Fatih Cetinkaya
@devfacet
Followers
1,845
Following
3,380
Media
72
Statuses
436
Software Developer. Opinions are my own.
Massachusetts, USA
Joined July 2009
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Happy Pride
• 1904095 Tweets
Real Madrid
• 1234932 Tweets
Champions
• 1175873 Tweets
Dortmund
• 727834 Tweets
#UCLfinal
• 646052 Tweets
Vini
• 393374 Tweets
Kroos
• 374086 Tweets
Neymar
• 272973 Tweets
Bellingham
• 214357 Tweets
مدريد
• 202350 Tweets
Wembley
• 177294 Tweets
Reus
• 167965 Tweets
Carvajal
• 160761 Tweets
WNBA
• 154284 Tweets
La 15
• 152051 Tweets
Ancelotti
• 150050 Tweets
Clark
• 131189 Tweets
Modric
• 112501 Tweets
Sancho
• 91037 Tweets
Bola de Ouro
• 90349 Tweets
Ballon
• 90308 Tweets
Nacho
• 86628 Tweets
Mbappe
• 85051 Tweets
دورتموند
• 84576 Tweets
Adeyemi
• 70184 Tweets
Courtois
• 50744 Tweets
Camavinga
• 44596 Tweets
الريال
• 43346 Tweets
Dubois
• 39942 Tweets
Bivol
• 34958 Tweets
Hrgovic
• 34902 Tweets
Balón de Oro
• 28249 Tweets
Hummels
• 27401 Tweets
ドルトムント
• 23868 Tweets
#5vs5
• 23246 Tweets
كروس
• 22100 Tweets
#precure
• 21105 Tweets
Maatsen
• 19967 Tweets
#25thTGMA
• 17862 Tweets
Don Carlo
• 17654 Tweets
Fábio
• 15741 Tweets
マドリー
• 15451 Tweets
クロース
• 13542 Tweets
Congratulations FC
• 10970 Tweets
كارفخال
• 10865 Tweets
This is how I save over $1000/mo on compute, storage, and bandwidth costs for my AI side projects: a total of 52 CPU cores, 104GB RAM, 3TB NVMe, unlimited ingress/egress, and less than 100W power consumption.
The taller cluster runs API servers built with the Transformer.js
149
323
3K
@shanginn
13 x Raspberry Pi 5 8GB = $80/ea
13 x Active Cooler = $5/ea
13 x PSU = $12/ea
3 x HatDrive = ~$27/ea
3 x 1TB NVMe = ~$60/ea
+ network equipment, cables, etc.
It's around $1600
To put that in perspective, the cheapest Amazon EC2 instance with 4 CPUs and 8 GB of memory
8
12
151
I finally had a chance to rearrange my Pi5 cluster today.
For the compute-only Pi5s (fan only, no USB/PCIe/NVMe), I got rid of the official Pi5 power supplies and replaced them with Anker PowerPort 6, which is common for Pi clusters. One of the reasons was that no matter what I
6
12
96
@NullMasterX
I’m sorry, but even the new efficient Intel CPUs are nowhere near ARM CPUs in terms of power consumption. How can you compare them with i5s/i7s? I can query 50 embeddings/sec at approximately 7W on a Pi, while most efficient Intel CPUs use >10W even when idle, do nothing
7
0
64
Introducing Embedefy 🎉 Embeddings for everyone.
A couple of months ago, I was working on an AI side project aimed at finding relevant information in a PostgreSQL database based on user prompts. My goal was to combine this information with the user prompts and send it to a
2
8
44
@sagitz_
@huggingface
TL;DR
@huggingface
infra team doesn’t know how to run a secured Kubernetes cluster.
1
0
42
@WatchShipSink
“13 x Pi 5s and the network equipment use less than 100W in total. Each Pi 5 consumes between 1.8W and 7W, making them very energy-efficient.
Having multiple servers helps avoid a single point of failure. I can update and reboot them individually and swap them out as needed.
