to adopt, for programmers, tools and compilers.

For those exploring new algorithms for AI, including machine learning and HPDA workloads, the inclusion for bit, byte, and word operations alongside floating point opens up exciting new possibilities. The vector length extensions make them immediately useful to SSE/AVX/AVX2 programmers without forcing an immediate shift to 512-bits.

Three Highlights in the extended AVX-512
Intel documentation and the CPUID enabling bits divide up the “SKX” extensions to into AVX512DQ, AVX512BW, and AVX512VL.

  • AVX512DQ is vector support for Double-word and Quad-word integers, also commonly thought of as int32/int and int64/long. In addition to integer arithmetic and bitwise operations, there are instructions for conversions to/from floating-point vectors. Masking is supported down to the byte level which offers amazing flexibility in using these instructions.
  • AVX512BW is vector support for Byte (half-words) and Words, also commonly thought of as char/int8 and short/int16. A rich set of instructions for integer arithmetic and bitwise operations are offered. Masking is supported down to the byte level with AVX512BW as well.
  • AVX512VL ups the ante enormously, by making almost all of the AVX512 instructions available as SSE and AVX instructions, but with a full 32 register capability (at least double the registers that SSE or AVX instructions have to offer). AVX512VL is not actually a set of new instructions. It is an orthogonal feature that applies to nearly all AVX-512 instructions (the exceptions make sense – a few AVX512F and AVX512DQ instructions with implicit 256 or 128 bit widths such as those explicitly working on 32×4 and 64×2 data).

The trend with vector instructions to grow to longer and longer lengths has not been without its disadvantages and difficulties for programmers. While longer vectors are often a great thing for many supercomputer applications, longer vector lengths are often more difficult to use all the time when handling compute problems that do not always have long vectors to process. The VL extensions to AVX-512 bring flexibility to Intel’s AVX-512 that broaden its applicability.

AVX-512 instructions offer a rich collection of operations but they have always operated on the 512-bit registers (ZMM). The downside of a 512-bit register is that it wastes bandwidth and power to use a 512-bit instructions and registers for 256 or 128-bit operations.  Well optimized code (such as that emitted by compilers, or experienced intrinsic or assembly programmers) would seemingly have a careful mix of SSE (128-bit SIMD) and AVX (256-bit SIMD). Doing this is clumsy at best, and complicated by a performance penalty when mixing SSE and AVX code (fortunately there is no performance penalty when mixing AVX and AVX-512 instructions).

The VL extension enables AVX-512 instructions to operate on XMM (128-bit) and YMM (256-bit) registers, and are not limited to just the full ZMM registers. This symmetry definitely is good news. AVX-512, with the VL extension, seems well set to be the programming option of choice for compilers and hand coders because it unifies so many capabilities together along with access to 32 vector registers regardless of their size (XMM, YMM or ZMM).

More AVX-512 features after Skylake?
There is further evidence in the open source enabling work of additional instructions coming after Skylake. Intel has placed documentation of these into its Software Developer Guides, and helped add support to open source projects. Again, this is all good news for software developer because it means software tools do not need to lag the hardware. The four categories of instructions that are documented by Intel thus far, and are not enabled by compiler options for “KNL” (Intel Xeon Phi processors) or “SKX” (Skylake Xeon processors) are:

  • AVX512IFMA: 2 instructions for high/low result of Fused Multiply-Add for 2/4/8-element vectors of 52-bit integers stored in 64-bit fields of 128/256/512-bit vectors.
  • AVX512VBMI: Byte-level vector permute (on 128/256/512/1024-bit vectors) and a select+pack instruction — 4 instructions (with multiple argument types)
  • AVX512_4VNNIW: Vector instructions for deep learning enhanced word variable precision.
  • AVX512_4FMAPS: Vector instructions for deep learning floating-point single precision.

Speculation, supported by open source documentation, says that the first two will appear in a Xeon processor after Skylake, and the latter two will appear in a future Intel Xeon Phi processor (Knights Mill is commonly suggested).

Summary
Regardless of how perfect my speculations are on timing, it is clear that Intel is investing heavily in their expansion of vector capabilities to much more than floating-point operations.  This gives a much-expanded capability for algorithms to be developed with a much wider variety of arithmetic and bitwise operations than floating-point alone can offer.  Who will take advantage of these remains to be seen – but ML/AI and cryptographic programmers seem to be obvious candidates.

With Skylake-architecture based Intel Xeon processors coming soon, it is a great time for programmers to take a closer look.  The instructions are already well supported in tools, and the Intel Software Development Emulator (SDE) makes it easy to run the instructions today.

