Binary Blobs Continue To Prove Challenging For POWER10 Plus Very Expensive Motherboards

[phoronix]

Phoronix: Binary Blobs Continue To Prove Challenging For POWER10 Plus Very Expensive Motherboards

While POWER CPUs have generally been well received by the free software community for being open-source friendly especially with the OpenPOWER Foundation, IBM's latest-generation POWER10 processors are continuing to be an upset...

https://www.phoronix.com/scan.php?page=news_item&px=POWER10-Blobs-Raptor

[ServerGarbage]

I like the way Raptor Computing behaves with the community and customers. Sad i can't afford any of their systems because i would enjoy using it and learning about power architecture. But I understand they're the first ones that would like to make their products cheaper and in anyone's budget.

[tildearrow]

I don't understand why so much firmware these days...

[milkylainen]

DDR blobs are probably the usual Synopsys DDR IP block bullshittery.
But that's just a guess.
Same for a lot of ARM DDR4 constructions...

[Cape]

I don't understand why so much firmware these days...

Capitalism

[Viki Ai]

I don't understand why so much firmware these days...

Basically (and over-simplifying), a lot of high-speed interfaces (like the mentioned PCIe and DDR) are so highly-engineered these days that: 1) specialist design houses create them as IP-modules that other chip makers (IBM, AMD, etc.) buy and plug in to the edges of their CPUs, either because it is cheaper, or the CPU designer simply doesn't have the in-house specialist knowledge to do it (well) even if they wanted to; 2) these highly-engineered IP blocks are often effectively very-specialised-function micro-controllers in their own right, running their own custom code on their own proprietary instruction sets, either for bringing up the interface and attached devices at cold-boot, or even for ongoing running and real-time tuning of the buses; 3) the only choice the CPU maker often has over the 'openness' of the specialised code of these externally-sourced IP blocks is to choose to use a different IP-block, assuming such is even available, and is competitive on either/both cost or/and performance metrics; 4) These IP suppliers are so many levels removed from the end user that they may as well be operating from another local solar system - if you think it it is hard convincing the average general CPU maker to be properly open, chip-peripheral-IP design houses are on a whole new level of recalcitrance again!

In this case, IBM might have been able to buy or in-house blob-less bus IP blocks for their Power10 chips (they presumably did for power 8 and 9), but I'm guessing there would have been a price, or more likely considering their commercial target demographic, a noticable performance penalty for doing so, compared to the blobby IP blocks they went with. Disappointing from an OSS perspective, of course, but its value as a commercial decision will depend on how the bulk of their customers weigh openness vs performance: I may be in the market for an entry-level Power system in the next year,and this pretty much ensures I would be going with a Power9 since for me open>performance. But I'm am a single-unit customer, so that isn't even a blip on IBM's balance sheets.

[chocolate]

If they decide to go forward with a Power10 system that relies on some blobs, let's hope existing Power9 systems remain available for a long time as well. There are no other performant and reasonably trusted (also, reasonably efficient) machines on the market, making the current Talos II and Blackbird systems unique.
Raptor's website is very nice as well, doesn't seem to contain any third-party tracking JavaScript such as G. Analytics, an increasingly rare thing on this cursed web. It's the kind of thing that makes the company's statements consistent with their actions.

[Termy]

With that cost projections, blobs would be an absolute no-go for me.
Being open/libre is the USP of raptor, without that it would be really hard to justify the downsides compared to standard x86 systems...

[rjzak]

There's the recent article about Ubuntu dropping POWER8. https://www.phoronix.com/scan.php?pa....04-LTS-POWER9
At what point do distros start dropping POWER9? How much life is left in P9? What's Raptor's next step when P9 is deemed too old, or can P9 (as part of an OpenPOWER thing) be improved upon?

[phoron]

So the poll is only on twitter ? Odd. Well I'm not going to use twitter just to answer Raptor. In any case they used to hang around here. I'm 0 interested in power10 offerings or anything that's not 100% free of blobs. The performance of power9 is plenty for me. If they could somehow lower prices much the better (no idea how, though, and it's not so important as freedom, but still) I'd concentrate in production and sales channels and staff, even if they didn't engineer anything new. I think they're working on new BMCs, that's OK, but I've been waiting for the webshop of Vikings to come online for ages, and I wonder why should it be so hard to buy.

If Ubuntu ever drops Power9 we'll have another distro, and if those also drop it then we'll have to maintain our own, but at least it'll be always possible because it's all free. It'll be more likely if Raptor manages to sell more units of Power9 systems, than if they now divert into blob-ridden Power10 and divide their base. Never mind were IBM is taking Power10. Invest in production, distribution and support, and try to become bigger and appear a little in the IBM radar to hopefully influence future decisions. Grow while Power9 is available, and keep an eye on any other possible 100% free platform in the future.