AMD: We Stand Ready to Make Arm Chips

Renoir
(Image credit: Fritzchens Fritz)

AMD's CFO Devinder Kumar recently commented that AMD stands ready to manufacture Arm chips if needed, noting that the company's customers want to work with AMD on Arm-based solutions. Kumar's remarks came during last week's Deutsche Bank Technology Conference, building on comments from AMD CEO Lisa Su earlier in the year that underscored the company's willingness to create custom silicon solutions for its customers, be they based on x86 or Arm architectures. Intel also intends to produce Arm and RISC-V chips, too, meaning that the rise of non-x86 architectures will be partially fueled by the stewards of the dominant x86 ecosystem.

"But I'll tell you from my standpoint, when you look at compute solutions, whether it's x86 or ARM or even other areas, that is an area for our focus on investment for us," AMD CFO Devinder Kumar responded to a question about the company's view of competing Arm chips. "We know compute really well. Even ARM, as you referenced, we have a very good relationship with ARM. And we understand that our customers want to work with us with that particular product to deliver the solutions. We stand ready to go ahead and do that even though it's not x86, although we believe x86 is a dominant strength in that area."

Kumar's comments don't give us any concrete indication that the company has specific Arm projects underway. Still, they do confirm that the company's customers are interested in purchasing Arm-based processors from AMD, and that the company is willing to fulfill that need.

AMD could deliver Arm chips through several possible go-to-market avenues, such as through standard off-the-shelf data center or desktop PC processor lineups as an AMD-branded solution, or through its semi-custom business, wherein it would design Arm processors to meet certain customer criteria. Given the trend we've seen of an increasing number of large hyperscalers developing their own chips, like AWS with Graviton, it's reasonable to expect that AMD could attract plenty of interest for Arm designs through its semi-custom business.

AMD CEO Lisa Su set the tone earlier this year at the JPMorgan Global Technology event, saying, "I think AMD has a lot of experience with the ARM architecture. We have done quite a bit of design in our history with ARM as well. We actually consider ARM as a partner in many respects."

"From an AMD standpoint, we consider ourselves sort of the high-performance computing solution working with our customers, and that that is certainly the way we look at this. And if it means ARM for certain customers, we would certainly consider something in that realm as well," Su explained. 

AMD is an Arm licensee and has quite a bit of experience with Arm architectures, dating back to its K12 architecture that never came to market as planned back in 2017. Fast forward to today, and AMD does ship Arm cores, but they come as small microcontrollers for relatively simple tasks, like the company's in-built Platform Security Processors (PSP) that perform security functions to harden the company's CPUs. Those chips pale in comparison to the company's high-performance Ryzen 5000 and EPYC Milan lineups.

AMD is willing to diversify to remain competitive in all possible workloads, and it has an expansive IP canvas to paint on. For instance, AMD uses its Infinity Fabric to tie numerous elements, such as chiplets, GPU cores, and various memories, into one cohesive chip, and that same design methodology could be applied to Arm designs. Additionally, AMD's pending Xilinx acquisition could provide plenty of avenues for FPGA additives.

It's easy to imagine AMD creating powerful data center-class Arm chips for its clients, but even though Arm has reached record market share in PCs, it doesn't seem likely that we'll see Arm-powered PC chips from AMD any time soon. 

Naturally, the fog of Nvidia's pending $45 billion Arm acquisition hangs thick over any conversation about Arm, but that endeavor has hit more than a few snags and might not pass regulatory muster. AMD and Nvidia haven't always been the best of chums, but Nvidia has pledged to honor Arm's commitments and to continue licensing the architecture, even to its rivals, if the deal does go through. That could shape up for an odd encounter in the data center, with AMD-designed Arm chips facing off with Nvidia's Arm-based Grace CPUs.

It could get even stranger, though. Intel, the other half of what is effectively an x86 duopoly, has pledged to produce chips for its foundry clients based on just about any architecture through its Intel Foundry Services, including Arm and RISC-V. Intel may take things a step further, though, and build its own branded RISC-V chips after its rumored $2 billion acquisition attempt of RISC-V chipmaker SiFive completes, proving that the coming years could completely redefine the semiconductor market. 

We shouldn't expect the x86 architecture to fade into the shadows, far from it, but there will soon be many more options on the market, and some of those will obviously come from unexpected places. Meanwhile, AMD stands ready to punch out Arm-based chips for its customers, so it is probably more of a question of when it will happen rather than if.

Paul Alcorn
Managing Editor: News and Emerging Tech

Paul Alcorn is the Managing Editor: News and Emerging Tech for Tom's Hardware US. He also writes news and reviews on CPUs, storage, and enterprise hardware.

  • digitalgriffin
    Admin said:
    AMD's CFO Davinder Kumar says the company stands ready to produce Arm processors for its customers.

    AMD: We Stand Ready To Make Arm Chips : Read more

    Curious

    ARM needs to really stand an individual company. With cloud services moving to ARM for efficiency reasons it's vitally important that NVIDIA doesn't control this market.

    While NVIDIA claims they will be neutral, there's no mention of licensing fees for competitors.

    This is akin to Facebook claiming they would not force customers of Occulus into the Facebook eco system. Well they outright lied several years after the acquisition, by forcing Facebook accounts to be linked.

    Nvidia also completely dropped support for PhysX hardware after they acquired the company, leaving owners out in the cold.

