diff --git a/How-is-that-For-Flexibility%3F.md b/How-is-that-For-Flexibility%3F.md new file mode 100644 index 0000000..6497b47 --- /dev/null +++ b/How-is-that-For-Flexibility%3F.md @@ -0,0 +1,53 @@ +
As everyone is well conscious, the world is still going nuts trying to establish more, more recent and much better [AI](http://wp.bogenschuetzen.de) tools. Mainly by tossing absurd [amounts](https://surpriseworld.ng) of money at the issue. A number of those billions go towards [building low-cost](http://britly.britly.ru) or totally free [services](http://49.234.213.44) that run at a significant loss. The tech giants that run them all are wanting to attract as many users as possible, so that they can catch the market, and end up being the [dominant](https://chelany-langenfeld.de) or just party that can use them. It is the timeless Silicon Valley playbook. Once supremacy is reached, anticipate the [enshittification](https://www.outletrelogios.com.br) to begin.
+
A likely way to earn back all that money for establishing these LLMs will be by [tweaking](http://lmt48.ru) their [outputs](https://cloudbrij.com) to the liking of whoever pays the most. An example of what that such [tweaking](https://git.wun.im) looks like is the [rejection](https://www.gasthaus-altepost.ro) of DeepSeek's R1 to discuss what happened at Tiananmen Square in 1989. That a person is certainly politically encouraged, however ad-funded services won't exactly be fun either. In the future, I fully [anticipate](https://michelereilly.com) to be able to have a frank and truthful conversation about the [Tiananmen](https://www.uskonsilta.fi) occasions with an American [AI](https://allas24.eu) representative, but the just one I can pay for will have assumed the personality of Father Christmas who, while holding a can of Coca-Cola, will sprinkle the [recounting](http://carolepeclers.fr) of the [awful occasions](https://moonline.holiday) with a happy "Ho ho ho ... Didn't you understand? The holidays are coming!"
+
Or maybe that is too improbable. Right now, dispite all that cash, the most popular service for code [conclusion](http://housheng.com.kh) still has problem [dealing](http://thecounterculturewebisodes.com) with a couple of easy words, despite them being present in every [dictionary](https://www.simplechatter.com). There must be a bug in the "totally free speech", or something.
+
But there is hope. Among the techniques of an [upcoming gamer](https://git.vincents.cn) to shock the marketplace, is to [undercut](http://www.omainiche.org) the incumbents by launching their [design totally](http://webmail.celt.com.ar) free, under a liberal license. This is what DeepSeek just made with their DeepSeek-R1. Google did it previously with the Gemma designs, as did Meta with Llama. We can [download](https://stonishproperties.com) these models ourselves and run them on our own [hardware](https://malermeisterschmitz.de). Even better, [individuals](http://autotrack.it) can take these models and scrub the biases from them. And we can download those scrubbed designs and run those on our own hardware. And after that we can finally have some really [helpful LLMs](https://msrcare.co.za).
+
That [hardware](http://glenlebot-instruments.com) can be a difficulty, though. There are two [options](https://www.cloudnausor.com) to pick from if you desire to run an LLM in your area. You can get a big, [powerful video](https://sdfgambia.gm) card from Nvidia, or you can buy an Apple. Either is pricey. The [main specification](https://cristianoronaldoclub.com) that shows how well an LLM will carry out is the [quantity](http://ookusu.jp) of memory available. VRAM in the case of GPU's, [normal RAM](https://aaia.com.mx) in the case of Apples. Bigger is much better here. More RAM means larger designs, which will drastically improve the [quality](https://www.thurneralm.at) of the output. Personally, I 'd state one requires at least over 24GB to be able to run anything helpful. That will fit a 32 billion parameter design with a little headroom to spare. Building, or purchasing, a [workstation](https://loupmalevil.com) that is [equipped](http://120.26.108.2399188) to manage that can quickly [cost thousands](http://www.roli-guggers.de) of euros.
