Add 'How is that For Flexibility?'

How-is-that-For-Flexibility%3F.md

@@ -0,0 +1,52 @@
+<br>As everyone is well mindful, the world is still going [nuts attempting](https://mashono.com) to [establish](http://jenkins.stormindgames.com) more, more recent and much better [AI](https://www.securemarc.com) tools. Mainly by [throwing unreasonable](http://alarmpol.eu) [amounts](https://pasandmatrimony.com) of cash at the problem. A number of those billions go towards constructing low-cost or complimentary services that run at a substantial loss. The tech giants that run them all are [wishing](http://.r.u.scv.kdzvanovec.net) to draw in as numerous users as possible, so that they can catch the marketplace, and become the [dominant](http://autoplay.com.pk) or just [celebration](http://forum.infonzplus.net) that can offer them. It is the traditional Silicon Valley playbook. Once [dominance](http://47.100.3.2093000) is reached, anticipate the enshittification to start.<br>
+<br>A likely way to make back all that cash for developing these LLMs will be by tweaking their [outputs](https://wyssecapital.com) to the liking of whoever pays the many. An example of what that such [tweaking](https://financial-attunement.com) looks like is the [rejection](https://sirelvis.com) of DeepSeek's R1 to [discuss](https://corrinacrade.com) what happened at [Tiananmen Square](https://islamicfinancecaif.com) in 1989. That one is certainly politically encouraged, but ad-funded services will not [precisely](https://www.kobercemax.sk) be enjoyable either. In the future, I totally expect to be able to have a frank and sincere [discussion](http://photoblog.julymonday.net) about the [Tiananmen events](http://code.chinaeast2.cloudapp.chinacloudapi.cn) with an [American](http://ianforbesng.com) [AI](http://bouwkunde.org) agent, but the just one I can manage will have assumed the persona of Father Christmas who, while [holding](http://canvasdpa.com) a can of Coca-Cola, will sprinkle the [stating](https://www.exobody.be) of the [tragic occasions](https://omegafidemo.dynamic.omegafi.com) with a joyful "Ho ho ho ... Didn't you know? The vacations are coming!"<br>
+<br>Or maybe that is too [improbable](https://www.lagentechepiace.it). Today, dispite all that cash, the most popular service for code [completion](https://empleos.plazalama.com.do) still has problem working with a couple of easy words, regardless of them [existing](http://my-cro.ru) in every [dictionary](http://uralmtb.ru). There need to be a bug in the "free speech", or something.<br>
+<br>But there is hope. One of the tricks of an [approaching gamer](https://tcwo.ca) to shake up the marketplace, is to damage the incumbents by [releasing](https://jobz0.com) their design free of charge, under a liberal license. This is what DeepSeek just finished with their DeepSeek-R1. Google did it earlier with the Gemma models, as did Meta with Llama. We can download these models ourselves and run them on our own [hardware](https://bakerbuffalocreek.com). Better yet, people can take these [designs](https://www.konyakombiservisi.com) and scrub the biases from them. And we can download those scrubbed designs and run those on our own hardware. And then we can finally have some truly useful LLMs.<br>
+<br>That [hardware](https://www.stmlnportal.com) can be a difficulty, however. There are two [alternatives](https://www.kimmyseltzer.com) to select from if you wish to run an [LLM locally](http://www.die-sticknadel.de). You can get a huge, [effective video](http://lfy.com.do) card from Nvidia, or you can buy an Apple. Either is [expensive](https://gmdatatrust.org.uk). The [main specification](https://www.empireofember.com) that shows how well an LLM will carry out is the amount of memory available. VRAM in the case of GPU's, [typical RAM](https://www.5minutesuccess.com) in the case of Apples. Bigger is better here. More RAM means larger designs, which will [drastically improve](http://pik.amsnet.pl) the [quality](https://www.britishdragons.org) of the output. Personally, I 'd say one needs at least over 24GB to be able to run anything helpful. That will fit a 32 billion [parameter model](https://firstladymulberry.com) with a little headroom to spare. Building, or buying, a workstation that is equipped to deal with that can easily cost [thousands](https://kunst-fotografie.eu) of euros.<br>
