[19:21:13] Definitely not too late. [19:21:36] I understand that there's a problem with the size of the case that would prevent us from putting a beefy GPU in there. [19:22:00] right [19:22:02] the current stat1002 [19:22:13] but, we haven't started the procurement process for the other ones yet [19:22:18] so, we can take that into consideration [19:22:32] but, robh has worries about it not being worth it, since it will probably up the cost a lot [19:22:43] also, i don't think we (analytics) budgeted for that cost increase [19:23:07] i don't really know how budgeting works between teams, but does research have some hw budget reserved it could contribute? or do we just have to convince ops it is worth it? [19:24:13] we budgeted about $10K for a stat1002 replacement, but this is because it also needs a lot of storage space [19:25:00] the GPU ellery linked to is $1200, not sure how much more it would be for a chassis that could hold it [19:25:36] Robh had a question for ellery: [19:25:39] 'Would the GT line still be a viable use and is it preferable to purely CPU driven crunching like we do now?' [19:25:53] the GT line is not 'top of the line', but has a smaller form factor that would more easily fit into the boxes we usually order [19:26:38] ottomata, gotcha. My immediate concern right now is whether or not we should budget for this in the next annual plan. I think we'd need DarTar to answer questions about the budget for this year. I can prod him to make a statement about that in 30 minutes when I meet with him. [19:26:53] ok [19:27:20] Re. the GT line, I imagine that would be several orders of magnitude better than a CPU-only approach -- even though it's not the big fancy one, but I think that deserves some research. [19:27:32] hm, quick answer then: i think (unless it works differently than I know), research would need to contribute budget some for the procurement. [19:28:06] Gotcha. If not research this year, I'll be looking to add capex to the proposal I'm working on for next year. [19:28:21] if we: A. combine it with the stat1002 replacement or B. fit a smaller (GT) one into stat1004 now, I'm going to guess the cost research would need to contribute ~$5000, give or take a few thousand :) [19:28:37] That sounds very reasonable. [21:03:29] halfak: so we were talking about wikidata's backlog being "simple", do you know if there's a page like https://en.wikipedia.org/wiki/Wikipedia:Backlog for wikidata? [21:03:40] or something similar to link to that shows a lot of their work? [21:13:02] (well, I just sent an email on the topic, halfak, so feel free to just edit the task as I say there) [21:19:50] o/ milimetric. In meeting. Will respond to email :) [23:19:22] Holy freakin' moley, we have 11,680 users with ORES enabled on English Wikipedia!!! [23:31:49] 17,931 installations across 18 wikis