HNHacker News
TopNewBestAskShowJobs

thangngoc89

1,369 karma · joined November 18, 2015

Email: hi [at] khoanguyen.me
submissionscomments
thangngoc89··on Show HN: An SQS Alternative on Postgres
I’m wondering if there are language agnostic queues where the queue consumers and publishers could be written in different languages?
thangngoc89··on Fly.io Infra log: week-by-week record of what the team does
What do you mean by unbounded pricing model? Just curios because I'm in the process of jumping to fly's GPU offering. I can see that you could set auto scale with min and max instances. So from my thinking, you could know the absolute maximum spending per month.
thangngoc89··on Show HN: Bloom – A shadcn like UI library for Elixir and Phoenix
It's a React library.
thangngoc89··on [dead]
TL;DR: you could require() a ES module and Websockets is built-in
thangngoc89··on Py2wasm – A Python to WASM Compiler
For popular ML/DL models, you could already export model to ONNX format for interference. However, glue codes are still python and you may need to replace that part with the host’s language
thangngoc89··on Ruby vs. Python comes down to the for loop (2021)
Isn’t the double underscores __name would makes the method private?
thangngoc89··on Hacked Nvidia 4090 GPU driver to enable P2P
Yep. Price/performance of multiple 4090s system are way better than the professional cards (Axxx). Also deep learning outside of LLM has many different usage.
thangngoc89··on Hacked Nvidia 4090 GPU driver to enable P2P
8 GPUs x 16 PCIe lanes each = 128 lanes already.

That’s the limit of single CPU platforms.

thangngoc89··on Hacked Nvidia 4090 GPU driver to enable P2P
> You may need to uninstall the driver from DKMS. Your system needs large BAR support and IOMMU off.

Can someone point me to the correct tutorial on how to do these things?

thangngoc89··on Hacked Nvidia 4090 GPU driver to enable P2P
I believe this is mostly for memory capacities. PCIe access between GPUs is slower than soldered RAM on a single GPU
thangngoc89··on Hacked Nvidia 4090 GPU driver to enable P2P
training on MPS backend is suboptimal and really slow.
thangngoc89··on Blocky – a DNS proxy and ad-blocker for the local network
What annoyed me the most is actually clicking on ads from Google Search. Sometime I searched a product and I couldn't clicked on the first result because they are advertisements.
thangngoc89··on Blocky – a DNS proxy and ad-blocker for the local network
Usually pi-hole is installed together with Unbound [1] for the function of Blocky.

[1] https://docs.pi-hole.net/guides/dns/unbound

thangngoc89··on 7.4 earthquake in Taiwan, 34km depth
It happened too early 8:03 AM. Probably no visitor or even staff in Taipei 101. Last time videos are all from visitor.
thangngoc89··on 7.4 earthquake in Taiwan, 34km depth
I’m also in Taipei. It’s the worst one ever I’ve felt in my life. Staying on the 7th floor with noway to escape in time is probably the worst feeling
thangngoc89··on Python Cloudflare Workers
> does Cloudflare provide container hosting service agnostic of language -- something like Google Cloud Run?

Nope. Their Workers are V8 based so JS or Wasm

thangngoc89··on Building WOPR: A 7x4090 AI Server
I come you can find the Foundation model with blower style fan. I’ve been scouring 4 of them for a 4x build. I have the other style of fan in a dual build and the heat dispensation is bad.
thangngoc89··on Pyenv – lets you easily switch between multiple versions of Python
Before CUDA 12.0 you have to specify a field in pyproject.toml like this

    [tool.poetry.dependencies]
    python = ">=3.10,<3.12.0"
    torch = {version = "^2.0.1+cu118", source = "torch118"}
    torchvision = {version = "^0.15.2+cu118", source =     "torch118"}

    [[tool.poetry.source]]
    name = "torch118"
    url = "https://download.pytorch.org/whl/cu118"
    priority = "explicit"
However, since CUDA 12.0 and Pytorch 2.1.0, just install like normal

    poetry add torch torchvision
thangngoc89··on Pyenv – lets you easily switch between multiple versions of Python
I use poetry and direnv. Coming from node/npm, it feels natural for me to just do this. I have really no troubles of installing Pytorch with poetry
thangngoc89··on Supervision: Reusable Computer Vision
To achieve realtime speed, you need to at least convert it to ONNX or similar, applying quantization for 30fps mode.
thangngoc89··on Cloudflare loses 22% of its domains in Freenom .tk shutdown
I remember there was scripts that remove the ads .tk injected
thangngoc89··on Cloudflare loses 22% of its domains in Freenom .tk shutdown
Unrelated to the article but seeing .tk brings back many memories. As a kid without a bank account let's alone an international credit card (VISA/Mastercard), dot.tk is the only way to put a website online with your name. I created countless of websites with .tk for classmates, school and families.
thangngoc89··on Nvtop: Linux Task Monitor for Nvidia, AMD and Intel GPUs
My favorite would be gpustat [1]. This shows the bare minimum amount of information to let's me know that the training has problems/running well

[1] https://github.com/wookayin/gpustat

thangngoc89··on Nvtop: Linux Task Monitor for Nvidia, AMD and Intel GPUs
Of course it doesn't. `nvidia-smi` needs to be available
thangngoc89··on Replicate vs. Fly GPU cold-start latency
banana.dev has announced sunsetting of their server GPUs offering on March 31st, 2024

https://www.banana.dev/blog/sunset

thangngoc89··on Fly.io has GPUs now
About 1-1.5 hours on a Core i9-13900K. Which is unacceptable for my use case.
thangngoc89··on Fly.io has GPUs now
This is right on time. I'm evaluating "severless" GPU services for my upcoming project. I see on the announcement that pricing is per hours. Is scaling to zero priced based on minutes/seconds? For my workflow, medical image segmentation, one file takes about 5 minutes.
thangngoc89··on Fly.io has GPUs now
It's slow but still decent since it has 8GB of RAM.
thangngoc89··on Fly.io has GPUs now
If it's a link to nvidia.com then it's expected to be broken. Seriously, I've never seen a valid link to nvidia.com
thangngoc89··on [dead]
This looks like Spam to me. There are only 4/180 total repository are sources. The rest are forks of popular packages and I don't see any of your own commits.
← PreviousPage 4 of 12Next →