    You can't trust a company once they have the goods.
    Reply
  • zodiacfml
    this more interesting than I thought. i was expecting Arm based products already in the works considerign Apple M1 but they're pretty much divided considering large gaming consoles and some handhelds are x86. giving this minutes of thought, I think AMD should go for it because the future will be graphics and Apple doesn't have that. Apple might want to outsource their Arm based SoC in the future to AMD to get better graphics. once AMD gets Apple as a customer, consoles be converted to Arm. all this is assuming Nvidia doesn't get Arm which is probably the reason Nvidia wants Arm so badly.
    Reply
  • ezst036
    Hopefully Nvidia fails to acquire ARM.
    Reply
  • spongiemaster
    digitalgriffin said:
    Nvidia also completely dropped support for PhysX hardware after they acquired the company, leaving owners out in the cold.

    You can't trust a company once they have the goods.
    I'm sure that disappointed both people that owned one of those PPU's. Did you know anyone that had one? I didn't, which isn't something I can say often. They dropped support 3 years after the acquisition which was 6 years after the hardware was initially released. Pretty much any Cuda capable GPU outperformed the PPU at that point. Nvidia made significant improvements PhysX then released the source code in 2015 before open sourcing it in 2018. They just gave it all away for free. Can't trust those bastards at Nvidia.
    Reply
  • ddcservices
    What many do not realize, is that with the whole "chiplet" design, AMD could throw some ARM cores into a processor and suddenly, native phone/tablet app execution in a x86-64 computer. I know that being able to run some Android apps will be welcome when it comes to using a number of devices that require a phone/tablet to set them up because they use apps for those, and don't have a true program for a computer to do the same thing.
    Reply
  • ddcservices
    zodiacfml said:
    this more interesting than I thought. i was expecting Arm based products already in the works considerign Apple M1 but they're pretty much divided considering large gaming consoles and some handhelds are x86. giving this minutes of thought, I think AMD should go for it because the future will be graphics and Apple doesn't have that. Apple might want to outsource their Arm based SoC in the future to AMD to get better graphics. once AMD gets Apple as a customer, consoles be converted to Arm. all this is assuming Nvidia doesn't get Arm which is probably the reason Nvidia wants Arm so badly.
    The power of the M1 is that it has silicon dedicated to specific tasks, and that is why it is excellent in many ways. Picture dedicated chips for this and that feature, merge them all into "the CPU", and there you go.
    Reply
  • artk2219
    spongiemaster said:
    I'm sure that disappointed both people that owned one of those PPU's. Did you know anyone that had one? I didn't, which isn't something I can say often. They dropped support 3 years after the acquisition which was 6 years after the hardware was initially released. Pretty much any Cuda capable GPU outperformed the PPU at that point. Nvidia made significant improvements PhysX then released the source code in 2015 before open sourcing it in 2018. They just gave it all away for free. Can't trust those bastards at Nvidia.

    You can't trust them, and they "gave it all away for free" 10 years later, and basically killed any future development by absorbing them and close sourcing the code for years when it could have been used by other companies. You're right though, not many people had those cards, but i knew a few people that did pick them up. It was also stupidly annoying that even if you did buy an nvidia gpu just for the physx capabilities, they made you jump through hoops to use it as a coprocessor, i did that with a GTS 250 i picked up for 50 bucks and my Radeon 6950 for a couple of years, just about every driver update broke it.
    Reply
  • hotaru.hino
    ddcservices said:
    What many do not realize, is that with the whole "chiplet" design, AMD could throw some ARM cores into a processor and suddenly, native phone/tablet app execution in a x86-64 computer. I know that being able to run some Android apps will be welcome when it comes to using a number of devices that require a phone/tablet to set them up because they use apps for those, and don't have a true program for a computer to do the same thing.
    Android apps don't require an ARM processor to run. The real problem with running Android apps natively outside of Android is ART/Dalvik hasn't been ported to any other OS as far as I know.

    Also Android has run on x86 devices such as the first Asus Zenfone and Lenovo's K80.
    Reply
  • spongiemaster
    artk2219 said:
    You can't trust them, and they "gave it all away for free" 10 years later, and basically killed any future development by absorbing them and close sourcing the code for years when it could have been used by other companies.

    They didn't close source the code. It was closed sourced when they bought it. They were under no obligation to open source technologies they purchased. How often do companies do that? Why is Nvidia held to a different standard than other companies? It's not really debatable that PhysX was never popular with developers or gamers at any point in its history, so Nvidia didn't kill anything. Before Nvidia, PhysX required the PPU you mentioned to work. That more than anything killed any chance of it being popular. Nvidia removing that requirement increased adoption of the technology, not the other way around. PhysX is still alive and well today as it and Havoc are basically the only physics models used in gaming.
    Reply
  • hotaru.hino
    Having a dedicated PPU card was going to be dead in the water the moment NVIDIA took over anyway. Why bother supporting another piece of hardware when the stuff you already make can handle it just fine? And history can tell you that in most computer systems, if some whizz bang feature is optional, can't be shoehorned into what you already have, or doesn't provide an exponentially better result, chances are it's not going to last long.

    Also, most developers run PhysX on the CPU. The number of games that actually use the GPU or PPU to accelerate it is pretty small. Keep in mind that a lot of PC games also have a console version, so they can't make PhysX run on the GPU when there isn't a compatible one to begin with. If anything, the hardware accelerated aspects of PhysX are just to handle cosmetic things most of the time.
    Reply