Mastodawn

🎮 Story trailer for Halo: Campaign Evolved has dropped, a REMAKE of the main campaign on Unreal Engine 5 with upgraded visuals, expanded gameplay systems and 3 new missions.

Developers also posted PC specs: SSD, 100 GB. Minimum: Intel Core i7-10700K / AMD Ryzen 5 3600 + RTX 2060 SUPER / RX 6600 / Arc A580 + 16 GB RAM (8 GB VRAM); Recommended: i5-12600K / Ryzen 7 5700X + RTX 3070 / RX 7600 XT + 16 GB; High: i7-12700K / Ryzen 7 7700 + RT...

#SteamAndEpic #Ryzen #RTX #VRAM #Campaign #Recommended

vramshield 2d ago

Trying to run the new Gemma-4 26B locally on your laptop?

You might get a stable 20 t/s at first, but sustained memory-swapping will quickly push your VRAM junction to 105°C, causing sudden throttling. We wrote a practical guide on how to manage these thermal limits safely without relying on buggy hardware drivers.

https://vramshield.com/article-gemma4-laptop
#VRAM #LocalAI #Hardware

How to Run Gemma-4 26B on an 8GB Laptop Without Overheating | VRAM Shield

Learn how to run Google's Gemma-4 26B model on an 8GB VRAM laptop using llama.cpp and manage VRAM thermal throttling.

Show thread

Wordmark 6d ago

@jwildeboer i think it should be possible to combine CPU+RAM and GPU+VRAM but most models just say #nah if there is not enough expensive #VRAM question: what about distributed computing?
#gpt4all https://dwaves.de/2023/12/16/gnu-linux-how-to-install-gpt4all-on-debian-12-gpt4all-j-an-apache-2-licensed-assistant-style-chatbot-a-free-to-use-locally-running-privacy-aware-chatbot-no-gpu-or-internet-required-dialogu/
#deepseek can run on #cpu but its slow https://dwaves.de/2025/02/01/also-with-ai-opensource-for-the-win-cpu-only-benchmarking-deepseek-on-local-amd-server-can-it-be-funny/

» GNU Linux – how to install gpt4all on Debian 12 – GPT4All-J: An Apache-2 Licensed Assistant-Style Chatbot: A free-to-use, locally running, privacy-aware chatbot. No GPU or internet required – dialogue with a machine: AI: How to fix democracy? – THE NEW TOURING TEST: Humor | dwaves.de

Why keep all your results to yourself? - Blog with howtos and public free software and hardware OpenSource searchable knowledgebase about Linux and OpenSource - with a touch security, politics and philosophy.

N-gated Hacker News Jun 2

🎉 Ah, the joys of turning your overpriced RTX card into a glorified RAM extender because your laptop's memory is soldered tighter than a miser's wallet. 🤦‍♂️ Brilliant! Now you can feel like a true tech wizard, using #VRAM for swap space—because who needs efficiency when you have desperation? 🧙‍♂️💻
https://github.com/c0dejedi/nbd-vram #techhumor #gaminghardware #laptopissues #RTXswap #HackerNews #ngated

GitHub - c0deJedi/nbd-vram: Use your NVIDIA GPU's VRAM as swap space on Linux. Built for laptops with soldered memory and no upgrade path. If you have an RTX card sitting there with 8GB of VRAM and you're getting swapped to SSD, this puts that VRAM to work

Use your NVIDIA GPU's VRAM as swap space on Linux. Built for laptops with soldered memory and no upgrade path. If you have an RTX card sitting there with 8GB of VRAM and you're getting swap...