An
2
0
32
Check out SigLens, an open-source observability solution written in Go. It has zero external dependencies and is a single binary that can run on your laptop, processing up to 8 TB/day.
#golang
#observability
#logging
0
5
25
@fkardame
@shanginn
@orangepixunlong
@theradxa
I’ve already tried some of them, the software support (e.g. Linux kernel) and system stability (Google it yourself) is no where near Raspberry Pi. They were keep crashing.
1
1
23
2
2
23
The Art of WebAssembly by
@battagline
I really like the format of this book. Another nice one from
@nostarch
1
3
21
@JacobESalazarS
With 16 cores, the Ryzen 9 7950X would consume between 170W - 220W, in addition to the power used by other parts such as the graphics card, motherboard, etc.
In comparison, 4 x Raspberry Pi 5s (=16 cores) would use between 7.2W and 28W. Also I can easily swap out the Pi 5s,
6
0
19
@samlakig
Currently it runs 6 different embedding models simultaneously. Here are some stats
@AntoineGrondin
@dave_newton
@shanginn
I guess it depends on what you are doing. In my case, I mostly work on embeddings/classification, and most of the models () are nicely utilized with Pi 5s. Currently, a single Pi 5 can handle approximately 65 requests per second (including traffic between
1
0
15
2
0
16
@radamar
@xenovacom
Just as a side note, I'm using this cluster to offer free AI inference services to those who have a limited budget or no credit card for platforms like OpenAI. I'll be adding more stuff in the coming days. You can read more at
Introducing Embedefy 🎉 Embeddings for everyone.
A couple of months ago, I was working on an AI side project aimed at finding relevant information in a PostgreSQL database based on user prompts. My goal was to combine this information with the user prompts and send it to a
2
8
44
1
0
17
@radamar
@xenovacom
Feature Extraction (for RAG), Text classification, Image Classification, and Summarization. That being said you can run almost anything at
1
0
17
To software developers who are still skeptical about AI:
Last year, I developed and launched four different apps. All of them have one thing in common: I used programming languages which I had never used before.
The first app I developed was a macOS app written in Swift. It was
2
0
13
@AntoineGrondin
@dave_newton
@shanginn
I guess it depends on what you are doing. In my case, I mostly work on embeddings/classification, and most of the models () are nicely utilized with Pi 5s. Currently, a single Pi 5 can handle approximately 65 requests per second (including traffic between
1
0
15
0
1
14
Introducing the Embedefy PostgreSQL Extension 🎉
The Embedefy PostgreSQL Extension provides access to embeddings directly from your database, without building and maintaining additional applications. Once the extension is installed, you can query your database as you normally
0
2
11
Servers as Microservices with Node.js at Yieldbot -
http://t.co/WtOeK621js
-
@nodejs
@hapijs
#javascript
0
2
11
@dave_newton
@shanginn
The cheapest 16 vCPU EC2 instance is about ~$390 x 3 (if you need similar amount of cores) = $1170 per month
+ NAT gateway fee $0.045 per hour + storage + data transfer + IP address fee + etc. (source )
vs
In contrast, purchasing the necessary equipment
3
0
12
@Raistlin
Good question. The reason I use Transformer.js (besides its awesomeness) is because the API server is written in TypeScript, and I didn’t want to add another moving part to communicate between Python and Node.js.
I don’t think there is a significant performance difference
2
0
11
@tgraf__
@danwendlandt
: thanks for visiting the Mountain and great
@ciliumproject
presentation today.
@Bose
Auditorium
1
5
11
@ErrataRob
By the way this specific cluster runs the Embedefy API. A free AI inference service for those who have a limited budget or no credit card for platforms like OpenAI. I shared more details here
This is how I save over $1000/mo on compute, storage, and bandwidth costs for my AI side projects: a total of 52 CPU cores, 104GB RAM, 3TB NVMe, unlimited ingress/egress, and less than 100W power consumption.
