Water Cooler > General OpenPOWER Discussion
News?
pocock:
I think some of this is out of Raptor's control
There is widespread discussion elsewhere about IBM wrecking things in Red Hat, CentOS and Fedora, POWER10 not being 100% open and Google has been wrecking things in Debian. Coincidentally, IBM and Google are the two key players in OpenPOWER. I don't want to gripe about it, only to express what it may mean for everybody who already bought in to this platform.
They had an OpenPOWER room at FOSDEM, they accepted my talk and then they canceled it, this is incredibly nasty stuff. We go to FOSDEM every year as volunteers. It was online this year, the Google people don't have to look the volunteers in the eye because it is all remote so they brought this dirty behavior into FOSDEM.
On a platform like POWER where there are not so many developers to begin with you can't do stuff like that. If people see one volunteer being hurt then ten other volunteers will avoid the platform. People don't care who is right or wrong, when they get the feeling of conflict, they decide to give it a pass or come back and look again in 12 months. That makes it much harder for Raptor to sell to volunteer developers.
Debian has started spending a lot of money on a lawsuit recently. I don't want to hijack this thread with details of the lawsuit, you can download the dossier if you really want to, the only thing I want to emphasize is that 17 developers resigned. 15 of those developers were removed from the keyring on 7 June. Most of them will probably not make any statement about their resignation, most people don't like to take sides but there is a cause and effect: when conflict starts, some people quit.
This is all out of Raptor's control of course but it is a risk to the platform.
Personally, having contributed to patching packages for POWER9, I'm optimistic that lessons learnt from porting work on this platform will be easily transposed to ARM64 and RISC V. Raptor has made the motherboards to a high standard and most people seem to agree they will last a long time, hopefully long enough for RISC V to appear on our desktops too.
SiteAdmin:
We would respectfully ask that the discussion here remain on topic as much as possible. While RISC-V and ARM remain attractive from a cost perspective, we do want to remind everyone involved that all current ASIC implementations of RISC-V -- and all reasonably powerful ARM devices -- are actually more closed and require more blobs than POWER10 does. While at least right now we won't build POWER10 systems due to the blobs resident in those CPUs and associated interface devices, it should also be apparent that current RISC-V and ARM implementations are a step even further in the wrong direction.
We believe there is a future for the POWER platform and its truly open, standardized ISA -- especially with projects like LibreSoC already underway, and internal projects (some of which are already yielding public results in the Kestrel OpenPOWER soft BMC project), we would advise against switching to blob-filled, proprietary platforms just to "hit back" at IBM.
For clarity: we do not agree with where IBM is going with the POWER platform with POWER10. We also do not believe that IBM has the ability to restrict what can be done with the OpenPOWER ecosystem now or in the future. They are (were) a major player, yes, but at the end of the day they are unable to call back or restrict what other companies and individuals are able to do -- and are already starting to do -- with the ISA. The best way to keep things open in the short term, when also looking at long-term effects, would be to simply standardize the open software ecosystem on the POWER9 ISA vs. the POWER10 ISA. At least for now, as this would further encourage the use of open chips vs. chips that require binary components, since POWER9 would remain the baseline ISA compatibility level for said new devices.
Hopefully this will spark further on-topic discussion. We're not going anywhere, and we refuse to build and ship blob-infested hardware, so RISC-V remains a no-go for us. Hopefully Debian will also continue to show this blob-free focus in the future, as Raptor and Debian have been fairly well aligned on this topic for many years.
...typing this from a Talos II workstation with Debian and Ungoogled Chromium installed.
pocock:
--- Quote from: pocock on June 11, 2022, 04:49:41 pm ---
Raptor has made the motherboards to a high standard and most people seem to agree they will last a long time, hopefully long enough for RISC V to appear on our desktops too.
--- End quote ---
If you look at that comment, there is nothing in it to suggest an open RISC V will arrive imminently. In fact, given that nobody is holding their breath for open RISC V, the comment could be seen as deliberately optimistic about Talos II boards running for years to come due to their quality.
--- Quote from: SiteAdmin on June 12, 2022, 01:31:50 am --- we would advise against switching to blob-filled, proprietary platforms just to "hit back" at IBM.
