| programmerjake | lkcl, reverse dns is still broken (idk if you tried to fix it again) | 02:20 |
|---|---|---|
| emanuel | Hello, I wanted to ask in which countries Red Semiconductor woud produce the LibreSOC | 03:21 |
| tplaten | I tried to run a verilator simulation over night, in my microwatt branch, but as expected, I did not get any uart output. I have heard that there is a bug that needs to be fixed in microwatt related to wishbone addresses. | 08:03 |
| lkcl | emanuel: you'll see this in the archives - answer is, world-wide distribution and we will use available Foundries. | 10:44 |
| tplaten | How do I stop microwatt-verilator after a short time amount? after running about one minute, I got a broken vcd with this kind of error: | 14:44 |
| tplaten | Near byte 14323, $VAR parse error encountered with 'TOP.toplevel.soc0.processor_external.processor.\$auto$verilog_backendcc' | 14:44 |
| tplaten | I saw that there is TERMINATE_AT_COUNTDOWN, im using that | 15:07 |
| tplaten | So even if I terminate microwatt-verilator after n cycles, I get that error from gtkwave. | 15:30 |
| emanuel | lkcl: Since it is very costly to get the production started I concluded it would be difficult to move it someplace else. Is this correct? I do not think that there could be many chips manufactured in countries where work conditions are better than let's say in China or India, as that would be way too expensive. However, I am not sure if my thoughts are correct. If they are, on the other hand, wouldn't such a large dependency on | 15:38 |
| emanuel | (often undemocratic) states like China be problematic? I think you know way more about this than me, therefore I'm asking what you think about this. | 15:38 |
| tplaten | With libre-silicon one could fabricate in any country, at least theoretically. | 15:40 |
| emanuel | I know, but quite some time ago I was told on this channel that getting the production started is extremely expensive and that it is planned for Red Semiconductor to produce the Libre-SOC so I would like to know where Red Semiconductor would be planned to do the crucial steps of manufacturing the Libre-SOC (those that are so expensive to get started) and how problematic you think | 15:56 |
| emanuel | the production in undemocratic countries could be. | 15:56 |
| tplaten | I don't have any information about where Red Semiconductor could produce. Maybe in the EU, or in Silicon Saxony. | 15:59 |
| tplaten | I had a look at my version of microwatt-verilator, it seems that instruction fetch fails. | 16:00 |
| emanuel | Wouldn't it be even more expensive to produce in the EU, are there any high-speed processors produced in the EU? How would this be possible to sell at an acceptable price? I am trying to understand. | 16:03 |
| tplaten | I know that Samsung produces flash-memory in South Korea, but I am unable to compare South Korea with the EU. Same for the US and Taiwan. | 16:07 |
| tplaten | When I run microwatt-verilator from the verilator_trace branch I get: pc 700 insn 48000000 msr 8000000000000001, pc never changes | 16:12 |
| tplaten | when I run my version top->nia_req never becomes true | 16:12 |
| tplaten | In both cases the vcd files are broken. | 16:13 |
| emanuel | So we both don't know if there are any high-speed processors produced in other countries than China, India and South Korea. Let's leave the question here like this, maybe someone who knows the answer will see it. I will check back later, thanks for answering and sorry I consumed your time. | 16:14 |
| tplaten | For me that is not a problem, as that is my interest too. | 16:15 |
| tplaten | most likely I have a reset address mismatch, but I'm not sure here | 16:23 |
| tplaten | I had forgotten to wire out some signals in fpga/top-generic.vhdl | 17:26 |
| tplaten | Instruction fetch seems to be working | 17:44 |
| tplaten | pc 700 insn 2020200a msr 8000000000000001 | 17:44 |
| tplaten | pc 704 insn 4f4f6f2e msr 8000000000000001 | 17:44 |
| lkcl | emanuel: china is complex. it's a billion people, and it's not controlled by the "government", it's controlled (owned) by the Five Families. | 18:14 |
| lkcl | each Family is responsible for an astounding number of people, basically over a billion divided by five is the same population as the United States. | 18:15 |
| lkcl | bottom line is that everything you've been told through a Western upbringing is pretty much flat-out wrong. | 18:16 |
| lkcl | what we *do* have is a problem not caused by China but by the United States! | 18:16 |
| emanuel | Which problem? | 18:17 |