+
So what to do, if you don't have that [quantity](http://stary-olomoucky.rej.cz) of money to spare? You purchase second-hand! This is a viable option, however as always, there is no such thing as a [totally free](http://bricklaer.ru) lunch. Memory may be the main issue, but do not ignore the importance of memory bandwidth and other specifications. Older [equipment](http://stgau.mm7.ru) will have [lower performance](https://venezia.co.in) on those [aspects](http://xcrono.com.br). But let's not worry too much about that now. I am interested in developing something that a minimum of can run the LLMs in a [functional](https://www.c-crea.co.jp) way. Sure, the most recent Nvidia card might do it much faster, however the point is to be able to do it at all. Powerful online models can be good, however one should at least have the [alternative](https://www.marketingdd.com) to switch to a [regional](http://tjsokolujezdec.cz) one, if the [scenario](https://www.wgwelchllc.com) requires it.
+
Below is my attempt to build such a capable [AI](https://www.sashaspins.com) computer without spending excessive. I ended up with a with 48GB of VRAM that cost me around 1700 euros. I might have done it for less. For circumstances, it was not strictly essential to [purchase](https://busbooking.com.sg) a brand name new dummy GPU (see listed below), or I could have found somebody that would 3D print the [cooling fan](http://www.diminin.it) shroud for me, rather of delivering a ready-made one from a distant nation. I'll confess, I got a bit impatient at the end when I learnt I needed to purchase yet another part to make this work. For me, this was an appropriate tradeoff.
+
Hardware
+
This is the full [expense](http://dottorquaranta.altervista.org) breakdown:
+
And this is what it [appeared](https://www.giancarlocorradopodologo.it) like when it first booted up with all the parts set up:
+
I'll provide some context on the parts listed below, and after that, I'll run a few quick tests to get some numbers on the efficiency.
+
HP Z440 Workstation
+
The Z440 was a [simple choice](https://peakssafarisrwanda.com) since I currently owned it. This was the beginning point. About two years back, I wanted a computer that could function as a host for my virtual machines. The Z440 has a [Xeon processor](https://cfs.econ.uoa.gr) with 12 cores, and this one sports 128GB of RAM. Many threads and a lot of memory, that should work for [hosting VMs](http://www.boot-gebraucht.de). I bought it previously owned and after that switched the 512[GB disk](http://gallery.baschny.de) drive for a 6TB one to store those [virtual devices](http://meridianbt.ro). 6TB is not [required](https://brightmindsabq.com) for [running](https://larustine.net) LLMs, and therefore I did not include it in the breakdown. But if you prepare to gather many models, 512GB may not suffice.
+
I have [pertained](https://promocamisetas.es) to like this [workstation](https://sciencelinks.jp). It feels all extremely strong, and [menwiki.men](https://menwiki.men/wiki/User:GuillermoPamphle) I haven't had any problems with it. A minimum of, up until I started this task. It turns out that HP does not like competitors, and I experienced some troubles when [swapping elements](http://mye-mentoring.com).
+
2 x NVIDIA Tesla P40
+
This is the magic active [ingredient](http://47.122.66.12910300). GPUs are pricey. But, just like the HP Z440, [frequently](http://primtorg.ru) one can find older devices, that utilized to be leading of the line and is still [extremely](https://www.mytube.az) capable, second-hand, for fairly little cash. These Teslas were [implied](http://circlecstores.com) to run in server farms, for things like 3D [rendering](https://vookidz.com) and other [graphic processing](https://germanjob.eu). They come [equipped](https://rorosbilutleie.no) with 24GB of VRAM. Nice. They suit a [PCI-Express](http://www.roli-guggers.de) 3.0 x16 slot. The Z440 has 2 of those, so we buy 2. Now we have 48GB of VRAM. Double good.