The taller cluster runs API servers built with the Transformer.js
149
323
3K
1
1
11
@kelseyhightower
Agree. We built managed database services top of Kubernetes using statefulsets. That being said I wouldn’t recommend to anyone. Operational cost is high, versioning/upgrades are painful, multi-region is hard, backups are tricky. And it’s a distraction against your real focus.
1
2
9
@TheMahnaMahna
@NullMasterX
The Pi 5 cluster consumes less than 100W under load (running AI inference and a Postgres server), not while idle. It's rarely idle, unlike NAS or similar systems. Why don't you compare power consumption per watt? Also, if you want to compare it with ThinkCentre and EliteDesk,
0
0
8
If you're using bge-base and uae-large embeddings models in your projects, then you should check out It's 100% cheaper than OpenAI and 100% cheaper than Together AI. Although it's slower, it's free and doesn't require a credit card. Get started at
Save up to 8x on embeddings.
We can’t wait to see what you build on Together AI.
14
28
297
0
0
6
I saw this variation of the pong game (link below) where two balls continuously move, aiming to capture the opposing player's area. My first question was how this game could end, if at all. I didn't see a possibility (correct me if I'm wrong) without modifying the game. So, I'm
3
1
5
@joao_c137
@JacobESalazarS
His question would be much more interesting if he asked “why not single desktop with
@AmpereComputing
“
1
0
6
0
0
6
I got some
@StarTrek
pins from
@FanSets
and they look awesome. I really like Picard/Locutus (
@SirPatStew
) and Captain Janeway (
@TheKateMulgrew
) pins.
#StarTrek
#IDIC
1
1
6
@itakenaps
AI inference. I shared some details here
@radamar
@xenovacom
Feature Extraction (for RAG), Text classification, Image Classification, and Summarization. That being said you can run almost anything at
1
0
17
0
0
6
I'm doing some AI training for the Duel of the Paddles game so it won't require two players to play. So far, after 250K+ steps (RL/PPO), it's still performing terribly. 😅🕹️🤖
#ai
#artificalintelligence
0
0
4
@CameronPak
I posted some details here
@shanginn
13 x Raspberry Pi 5 8GB = $80/ea
13 x Active Cooler = $5/ea
13 x PSU = $12/ea
3 x HatDrive = ~$27/ea
3 x 1TB NVMe = ~$60/ea
+ network equipment, cables, etc.
It's around $1600
To put that in perspective, the cheapest Amazon EC2 instance with 4 CPUs and 8 GB of memory
8
12
151
0
1
6
Here we go... “Serious Cryptography - A Practical Introduction to Modern Encryption” by
@veorq
#cryptography
#encryption
1
0
5
0
0
5
My Google search usage has dropped by more than 50% since I started using ChatGPT.
0
0
5
It seems that launching a wireless microphone for $99 and labeling it as AI hardware has become the new trend. 🤖
#ai
#artificiaiIntelligence
1
0
4
"Two people finish a job in two months, five people in five months."
A variation of Brooks' Law, which states: "Adding manpower to a late software project makes it later."