| lkcl | the *United States* is acting in a deeply unethical and heavy-handed fashion, causing embargoes, creating Trade "wars" and so on, in its usual fashion that involves Oil and ensuring that its grip on world finance is kept by forcing everyone to use the U.S. as the world's Reserve Currency | 18:18 |
| lkcl | RED Semiconductor has to navigate that, where Libre-SOC does not. | 18:19 |
| lkcl | Libre-SOC is a pure Libre/Open venture, funded by NLnet under their mandate "Works for the Public Good". | 18:20 |
| lkcl | and it will stay that way. | 18:20 |
| lkcl | but, obviously, a pure R&D FOSS Project ain't exactly gonna have the USD 16 million dropped into its lap to pay for 7 nm Mask Charges. | 18:21 |
| lkcl | i mean, it'd be really nice: none of us would say "no" :) | 18:21 |
| lkcl | which is where RED Semiconductor Ltd comes in, and it will be a "Practical Delivery Machine" for the production and sale of SoCs which "happen to use Libre-SOC Cores" | 18:22 |
| lkcl | as such it will be subject to "the usual practical matters surrounding fabrication and sale of ASICs" | 18:23 |
| lkcl | basically, RED Semiconductor Lts faces exactly the same problems faced by | 18:24 |
| lkcl | Texas Instruments | 18:24 |
| lkcl | Freescale | 18:24 |
| lkcl | Samsung | 18:24 |
| lkcl | Intel | 18:24 |
| lkcl | AMD | 18:24 |
| lkcl | IBM | 18:24 |
| lkcl | Mediatek | 18:25 |
| lkcl | etc. etc. etc. etc. etc. | 18:25 |
| lkcl | bottom line is, RED Semiconductor will make and sell commercial ASICs, world-wide, just like everyone else. they will happen to be entirely transparent designs, and they will happen to use a Libre-SOC core. | 18:30 |
| lkcl | emanuel: https://www.linkedin.com/posts/marc-sukkar-fciarb-38b272148_petro-dollar-russia-ugcPost-6915972413805527040-njxC?utm_source=linkedin_share&utm_medium=member_desktop_web | 18:35 |
| emanuel | I don't think either of those two countries are doing much better deeds to the world than the other but I definitely prefer to live in a democracy than in a country "ruled by five families", to cite you. I generally would trust western media than other media since western countries are normally democracies where you don't have to fear that you might get arrested for saying something that some high people don't like. On the other | 19:04 |
| emanuel | hand, western media can be just as wrong as anything else since people might not be as well-informed or might lie. This problem is addressed by giving everyone the freedom of speech meaning if someone says something wrong another person will publicly prove it wrong. My view is that Chinese media is (deliberately) spreading an enormous amount of false information about western concepts like democracy with the intention to make pe | 19:04 |
| emanuel | ple believe they where all bad so that people don't try to adopt them because that would mean those "five families" would loose power while the people would would have more to say. I also have a very critical view regarding the amount of surveillance and the social credit system in China as well as how these things are (ab-)used to keep people quiet. I am not some person who just watches the news, I sometimes do my own research | 19:04 |
| emanuel | s well (using tools such as anonymisation networks, looking very close at things and going to less crowded places on the internet) so I think I know a little more than what you might think I know | 19:04 |
| emanuel | Nonetheless there also have been quite some leaks about things going very wrong in the western world. The difference is that here we know and speak about them. | 19:06 |
| emanuel | The leaks that I am talking about are not what Chinese media says about the western world but leaks like those from Edward Snowden. | 19:10 |
| lkcl | have you by chance read any Jack Reacher books, by Lee Child? | 19:11 |
| lkcl | i just read "The Sentinel" (the 35th book) | 19:11 |
| lkcl | it's pretty clear that the author is receiving some very good anonymous intel, "made up" as a "story". | 19:12 |
| lkcl | a bit like how Anne McCaffrey used to co-author books with someone covering a particular Social Justice issue | 19:12 |
| lkcl | the Five Families own literally everything, including the banks. this is not easy for Westerners to comprehend, the sheer scope and scale, of any Family being in power in a country for literally thousands of years. | 19:14 |
| emanuel | What about the CCP? How much power would you say it possess in China? | 19:24 |
| lkcl | honestly i have no idea about the CCP. | 19:27 |
