| justinrestivo[m] | lkcl: just double checking: is https://libre-soc.org/HDL_workflow/ an up to date? Try to pattern match between that and the dev-env-setup repo. | 01:39 |
|---|---|---|
| lkcl | justinrestivo[m]: it should be. we just noted a discrepancy yesterday, programmerjake was unaware of the pia install script | 10:51 |
| lkcl | because dev-env-setup is the "automated" version of the manual descriptions on HDL_workflow, it's dev-env-setup that takes top precedence and priority as far as correctness is concerned | 12:02 |
| lkcl | because dev-env-setup is (a) automated (b) saves so much time (c) is part of the (critical) reproducible builds | 12:03 |
| lkcl | fascinating discussion on #nmigen about LGPLv3 and HDL licensing in general https://libera.irclog.whitequark.org/nmigen/2021-08-24#30728292 | 13:37 |
| EmanuelLoos[m] | Hello, how far is the libre-soc (the hardware part) from being usable in a normal pc (with free software like Debian GNU/Linux of course)? I am thinking of a normal setup with for example XFCE, Firefox, Thunderbird, LibreOffice, GIMP, MPV (a video player) ond other desktop applications working fast enough for normal use cases with a libre-soc at a reasonable price. How far is the libre-soc from that hardware-wise? What are the greatest | 14:32 |
| EmanuelLoos[m] | difficulties? Just asking. | 14:32 |
| veera[m] | Libre-soc is in design phase. Initially a test sample is being made. The final product will support all your are asking. | 14:37 |
| EmanuelLoos[m] | veera[m]: Thanks! When do you expect the final product to be ready? | 14:39 |
| veera[m] | A test OPENPOWER 3.0B microwatt chip has been sent to TSMC Fab. | 14:41 |
| veera[m] | When it will be ready; lkcl can give correct estimate; I say not this year, may be next year. | 14:42 |
| EmanuelLoos[m] | <veera[m]> "When it will be ready; lkcl can..." <- While we are at estimating when it will be ready, in a few years the standard pc user might expect more CPU/GPU/VPU power than right now. Are you planing on offering more and less powerful chips or would that be much more work? How powerful do you plan and expect the libreSOC to be? | 14:53 |
| jn | AFAIK, some degree of performance parametrization (making more/less powerful chips) is planned | 14:56 |
| EmanuelLoos[m] | jn: By the way, what makes less powerful chips cheaper? Less manufacturing steps? | 15:00 |
| jn | possibly any of the following: | 15:01 |
| jn | - less die area | 15:01 |
| jn | - less complex (bigger) process | 15:01 |
| jn | - less partially broken parts (see binning) | 15:02 |
| jn | - lower development and verification costs | 15:02 |
| jn | in a specific situation, the factors will combine way, but these are some general factors | 15:03 |
| EmanuelLoos[m] | Is the market large enough for producing at a reasonable price? | 15:05 |
| lkcl | EmanuelLoos[m], hi | 15:05 |
| EmanuelLoos[m] | lkcl: Hi | 15:05 |
| lkcl | it's down to funding. | 15:05 |
| lkcl | basically to design and manufacture something in. e.g. 22nm would require about 18 months and around USD 8 million. | 15:06 |
| jn | s/will combine way/ will combine in a specific way/ (oops) | 15:06 |
| lkcl | there's a page somewhere... | 15:06 |
| lkcl | here it is https://libre-soc.org/22nm_PowerPI/ | 15:06 |
| lkcl | 8 months of that is spent doing simulations and verification | 15:07 |
| jn | a massive undertaking | 15:07 |
| lkcl | because, if you're doing to drop USD 4 million on masks (16 million for 7nm) you sure as s*** want to know that what you're putting your money down for is going to work :) | 15:07 |
| lkcl | so we're looking for VC funding, and that's what Red Semiconductor is for | 15:08 |
| lkcl | Red Semiconductor will put Libre-SOC cores into actual products, in exactly the same way that e.g. qualcomm puts ARM cores into actual products | 15:08 |
| lkcl | and we [putting Red Semi hat on] have a roadmap going single-core, quad-core, 8-core, 16/64-core chiplets, 1024-core supercomputer on substrate using 16/64-core chiplets | 15:09 |
