00:01:31 btw Wallet bot supports output redirection.. for example say there is a dedicated bot channel & a general chat channel. the bot can be in both but if used in chat the bot displays a rate limited "please use in #monero-bots" message & then outputs in that other channel instead 00:03:37 it's a handy feature for chat channels where bots can be a distraction especially for noisy commands 00:06:44 that feature can be always on or automatically enabled when thresholds are met such as during busy chat times or busy bot use times. whole lot of features that never made it into docs. will have to update some day 06:49:17 Oh nooo, the chineses released big llms models for free <- I only know Deepseek and qwen though, and I got the impression those families are not for the lowest-end hardware, with Ollama being more accessible? 07:00:58 What is the underlying reason behind the increase in the value of the Monero currency? Increase in the number of transactions on the chain?. Increase in the average value of transactions on the chain? Buy and hold as store of value i.e. speculation? 07:43:02 I'd say that it's the energy needed to calculate the chain. 12:13:13 Hello, I need advice . I am a super fan of cybersecurity and opsec and regulation for IT. Yet i want to ask is there an opportunity to open cybersecurity consulting firm . Is this profitable business model ? 12:13:41 It’s less monero related but need monero community advise how they loook at this 13:03:01 So to confirm using anonymous inbound does not provide syncing over mixnet (i2p/tor), and is only for transaction proxying? 13:25:29 Correct 16:18:27 Last chance to register: https://magicgrants.org/2025/12/05/Monero-Fund-2026-Election 16:20:45 There are plenty of Deepseek and qwen flavor available on Ollamal, that you can get according to your hardware. > Oh nooo, the chineses released big llms models for free <- I only know Deepseek and qwen though, and I got the impression those families are not for the lowest-end hardware, with Ollama being more accessible? 16:20:45 They also have MoE options. 16:20:45 Same for many other models. 16:20:45 Also you can run there big model fast even if you don't have the VRAM to load it all, only thing you needed, is the ram. 16:22:18 Like, if I only have an integrated GPU but 16 gigs of RAM, would I be able to run some decent Qwen flavor? I know there are some super-miniature ones, but not talking about that. 16:26:22 https://mrelay.p2pool.observer/m/monero.coffee/mXbLNCLytZXhAzdgIsinsVET.png (george.png) 16:26:35 @shitpost:monero.coffee: 16:26:54 don't shitpost here lmao 16:27:10 oh 16:27:29 look im bridged to the discord 16:27:44 check the monero bridge 16:27:51 thats cool 16:27:55 yes but don't use the offtopic channel, this is not for that 16:28:05 ah ok 16:36:33 is this a two way discord bridge now? 16:36:52 given that's monero.coffee 16:37:00 no it's not two way, he just registered on my matrix instance 16:37:13 it's read only channel, I already told him these channels are not for that, sorry 16:37:23 the discord leaks like thin oil 16:37:25 :D 16:37:42 first thing they do out of containment is @everyone SEE MY MEME 16:38:22 🤣🤣 16:42:40 BlueyHealer: With 16gb vram you can run the 30GB one with llamacpp decently (its a MoE model). 16:42:40 I can even run gpt-oss-120b decently (surprisingly) with 16GB vram, but that because I have 64GB ram and it use almost all of with for that specific one. 16:43:20 https://mrelay.p2pool.observer/m/monero.coffee/HDAnbhuJdCNZvTnyGfSPggaE.png (image.png) > is this a two way discord bridge now? 16:44:22 oh no 16:44:24 it's discord people 16:44:47 But I'm not talking VRAM - just regular RAM. 16:45:34 BlueyHealer: you need VRAM, you just don't need the whole model to fit in it when it's a MoE model 16:46:42 You can run 300B parameters models with like 24GB vram.. 16:46:42 I mean... Why do you think they buy all the ram... oh... You can jut load the rest of the 300B model in that 256GB RAM of yours 16:47:08 If you can get the ram of couse, good luck 16:47:23 That's all planned, else they would not have even brought useless unfinished wafers 16:47:24 Ah, so me with 16G can still run none of that 16:47:53 BlueyHealer: if you have no vram, no.. 16:47:53 except if you like it ultra slowmo 16:49:49 Good thing I don't even have any use for LLMs in the first place lol 18:59:30 why does syncing the blockchain take so long even with a good internet connection amd high end hardware? 18:59:30 i noticed that the only real load is disk IO 18:59:30 cpu, ram & network are doing almost nothing while a lot of data gets written, way more than the actual filesize[... more lines follow, see https://mrelay.p2pool.observer/e/6OCB0dcKTS1rTm5o ] 19:00:01 luzidd: you need a SSD for good performance 19:00:22 monerod is probably getting bottlenecked on the hard drive speed 19:00:24 i have a good ssd 19:01:47 kingston kc3000 2TB 19:01:50 most of monero looks up random inputs 19:02:07 so effectively you need the full dataset loaded most of the time to lookup those old transactions 19:02:19 it does not make sense to me that it's writing so much 19:02:46 building up the LMDB cache :D 19:03:07 but really, some is legacy, other was worked on 19:15:24 disk read is ~10-20MB/s, disk write is 100-200MB/s 19:15:24 wtf?! 19:26:21 i wonder if it's worth it to just download the raw blockchain from the website and import it? 19:27:38 that's not how it works 19:29:46 okay, why not? 19:32:18 how long are you syncing already btw? 19:35:50 i was syncing for multiple weeks on a hdd raid before i copied my file over to the nvme 19:37:00 20:26:21 i wonder if it's worth it to just download the raw blockchain from the website and import it? 19:37:11 the importing will take the same time 19:37:40 though for me on NVMe it was synced after a couple of hours from scratch, for a full node 19:37:53 DataHoarder: This ^ 19:37:55 what matters is IOPS 19:38:04 i'm also getting exceptions and stack traces in the bitmonero.log, is that normal? 19:38:14 not speed, consumer ssds are meh on random write iops 19:38:45 I managed to sync pruned node in 24 hours on a consumer ssd 19:38:56 DataHoarder: yeah, extrapolating at the current rate should get me fully synced in a few hours 19:38:59 and ethernet 100mbit 19:39:20 that model claims good IOPS tbh, but it's probably on the fast layer 19:39:40 should be fine 19:41:09 @luzidd:matrix.org: Depends on what the exceptions are 19:41:25 @luzidd:matrix.org: https://github.com/monero-project/monero/issues/8341 19:41:39 it's the same issue 19:41:49 then yes its normal 19:42:00 Not ideal, but is normak 20:02:42 I should bench how much time it take on HDD+128GB nvme write-only cache 20:02:42 The bench I did 2-3 years ago (~12Hr on SMR piece of shit + 32GB nvme cache) is surely outdated, especially with the stress test and all 20:27:20 Yeah LMDB has quite a bit of write amplification, the next update to cuprate we will probably be switching out LMDB for a different DB. > <@luzidd:matrix.org> it does not make sense to me that it's writing so much 20:48:46 is there a common format suitable for moneros workload? 21:11:13 @luzidd:matrix.org: Not really, the current solution I am at with cuprate is a dual database with fjall (a rust database) and a custom one. 21:11:13 I managed to get a 1hr 25 minute fast sync with this setup on a computer with only a HDD. 22:44:35 @boog900: I didn't know about cuprate, looks interesting