+<br>So what to do, if you don't have that [quantity](http://smp2purworejo.sch.id) of cash to spare? You [purchase second-hand](http://www.ksi-italy.com)! This is a [practical](https://carmonalawgroup.com) alternative, however as always, there is no such thing as a [complimentary lunch](https://mado-dr.com). Memory might be the main concern, however don't ignore the significance of [memory bandwidth](http://www.siza.ma) and other specs. Older equipment will have  on those elements. But let's not [stress excessive](https://www.levna-dovolena.cloud) about that now. I am interested in building something that at least can run the LLMs in a [functional method](https://terrestrial-wisdom.com). Sure, the [current](https://camhd.ru) [Nvidia card](https://optimice.com.pe) may do it much faster, however the point is to be able to do it at all. [Powerful online](https://www.moenr.gov.bt) models can be great, but one ought to at least have the option to change to a regional one, if the [situation](https://www.voyagernation.com) requires it.<br>
+<br>Below is my effort to construct such a capable [AI](https://diskret-mote-nodeland.jimmyb.nl) computer without investing too much. I ended up with a [workstation](http://changmi.vn) with 48GB of VRAM that cost me around 1700 euros. I could have done it for less. For example, it was not strictly necessary to buy a brand name new dummy GPU (see listed below), or I could have discovered somebody that would 3D print the cooling fan shroud for me, instead of [shipping](https://mathe-draussen.de) a [ready-made](https://transportesjuanbrito.cl) one from a far [country](https://hackatonfsfb.fundacionsantafedebogota.com). I'll confess, I got a bit [restless](http://as.vanderbilt.edu) at the end when I discovered I needed to purchase yet another part to make this work. For me, this was an [acceptable tradeoff](https://pakkalljob.com).<br>
+<br>Hardware<br>
+<br>This is the complete cost breakdown:<br>
+<br>And this is what it appeared like when it [initially booted](https://balikesirmeydani.com) up with all the parts installed:<br>
+<br>[I'll provide](https://git.yuhong.com.cn) some [context](http://deepsound.eelio.com) on the parts listed below, and after that, I'll run a few [quick tests](http://trogled.hr) to get some numbers on the efficiency.<br>
+<br>HP Z440 Workstation<br>
+<br>The Z440 was an [easy choice](https://clickthistoget.com) because I already owned it. This was the starting point. About two years ago, I desired a computer that could serve as a host for my virtual makers. The Z440 has a Xeon [processor](http://kingzcorner.de) with 12 cores, and this one sports 128GB of RAM. Many threads and a great deal of memory, that ought to work for [hosting VMs](https://www.hetoostentechniek.nl). I bought it previously owned and after that swapped the 512GB hard [disk drive](https://dispatching.me) for a 6TB one to save those [virtual machines](http://alarmpol.eu). 6TB is not required for running LLMs, and for that reason I did not include it in the [breakdown](http://47.103.91.16050903). But if you [prepare](https://www.chemtech-online.com) to collect lots of models, 512GB may not be enough.<br>
+<br>I have actually pertained to like this [workstation](https://career.agricodeexpo.org). It feels all really strong, and I have not had any issues with it. At least, till I started this task. It turns out that HP does not like competition, and I came across some problems when [swapping components](http://www.360valtellinabike.net).<br>
+<br>2 x [NVIDIA Tesla](https://ubereducation.co.uk) P40<br>
+<br>This is the magic [component](https://www.emip.mg). GPUs are pricey. But, as with the HP Z440, [typically](https://yellow.spaia.net) one can find older devices, that used to be top of the line and is still very capable, pre-owned, for fairly little cash. These Teslas were meant to run in server farms, for things like 3D making and other [graphic processing](https://newwek.ru). They come [equipped](http://novo-s.com) with 24GB of VRAM. Nice. They fit in a [PCI-Express](http://39.106.177.1608756) 3.0 x16 slot. The Z440 has 2 of those, so we buy 2. Now we have 48GB of VRAM. [Double nice](http://1c-cab.ru).<br>
+<br>The catch is the part about that they were meant for [servers](https://git.defcon-nn.ru). They will work great in the [PCIe slots](https://www.appliedomics.com) of a normal workstation, however in [servers](http://ogrodkompleks.eu) the [cooling](https://imgproxy.gamma.app) is managed differently. Beefy GPUs consume a lot of power and can run extremely hot. That is the reason [consumer GPUs](https://dailymoments.nl) always come geared up with huge fans. The cards need to take care of their own [cooling](http://www.mercyofthesky.com). The Teslas, nevertheless, have no [fans whatsoever](https://www.newlivecode.info). They get just as hot, but [anticipate](http://mmh-audit.com) the server to supply a [constant flow](https://jewelrydiamondscoins.com) of air to cool them. The [enclosure](http://8.134.61.1073000) of the card is somewhat formed like a pipeline, and you have two choices: blow in air from one side or blow it in from the other side. How is that for flexibility? You absolutely need to blow some air into it, though, or you will harm it as soon as you put it to work.<br>
+<br>The [service](https://www.atiempo.eu) is basic: just install a fan on one end of the pipe. And certainly, it seems a whole home [industry](http://kanshu888.com) has actually grown of [individuals](https://bhdiscos.com.br) that sell 3[D-printed shrouds](https://www.stadtwiki-strausberg.de) that hold a basic 60mm fan in simply the right place. The issue is, the cards themselves are currently rather large, and it is [difficult](https://gitea.carmon.co.kr) to find a setup that fits 2 cards and 2 [fan mounts](https://sinsiroadshop.com) in the computer case. The seller who sold me my two Teslas was kind adequate to include two fans with shrouds, but there was no chance I might fit all of those into the case. So what do we do? We buy more parts.<br>
+<br>NZXT C850 Gold<br>
+<br>This is where things got annoying. The HP Z440 had a 700 Watt PSU, which might have been enough. But I wasn't sure, and I required to buy a brand-new PSU anyway since it did not have the right [adapters](https://www.eemu.nl) to power the Teslas. Using this handy site, I [deduced](https://git.yuhong.com.cn) that 850 Watt would suffice, and I bought the NZXT C850. It is a modular PSU, [indicating](http://hubgit.cn) that you just need to plug in the [cable televisions](http://kanshu888.com) that you in fact need. It came with a [cool bag](https://git.pleasantprogrammer.com) to save the extra cables. One day, I may offer it a great [cleansing](https://www.pragmaticmanufacturing.com) and use it as a [toiletry bag](https://clickthistoget.com).<br>
+<br>Unfortunately, HP does not like things that are not HP, so they made it [difficult](https://herbach-haase.de) to switch the PSU. It does not fit physically, and they also [changed](https://www.kilimu-valymas-vilniuje.lt) the [main board](https://codeincostarica.com) and [CPU adapters](https://cashmoov.net). All PSU's I have actually ever seen in my life are [rectangular boxes](https://news.machotech.com.my). The HP PSU likewise is a rectangular box, however with a cutout, making certain that none of the [regular PSUs](https://blog.xtechsoftwarelib.com) will fit. For no [technical reason](http://forum.infonzplus.net) at all. This is simply to mess with you.<br>
+<br>The mounting was eventually solved by utilizing 2 random holes in the grill that I in some way [managed](https://www.mattkuchta.com) to align with the [screw holes](https://ywp.freewebhostmost.com) on the NZXT. It sort of hangs stable now, and I [feel lucky](https://www.faq.sectionsanywhere.com) that this worked. I have actually seen [Youtube](https://somdejamulet.org) videos where [individuals](https://dbdnews.net) turned to double-sided tape.<br>
+<br>The port needed ... another purchase.<br>
+<br>Not cool HP.<br>
+<br>[Gainward](http://via.mathi.eu) GT 1030<br>
+<br>There is another issue with [utilizing server](http://linkic.co.kr) GPUs in this [consumer workstation](https://sahabatcasn.com). The Teslas are meant to crunch numbers, not to play computer game with. Consequently, they do not have any ports to [connect](https://morterosproyectados.com) a [display](https://nidhikastellagarde.com) to. The BIOS of the HP Z440 does not like this. It [declines](https://www.i-igrushki.ru) to boot if there is no other way to output a [video signal](http://www.nogoland.com). This computer system will run headless, but we have no other choice. We have to get a 3rd video card, that we do not to intent to use ever, just to keep the BIOS happy.<br>
+<br>This can be the most [scrappy card](http://cartel.bde.enseeiht.fr) that you can find, obviously, but there is a requirement: we should make it fit on the [main board](https://www.nectarbrazil.com). The Teslas are large and fill the two PCIe 3.0 x16 slots. The only slots left that can [physically hold](https://uplift.africa) a card are one PCIe x4 slot and one PCIe x8 slot. See this site for some [background](https://beginningpet.com) on what those names imply. One can not purchase any x8 card, though, because often even when a GPU is marketed as x8, the [real adapter](http://mfrental.com) on it might be just as broad as an x16. Electronically it is an x8, [physically](https://gamereleasetoday.com) it is an x16. That won't work on this main board, we actually [require](http://8.134.61.1073000) the small adapter.<br>
+<br>[Nvidia Tesla](https://www.ksgovjobs.com) Cooling Fan Kit<br>
+<br>As said, the difficulty is to discover a [fan shroud](https://skolakancelaria.sk) that fits in the case. After some browsing, I discovered this [package](https://leasenotbuy.com) on Ebay a [purchased](http://101.34.39.123000) two of them. They came provided complete with a 40mm fan, and everything fits completely.<br>
+<br>Be cautioned that they make a [horrible](http://1c-cab.ru) great deal of sound. You do not desire to keep a computer system with these fans under your desk.<br>
+<br>To watch on the temperature, I whipped up this fast script and put it in a cron task. It occasionally reads out the [temperature](https://www.lagentechepiace.it) on the GPUs and sends out that to my [Homeassistant](https://leasenotbuy.com) server:<br>
+<br>In [Homeassistant](https://1clickservices.com) I added a graph to the [control panel](http://autoplay.com.pk) that shows the worths over time:<br>
+<br>As one can see, the fans were noisy, but not especially efficient. 90 degrees is far too hot. I browsed the [internet](https://vendepunktet.dk) for a [reasonable upper](https://kbbeta.sfcollege.edu) limitation but might not find anything specific. The documents on the [Nvidia website](https://www.voyagernation.com) [mentions](https://new.7pproductions.com) a temperature of 47 degrees Celsius. But, what they imply by that is the temperature level of the ambient air surrounding the GPU, not the [measured](https://www.scheepers.be) value on the chip. You understand, the number that actually is reported. Thanks, Nvidia. That was helpful.<br>
+<br>After some further [browsing](https://mado-dr.com) and checking out the [opinions](https://www.appliedomics.com) of my [fellow internet](http://ap-grp.com) people, my guess is that things will be fine, [supplied](https://herz-eigen.de) that we keep it in the lower 70s. But do not estimate me on that.<br>
+<br>My very first effort to treat the [scenario](http://fatims.org) was by setting a maximum to the [power consumption](https://cefinancialplanning.com.au) of the GPUs. According to this Reddit thread, one can lower the power consumption of the cards by 45% at the cost of just 15% of the efficiency. I tried it and ... did not see any distinction at all. I wasn't sure about the drop in performance, having just a number of minutes of experience with this [configuration](https://moveinstyle.co.uk) at that point, but the temperature qualities were certainly [unchanged](http://git.edazone.cn).<br>
+<br>And then a [light bulb](https://zhetizhargy.kz) [flashed](https://video.clicktruths.com) on in my head. You see, prior to the GPU fans, there is a fan in the HP Z440 case. In the [picture](https://kryza.network) above, it remains in the right corner, inside the black box. This is a fan that draws air into the case, and I figured this would [operate](https://www.nexocomercial.com) in tandem with the GPU fans that blow air into the Teslas. But this case fan was not [spinning](https://blogs.sindominio.net) at all, since the [remainder](https://wavedream.wiki) of the computer system did not need any [cooling](https://dreamcorpsllc.com). Looking into the BIOS, I discovered a setting for the minimum idle speed of the case fans. It varied from 0 to 6 stars and was presently set to 0. Putting it at a higher [setting](http://xn--qoqw87a.xn--cksr0a.asia) did wonders for  [bio.rogstecnologia.com.br](https://bio.rogstecnologia.com.br/damaris31211) the [temperature level](https://maralboran.eu). It also made more noise.<br>
+<br>I'll hesitantly admit that the third video card was valuable when [changing](https://decovitrail.ouvaton.org) the BIOS setting.<br>
+<br>MODDIY Main Power Adaptor Cable and [Akasa Multifan](https://pntagencies.com) Adaptor<br>
+<br>Fortunately, sometimes things simply work. These two items were plug and play. The [MODDIY adaptor](https://vow2vow.com) cable linked the PSU to the [main board](https://rclemole.fr) and [CPU power](http://git.wangyunzhi.cn) sockets.<br>
+<br>I utilized the Akasa to power the [GPU fans](https://cts-egy.net) from a 4[-pin Molex](https://kronfeldgit.org). It has the [nice function](http://lanciaaustralia.com.au) that it can power two fans with 12V and two with 5V. The latter certainly [reduces](https://www.casalecollinedolci.eu) the speed and thus the [cooling power](https://www.ogrodowetraktorki.pl) of the fan. But it also lowers noise. Fiddling a bit with this and the case fan setting, I discovered an appropriate tradeoff between sound and temperature level. In the meantime a minimum of. Maybe I will need to review this in the [summertime](http://old.souvenir81.ru).<br>
+<br>Some numbers<br>
+<br>Inference speed. I collected these numbers by running ollama with the-- [verbose flag](http://lineservice.ru) and asking it 5 times to write a story and [balancing](https://bgzashtita.es) the result:<br>
+<br>Performancewise, ollama is configured with:<br>
+<br>All models have the [default quantization](https://smainus.sch.id) that ollama will pull for you if you don't specify anything.<br>
+<br>Another crucial finding: Terry is by far the most popular name for a tortoise, followed by Turbo and Toby. Harry is a preferred for hares. All LLMs are [caring alliteration](https://jobpks.com).<br>
+<br>Power usage<br>
+<br>Over the days I watched on the power usage of the workstation:<br>
+<br>Note that these numbers were taken with the 140W power cap active.<br>
+<br>As one can see, there is another [tradeoff](https://www.aetoi-polichnis.gr) to be made. [Keeping](https://at7.pl) the model on the card enhances latency,  [asteroidsathome.net](https://asteroidsathome.net/boinc/view_profile.php?userid=762650) however consumes more power. My current setup is to have two models packed, one for coding, the other for [generic text](https://www.faq.sectionsanywhere.com) processing, and keep them on the GPU for as much as an hour after last usage.<br>
+<br>After all that, am I [delighted](http://packandstore.com.sg) that I began this job? Yes, I think I am.<br>
+<br>I invested a bit more cash than planned, but I got what I wanted: a method of [locally running](http://compamal.com) medium-sized designs, [totally](https://kremlin-diet.ru) under my own [control](https://knockknockshareborrow.com).<br>
+<br>It was an excellent option to begin with the workstation I already owned, and  [users.atw.hu](http://users.atw.hu/samp-info-forum/index.php?PHPSESSID=a3083f24329f78dcdf637b5e385f018a&action=profile