--- End quote ---
There was nothing in the post to suggest people "hit back" at IBM
It is just a reality check, that is all. We can see the way people started building Alma Linux and other continuations of CentOS. That is not to "hit back" at Red Hat or IBM, they are building that because they need it and it was the right thing to do.
Given the communities are not so big, these situations do lead to people duplicating effort (e.g. Rocky Linux) and other inefficiencies that may have been avoidable. Rather than having 3 or 4 forks of CentOS, developers doing exactly the same thing in parallel, it would be interesting if some of that developer effort went to POWER9 porting.
Speaking as a developer, I fully respect the right of any developer, whether it is a lone volunteer or a giant company like IBM to change their direction. It then raises the question about how other people work around that.
For example, do you see the IBM POWER9 chips continuing to be available in sufficient quantities for the Raptor ecosystem?
Do you see any other manufacturer coming along with a 100% open chip to continue from POWER9?
--- Quote from: SiteAdmin on June 12, 2022, 01:31:50 am ---The best way to keep things open in the short term, when also looking at long-term effects, would be to simply standardize the open software ecosystem on the POWER9 ISA vs. the POWER10 ISA.
--- End quote ---
Could you define that? For example, which key software products need to commit to that statement? In practice, how many of the developers on that part of the open source ecosystem are employed by IBM / Red Hat and could that take them down a POWER10 path?
--- Quote from: SiteAdmin on June 12, 2022, 01:31:50 am ---...typing this from a Talos II workstation with Debian and Ungoogled Chromium installed
--- End quote ---
Does this mean the Chromium libs are now available for building other things? If so, I might have another look at some projects that were pending for me to port.
SiteAdmin:
--- Quote from: pocock on June 12, 2022, 06:47:45 am ---If you look at that comment, there is nothing in it to suggest an open RISC V will arrive imminently. In fact, given that nobody is holding their breath for open RISC V, the comment could be seen as deliberately optimistic about Talos II boards running for years to come due to their quality.
--- End quote ---
Out main point was that the first part could be taken as RISC-V being the inevitable long-term succesor to POWER, when we're fairly certain that would not be the case, especially in the blob-free / owner-controlled sphere. Regardless, we do appreciate the note on quality -- that's an area we've put a lot of time, effort, and resources into.
--- Quote from: pocock on June 12, 2022, 06:47:45 am ---There was nothing in the post to suggest people "hit back" at IBM
It is just a reality check, that is all. We can see the way people started building Alma Linux and other continuations of CentOS. That is not to "hit back" at Red Hat or IBM, they are building that because they need it and it was the right thing to do.
--- End quote ---
This wasn't directed so much to you as it was to the general developer community. We've seen some knee-jerk reactions when bad news comes out, including some people that basically just gave up and said "I'm doing x86 only, ME and vendor blobs are inevitable".
--- Quote from: pocock on June 12, 2022, 06:47:45 am ---Speaking as a developer, I fully respect the right of any developer, whether it is a lone volunteer or a giant company like IBM to change their direction. It then raises the question about how other people work around that.
--- End quote ---
Indeed. For us, that's why we have such a stong requirement for owner control, including opening and standardization of the ISA itself. At least this removes all of the artifical barriers that would otherwise be thrown up. For example, Intel and AMD want to make gaming and streaming consumer systems -- they need the ME and PSP to do that, but since the ISA is not open that now steps on everyone else's rights to use the systems in other ways without effectively allowing Intel/AMD access to the data on them, because no one else can (legally) make compatible CPUs that also meet the blob-free / owner-controlled requirements.
--- Quote from: pocock on June 12, 2022, 06:47:45 am ---For example, do you see the IBM POWER9 chips continuing to be available in sufficient quantities for the Raptor ecosystem?
--- End quote ---
Yes, there are many, many years worth of the CPUs available.
--- Quote from: pocock on June 12, 2022, 06:47:45 am ---Do you see any other manufacturer coming along with a 100% open chip to continue from POWER9?
--- End quote ---
Short answer: yes. Long answer: [redacted] skunkworks [redacted].
They're still planned, just delayed from COVID and the recent macroeconomic meltdown. LibreSoC is also working in the background and contining to make progress, we're not looking at a single pathway to the required silicon here, just that all pathways produce 100% blob free ppc64le ISA 2.07+ compatible devices.
