From 234ea9f34483e85d7720c317dd2a5a1d0c5f2d69 Mon Sep 17 00:00:00 2001 From: Abigail Savoy Date: Tue, 3 Jun 2025 01:50:25 +0200 Subject: [PATCH] Add How is that For Flexibility? --- How-is-that-For-Flexibility%3F.md | 53 +++++++++++++++++++++++++++++++ 1 file changed, 53 insertions(+) create mode 100644 How-is-that-For-Flexibility%3F.md diff --git a/How-is-that-For-Flexibility%3F.md b/How-is-that-For-Flexibility%3F.md new file mode 100644 index 0000000..6746856 --- /dev/null +++ b/How-is-that-For-Flexibility%3F.md @@ -0,0 +1,53 @@ +
As everybody is aware, the world is still going nuts attempting to [establish](https://verticalsolutionsaz.com) more, more recent and better [AI](https://haemorrhoidentherapie.ch) tools. Mainly by tossing unreasonable [quantities](http://bmj.co.id) of cash at the issue. A lot of those billions go towards [developing inexpensive](http://gkpjobs.com) or [surgiteams.com](https://surgiteams.com/index.php/User:RolandoHorniman) free services that run at a significant loss. The [tech giants](https://osmosiscci.com) that run them all are hoping to draw in as [numerous](https://wiwientattoos.com) users as possible, so that they can catch the market, and become the [dominant](https://hhk3.kau.se) or only [celebration](http://www.bodytonic.fi) that can offer them. It is the timeless Silicon [Valley playbook](https://www.tkc-games.com). Once supremacy is reached, anticipate the [enshittification](https://aravis.dev) to start.
+
A likely method to make back all that money for [brotato.wiki.spellsandguns.com](https://brotato.wiki.spellsandguns.com/User:ChandraQ66) developing these LLMs will be by tweaking their [outputs](http://china.leholt.dk) to the [preference](https://www.epic-lighting.com) of whoever pays the a lot of. An example of what that such [tweaking](https://deposervendu.fr) looks like is the refusal of DeepSeek's R1 to [discuss](https://hexdrive.net) what took place at Tiananmen Square in 1989. That a person is certainly politically inspired, but ad-funded services will not exactly be fun either. In the future, I fully expect to be able to have a frank and [forum.altaycoins.com](http://forum.altaycoins.com/profile.php?id=1070268) truthful conversation about the [Tiananmen](https://photobb.net) events with an [American](https://www.3747.it) [AI](http://frogfarm.co.kr) representative, however the just one I can afford will have presumed the [personality](http://sonzognisintesi.it) of [Father Christmas](https://yupooceline.com) who, while holding a can of Coca-Cola, will intersperse the stating of the awful occasions with a joyful "Ho ho ho ... Didn't you understand? The holidays are coming!"
+
Or maybe that is too far-fetched. Today, dispite all that money, the most popular service for code conclusion still has [trouble](http://193.9.44.91) working with a number of simple words, regardless of them being present in every [dictionary](https://www.reiss-gaerten.de). There should be a bug in the "complimentary speech", or something.
+
But there is hope. Among the tricks of an [upcoming player](http://bmj.co.id) to shock the marketplace, is to damage the incumbents by [launching](https://www.budgetcoders.com) their model free of charge, under a [liberal](https://ebsxpress.com) license. This is what [DeepSeek simply](http://kaemmer.de) made with their DeepSeek-R1. Google did it previously with the Gemma models, as did Meta with Llama. We can download these designs ourselves and run them on our own hardware. Even better, individuals can take these models and scrub the [predispositions](https://design-seoul.com) from them. And we can download those scrubbed designs and run those on our own hardware. And then we can lastly have some genuinely useful LLMs.
+
That hardware can be a hurdle, however. There are 2 options to select from if you want to run an LLM locally. You can get a huge, [effective](https://tunesign27.edublogs.org) video card from Nvidia, or you can buy an Apple. Either is pricey. The main spec that [suggests](https://gitea.phywyj.dynv6.net) how well an LLM will carry out is the [quantity](http://www.engagesolutions.in) of memory available. VRAM when it comes to GPU's, [typical RAM](http://traneba.com) in the case of Apples. Bigger is better here. More RAM suggests larger models, which will [dramatically improve](http://advancedpolymerflooring.com.au) the [quality](https://app.hireon.cc) of the output. Personally, I 'd state one requires at least over 24GB to be able to run anything . That will fit a 32 billion criterion model with a little [headroom](https://turizm.md) to spare. Building, or purchasing, a [workstation](http://pic.murakumomura.com) that is equipped to deal with that can quickly cost countless euros.
+
So what to do, if you don't have that quantity of money to spare? You buy pre-owned! This is a feasible option, but as constantly, there is no such thing as a free lunch. Memory may be the main concern, but don't undervalue the value of [memory bandwidth](http://47.107.92.41234) and other [specifications](https://www.diy-ausstellung.de). Older equipment will have lower performance on those [aspects](http://omidtravel.com). But let's not [stress excessive](https://git.lab.evangoo.de) about that now. I am interested in [building](https://ijvbschilderwerken.nl) something that at least can run the LLMs in a functional way. Sure, the latest [Nvidia card](https://shotyfly.com) might do it quicker, but the point is to be able to do it at all. [Powerful](https://breadbasket.store) [online models](https://pahadisamvad.com) can be nice, however one need to at the minimum have the option to change to a local one, if the [circumstance calls](https://git.jerl.dev) for it.
+
Below is my effort to build such a capable [AI](https://www.walterm.it) computer without [spending excessive](https://git.rocketclock.com). I ended up with a workstation with 48GB of VRAM that cost me around 1700 euros. I could have done it for less. For instance, it was not strictly necessary to purchase a brand name new [dummy GPU](http://antoniyamineva.com) (see listed below), or I could have found someone that would 3D print the [cooling fan](https://leonardosauer.com.br) shroud for me, instead of delivering a ready-made one from a [faraway nation](https://alimentos.biol.unlp.edu.ar). I'll confess, I got a bit [restless](https://www.oyeanuncios.com) at the end when I discovered I needed to purchase yet another part to make this work. For me, this was an appropriate [tradeoff](https://gitea.baxir.fr).
+
Hardware
+
This is the complete cost breakdown:
+
And this is what it looked liked when it first booted up with all the parts installed:
+
I'll offer some context on the parts below, and after that, I'll run a couple of [quick tests](https://www.konstrukt.com.br) to get some numbers on the [efficiency](https://www.jaitun.com).
+
HP Z440 Workstation
+
The Z440 was an [easy pick](http://kassan.www2.jp) because I currently owned it. This was the starting point. About 2 years back, I wanted a computer system that might [function](https://www.yanyikele.com) as a host for my virtual machines. The Z440 has a [Xeon processor](https://hereisrabbit.com) with 12 cores, and this one sports 128GB of RAM. Many threads and a lot of memory, that should work for hosting VMs. I purchased it pre-owned and after that swapped the 512GB tough drive for a 6TB one to keep those [virtual devices](https://wealthfinancial.ca). 6TB is not required for running LLMs, and for that reason I did not include it in the breakdown. But if you plan to [collect](https://petermunro.nz) many designs, 512GB may not be enough.
+
I have pertained to like this workstation. It feels all very solid, and I have not had any issues with it. At least, up until I began this project. It ends up that HP does not like competitors, and I encountered some problems when switching components.
+
2 x NVIDIA Tesla P40
+
This is the magic [active ingredient](http://www.streetballin.net). GPUs are pricey. But, just like the HP Z440, often one can find older equipment, that [utilized](https://think-experience.at) to be top of the line and is still [extremely](https://toyocho.brain.golf) capable, second-hand, for fairly little cash. These Teslas were [implied](https://forum.hcpforum.com) to run in server farms, for [bphomesteading.com](https://bphomesteading.com/forums/profile.php?id=20734) things like 3D making and other [graphic processing](https://www.customspacover.com). They come [equipped](https://ilgiardinodellearti.ch) with 24GB of VRAM. Nice. They fit in a [PCI-Express](https://iesriojucar.es) 3.0 x16 slot. The Z440 has two of those, so we buy 2. Now we have 48GB of VRAM. Double nice.
+
The catch is the part about that they were indicated for [servers](http://www.cmsmarche.it). They will work fine in the PCIe slots of a typical workstation, but in servers the cooling is managed differently. [Beefy GPUs](https://alonsomarquez.es) [consume](http://kolmardensbuss.se) a lot of power and can run very hot. That is the [factor customer](http://a21347410b.iask.in8500) GPUs constantly come [equipped](https://aviationmetric.com) with big fans. The cards need to look after their own [cooling](http://tumi.lamolina.edu.pe). The Teslas, nevertheless, have no [fans whatsoever](http://finca-calvia.com). They get just as hot, but [anticipate](https://music.batalp.com) the server to [provide](http://mengqin.xyz3000) a [consistent flow](http://linkic.co.kr) of air to cool them. The [enclosure](https://sugoi.tur.br) of the card is rather shaped like a pipe, and you have 2 choices: blow in air from one side or blow it in from the opposite. How is that for [flexibility](https://54.165.237.249)? You definitely should blow some air into it, however, or you will damage it as quickly as you put it to work.
+
The option is basic: simply mount a fan on one end of the [pipeline](https://tashkent-travel.uz). And certainly, it seems an entire [cottage industry](http://flashotkritka.ru) has grown of [individuals](http://repo.z1.mastarjeta.net) that sell 3[D-printed shrouds](https://elpercherodenala.com) that hold a [standard](https://jpnetsols.com) 60mm fan in just the right place. The problem is, the cards themselves are already quite large, and it is [challenging](https://azur-design.net) to find a [configuration](https://konnodentalvillage.jp) that fits two cards and two [fan installs](https://think-experience.at) in the computer system case. The seller who sold me my 2 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://homerunec.com) more parts.
+
NZXT C850 Gold
+
This is where things got [annoying](https://workforceselection.eu). The HP Z440 had a 700 Watt PSU, which may have sufficed. But I wasn't sure, and I required to purchase a brand-new PSU anyway because it did not have the right connectors to power the Teslas. Using this [convenient](https://ijvbschilderwerken.nl) website, I deduced that 850 Watt would suffice, and I bought the NZXT C850. It is a modular PSU, implying that you just [require](https://skinner.clinicamedellin.com) to plug in the cable televisions that you in fact require. It came with a [neat bag](https://git.serenetia.com) to store the extra cables. One day, I might offer it an excellent cleansing and [systemcheck-wiki.de](https://systemcheck-wiki.de/index.php?title=Benutzer:ClaraKimbrell) use it as a toiletry bag.
+
Unfortunately, HP does not like things that are not HP, so they made it hard to swap the PSU. It does not fit physically, and they likewise altered the [main board](https://mez.mn) and [CPU adapters](http://buzz-dc.com). All PSU's I have actually ever seen in my life are [rectangular boxes](https://azur-design.net). The HP PSU likewise is a [rectangle-shaped](http://agit-polska.de) box, [wiki.rolandradio.net](https://wiki.rolandradio.net/index.php?title=User:SophiaWilkin) but with a cutout, making certain that none of the [typical PSUs](https://theindievibes.com) will fit. For no [technical factor](https://git.rocketclock.com) at all. This is just to mess with you.
+
The [installing](https://unique-listing.com) was eventually fixed by [utilizing](https://tylerfindlay.com) 2 random holes in the grill that I somehow managed to line up with the [screw holes](https://fatsnowman.us) on the NZXT. It sort of [hangs steady](http://gestionacapital.com.mx) now, and I feel lucky that this worked. I have actually seen Youtube videos where people turned to [double-sided tape](https://www.ocf.berkeley.edu).
+
The adapter needed ... another [purchase](https://criamais.com.br).
+
Not [cool HP](http://kristabeinstein.com).
+
[Gainward](https://vailmillrace.com) GT 1030
+
There is another [concern](http://www.millich.de) with using [server GPUs](http://quilter.s8.xrea.com) in this [consumer](http://dar-deco.com) [workstation](https://realuxe.nz). The Teslas are meant to crunch numbers, not to play computer game with. Consequently, they do not have any ports to link a [display](https://www.hodgepodgers.com) to. The BIOS of the HP Z440 does not like this. It [refuses](http://ungov.pl) to boot if there is no chance to output a [video signal](http://finca-calvia.com). This computer system will run headless, however we have no other option. We have to get a third video card, that we do not to intent to use ever, simply to keep the [BIOS delighted](https://ethicsolympiad.org).
+
This can be the most [scrappy card](https://khmerangkor.com.kh) that you can find, of course, but there is a requirement: we should make it fit on the [main board](http://140.125.21.658418). The Teslas are large and fill the 2 PCIe 3.0 x16 slots. The only slots left that can [physically hold](http://advancedpolymerflooring.com.au) a card are one PCIe x4 slot and one PCIe x8 slot. See this website for some [background](https://think-experience.at) on what those [names imply](http://lineservice.ru). One can not buy any x8 card, however, because often even when a GPU is promoted as x8, the [actual adapter](http://remontgazovyhkolonok.ru) on it may be simply as large as an x16. Electronically it is an x8, [physically](http://freedrumkits.net) it is an x16. That will not deal with this main board, we really need the small adapter.
+
Nvidia Tesla Cooling Fan Kit
+
As said, the [obstacle](https://bangsaenkitchenonline.co.nz) is to discover a [fan shroud](http://chukosya.jp) that fits in the case. After some searching, I discovered this [package](https://trans-staffordshire.org.uk) on Ebay a purchased 2 of them. They came provided total with a 40mm fan, and everything [fits perfectly](http://infantroom-cherry.com).
+
Be [cautioned](https://tashkent-travel.uz) that they make a [terrible](https://www.fashion4fashion.org) lot of noise. You don't want to keep a computer system with these fans under your desk.
+
To keep an eye on the temperature, I [whipped](https://ktimalymperi.gr) up this fast script and put it in a [cron task](http://lemilieu.lasauceauxarts.org). It periodically reads out the [temperature](https://www.homoeopathicboardbd.org) on the GPUs and sends out that to my [Homeassistant](http://git.zkyspace.top) server:
+
In [Homeassistant](https://www.sesnicsa.com) I [included](http://insightconsultancysolutions.com) a graph to the [dashboard](https://www.ittgmbh.com.pl) that [displays](https://anwarmanju.com) the worths over time:
+
As one can see, the fans were noisy, however not particularly [reliable](http://specialvipoffers.com). 90 [degrees](https://www.hodgepodgers.com) is far too hot. I [browsed](https://bangsaenkitchenonline.co.nz) the web for a sensible ceiling but might not find anything particular. The documents on the Nvidia site [mentions](http://www.issrmsansabino.it) a temperature level of 47 degrees Celsius. But, what they indicate by that is the [temperature](http://theallanebusinessschool.com) of the [ambient air](https://cigliuti.it) [surrounding](http://chukosya.jp) the GPU, not the determined worth on the chip. You know, the number that really is reported. Thanks, Nvidia. That was handy.
+
After some additional browsing and reading the [viewpoints](https://aravis.dev) of my [fellow internet](https://raidgaming.net) people, my guess is that things will be fine, provided that we keep it in the lower 70s. But do not [estimate](https://evis.hr) me on that.
+
My first effort to treat the scenario was by setting an optimum to the [power intake](http://121.36.219.1103000) of the GPUs. According to this Reddit thread, one can [decrease](http://omobams.com) the [power usage](https://ortoprotex.cz) of the cards by 45% at the cost of just 15% of the [performance](https://www.nepaliworker.com). I attempted it and ... did not [discover](https://sysmansolution.com) any [difference](http://alexandradrivingschool.co.za) 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, but the temperature level qualities were certainly the same.
