I would like to request funding for the purchase of between 1 and 4, inclusive, 64-bit RISC-V systems based on the upcoming HiFive Unmatched mini-ITX board, with the aim of providing true-hardware (rather than emulated) infrastructure for supporting this architecture in Gentoo. I have floated this idea in #gentoo-dev and #gentoo-infra lately, and the responses have been favourable. THE PEOPLE The RISC-V Project and me (later on the RISC-V Project with me as member, if successful), other participants welcome. THE HARDWARE Each system should consist of the following: * one HiFive Unmatched board (https://www.sifive.com/boards/hifive-unmatched) * one M.2 NVMe solid-state drive, capacity in the order of 256 GB * one ATX Power Supply Unit rated at a minimum of 150 W * one Mini-ITX case [1] [1] The type depends on the hosting solution - if we find a proper data centre 1U rack-mounted cases would likely be the best, were these to stand in someone's office a small tower or a cube (possibly a dual-system one so that we only need one case per two boards) would be better. THE COST By my rough estimate we are talking about no more than 1,500 USD per system. Note that this is still an early draft and the exact costs of everything except the boards are yet to be figured out, depending to a large degree of where those machines will end up being hosted. THE HOSTING This still has to be figured out, indeed this is why I am - following the suggestion from antarus in #gentoo-infra - opening this ticket in such early stages: * I have asked my employer if we could host those machines, however it will take time for them to get back to me on this even (or possibly especially) in case of an affirmative response; * we could host them alongside existing Infra systems, for instance at Manitou in Germany (I would slightly prefer for those machines to be located in the EU); * failing all else, I could host ONE such machine personally.
TL;DR: Palmer offered us a board 2 years ago (us == riscv gentoo proj), but it got stuck somewhere (and Palmer himself had moved to Google since). A guy from RISCV Foundation approached me at FOSDEM '20 and asked me if we were in need of RISC-V hardware. That said, we may ask both entities and maybe they will help us getting it for free.
Sorry about that, the whole RISC-V this is kind of on the back burner for me these days and I guess I thought we'd already gotten the board sorted out. It should be sitting in the rack in Fremont, which means someone just needs VPN access to get in. That said, I am planning on shutting down the rack eventually (though the Fedora guys are sort of dragging their feet) so if you have someone I can send the board to that would be ideal. The new board is probably a better bet, and you might even be able to get them donated.
FWIW I support buying the new hardware on Gentoo expenses. Getting a donation would of course always be nice, but things may move quicker / easier when lubricated by money...
(In reply to Mikle Kolyada from comment #1) > That said, we may ask both entities and maybe they will help us getting it > for free. It's an option! Do you recall with whom from the RISC-V Foundation you spoke last year? Anyway, I guess this means we should focus on planning, and possibly allocating resources for, the hosting for now.
(In reply to Marek Szuba from comment #0) > I would like to request funding for the purchase of between 1 and 4, > inclusive, 64-bit RISC-V systems based on the upcoming HiFive Unmatched > mini-ITX board, with the aim of providing true-hardware (rather than > emulated) infrastructure for supporting this architecture in Gentoo. I have > floated this idea in #gentoo-dev and #gentoo-infra lately, and the responses > have been favourable. > > THE PEOPLE > > The RISC-V Project and me (later on the RISC-V Project with me as member, if > successful), other participants welcome. > > THE HARDWARE > > Each system should consist of the following: > * one HiFive Unmatched board > (https://www.sifive.com/boards/hifive-unmatched) > * one M.2 NVMe solid-state drive, capacity in the order of 256 GB > * one ATX Power Supply Unit rated at a minimum of 150 W > * one Mini-ITX case [1] > > [1] The type depends on the hosting solution - if we find a proper data > centre 1U rack-mounted cases would likely be the best, were these to stand > in someone's office a small tower or a cube (possibly a dual-system one so > that we only need one case per two boards) would be better. > > THE COST > > By my rough estimate we are talking about no more than 1,500 USD per system. > Note that this is still an early draft and the exact costs of everything > except the boards are yet to be figured out, depending to a large degree of > where those machines will end up being hosted. > > THE HOSTING > > This still has to be figured out, indeed this is why I am - following the > suggestion from antarus in #gentoo-infra - opening this ticket in such early > stages: > * I have asked my employer if we could host those machines, however it will > take time for them to get back to me on this even (or possibly especially) > in case of an affirmative response; > * we could host them alongside existing Infra systems, for instance at > Manitou in Germany (I would slightly prefer for those machines to be located > in the EU); > * failing all else, I could host ONE such machine personally. So my first question related to hosting is: We have existing hosting at the OSL. These units are 150W, so we could probably host them there. Is there any reason besides latency (since obvs. west coast US is far from most of our developers) that we could not host them there? -A
(In reply to Marek Szuba from comment #0) > I would like to request funding for the purchase of between 1 and 4, > inclusive, 64-bit RISC-V systems based on the upcoming HiFive Unmatched > mini-ITX board, with the aim of providing true-hardware (rather than > emulated) infrastructure for supporting this architecture in Gentoo. I have > floated this idea in #gentoo-dev and #gentoo-infra lately, and the responses > have been favourable. You requested "between 1 and 4 systems." Can you more clearly describe the need beyond 1 system? This will help with justifying the spend on this, thanks. > > THE PEOPLE > > The RISC-V Project and me (later on the RISC-V Project with me as member, if > successful), other participants welcome. > > THE HARDWARE > > Each system should consist of the following: > * one HiFive Unmatched board > (https://www.sifive.com/boards/hifive-unmatched) > * one M.2 NVMe solid-state drive, capacity in the order of 256 GB > * one ATX Power Supply Unit rated at a minimum of 150 W > * one Mini-ITX case [1] > > [1] The type depends on the hosting solution - if we find a proper data > centre 1U rack-mounted cases would likely be the best, were these to stand > in someone's office a small tower or a cube (possibly a dual-system one so > that we only need one case per two boards) would be better. > > THE COST > > By my rough estimate we are talking about no more than 1,500 USD per system. > Note that this is still an early draft and the exact costs of everything > except the boards are yet to be figured out, depending to a large degree of > where those machines will end up being hosted. > > THE HOSTING > > This still has to be figured out, indeed this is why I am - following the > suggestion from antarus in #gentoo-infra - opening this ticket in such early > stages: > * I have asked my employer if we could host those machines, however it will > take time for them to get back to me on this even (or possibly especially) > in case of an affirmative response; > * we could host them alongside existing Infra systems, for instance at > Manitou in Germany (I would slightly prefer for those machines to be located > in the EU); > * failing all else, I could host ONE such machine personally.
(In reply to Alec Warner from comment #5) > So my first question related to hosting is: We have existing hosting at the > OSL. These units are 150W, so we could probably host them there. Is there > any reason besides latency (since obvs. west coast US is far from most of > our developers) that we could not host them there? None I can think of. While I would be hesitant to add new hosting within juristiction of the FISA (AKA "we rubberstamp every NSA demand and add a gag order on top of them") Court, for existing hosting this point is moot. (In reply to Alec Warner from comment #5) > You requested "between 1 and 4 systems." Can you more clearly describe the > need beyond 1 system? This will help with justifying the spend on this, thanks. My thinking has been that if/when we start using those systems for more than just experimental work, stage building for instance, on the one hand it definitely would be good to have some redundancy (i.e. 2 or more nodes) and on the other four cores might be too little. On the other hand, with cores spread across multiple machines there is also the network overhead of distcc/MPI/... to consider so there is no point in increasing this number too high. In the end, my educated guess (based on my working experience with HPC clusters, admittedly amd64 ones) is that we shouldn't need more than 16 cores spread across 4 nodes. That said, with the current number of physical RISC-V Gentoo systems being 0 having even one such machine would be a benefit.
Update: looks like we might, thanks to the BeagleV open beta programme, end up with quite a few RISC-V boards for free. We may or may not need a more powerful, official system of this sort in the future but for now, I reckon no need for buying any. Opinions? Feel free to close this ticket if you agree.
(In reply to Marek Szuba from comment #8) > Update: looks like we might, thanks to the BeagleV open beta programme, end > up with quite a few RISC-V boards for free. We may or may not need a more > powerful, official system of this sort in the future but for now, I reckon > no need for buying any. Opinions? Feel free to close this ticket if you > agree. Just to be clear i thought we were blocked on hosting. If we get some BeagleV's we still need a place for them, right? -A
Hah, *I* thought we were waiting for me to provide a more accurate cost estimate in the context of hosting these at the OSL! Anyway, I would say that in the long run we do still need more concrete information about hosting (who is responsible for communication between the OSL and Manitou? Infra?) - but for now, between BeagleV being considerably more lightweight than HiFive Unmatched (2 vs 4 cores, 8 vs 16 GB of RAM, no NVMe SSD support) and it only being in beta (meaning that there is a small but non-zero possibility that the current design will turn out not to be usable in production), we would for now have individual developers managing their own BeagleV systems.
Let's keep this open, turns out the JH7100 chip installed on BeagleV beta boards is very much NOT production-ready [1]. [1] https://wiki.seeedstudio.com/BeagleV-Getting-Started#FAQ
(In reply to Palmer Dabbelt from comment #2) > Sorry about that, the whole RISC-V this is kind of on the back burner for me > these days and I guess I thought we'd already gotten the board sorted out. > It should be sitting in the rack in Fremont, which means someone just needs > VPN access to get in. > > That said, I am planning on shutting down the rack eventually (though the > Fedora guys are sort of dragging their feet) so if you have someone I can > send the board to that would be ideal. > > The new board is probably a better bet, and you might even be able to get > them donated. If its Fremont, CA we can probably meeting (I'm San Jose based) and I can get it up and running somewhere.
releng: if RISCV hardware is still required; please open a new bug.