| lkcl | using each phase as a lower-cost proving-ground for the next | 15:10 |
| lkcl | so there isn't "one answer", there's multiple answers (multiple products) | 15:10 |
| * lkcl [taking Red Semi hat off, putting Libre-SOC one back on] | 15:11 | |
| lkcl | we've got NGI POINTER funding going through the approvals process at the moment for another single-core ASIC shuttle run | 15:12 |
| lkcl | this will be for a gigabit router IC, the goal is 300 mhz, in either 180nm or 130nm. | 15:12 |
| EmanuelLoos[m] | So, if I understand right, it all comes down to funding. Does the project have enough financial support? You pointed out costs for producing 22nm and at 7nm. For producing at how much nanometer does libreSOC have enough financial support? I suppose producing at less nanometers will make the chip more energy efficient, am I right? | 15:18 |
| EmanuelLoos[m] | * at 22nm and at 7 nm | 15:26 |
| EmanuelLoos[m] | * at how many nanometers | 15:26 |
| EmanuelLoos[m] | lkcl | 15:30 |
| lkcl | NGI POINTER is happy to consider EUR 200,000, which will help cover the 180/130 nm Shuttle | 15:35 |
| lkcl | and if that works, Red Semi would be able to consider picking it up as a commercial product - a **PROPERLY** Open OpenWRT-compatible router ASIC | 15:36 |
| lkcl | Libre-SOC is not the appropriate vehicle for funding to create commercial ASICs | 15:36 |
| lkcl | a *company* is the appropriate vehicle for funding to create commercial ASICs | 15:36 |
| lkcl | Libre-SOC may at some point down the line apply for a EUR 15-17 million EU Innovation Grant | 15:37 |
| lkcl | however that has absolutely nothing to do with Red Semiconductor or any other company that happens to pick up Libre-SOC designs | 15:37 |
| lkcl | i very specifically mentioned "Libre-SOC is to RED Semi as ARM is to Qualcomm" for that very reason | 15:38 |
| lkcl | ARM has *NEVER* created or sold actual commercial silicon itself since the very first ASICs went into BBC Micros in the late 80s | 15:38 |
| lkcl | the reason is dead simple: it would be a commercial conflict of interest with their licensees. | 15:39 |
| lkcl | pissing off 2,000 customers by creating commercially competing products is a really good way to lose that 43 billion dollar valuation | 15:39 |
| lkcl | so you need to be asking the question, "does RED Semiconductor have the VC Funding to create a commercial ASIC" | 15:40 |
| lkcl | not "will Libre-SOC have enough VC funding or other financial support to create a commercial ASIC" | 15:40 |
| lkcl | Libre-SOC will *never* create a commercial ASIC. | 15:41 |
| lkcl | *RED SEMICONDUCTOR* will create a commercial ASIC | 15:41 |
| EmanuelLoos[m] | If a commercial company picks up LibreSOC would it be profitable for the company to make its own non-free chip designs or would important changes require to do much from scratch? Is libreSOC preventing non-free designs based on it with copyleft licenses? Will the RED Semiconductor chip be open hardware as well? | 15:49 |
| lkcl | RED Semi is specifically pitching to VCs on the transparency | 15:54 |
| lkcl | given the situation with total lack of transparency in ASIC manufacturing | 15:55 |
| lkcl | and how badly they're screwing up | 15:55 |
| lkcl | this is a very easy pitch / sell | 15:55 |
| lkcl | Supermicro delisted from NASDAQ after a Bloomberg report on rogue components | 15:56 |
| lkcl | Intel Master firmware key being reverse-engineered | 15:56 |
| lkcl | RED Semi *requires* that its designs be fully transparent and available for audit and inspection by its customers | 15:56 |
| lkcl | the only fly in the ointment there is the Foundry PDK NDAs | 15:57 |
| lkcl | but that's RED Semi's problem to solve | 15:57 |
| lkcl | the reason why we're putting the HDL under the LGPL-v3 is specifically so that all commercial companies would be required to make available all and any modifications to the Libre-SOC core as per the terms of the LGPL-v3 Licenes | 15:58 |
| EmanuelLoos[m] | lkcl: Why not the GPL? | 15:59 |
