Researchers ‘recognize that AI will never replace HPC,’ says AMD executive — high-precision workloads will ‘exist for a very long time’ even as AI usage expands

IBM quantum system alongside Umbra supercomputer.
(Image credit: IBM)
4

While AI continues to reshape industries across science and mathematics, traditional supercomputing has seemingly taken a back seat. However, AMD's Scott Tease told Tom's Hardware Premium that most HPC sites “recognize that AI will never replace HPC,” and suggests that building a heterogeneous compute stack is the best path forward to tackle bleeding-edge research, despite Nvidia pushing performance on lower-precision data types higher for AI workloads at the cost of native performance for the important double-precision FP64 format that’s critical for HPC workloads.

A heterogeneous compute stack brings together quantum, HPC, and AI under one roof, a fleet that AMD is helping pilot at Oak Ridge National Laboratory alongside IBM. Tease is AMD’s corporate vice president of HPC and sovereign AI, who joined the company earlier this year after more than a decade at Lenovo (and, before that, another 13 years at IBM). We spoke with Tease ahead of SC26 to get a read on the state of HPC, how it intersects with AI, and what’s actually driving the research community in a time when computer chips are driving headlines about international espionage plots.

Latest Videos FromTom's Hardware
Jake Roach
Senior Analyst, CPUs

Jake Roach is the Senior CPU Analyst at Tom’s Hardware, writing reviews, news, and features about the latest consumer and workstation processors.

  • Jame5
    “with at least 100 logical qubits that are capable of performing hundreds of millions fault-tolerant operations.”

    Time frame? seconds? hours? days? years?

    On a long enough timeline, current quantum computers easily hit hundreds of millions of operations.
    Reply
  • bit_user
    The article said:
    open-source schedulers like Slurm driving much of the TOP500 to properly schedule workloads on supercomputers.
    Yes, slurm is nice. I've had only limited exposure to it, but it does seem pretty close to the ideal that I had imagined.

    Decades ago, I worked at a chip company that used a homegrown workload scheduling and job management solution for their EDA simulation jobs. They even generalized it to the point that it could handle dispatching jobs to run on testbench machines and could route test automation jobs to the machines which had the appropriate hardware configuration to run them. Slurm can do all of that and much more.
    Reply
  • pf100
    What's HPC?
    Shouldn't the definition be in the article?
    Reply
  • bit_user
    pf100 said:
    What's HPC?
    High-Performance Computing (i.e. supercomputing). There's somewhat of a distinction between that and AI. HPC is the sort of scientific simulations and financial modelling that supercomputers have traditionally been used for. It requires higher-precision arithmetic than the consumer-oriented GPUs are designed for, hence it needs special chips.

    AI, on the other hand, uses lower-than-normal precision arithmetic, but relies on extremely high-density computations. So, there's a natural tension, when the most powerful GPUs would naturally tend to specialize in one or the other, but not both.

    pf100 said:
    Shouldn't the definition be in the article?
    Eh, maybe. Regular readers of the site would probably know it, but I'll grant you that it is sort of jargon.
    Reply