--- Quote from: pocock on June 12, 2022, 06:47:45 am ---Could you define that? For example, which key software products need to commit to that statement? In practice, how many of the developers on that part of the open source ecosystem are employed by IBM / Red Hat and could that take them down a POWER10 path?
--- End quote ---
Great question! We'd say just having the main binary distros (Debian, Fedora, SuSE?) ensure they keep their package archives compatible with POWER9 vs. requiring POWER10 is enough. Then the only other group that needs to be on-board is the JIT writers -- don't use POWER10-specific instructions. To be honest, this is going to happen naturally anyway, since no one that we know of is making POWER10 systems that are would be desktop-class or exist outside of a cloud environment, let alone open systems. This all solely due to IBM's poor decision to close parts of the POWER10 platform, it's quite sad.
--- Quote from: pocock on June 12, 2022, 06:47:45 am ---Does this mean the Chromium libs are now available for building other things? If so, I might have another look at some projects that were pending for me to port.
--- End quote ---
Yes, we've been investing significant resources in keeping a POWER build of Chromium available, tracking upstream Debian including the security patches but also including the Ungoogled patchset. Our thought was that since Google absolutely refuses to upstream the POWER port (our suspicion is because it is an owner controlled, blob-free platform, this would interfere with the long-term goal of mandated low-level tracking for advertising), why enable any of the Google services at all?
The patches are also fairly easy to unroll in debian/series if you did want to just build a stock Debian Chromium for POWER. In any case, we'd love to see the base patches integrated into the official Debian builds, maybe you could help us get some traction there?
Builds are currently here:
https://quickbuild.io/~raptor-engineering-public/+archive/ubuntu/chromium/+packages
dget -x / apt-get source work against that repository.
Google's unfortunate stance on upstreaming is basically "we won't do it and we won't tell you why".
https://chromium-review.googlesource.com/c/chromium/src/+/3133001
Applying Occam's Razor against that yields a few concerning options, including the potential existence of a third party agreement that would prohibit it. As such, the remaining patches need to be carried downstream in various distros; they're already carried downstream for Gentoo etc., see e.g. https://bugs.gentoo.org/669748 .
AdamJoseph:
--- Quote from: SiteAdmin on June 12, 2022, 02:08:21 pm ---
--- Quote from: pocock on June 12, 2022, 06:47:45 am ---For example, do you see the IBM POWER9 chips continuing to be available in sufficient quantities for the Raptor ecosystem?
--- End quote ---
Yes, there are many, many years worth of the CPUs available.
--- End quote ---
In this respect POWER9 is far more secure than POWER10 or any of its competitors.
Three years ago 14LPP wafers were running in New York (East Fishkill, I think, but maybe Malta) and production was starting up in Dresden as a second site. There are probably additional sites by now; my attention has been focused elsewhere lately. In any case, just those two locations alone are a whole lot more geopolitical security than the rest of the industry has. 14LPP isn't the same as IBM's special eDRAM-cap variant of 14nm, but it would be very unusual for it to not be running from the same sites as the less-flashy merchant foundry wafers.
Don't worry about supply of chips. Supply of blackbird motherboards, on the other hand...
--- Quote from: SiteAdmin on June 12, 2022, 02:08:21 pm ---
--- Quote from: pocock on June 12, 2022, 06:47:45 am ---Do you see any other manufacturer coming along with a 100% open chip to continue from POWER9?
--- End quote ---
Short answer: yes. Long answer: [redacted] skunkworks [redacted].
LibreSoC is also working in the background and contining to make progress,
--- End quote ---
I've been having a hard time finding information about their 180nm tapeout. The chips came back almost a year ago; are any data available? How did things turn out? Surely there's been time for several iterations with a FIB by now...
--- Quote from: SiteAdmin on June 12, 2022, 02:08:21 pm ---we're not looking at a single pathway to the required silicon here, just that all pathways produce 100% blob free ppc64le ISA 2.07+ compatible devices.
--- End quote ---
Unfortunately getting the foundries to care about anything other than your organization's market cap is a gigantic obstacle these days. Design quality and engineering talent are not taken into consideration anymore.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version