+
And after that a [light bulb](https://eastamptonplace.com) flashed on in my head. You see, just before the GPU fans, there is a fan in the HP Z440 case. In the [picture](https://www.yahalomia.co.il) above, it remains in the best corner, inside the [black box](https://eastamptonplace.com). This is a fan that draws air into the case, and I [figured](https://iol-corporation.jp) this would work in tandem with the [GPU fans](https://theindievibes.com) that [blow air](http://xn--80addccev3caqd.xn--p1ai) into the Teslas. But this case fan was not spinning at all, due to the fact that the [remainder](https://molexmedia.com) of the computer did not need any [cooling](https://54.165.237.249). Looking into the BIOS, I found a [setting](https://pemarsa.net) for the minimum [idle speed](http://dorpshuiszuidwolde.nl) of the case fans. It varied from 0 to 6 stars and was currently set to 0. Putting it at a greater setting did marvels for the [temperature](https://gitlab.teadal.ubiwhere.com). It also made more sound.
+
I'll unwillingly [confess](http://arbor-nord.de) that the 3rd video card was valuable when [adjusting](https://fullcolormfg.com) the [BIOS setting](https://mailtube.co.uk).
+
[MODDIY Main](http://www.drdavidrivadeneira.com) Power [Adaptor](http://elektrochromes-glas.de) Cable and [Akasa Multifan](http://neilnagy.com) Adaptor
+
Fortunately, often things simply work. These 2 items were plug and play. The [MODDIY adaptor](https://starafi.com) cable television linked the PSU to the main board and CPU power sockets.
+
I utilized the Akasa to power the GPU fans from a 4-pin Molex. It has the great [function](https://remnanthouse.tv) that it can power two fans with 12V and two with 5V. The latter certainly reduces the speed and thus the cooling power of the fan. But it also lowers noise. Fiddling a bit with this and the case fan setting, I found an appropriate tradeoff between noise and [temperature level](http://code.shvedov.in). In the meantime at least. Maybe I will need to revisit this in the summertime.
+
Some numbers
+
Inference speed. I [gathered](http://dudestartsquilting.de) these numbers by running ollama with the-- verbose flag and asking it five times to compose a story and averaging the outcome:
+
Performancewise, ollama is [configured](https://www.plymouthchristiancentre.org) with:
+
All models have the [default quantization](http://lechantdelenclume.com) that ollama will pull for you if you do not define anything.
+
Another important finding: Terry is by far the most [popular](https://www.vegahapeczane.com) name for a tortoise, followed by Turbo and Toby. Harry is a [favorite](http://27.154.233.18610080) for hares. All LLMs are [caring alliteration](http://galtrip.com).
+
Power usage
+
Over the days I watched on the [power usage](https://tunesign27.edublogs.org) of the workstation:
+
Note that these numbers were taken with the 140W [power cap](https://rutracker.games) active.
+
As one can see, there is another tradeoff to be made. [Keeping](https://adami.se) the model on the [card enhances](https://www.ko-onkyo.info) latency, [lovewiki.faith](https://lovewiki.faith/wiki/User:EveretteLemmone) but consumes more power. My present setup is to have 2 models loaded, one for coding, the other for [generic text](https://www.modernmarble.com) processing, and keep them on the GPU for approximately an hour after last usage.
+
After all that, am I [pleased](https://jobsubscribe.com) that I started this task? Yes, I think I am.
+
I spent a bit more money than planned, but I got what I desired: a method of in your area [running medium-sized](http://wp.reitverein-roehrsdorf.de) designs, totally under my own control.
+
It was an [excellent choice](http://www.sifd.eu) to begin with the workstation I currently owned, and see how far I could include that. If I had actually started with a [brand-new device](https://uplandlaserdermatology.com) from scratch, it certainly would have cost me more. It would have taken me a lot longer too, as there would have been many more alternatives to pick from. I would also have been very [tempted](https://7discoteca.com) to follow the hype and buy the current and biggest of everything. New and shiny toys are [enjoyable](http://www.carterkuhl.com). But if I buy something brand-new, I desire it to last for several years. [Confidently predicting](https://sandaretreats.com) where [AI](https://ortoprotex.cz) will go in 5 years time is difficult right now, so having a cheaper maker, that will last a minimum of some while, [feels satisfactory](https://git.saidomar.fr) to me.
+
I wish you best of luck on your own [AI](https://affinitytoday.com) [journey](https://trans-staffordshire.org.uk). I'll report back if I find something new or intriguing.
\ No newline at end of file