For More Information
I recommend the following sites for more detailed information:

  • Intel’s online guide to AVX-512 instructions as they are best accessed in C/C++ (intrinsics) has a detailed guide (click on instructions to expand) for AVX-512 instructions.
  • The Intel Software Development Emulator (SDE) allows us to run programs using these Intel AVX-512 instructions on our current x86 systems. The SDE runs code via emulation, which is accurate but slower than on hardware with support for the instructions built-in.
  • Intel documentation is massive, a complete list “Intel® 64 and IA-32 Architectures Software Developer Manuals” covers everything, the specific documents to learn about AVX-512VL are the “Intel® 64 and IA-32 architectures software developer’s manual combined volumes: 1, 2A, 2B, 2C, 2D, 3A, 3B, 3C, 3D, and 4” and “Intel® architecture instruction set extensions programming reference.” Normally, information in the latter document migrates to the former when support appears in a shipping product – so eventually everything about VL will be in the first document. I did not link the specific documents because their links change. We can find them on the main documentation page and click there to get the latest documents. I scroll through them by searching for “AVX512VL” in my PDF viewer and moving from match to match.

About the Author
James Reinders likes fast computers and the software tools to make them speedy. Last year, James concluded a 10,001 day career at Intel where he contributed to projects including the world’s first TeraFLOPS supercomputer (ASCI Red), compilers and architecture work for a number of Intel processors and parallel systems. James is the founding editor of The Parallel Universe magazine and has been the driving force behind books on VTune (2005), TBB (2007), Structured Parallel Programming (2012), Intel Xeon Phi coprocessor programming (2013), Multithreading for Visual Effects (2014), High Performance Parallelism Pearls Volume One (2014) and Volume Two (2015), and Intel Xeon Phi processor (2016). James resides in Oregon, where he enjoys both gardening and HPC and HPDA consulting.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, code-named Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from its predecessors, including the red-hot H100 and A100 GPUs. Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. While Nvidia may not spring to mind when thinking of the quant Read more…

2024 Winter Classic: Meet the HPE Mentors

March 18, 2024

The latest installment of the 2024 Winter Classic Studio Update Show features our interview with the HPE mentor team who introduced our student teams to the joys (and potential sorrows) of the HPL (LINPACK) and accompany Read more…

Houston We Have a Solution: Addressing the HPC and Tech Talent Gap

March 15, 2024

Generations of Houstonian teachers, counselors, and parents have either worked in the aerospace industry or know people who do - the prospect of entering the field was normalized for boys in 1969 when the Apollo 11 missi Read more…

Apple Buys DarwinAI Deepening its AI Push According to Report

March 14, 2024

Apple has purchased Canadian AI startup DarwinAI according to a Bloomberg report today. Apparently the deal was done early this year but still hasn’t been publicly announced according to the report. Apple is preparing Read more…

Survey of Rapid Training Methods for Neural Networks

March 14, 2024

Artificial neural networks are computing systems with interconnected layers that process and learn from data. During training, neural networks utilize optimization algorithms to iteratively refine their parameters until Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, code-named Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Houston We Have a Solution: Addressing the HPC and Tech Talent Gap

March 15, 2024

Generations of Houstonian teachers, counselors, and parents have either worked in the aerospace industry or know people who do - the prospect of entering the fi Read more…

Survey of Rapid Training Methods for Neural Networks

March 14, 2024

Artificial neural networks are computing systems with interconnected layers that process and learn from data. During training, neural networks utilize optimizat Read more…

PASQAL Issues Roadmap to 10,000 Qubits in 2026 and Fault Tolerance in 2028

March 13, 2024

Paris-based PASQAL, a developer of neutral atom-based quantum computers, yesterday issued a roadmap for delivering systems with 10,000 physical qubits in 2026 a Read more…

India Is an AI Powerhouse Waiting to Happen, but Challenges Await

March 12, 2024

The Indian government is pushing full speed ahead to make the country an attractive technology base, especially in the hot fields of AI and semiconductors, but Read more…

Charles Tahan Exits National Quantum Coordination Office

March 12, 2024

(March 1, 2024) My first official day at the White House Office of Science and Technology Policy (OSTP) was June 15, 2020, during the depths of the COVID-19 loc Read more…

AI Bias In the Spotlight On International Women’s Day

March 11, 2024

What impact does AI bias have on women and girls? What can people do to increase female participation in the AI field? These are some of the questions the tech Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Analyst Panel Says Take the Quantum Computing Plunge Now…

November 27, 2023

Should you start exploring quantum computing? Yes, said a panel of analysts convened at Tabor Communications HPC and AI on Wall Street conference earlier this y Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Training of 1-Trillion Parameter Scientific AI Begins

November 13, 2023

A US national lab has started training a massive AI brain that could ultimately become the must-have computing resource for scientific researchers. Argonne N Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire