Newest at the top
| 2025-01-10 23:11:23 +0100 | <mchav> | Well if you try it and get the performance under control it'll make for a lot of learning. I would still say do a prototype on your machine and profile it. |
| 2025-01-10 23:10:46 +0100 | merijn | (~merijn@128-137-045-062.dynamic.caiway.nl) merijn |
| 2025-01-10 23:10:21 +0100 | <euouae> | because the paper has Haskell pseudo-code to explain its concepts... call me easily influenceable |
| 2025-01-10 23:09:54 +0100 | <euouae> | I was reading this paper <https://www.cs.ox.ac.uk/people/jeremy.gibbons/publications/rationals.pdf> which I need in some small part in my code, and it sort of made me think about doing it all in Haskell for the fun of it |
| 2025-01-10 23:09:13 +0100 | <euouae> | Yeah fair enough. I'm fairly good at reasoning with C++ so I don't need to involve Haskell to end up having to write expert-level Haskell for it |
| 2025-01-10 23:08:20 +0100 | <mchav> | Well I would still profile the run. I've been working on a dataframe library and I find I REALLY have to be careful to avoid copies. Intuitive/readable Haskell ends up being a memory hog. Or maybe I'm not that good at writing it. I don't know how this changes when you use something that FFIs to C and how GHC will then deal with pinned memory etc. |
| 2025-01-10 23:07:35 +0100 | <mari51520> | the interface will not need much logic anyways |
| 2025-01-10 23:06:01 +0100 | <euouae> | Ah -- memory-wise. hm... OK fine. Maybe I'll do it in C++. |
| 2025-01-10 23:05:31 +0100 | <mchav> | Memory and as result money, I guess. Have you profiled a sample implementation on your computer? |
| 2025-01-10 23:05:00 +0100 | <euouae> | does the jumpstart matter? |
| 2025-01-10 23:04:49 +0100 | <euouae> | mchav, OK but what do you mean costly? If my computations need 1+hr each? |
| 2025-01-10 23:04:37 +0100 | <mari51520> | well but it makes sense to have a server for crunching and another for haskell |
| 2025-01-10 23:04:32 +0100 | <mchav> | *GHC |
| 2025-01-10 23:04:24 +0100 | <mchav> | Haskell is really heavy-handed for this sort of thing. It'll be costly just to have Haskell running on each instance. |
| 2025-01-10 23:03:43 +0100 | <euouae> | also, for fun |
| 2025-01-10 23:03:29 +0100 | <euouae> | I'm expecting Haskell to simplify the MPI integration |
| 2025-01-10 23:03:24 +0100 | <euouae> | openMP is for single-host. if you want multi-host you need MPI. I don't have that code |
| 2025-01-10 23:03:18 +0100 | tromp | (~textual@92-110-219-57.cable.dynamic.v4.ziggo.nl) (Quit: My iMac has gone to sleep. ZZZzzz…) |
| 2025-01-10 23:03:14 +0100 | <haskellbridge> | <magic_rb> But id recommend equinix or hetznee, theyre cheaper and more simple. Also youre not helping jeff bazos |
| 2025-01-10 23:03:02 +0100 | <mari51520> | why do you need haskell-friendly then? |
| 2025-01-10 23:02:55 +0100 | <euouae> | magic_rb: a single dedicated machine though? |
| 2025-01-10 23:02:40 +0100 | <euouae> | mari51520: I don't depend on anything, it's all my own code. In C++ it was with OpenMP and libgmp/libmpfr. |
| 2025-01-10 23:02:31 +0100 | <haskellbridge> | <magic_rb> On amazon cloud if you get a dedicated machine. You get a remote kvm and you can do whatever. Install freebsd if you want |
| 2025-01-10 23:02:00 +0100 | <euouae> | but it depends on the level of control, that's what I mean by friendly |
| 2025-01-10 23:01:50 +0100 | <euouae> | and it might not be supercomputers I need, maybe just amazon's cloud is fine |
| 2025-01-10 23:01:35 +0100 | <euouae> | magic_rb: some supercomputers don't run ghc ... it depends on their setups |
| 2025-01-10 23:01:28 +0100 | <mari51520> | huh for euouae prob |
| 2025-01-10 23:01:18 +0100 | <haskellbridge> | <magic_rb> Or equinix |
| 2025-01-10 23:01:03 +0100 | <haskellbridge> | <magic_rb> mari51520 i dont know what you mean by compute friendly hardware. Hardware is hardware, but for compute id look at hetzner dedicated servers. |
| 2025-01-10 23:00:23 +0100 | <haskellbridge> | <sm> sim590: 👍️ |
| 2025-01-10 23:00:06 +0100 | <mari51520> | cool. Well i do not know many providers, everyone obviously is craze about amazon services while i have heard good reports about haskell&digitalocean in the past. I am too curious not to ask though: which lib are you relying on? |
| 2025-01-10 22:59:18 +0100 | <euouae> | Specifically those were the numbers with C++, fixed width integers, and openmp. With MPI in place and bignum stuff, it'll probably take longer |
| 2025-01-10 22:57:47 +0100 | <euouae> | it would be nice to go higher too, I haven't yet done any calculations that tell me roughly what n=7 or general 'n' look like, n=8 might not be feasible e.g. |
| 2025-01-10 22:57:24 +0100 | <euouae> | I know that on 6 cores my operation for (say n=6) takes around an hour? I want to run it for say 1000 data points |
| 2025-01-10 22:56:56 +0100 | <euouae> | I mean I've done that much |
| 2025-01-10 22:56:19 +0100 | <mari51520> | huh i would advice starting from your personal machine |
| 2025-01-10 22:55:01 +0100 | <euouae> | Hello, I've never done this before & I'm wondering if anyone has advice: If want to try some scientific computations of very specific number-crunching stuff I have in mind, how can I pay a reasonable amount to get access to some computing power? In particular haskell-friendly becuase I want to write it in haskell |
| 2025-01-10 22:53:46 +0100 | euouae | (~euouae@user/euouae) euouae |
| 2025-01-10 22:49:55 +0100 | merijn | (~merijn@128-137-045-062.dynamic.caiway.nl) (Ping timeout: 252 seconds) |
| 2025-01-10 22:47:47 +0100 | Smiles | (uid551636@id-551636.lymington.irccloud.com) (Quit: Connection closed for inactivity) |
| 2025-01-10 22:46:17 +0100 | housemate | (~housemate@146.70.66.228) (Max SendQ exceeded) |
| 2025-01-10 22:46:06 +0100 | housemate | (~housemate@146.70.66.228) housemate |
| 2025-01-10 22:45:25 +0100 | housemate | (~housemate@146.70.66.228) (Max SendQ exceeded) |
| 2025-01-10 22:45:13 +0100 | housemate | (~housemate@146.70.66.228) housemate |
| 2025-01-10 22:45:08 +0100 | merijn | (~merijn@128-137-045-062.dynamic.caiway.nl) merijn |
| 2025-01-10 22:44:53 +0100 | SlackCoder | (~SlackCode@64-94-63-8.ip.weststar.net.ky) (Quit: Leaving) |
| 2025-01-10 22:44:37 +0100 | housemate | (~housemate@146.70.66.228) (Max SendQ exceeded) |
| 2025-01-10 22:44:26 +0100 | housemate | (~housemate@146.70.66.228) housemate |
| 2025-01-10 22:43:47 +0100 | housemate | (~housemate@146.70.66.228) (Max SendQ exceeded) |
| 2025-01-10 22:43:40 +0100 | housemate | (~housemate@146.70.66.228) housemate |