GitHub

Hacker News Jun 2

Use your Nvidia GPU's VRAM as swap space on Linux

https://github.com/c0dejedi/nbd-vram

#HackerNews #NvidiaGPU #VRAM #LinuxSwap #GPUComputing #TechTips #OpenSource

GitHub - c0deJedi/nbd-vram: Use your NVIDIA GPU's VRAM as swap space on Linux. Built for laptops with soldered memory and no upgrade path. If you have an RTX card sitting there with 8GB of VRAM and you're getting swapped to SSD, this puts that VRAM to work

Use your NVIDIA GPU's VRAM as swap space on Linux. Built for laptops with soldered memory and no upgrade path. If you have an RTX card sitting there with 8GB of VRAM and you're getting swap...

GitHub

Habr Jun 1

А что если бы видеопамять можно было использовать как обычную RAM?

У меня 8 гигабайт оперативки. На бумаге звучит терпимо — ровно до того момента, пока не откроешь десятка два вкладок в хроме, рядом Figma, Slack, ещё вкладку со Stack Overflow, и поверх всего этого попробуешь что-то писать в VS Code. Сначала система начинает подтормаживать. Потом подтормаживать сильнее. В какой-то момент я просто встаю и иду за чаем — и, возвращаясь, нередко застаю её всё в той же задумчивости.. Тут-то вопрос и нарисовался: а нельзя ли, хотя бы теоретически, приспособить видеопамять под расширение оперативки?

https://habr.com/ru/articles/1042028/

#RAM #VRAM #Swap #Оптимизация_памяти #Видеокарты #GPU #Chrome #Ноутбуки #GpuRamDrive #Файл_подкачки

А что если бы видеопамять можно было использовать как обычную RAM?

С чего всё началось У меня 8 гигабайт оперативки. На бумаге звучит терпимо — ровно до того момента, пока не откроешь десятка два вкладок в хроме, рядом Figma, Slack, ещё вкладку со Stack Overflow, и...

Хабр

Ramon Marquez May 29

Con Linux 7.2 se habilitará el soporte completo para la propiedad de color "BACKGROUND_COLOR" lo cual permitirá ahorrar ancho de banda en la VRAM.

#Linux
#Intel
#VRAM

https://notasrem.com/intel-anadir-soporte-color-fondo-linux-72/?utm_source=mastodon&utm_medium=jetpack_social

Intel añade soporte para el color de fondo en Linux 7.2

Linux 7.2 soportará la propiedad de color "BACKGROUND_COLOR" tanto en el controlador gráfico de Intel como en el driver DRM de la pantalla

NotasREM

PsychoticSheep May 26

My OpenCode experience: From local LLM to free cloud

I tried OpenCode because I wanted an AI agent for coding.

One of the cool things about an agent is that it automatically checks if all dependencies are installed. No more "module not found" stress.

My journey:
- I started with Qwen2.5 Coder 14B. Tool calling does not work, so it is useless for an agent framework.
- Then I attempted local Qwen3:14B. The problem here is the context window. Tool calling requires num_ctx to be increased to 16384. At that size, the model no longer fits into VRAM and gets partially offloaded to the CPU, which kills performance.
- Next I tried local Qwen3:8b. It works fine for simple Nginx configs and static HTML/CSS. But with JavaScript, it struggles without very precise instructions.
- I also looked at Qwen3 Coder. Even the smallest model has 30B parameters, which is too heavy for my PC.

Then DeepSeek V4 Flash Free changed everything

This is a full cloud model, completely free. No API key, no credit card, no subscription.

About the limits: According to my research, the community proxy services behind OpenCode allow about 50 to 200 million tokens per day, which is more than enough. The counter resets daily. The context is around 256k instead of 1M, but that is still plenty for coding.

Local models are nice, but let's be honest, running 30B+ models on normal hardware is still a pain. Until that changes, DeepSeek V4 Flash Free in OpenCode is the perfect bridge solution. It just works. It is fast, simple, and costs nothing. Highly recommended!

#OpenCode #DeepSeekV4 #FreeAI #AIAgent #Coding #LocalLLM #VRAM #Qwen3 #LLM #DevTools #OpenSourceAI #AICoding #Fediverse #Programming