| markos | lkcl, to add my 2c to the discussion, the West also have a couple of families that in total own pretty much the whole world (see Rothchild, Rockefeller, and some other not so known names) | 19:27 |
| lkcl | i have a reasonable handle on the West, and Asia, but virtually none on the CCP. | 19:27 |
| markos | problem is if you start discussing this, you are immediately a conspiracy theorist and outright dismissed | 19:27 |
| lkcl | markos, yeah my slightly-clueless-but-well-meaning friend tells me aaalll about the Rothchilds :) | 19:27 |
| markos | well, there is a saying, "follow the money", do that and the whole world situation becomes pretty clear | 19:28 |
| lkcl | all of these things we have to either be blissfully ignorant of, or filter-out-the-bullshit and stay on focus | 19:28 |
| markos | it's definitely not because of "democracy" that the whole of NATO is against Russia, nor do they care about Ukraine, in fact they put Ukraine in that position knowingly, they knew exactly how Putin would react and they did anyway | 19:29 |
| lkcl | markos: yyep. problem is for most people the amounts involved are so vast that it's beyond their scope to imagine, and consequently they dismiss it (by any means) | 19:30 |
| markos | the US and NATO is spending billions of USD on so-called "think-tanks" that estimate every possible outcome for such scenarios, so there are only 2 possible outcomes: a) they did not predict it which makes them stupid and useless -so dangerous b) they did predict it and let it unfold -also dangerous | 19:31 |
| lkcl | i saw the figures on both the agricultural supplies, fertilizer, and the chemicals used for silicon ASIC manufacture | 19:31 |
| lkcl | my money's on the "Charlie Wilson's War" theory: (c) some Analysts *did* warn them... and the politicians decided to ignore it | 19:32 |
| markos | the problem as usual with all such conflicts, it's always the innocents that pay the price | 19:32 |
| * lkcl sighs | 19:32 | |
| lkcl | well they're all "useless eaters" anyway, right? | 19:32 |
| markos | depends who they eat, in Schwab's case I'd make an exception on cannibalism | 19:33 |
| lkcl | again i'm reminded of that slashdot article, the one about how power literally creates a reduction in empathy | 19:34 |
| lkcl | :) | 19:34 |
| lkcl | https://politics.slashdot.org/story/21/10/03/233256/andrew-yang-suggests-power-may-affects-politicians-brain-neurons | 19:34 |
| markos | "may"? | 19:34 |
| lkcl | This even shows up in brain scans. Sukhvinder Obhi, a neuroscientist at McMaster University in Ontario, recently examined the brain patterns of the powerful and the not so powerful in a transcranial-magnetic-stimulation machine. He found that those with power are impaired in a specific neural process — mirroring — that leads to empathy... | 19:35 |
| lkcl | Perhaps most distressing is that in lab settings the powerful can't address this shortcoming even if told to try. Subjects in one study were told that their mirroring impulse was the issue and to make a conscious effort to relate to the experiences of others. They still couldn't do it. Effort and awareness made no difference in their abilities... | 19:35 |
| emanuel | Now I see what you meant. It is true that money is very influential in everything that happens in the world. However, in a democracy people know their rights better and what these concepts try to do is separating the power across multiple institutions and multiple people. Democracy alone can't solve all issues but it at least requires the powerful ones to first "sell" what to do they want to people so they get their okay. It is | 19:51 |
| emanuel | very unlikely that you will be persecuted and imprisoned for saying something they don't want you to say in the EU or the US while it is much more likely in China. (Exceptions apply for extreme cases: i.e. Snowden) The reason for this is that in western countries people are thought they would be in power since they live in a democracy. This makes them be tougher and one person can rely on the toughness of many others. | 19:51 |
| lkcl | so, therefore, the most obvious "solution" to that is: to get people to fight each other. constantly be divided | 20:01 |
| lkcl | and if you can say, ooo, i dunno, set up 100,000 "fake social media" accounts and/or hack into a few Diebold machines, then the underpinnings of that "democracy" are shot to s*** | 20:02 |
| lkcl | that's why i recommended the Jack Reacher books (35 and 36) | 20:03 |
| lkcl | and yet, all of this is still completely irrelevant to the goal of manufacturing ASICs. | 20:04 |
| lkcl | it's simply... "background" as far as i'm concerned. | 20:04 |
