Authors hit by bad reviews on Goodreads before review copies are even circulated (thebookseller.com)

^ title. I've been renting MI300Xs coz they are cheaper than H100s and my experience has been generally OK (smoother than i expected based on people shitting on AMD so much online). ROCm 6.x seems decent out of the box now, and I'll happily spend 30 more minutes setting up my GPU if it means 20% cheaper. that being said, it's still annoying to run inference for LLMs on AMD's hardware (e.g. You have to install vLLM from source). And there are some other small details which still suck. As a small example, nvidia-smi gives you a nice clear interface while rocm-smi dumps 3 pages of context that's hard to navigate.

would be curious to hear experiences from other folks experimenting with AI workloads.

Comments (2)

dlcarrier · 9h ago

I'm using an MI25, flashed as a PRO WX 9100, which requires an older version of ROCm to work. That's expectedm, because my GPU is depricated in future versions of ROCm, but what irks me is that everything neural network related barely works. You need the exact version of every interpreter and library, which ends up working on some distributions but not others. I've noticed that when people program in compiled languages, they seem to make a concerted effort to do some kind of bounds testing, but anything in Python or Node.js seems to be released as soon as it kind-of-sort-of works, some of the time.

technoabsurdist · 7h ago

oh yeah, in my experience anything below ROCm6.x really sucks.

I tried to run qwen2.5-32B on ROCm5.x and it was running at <15tok/s lol.

Have you tried running any sort of LLM inference on your MI25, or what NN workloads are you running?

A new PNG spec (programmax.net)

Reading NFC Passport Chips in Linux (shkspr.mobi)

Kid gamers to adult gamblers? Investigation of childhood gaming and YA gambling (tandfonline.com)

Show HN: I built a tool that blocks social media until you scream "I'm a loser"

Microsoft Edit (github.com)

Introducing Qodo Gen CLI: Build and Run Coding Agents Anywhere in the SDLC (qodo.ai)

Thnickels (thick-coins.net)

Fun with uv and PEP 723 (cottongeeks.com)

Yarn (YC W24) is hiring engineers in NYC (ycombinator.com)

The probability of a hash collision (2022) (kevingal.com)

Third places and neighborhood enterpenuership: Evidence from Starbucks cafes (thetreeoflife.cc)

Web Translator API (developer.mozilla.org)

A Dictionary of the Language of Myst's D'ni (eldalamberon.com)

Bill Atkinson: Polaroids Showing the Evolution of the Lisa GUI [video] (youtube.com)

Thoughts on Asunción, Paraguay (cpsi.media)

ChatGPT's enterprise success against Copilot fuels OpenAI/Microsoft rivalry (bloomberg.com)

The Fairphone (Gen. 6) (shop.fairphone.com)

Authors hit by bad reviews on Goodreads before review copies are even circulated (thebookseller.com)

PlasticList – Plastic Levels in Foods (plasticlist.org)

Show HN: I Built AskMedically – Get Research-Backed Answers to Medical Queries

XBOW, an autonomous penetration tester, has reached the top spot on HackerOne (xbow.com)

MCP is eating the world (stainless.com)

Managing time when time doesn't exist (multiverseemployeehandbook.com)

Ancient X11 scaling technology (flak.tedunangst.com)

CareerBuilder and Monster job boards, file for bankruptcy (reuters.com)

How to Think About Time in Programming (shanrauf.com)

Subsecond: A runtime hotpatching engine for Rust hot-reloading (docs.rs)

The bitter lesson is coming for tokenization (lucalp.dev)

Writing toy software is a joy (blog.jsbarretto.com)

Canal Boat Simulator (jacobfilipp.com)

Sourcehut Moving to Europe

Assembly Theory of Time (faculty.ucr.edu)

Gemini Robotics On-Device brings AI to local robotic devices (deepmind.google)

Playing First Contact in Eclipse, a 3-Day Sci-Fi Larp (mssv.net)

Few Americans pay for news when they encounter paywalls (pewresearch.org)

PicoEMP – A low-cost Electromagnetic Fault Injection (EMFI) tool (github.com)

How Cloudflare blocked a monumental 7.3 Tbps DDoS attack (blog.cloudflare.com)

Build an iOS app on Linux or Windows (xtool.sh)

Show HN: Autumn – Open-source infra over Stripe (github.com)

Switching Pip to Uv in a Dockerized Flask / Django App (nickjanetakis.com)

Mapping LLMs over excel saved my passion for game dev (danieltan.weblog.lol)

Jane Street Boss Says He Was Duped into Funding AK-47s for Coup (bloomberg.com)

Show HN: Oasis – An open-source, 3D-printed smart terrarium (github.com)

Lyon Drops Microsoft to Boost Digital Sovereignty (digitrendz.blog)

Show HN: Windowfied

ReBarUEFI: Resizable BAR for almost any UEFI system (github.com)

Nordic Semiconductor Acquires Memfault (nordicsemi.com)

National Archives at College Park, MD, will become a restricted federal facility (archives.gov)

FICO to incorporate buy-now-pay-later loans into credit scores (axios.com)

Finding a 27-year-old easter egg in the Power Mac G3 ROM (downtowndougbrown.com)

Ask HN: Is anyone using AMD GPUs for their AI workloads?

Comments (2)