IBM’s open sourcing of key parts of its Power architecture is either big news or a big nothing, depending on how much you trust in open source.
OK, I admit it: I’m confused. In late August 2019, IBM open sourced its Power Instruction Set Architecture (ISA), thereby making it easier (and cheaper) to build systems with IBM’s Power architecture.
The announcement seemed to garner mostly tepid reviews, but merited at least one caustic response by Jim Turley, “Remember PowerPC? Nah, me neither. But now it can be yours. For free. Srsly.” Balanced against the indifference or criticism, however, is Jason Perlow’s optimistic assessment on sister site ZDNet that “All this [Power] intellectual property for creating reference designs, which includes the patents themselves, is going to be royalty-free,” which could spark mass euphoria for IBM, Russia, IoT developers, and more while inciting mass hysteria for Intel, Qualcomm, and other competitors.
So is it big news or a big nothing? Spoiler alert: You won’t find any easy answers (just lots of opinions) on HackerNews. Double spoiler alert: The real answer probably depends on how much credence you give to the power of open source, and whether you think an open source license is enough to counteract decades of market momentum swinging against Power.
So, yes, I’m telling you there’s a chance.
But first, the bad news
Turley has been tracking movements in processors for a long time and isn’t easily impressed. By his reckoning, this IBM move should be weighed against the bulk of its Power history.
IBM early on went the licensing route with Power, but failed to get many licensees that ran at significant scale. Partly this may have been because IBM was allegedly a poor partner: “A PowerPC license was far more expensive than one for MIPS or ARM, and IBM was reputed to be a tough partner to deal with.”
Power was supposed to be an Intel-killer but “the PowerPC market bifurcated, with high-end silicon driving exotic IBM iron and low-end chips in various embedded systems, but with nothing much in between.” IBM notched a few high-profile design wins (e.g., PlayStation 3), but failed to make a dent in the PC market.
Even Power’s once-dominant position in high-end supercomputers has fallen, with just 13 of the world’s 500 fastest supercomputers running Power (down from roughly 200 in its heyday).
IBM arguably didn’t have much choice: “The company could continue to demand seven- and eight-figure licensing fees, or it could throw in the towel and hope that PowerPC catches its second (third?) wind among SoC developers. Besides, with [open source] RISC-V garnering so much attention of late, there wasn’t much time left before PowerPC completely missed the boat.”
Even with so much on the negative side of Turley’s tally, he still offers some hope:
There’s little downside to offering the ISA for free, and some potential upside. If PowerPC makes even a little bit of headway in the form of hardware users and software developers, that’s good for everybody concerned. A rising tide lifting all boats, and all that. PowerPC could potentially become the next RISC-V (or ARM… or 8051… or PDP/11…) and grow into a popular and well supported product family. It’s unlikely to ever compete with desktop CPUs like originally intended, but it’s a fine embedded processor and one with some history and provenance, and an impressive family tree. That’s more than most free CPUs can claim.
In other words, there’s hope. And with open source, perhaps more hope than Turley credits. He thinks, after all, that “You don’t drop your price to zero when business is good,” suggesting that open sourcing Power indicates weakness. It does, on one hand, but as I’ve long argued, open source is an effective tool for underdogs. Microsoft, for example, has used it effectively in its successful bid for renewed relevance. Who’s to say it can’t do the same for IBM’s chip ambitions?
And now, the good news
One person who emphatically argues that open source Power is a Very Shrewd Move is Perlow. While Perlow acknowledged a lack of media interest in IBM’s move, he doubled down on its importance: “[T]his tech is being overlooked by a lot of industry media but this is being examined very closely by the technology giants to move into the next phase of processor architecture for many solutions including 5G, networking, cloud computing and mobile.” Thus far no technology giants have publicly expressed that interest, but Perlow’s a smart, connected guy—let’s assume there’s interest.
Where would we be most likely to find that interest? According to Perlow, the list is long, starting but not exclusive to IBM, including:
Chinese companies. With the Trump administration cracking down on exports to China, open sourcing Power makes it easier for companies like Huawei to “now build 5G infrastructure, network switches, and IoT components using Power architecture.”
All IoT developers. “They now have a better choice than ARM or Intel. Microsoft Xbox could become PowerPC again—so could PlayStation and everyone making Wi-Fi routers, residential gateways, Alexa smart speakers, smart anything, etc. In other words, everything that is ‘made in China.'” (Note: It’s not clear why Power is a better choice than ARM or Intel, or why companies like Sony would return to a platform they clearly opted to abandon, but maybe the price point would do that.)
Apple. The company has been taking more and more chip design in-house. Now with Power, “Apple gets to completely throw Intel and ARM out in terms of required licensing and could own all its DNA again.”
And more. Likely losers, according to Perlow, would be Intel, Qualcomm, and every other major chip company that suddenly would find a freely licensed rival in the market. Other open source options exist, but none with IBM’s backing. Open source plus IBM’s backing could be just the shove Power has needed to return to relevance.
Of course, that’s a massive “if.” Power, for example, has a lot of work to do before it’s ready for mobile computing, given its poor record with power consumption. There’s a reason Intel spent a decade trying to get clock management and power consumption correct, as one commenter noted to Perlow. If it were easy, Intel would have fixed it immediately. Such things are hard, and won’t be solved by the Power community overnight.
But where there’s open source, there’s at least the hope that a community will form to shape the future of Power. Given enough self-interest, and an open source license, all technology problems are solvable. The future may not be as rosy as Perlow expects, or as dire as Turley might imagine, but a healthy middle might be more than enough to shake up the chip industry and make for competition that helps consumers.