| lkcl | however if they happen to mix it in with some proprietary HDL, we neither want to prevent and prohibit that nor do we want them to feel that they have to get the proprietary HDL released as Libre-Licensed | 15:59 |
| lkcl | because that would prevent and prohibit use of proprietary peripherals by N E Other Corporation | 15:59 |
| lkcl | ARM doesn't force its licensees to use ARM-mandated peripherals | 15:59 |
| lkcl | you license the ARM core, you lay it down with whatever peripherals *you* want | 16:00 |
| lkcl | and pay USD 250k to 2 million for | 16:00 |
| EmanuelLoos[m] | Will the libreSOC include hardware optimization for free video codecs like AV1, VP9, VP8, Daala, Theora, Thor and/or some old MPEG codecs where the patents are expired? | 16:18 |
| EmanuelLoos[m] | lkcl | 16:29 |
| lkcl | EmanuelLoos[m], this is the basis of the NLnet Video/Audio grant | 16:38 |
| lkcl | https://bugs.libre-soc.org/show_bug.cgi?id=137 | 16:38 |
| lkcl | we started with MP3 and it resulted in a massive 4.5x reduction in the code size for the inner loop of one of the key functions in MP3 | 16:39 |
| EmanuelLoos[m] | h264 and h265? Doesn't the MPEG LA charge huge patent fees for these codecs? | 17:01 |
| openpowerbot | [irc] <lkcl> EmanuelLoos[m], that won't be Our Problem (tm) because we're implementing general-purpose instructions | 17:09 |
| openpowerbot | [irc] <lkcl> which, oh look! what a coincidence! this software algorithm happens to run really really fast! | 17:09 |
| EmanuelLoos[m] | openpowerbot: Are general-purpose instructions as good as a specific hardware implementaion? | 17:11 |
| openpowerbot | [irc] <lkcl> if you go ultra-specialist then no, you can never beat ultra-specialist custom silicon with general-purpose instructions | 17:21 |
| openpowerbot | [irc] <lkcl> however the problem with ultra-specialist custom silicon: it's literally incapable of doing anything else | 17:21 |
| openpowerbot | [irc] <lkcl> so you add more and more custom silicon blocks to "solve" the problem | 17:22 |
| openpowerbot | [irc] <lkcl> and usually run smack into hardware patents as a result | 17:22 |
| openpowerbot | [irc] <lkcl> the Cray-style Vectors we can take a different approach | 17:22 |
| EmanuelLoos[m] | openpowerbot: Except with free codecs | 17:22 |
| openpowerbot | [irc] <lkcl> yes. theora, dirac etc. | 17:22 |
| EmanuelLoos[m] | openpowerbot: AV1, VP8, VP9, FFv1, Daala, Thor too (video) | 17:23 |
| EmanuelLoos[m] | openpowerbot: FLAC, Opus, Vorbis too (audio) | 17:24 |
| EmanuelLoos[m] | openpowerbot: old MPEG codecs where the patents are expired too | 17:25 |
| EmanuelLoos[m] | How about creating a ultra-specialist custom silicon (especially for encoding) as soon as AOMedia (Google, Netflix, etc) publishes a new free video codec? If libreSOC has one of the first fast custom silicon implementations I think there would be many buyers (especially around servers [where free software like Debian is used quite often] as AV1 is an internet standard supported by Firefox and Chromium in contrast to h265). | 17:44 |
| EmanuelLoos[m] | lkcl | 18:28 |
| programmerjake | we're not currently planning on creating custom silicon specifically dedicated to any particular video codec -- that said, being the first one to implement a particular codec may be difficult since the companies actually authoring the spec will probably have a headstart | 18:36 |
| programmerjake | interesting idea nonetheless | 18:36 |
| EmanuelLoos[m] | <programmerjake> "we're not currently planning on..." <- I know, it is just an idea I wanted to share. | 18:47 |
| EmanuelLoos[m] | When supporting a video codec that will be an internet standard right after it comes out the custom silicon probably would be usable for a very long time. | 18:50 |
| lkcl | lxo, programmerjake richardwilbur[m] jn klys meeting jitsi | 22:50 |
| programmerjake | yup, sec.. | 22:51 |
Generated by irclog2html.py 2.17.1 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!