Tag: GPU

  • The Opportunity in Lagging Edge Semiconductors

    While much attention is (rightly) focused on the role of TSMC (and its rivals Samsung and Intel) in “leading edge” semiconductor technology, the opportunity at the so-called “lagging edge” — older semiconductor process technologies which continue to be used — is oftentimes completely ignored.

    The reality of the foundry model is that fab capacity is expensive to build and so the bulk of the profit made on a given process technology investment is when it’s years old. This is a natural consequence of three things:

    1. Very few semiconductor designers have the R&D budget or the need to be early adopters of the most advanced technologies. (That is primarily relegated to the sexiest advanced CPUs, FPGAs, and GPUs, but ignores the huge bulk of the rest of the semiconductor market)
    2. Because only a small handful of foundries can supply “leading edge” technologies and because new technologies have a “yield ramp” (where the technology goes from low yield to higher as the foundry gets more experience), new process technologies are meaningfully more expensive.
    3. Some products have extremely long lives and need to be supported for decade-plus (i.e. automotive, industrial, and military immediately come to mind)

    As a result, it was very rational for GlobalFoundries (formerly AMD’s in-house fab) to abandon producing advanced semiconductor technologies in 2018 to focus on building a profitable business at the lagging edge. Foundries like UMC and SMIC have largely made the same choice.

    This means giving up on some opportunities (those that require newer technologies) — as GlobalFoundries is finding recently in areas like communications and data center — but provided you have the service capability and capacity, can still lead to not only a profitable outcome, but one which is still incredibly important to the increasingly strategic semiconductor space.

    When GlobalFoundries abandoned development of its 7 nm-class process technology in 2018 and refocused on specialty process technologies, it ceased pathfinding, research, and development of all technologies related to bleeding-edge sub-10nm nodes. At the time, this was the correct (and arguably only) move for the company, which was bleeding money and trailing behind both TSMC and Samsung in the bleeding-edge node race. But in the competitive fab market, that trade-off for reduced investment was going to eventually have consequences further down the road, and it looks like those consequences are finally starting to impact the company. In a recent earnings call, GlobalFoundries disclosed that some of the company’s clients are leaving for other foundries, as they adopt sub-10nm technologies faster than GlobalFoundries expected.

    GlobalFoundries: Clients Are Migrating to Sub-10nm Faster Than Expected
    Anton Shilov | Anandtech

  • Going from Formula One to Odd One Out

    Market phase transitions have a tendency to be incredibly disruptive to market participants. A company or market segment used to be the “alpha wolf” can suddenly find themselves an outsider in a short time. Look at how quickly Research in Motion (makers of the Blackberry) went from industry darling to laggard after Apple’s iPhone transformed the phone market.

    Something similar is happening in the high performance computing (HPC) world (colloquially known as supercomputers). Built to do the highly complex calculations needed to simulate complex physical phenomena, HPC was, for years, the “Formula One” of the computing world. New memory, networking, and processor technologies oftentimes got their start in HPC, as it was the application that was most in need of pushing the edge (and had the cash to spend on exotic new hardware to do it).

    The use of GPUs (graphical processing units) outside of games, for example, was a HPC calling card. NVIDIA’s CUDA framework which has helped give it such a lead in the AI semiconductor race was originally built to accelerate the types of computations that HPC could benefit from.

    The success of Deep Learning as the chosen approach for AI benefited greatly from this initial work in HPC, as the math required to make deep learning worked was similar enough that existing GPUs and programming frameworks could be adapted. And, as a result, HPC benefited as well, as more interest and investment flowed into the space.

    But, we’re now seeing a market transition. Unlike with HPC which performs mathematical operations requiring every last iota of precision on mostly dense matrices, AI inference works on sparse matrices and does not require much precision at all. This has resulted in a shift in industry away from software and hardware that works for both HPC and AI and towards the much larger AI market specifically.

    Couple that with the recent semiconductor shortage (making it harder and more expensive to build HPC system with the latest GPUs) and the fact that research suggests some HPC calculations are more efficiently simulated with AI methods than actually run (in the same way that NVIDIA now uses AI to take a game rendered at a lower resolution and simulate what it would look like at a higher resolution more effectively than actually rendering the game at a higher resolution natively), I think we’re beginning to see traditional HPC shift from “Formula One of computing” to increasingly the “odd one out”.

    The HPC community is used to being first, and we always considered ourselves as the F1 racing team of computing. We invent the turbochargers and fuel injection and the carbon fiber and then we put that into more general purpose vehicles, to use an analogy. I worry that the HPC community has sort of taken the backseat when it comes to AI and is not leading the charge. Like you, I’m seeing a lot of this AI stuff being led out of the hyperscalers and clouds. And we’ve got to find a way to take that back and carve our own use cases. There are a lot more HPC sites around the world than there are cloud sites, and we have got access to all a lot of data.

    Trying to Do More Real HPC in an Increasingly AI World
    Timothy Prickett Morgan | The Next Platform

  • The Marketing Glory of NVIDIA’s Codenames

    While code names are not rare in the corporate world, more often than not, the names tend to be unimaginative. NVIDIA’s code names, however, are pure marketing glory.

    Take NVIDIA’s high performance computing product roadmap (below) – these are products that use the graphics processing capabilities of NVIDIA’s high-end GPUs and turn them into smaller, cheaper, and more power-efficient supercomputing engines which scientists and researchers can use to crunch numbers. How does NVIDIA describe its future roadmap? It uses the names of famous scientists to describe its technology roadmap: Tesla (the great American electrical engineer who helped bring us AC power), Fermi (“the father of the Atomic Bomb”), Kepler (one of the first astronomers to apply physics to astronomy), and Maxwell (the physicist who helped show that electrical, magnetic, and optical phenomena were all linked).

    Source: Rage3D

    Who wouldn’t want to do some “high power” research (pun intended) with Maxwell? 

    But, what really takes the cake for me are the codenames NVIDIA uses for its smartphone/tablet chips: its Tegra line of products. Instead of scientists, he uses, well, comic book characters. For release at the end of this year? Kal-El, or for the uninitiated, that’s the alien name for Superman. After that? Wayne, as in the alter ego for Batman. Then, Logan, as in the name for the X-men Wolverine. And then Stark, as in the alter ego for Iron Man.

    Source: NVIDIA

    Everybody wants a little Iron Man in their tablet.