| lkcl | things to "navigate around". | 20:05 |
| lkcl | markos i'd be interested in your thoughts on SVP64 Branches at some point (one tuesday perhaps) https://libre-soc.org/openpower/sv/branches/ | 20:09 |
| lkcl | which is moved to 21:00 UTC, but there's usually 1-2 people on a little earlier by appx 15 mins | 20:10 |
| lkcl | the basic idea is to turn the BO bit-testing of CR Fields into a *vector* of CR Fields to be tested, allowing those bits to be ANDed, ORed, NANDed or NORed together | 20:12 |
| markos | lkcl, yes sure | 20:12 |
| lkcl | the simplest version of why they exist is to be able to skip over a batch of parallel/vector instructions where you know that the predicate mask is all zeros (or all 1s) | 20:14 |
| lkcl | no point wasting CPU cycles running a vector of nops when you *know* that they're all predicated out | 20:15 |
| lkcl | so you need an instruction that says "if all the predicate bits / condition bits are zero (or all 1s) then jump over" | 20:15 |
| lkcl | that's the simple justification | 20:15 |
| lkcl | the more comprehensive story is: there's an astounding 128 possible variants of Vectorised-branch options. | 20:16 |
| lkcl | you'll love it :) | 20:17 |
| emanuel | Democracy is there so people, if they really have to fight, don't do it physically but rather in a less destructive way. But you are right that this has gotten off topic so maybe we can get back to what was the topic before: Are there any high-speed processors produced in other countries than China, India and South Korea and where is RED Semiconductor planning to do the crucial parts of the manufacturing of the Libre-SOC? (You | 20:17 |
| emanuel | said it would be just as with other companies producing chips but you still did not mention any countries.) | 20:17 |
| lkcl | Intel fabs, Global Foundries. Skywater, many others. | 20:18 |
| lkcl | by the time we're ready there will be a glut of Fabs available | 20:18 |
| lkcl | there is absolutely no need - at all - to make any decisions or declare any hard intentions. | 20:18 |
| lkcl | i know you do not believe i have answered: i have. i answered "we will do exactly what every other Fabless Semi Company does: use what is available at the time". | 20:19 |
| lkcl | we will do exactly what IBM does | 20:20 |
| lkcl | we will do exactly what Intel does | 20:20 |
| lkcl | we will do exactly what Freescale does | 20:20 |
| lkcl | we will do exactly what Texas Instruments does | 20:20 |
| markos | emanuel, I'm sorry to disappoint, but we don't have democracy right now, at least the majority of countries don't | 20:20 |
| lkcl | markos, i didn't want to be the one to say it... | 20:21 |
| emanuel | I see, it is not decided yet as no one knows how the market will evolve. Thanks! | 20:21 |
| lkcl | pretty much, yes. | 20:21 |
| lkcl | {Insert Stupid Country} might decide to {Insert Stupid Action} that destroys, limits or affects ability of {Insert Company Here} to manufacture ASICs at {Insert Foundry Here} | 20:22 |
| lkcl | the only reason that the United States has not attacked Europe and terminated Licensing Access to Cadence and Synopsis VLSI Tools is because ASML, the world's *ONLY* supplier of top-end Steppers, is in Europe (Netherlands) | 20:23 |
| lkcl | fortunately, Mentor is now owned by Siemens (Germany) but that still isn't a guarantee | 20:24 |
| markos | lkcl, would the analogy to Arm be relevant? | 20:24 |
| lkcl | the entire situation's a goddamn f****g mess | 20:24 |
| lkcl | ARM is a Fabless Semi Design House | 20:24 |
| lkcl | they license world-wide (thousands of licensees) and do not manufacture, themselves | 20:25 |
| lkcl | ARM has other problems though. | 20:25 |
| emanuel | markos: This [21:20] is true but I am not sure we should continue discussing it here as it is off topic. | 20:26 |
| markos | emanuel, I agree, unfortunately even if we did, we could change a thing :( | 20:26 |
| lkcl | Softbank bought them as a way to make money. immediately they bought them, they jacked up royalties by USD 1.00 then borrowed heavily against the asset | 20:26 |
| lkcl | thus they're massively in debt and NVIDIA only tolerated that because they needed to be "like Intel" and "like AMD" who have their own GPU and CPU | 20:27 |
| lkcl | now that deal's been shot to s***, ARM is in deep s*** | 20:27 |
| lkcl | their only way out is an IPO | 20:28 |
| lkcl | but, absolutely nobody who does any due diligence is going to invest in ARM when it's got billions of dollars of debt due to Softbank sucking it dry | 20:28 |
| lkcl | the *only way* for ARM to get "back into the black" is to fire its most expensive staff. | 20:29 |