0
1
5
You are my sunshine... Just now me and my 5yo sing together. Thank you
@RobertPicardo
and
@JeriLRyan
for creating such a scene. It’s one of my favorite moments of
#StarTrekVoyager
#StarTrek
0
0
5
@kelseyhightower
@EnvoyProxy
You can achieve a similar functionality on AWS using VPC peering and
@ciliumproject
. The latency is near zero (other than inter region latency)
1
0
5
0
0
5
ReviewGPT Chrome Extension has arrived! 📣 🎉 Save time and money on Amazon with just one click. ReviewGPT helps you to avoid counterfeit products, pirated books, fake reviews, bad 3rd party sellers, retail arbitrage and products with safety and health risks by analyzing Amazon
0
1
5
@0xSparked
Get an Anker Power Station with solar panels and Starlink then call it Micro Datacenter. I thought about this setup for shed but a van would do. 😏
1
0
5
“There are only two kinds of languages: the ones people complain about and the ones nobody uses” - Bjarne Stroustrup
#programming
#languages
0
4
5
@danielmerja
I haven't run any LLM yet. Although you can find some numbers from other people here on X. I used it for
@embedefy
See
It can perform ~65 AI inference/sec with a single Pi 5 (including network overhead). But I'm also working on batch API to make it
@radamar
@xenovacom
Just as a side note, I'm using this cluster to offer free AI inference services to those who have a limited budget or no credit card for platforms like OpenAI. I'll be adding more stuff in the coming days. You can read more at
1
0
17
0
0
5
@kucikdebug
@AlexandrBesan
Pi 5 power management is different than previous generations. It gets more tricky with the ones which use PCIe (e.g., for NVMe). So I will wait for the official channels and the community for a while before being creative.
0
0
4
0
1
4
My Stack Overflow usage has dropped by more than 99% since I started using ChatGPT.
1
0
3
"Ask not what AI will do to you - ask what you can do with AI"
0
2
4
AWS Systems Manager is pretty cool. I’m using for managing my
@Raspberry_Pi
systems.
#aws
#RaspberryPi
0
0
3
@TheJasikaNicole
Fringe is one of my favorite tv series and some degree it changed my life. I’m sorry for the challenges and difficulties which you had. I wish you shared it back then so we (fans) could make sure that he/she never direct another Fringe episode again. What’s director’s name?
0
1
3
Ferret is a search engine that unifies Github, Slack, Trello and more in same interface.
#golang
#opensource
0
2
3
"You’ve got to start with the customer experience and work backwards for the technology. You can’t start with the technology and try to figure out where you’re going to try to sell it." - Steve Jobs (1997)
^ The current state of web3 and blockchain technologies.
1
0
1
0
0
3
The AI model finally learned how to hit the ball, but it still has a long way to go. 🤖🕹️
I'm doing some AI training for the Duel of the Paddles game so it won't require two players to play. So far, after 250K+ steps (RL/PPO), it's still performing terribly. 😅🕹️🤖
#ai
#artificalintelligence
0
0
4
1
0
3
@aiFunPolice
See
@radamar
@xenovacom
Just as a side note, I'm using this cluster to offer free AI inference services to those who have a limited budget or no credit card for platforms like OpenAI. I'll be adding more stuff in the coming days. You can read more at
1
0
17
0
0
2
There are some really points and thoughts by
@aaronwhite
in this "Who Will Win the Incumbent AI Battle?" video.
#ai
#artificialintelligence
0
1
2
@nathan_jcurtis
Sorry really busy with some other stuff. I posted some details, see and
Happy to answer if you have a specific question.
2
0
3
We use strong-supervisor library (by
@StrongLoop
) to run our
#nodejs
based services.
http://t.co/WtOeK621js
0
3
3
0
0
3
@fixersbk
@aallan
(I know it’s easy to say but) They could turn the company into a foundation and do fundraising once a year on social media (i.e.
#SaveMakerFaire
) I would donate some and perhaps I wouldn’t be the only one. Anyway.
0
1
3
@FiloSottile
Thank you for your work. I use the stream module. Btw I would chip in a few bucks every month and show appreciation/support somehow if you ever deicide to enable "Github Sponsors" for age.
0
0
2
@calorialejandro
@NullMasterX
Lots of similar messages today. Sorry if my tone wasn’t right. It seems weird to me lots of people trying to compare these credit card size computers with desktop grade system. If thats the case then let’s compare with
@AmpereComputing
or AWS Graviton ARM processors.
I think
1
0
2
The reason I stopped buying from them is because of this kind of attitude. Their obsession with competitor products is getting weird.
#raspberrypi
#pi5
A new HATDri… Oh wait, it’s not our board 😆
Glad to see
@Raspberry_Pi
finally joining the party!
3
1
63
0
0
2