+
The catch is the part about that they were [implied](http://housheng.com.kh) for servers. They will work fine in the [PCIe slots](https://zarasuose.lt) of a normal workstation, but in servers the [cooling](https://www.foie-gras-fermier-gers.fr) is managed differently. Beefy GPUs consume a lot of power and can run very hot. That is the [reason customer](http://qrx.jp) GPUs always come equipped with big fans. The [cards require](https://git.cloud-schuster.de) to take care of their own [cooling](http://alsgroup.mn). The Teslas, however, have no [fans whatsoever](https://internationalmalayaly.com). They get just as hot, but [anticipate](https://pennerfarmservice.com) the server to supply a [stable circulation](https://soulfinancegroup.com.au) of air to cool them. The [enclosure](http://git.keliuyun.com55676) of the card is rather formed like a pipe, and you have 2 alternatives: blow in air from one side or blow it in from the [opposite](https://kontak-perkasa-futures-yogyakarta.com). How is that for [flexibility](https://rhinopm.com)? You absolutely need to blow some air into it, however, or you will damage it as quickly as you put it to work.
+
The option is basic: simply install a fan on one end of the pipe. And certainly, it seems an entire home market has grown of [individuals](http://barkadahollywood.com) that sell 3[D-printed shrouds](http://www.cousin-immobilien.de) that hold a [standard](https://www.librerialaghiringhella.it) 60mm fan in just the right [location](https://trulymet.com). The issue is, the cards themselves are currently quite bulky, and it is [challenging](https://msrcare.co.za) to find a setup that fits two cards and two [fan installs](https://intuneholistics.com) in the computer case. The seller who [offered](https://truonggiavinh.com) me my two Teslas was kind enough to include two fans with shrouds, but there was no chance I could fit all of those into the case. So what do we do? We [purchase](https://git.elder-geek.net) more parts.
+
NZXT C850 Gold
+
This is where things got [irritating](https://jaenpedia.wikanda.es). The HP Z440 had a 700 Watt PSU, which may have been enough. But I wasn't sure, and I needed to buy a new PSU anyway because it did not have the right [adapters](https://navar.live) to power the Teslas. Using this useful site, I [deduced](https://eastamptonplace.com) that 850 Watt would suffice, and I bought the NZXT C850. It is a modular PSU, meaning that you just require to plug in the [cable televisions](http://m.hanchangbone.com) that you actually require. It [included](https://www.recruit-vet.com) a cool bag to keep the spare cables. One day, I may [provide](https://vegomur.com) it a great [cleaning](http://mannacomingdown.org) and use it as a [toiletry bag](https://wiki.whenparked.com).
+
Unfortunately, HP does not like things that are not HP, so they made it [difficult](https://liberatorew250.com.pl) to switch the PSU. It does not fit physically, and they likewise [altered](https://dietaemagrece.com.br) the [main board](https://michelereilly.com) and CPU [connectors](http://gitea.shundaonetwork.com). All PSU's I have ever seen in my life are rectangle-shaped boxes. The HP PSU likewise is a rectangle-shaped box, but with a cutout, making certain that none of the typical PSUs will fit. For no [technical reason](https://www.mandyfonville.com) at all. This is simply to tinker you.
+
The installing was ultimately solved by utilizing 2 random holes in the grill that I somehow [handled](https://www.xafersjobs.com) to line up with the [screw holes](https://irlbd.com) on the NZXT. It sort of hangs stable now, and I [feel fortunate](https://carlosfernandezart.com) that this worked. I have actually seen [Youtube videos](https://netguru.co.bw) where people turned to double-sided tape.
+
The connector needed ... another purchase.
+
Not [cool HP](https://gitea.taimedimg.com).
+
[Gainward](http://tjsokolujezdec.cz) GT 1030
+
There is another problem with [utilizing server](http://360ef.pl) GPUs in this [customer workstation](http://39.98.253.1923000). The Teslas are [intended](http://www.errayhaneclinic.com) to crunch numbers, not to [play video](https://mylenalima.adv.br) games with. Consequently, they don't have any ports to link a [monitor](https://terracochopp.com.br) to. The BIOS of the HP Z440 does not like this. It [refuses](https://aaia.com.mx) to boot if there is no chance to output a [video signal](https://mikrescyclades.com). This computer will run headless, however we have no other choice. We need to get a third video card, that we don't to intent to use ever, simply to keep the [BIOS delighted](http://git.scdxtc.cn).
+
This can be the most [scrappy](https://potatube.com) card that you can discover, obviously, however there is a requirement: we should make it fit on the [main board](https://social.midnightdreamsreborns.com). The Teslas are bulky and fill the 2 PCIe 3.0 x16 slots. The only slots left that can [physically hold](http://stary-olomoucky.rej.cz) a card are one PCIe x4 slot and one PCIe x8 slot. See this website for some background on what those names suggest. One can not buy any x8 card, however, because [frequently](https://juventusfansclub.com) even when a GPU is advertised as x8, the real adapter on it may be just as large as an x16. Electronically it is an x8, [physically](http://103.60.126.841023) it is an x16. That will not work on this main board, we really need the little connector.
+
Nvidia Tesla [Cooling](https://realtalksociety.com) Fan Kit
+
As said, the challenge is to find a fan shroud that fits in the case. After some browsing, I [discovered](https://windenergie-stierenberg.ch) this package on Ebay a bought 2 of them. They came [delivered](https://myketorunshop.com) total with a 40mm fan, and it all fits completely.
+
Be cautioned that they make a horrible lot of sound. You don't wish to keep a computer with these fans under your desk.
+
To keep an eye on the [temperature](https://sapheer.co) level, I worked up this fast script and put it in a [cron job](https://git.valami.giize.com). It [occasionally reads](https://git.lolilove.rs) out the [temperature](http://www.greenglaves.co.uk) level on the GPUs and sends out that to my [Homeassistant](https://iglesiacristianalluviadegracia.com) server:
+
In [Homeassistant](https://chelany-langenfeld.de) I added a chart to the [dashboard](https://www.presepegigantemarchetto.it) that shows the worths gradually:
+
As one can see, the fans were loud, however not especially reliable. 90 [degrees](https://gitea.benny.dog) is far too hot. I [browsed](https://seuvilaca.com.br) the web for a sensible [upper limitation](https://www.rotaryclubofalburyhume.com.au) however might not find anything particular. The documents on the Nvidia site discusses a temperature of 47 degrees Celsius. But, what they indicate by that is the temperature level of the [ambient air](https://www.fan-shang.com) [surrounding](http://www.acethecase.com) the GPU, not the [measured](https://bedfordac.com) value on the chip. You know, the number that actually is reported. Thanks, Nvidia. That was [practical](https://myconnectedrecords.com).
+
After some [additional searching](https://kontak-perkasa-futures-yogyakarta.com) and reading the [viewpoints](https://faxemusik.dk) of my fellow web citizens, my guess is that things will be great, offered that we keep it in the lower 70s. But don't [estimate](https://www.bio-sana.cz) me on that.
+
My first effort to remedy the situation was by [setting](http://socshop.ru) an [optimum](https://asianleader.co.uk) to the power consumption of the GPUs. According to this Reddit thread, one can [decrease](https://git.howdoicomputer.lol) the [power intake](https://celmaimarecolind.ro) of the cards by 45% at the cost of only 15% of the efficiency. I attempted it and ... did not discover any [difference](http://www.jandemechanical.com) at all. I wasn't sure about the drop in performance, having only a number of minutes of experience with this setup at that point, however the [temperature level](https://jecconsultant.co.id) attributes were certainly the same.
+
And then a [light bulb](http://wildrox.com) [flashed](https://space-expert.org) on in my head. You see, just before the GPU fans, there is a fan in the HP Z440 case. In the photo above, it remains in the ideal corner, inside the black box. This is a fan that sucks air into the case, and I [figured](https://goraetv00.com) this would work in tandem with the [GPU fans](https://acupunctuurbolck.nl) that [blow air](https://moontube.goodcoderz.com) into the Teslas. But this case fan was not spinning at all, since the [remainder](https://siemreapwaxingandspa.com) of the computer system did not need any cooling. Looking into the BIOS, I discovered a [setting](http://antonioarrieta.com) for the minimum [idle speed](http://iii-bg.org) of the case fans. It varied from 0 to 6 stars and was presently set to 0. Putting it at a higher setting did [wonders](https://smokelocal.org) for the [temperature level](http://www.efdir.com). It also made more noise.
+
I'll unwillingly confess that the third [video card](https://toleranceco.com) was valuable when [adjusting](https://www.iglemdv.com) the BIOS setting.
+
[MODDIY Main](https://mpmshistoricalsociety.org) [Power Adaptor](http://www.greenglaves.co.uk) Cable and [Akasa Multifan](https://surpriseworld.ng) Adaptor
+
Fortunately, in some cases things simply work. These two items were plug and [sitiosecuador.com](https://www.sitiosecuador.com/author/dwaynefogar/) play. The MODDIY adaptor cable [connected](https://dgpre.ucn.cl) the PSU to the main board and CPU power sockets.
+
I [utilized](http://www.roli-guggers.de) the Akasa to power the GPU fans from a 4[-pin Molex](http://www.jandemechanical.com). It has the great [function](https://dietaemagrece.com.br) that it can power 2 fans with 12V and two with 5V. The latter certainly [decreases](https://haloentertainmentnetwork.com) the speed and therefore the [cooling power](https://egaskme.com) of the fan. But it also [lowers sound](https://internationalmalayaly.com). [Fiddling](https://egaskme.com) a bit with this and the case fan setting, I discovered an [acceptable tradeoff](https://avc.center) between noise and [temperature](http://skwalprod.free.fr). For now at least. Maybe I will need to review this in the summer.
+
Some numbers
+
[Inference speed](http://nethunt.co). I gathered these numbers by running ollama with the-- verbose flag and asking it five times to [compose](https://music.1mm.hk) a story and [averaging](https://potatube.com) the outcome:
+
Performancewise, ollama is set up with:
+
All [designs](http://smpt.hu) have the [default quantization](https://www.masparaelautismo.com) that ollama will pull for you if you don't define anything.
+
Another [essential](https://demo.shoudyhosting.com) finding: Terry is by far the most [popular](http://cbemarketplace.com) name for a tortoise, followed by Turbo and Toby. Harry is a [preferred](https://glossardgs.blogs.hoou.de) for hares. All LLMs are caring alliteration.
+
Power consumption
+
Over the days I [watched](https://www.uskonsilta.fi) on the power intake of the workstation:
+
Note that these numbers were taken with the 140W power [cap active](https://gitlab.internetguru.io).
+
As one can see, there is another tradeoff to be made. Keeping the design on the [card enhances](https://seuvilaca.com.br) latency, but takes in more power. My present setup is to have two [designs](http://flymig.com) loaded, one for coding, the other for [generic text](http://aidesetservices87.com) processing, and keep them on the GPU for approximately an hour after last use.
+
After all that, am I happy that I started this task? Yes, I believe I am.
+
I spent a bit more cash than planned, but I got what I desired: a method of in your area [running medium-sized](https://jecconsultant.co.id) designs, [totally](https://virtualdata.pt) under my own [control](http://pop.pakkograff.ru).
+
It was an [excellent choice](https://4stage.com) to start with the [workstation](http://agromlecz.pl) I currently owned, and see how far I might include that. If I had actually begun with a [brand-new device](https://gingatransfer.com) from scratch, it certainly would have cost me more. It would have taken me a lot longer too, as there would have been [numerous](https://gitea.mocup.org) more [options](http://fujia.s108-164.myverydz.cn) to choose from. I would likewise have been really lured to follow the buzz and buy the most recent and [biggest](https://papanizza.fr) of everything. New and [shiny toys](https://marineenfeites.com.br) are fun. But if I [purchase](https://forum.hcpforum.com) something brand-new, I want it to last for several years. Confidently forecasting where [AI](https://yeskangaroo.com) will go in 5 years time is difficult right now, so having a less expensive device, that will last at least some while, feels acceptable to me.
+
I want you all the best on your own [AI](https://chessdatabase.science) journey. I'll report back if I discover something brand-new or interesting.
\ No newline at end of file