| lkcl | which happened last week | 20:29 |
| lkcl | they laid off a staggering 1/6th of the workforce | 20:29 |
| markos | really? | 20:29 |
| markos | never heard of that | 20:29 |
| lkcl | unnnfortunately, those "expensive staff" were all part of the R&D | 20:29 |
| markos | wow, that sucks | 20:29 |
| lkcl | yeah it's not exactly made news yet. | 20:29 |
| lkcl | well, it's not about "sucks", it means that ARM's competitiveness is going to be obliterated in about 3 years time | 20:30 |
| lkcl | because that's how long it takes to get R&D "ideas" through into actual silicon | 20:30 |
| markos | well, Intel would not buy them, Nvidia also, but IBM might :D | 20:30 |
| lkcl | their new CEO is an "Accountant Hatchet-job man" | 20:30 |
| lkcl | pffh, no chance | 20:30 |
| lkcl | IBM aren't stupid :) | 20:30 |
| markos | I wouldn't be surprised if they did | 20:30 |
| markos | it would actually be a very clever move from IBM | 20:31 |
| lkcl | they'd have to have a reaaallly good reason to write off tens of billions of dollars of debt. | 20:31 |
| lkcl | NVIDIA obviously thought they could tolerate that kind of hit | 20:31 |
| lkcl | for a strategic gain | 20:31 |
| markos | in one stroke they would have 2 architectures and own the entire industry, both low end and high end | 20:31 |
| lkcl | yyeahh but just like with NvIDIA they'd risk being investigated | 20:32 |
| markos | and IBM definitely can afford the money for that | 20:32 |
| lkcl | what would they do with it, though? | 20:32 |
| markos | get back in the game? | 20:32 |
| markos | they're starting to become irrelevant | 20:32 |
| lkcl | they're already making billions from Banking and Govt Contracts | 20:32 |
| markos | yeah, but sooner or later those cash cows will move to Arm/Intel | 20:33 |
| lkcl | ARM and Intel CPUs simply don't have the I/O bandwidth | 20:33 |
| lkcl | it's not a coincidence that IBM POWER CPUs have always had well over twice the I/O throughput of any Intel or AMD processor | 20:34 |
| markos | not right now, but unless IBM produce something really good after Power10, I don't see that continuing to be the case | 20:34 |
| markos | the Ampere I have right now is way faster than my Talos II | 20:34 |
| lkcl | m-hm... | 20:34 |
| markos | well, access, I don't actually have an Ampere box here | 20:34 |
| lkcl | wouldn't it be great if there was some small team that was, ooo, creating a massive modernisation of the Power ISA? :) | 20:35 |
| markos | and similarly Apple has proven that you can build incredibly powerful CPUs around Arm | 20:35 |
| markos | lkcl, hm, I wonder who would be so bold to go against such giants, must be some really driven, incredibly intelligent and possibly a little bit crazy people to do that :) | 20:36 |
| lkcl | what they proved was that if you drop variable-width instruction encoding, you can drastically improve the throughput and sustain near-100% multi-issue out-of-order execution | 20:36 |
| lkcl | and funnily enough, the Power ISA is (or was until v3.1 Prefixes were added) similarly a fixed-width ISA | 20:37 |
| lkcl | POWER9 has a dirty secret. actually two. | 20:37 |
| lkcl | 1) Packed SIMD is s*** | 20:37 |
| lkcl | 2) there's a bottleneck on the L1 D/I Caches | 20:37 |
| lkcl | if you make too heavy usage of "unrolling" such that your inner loop cannot fit within the L1 I-Cache, what you find is that the CPU stalls | 20:38 |
| lkcl | the reason is: the designers screwed up, and didn't make the pathways from L1 to L2 wide enough | 20:38 |
| lkcl | so you get massive resource contention. | 20:39 |
| lkcl | whoops | 20:39 |
| lkcl | also, yes, Apple also created a vast number of "accelerator co-processors" | 20:39 |
| lkcl | offloading main tasks to nicely-designed power-efficient specialist cores | 20:40 |
| lkcl | which you'd never do in a Mainframe-like environment (Banking, number-crunching) | 20:40 |
| lkcl | IBM bought Redhat though | 20:41 |
| lkcl | that *was* a strategic move that they needed to do. | 20:41 |
| lkcl | the entire Power ISA Software ecosystem has been slowly crumbling for over a decade because nobody in the dev-community can get affordable modern systems | 20:42 |
| lkcl | markos, i *very deliberately* haven't gone against IBM. why do you think i left out VSX? :) | 20:42 |
| lkcl | that will make them happy because it's VSX that gets them their bread-and-butter money | 20:43 |
| markos | if what I heard is true, that won't be true for much longer but I cannot mention that | 20:44 |
Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!