Hardware
Repurposing a $35 Mining Rig for Local AI Inference
A Reddit user acquired a 54GB VRAM mining rig for $35, successfully repurposing the legacy hardware to run 35B parameter AI models locally.
October 12, 2026, 01:00 PM·Tom
#vram
Hardware
Nvidia Introduces Mega Geometry 2.0 for Dynamic VRAM Management
Nvidia has launched the Mega Geometry 2.0 SDK, a developer tool that streams ray-tracing geometry into VRAM on demand to optimize memory usage and performance.
Tom
Hardware
Future GDDR7 Memory Chips May Reach 6 GB Capacity by 2028
Manufacturers are reportedly developing 4 GB and 6 GB GDDR7 memory chips for release between 2027 and 2028 to increase VRAM capacity without changing GPU bus architecture.
TechPowerUp
Hardware
New 4 GB and 6 GB GDDR7 Chips Expected by 2027-2028
Manufacturers are reportedly developing 4 GB and 6 GB GDDR7 memory chips, with a potential market release scheduled for 2027 or 2028.
TechPowerUp
Hardware
Intel Enhances VRAM Error Handling in Linux Xe Driver
Intel is updating its Linux Xe driver to detect memory errors and isolate defective VRAM pages, preventing system crashes on Battlemage GPUs.
IXBT.com
Hardware
Modder Creates Custom 16GB RTX 3070 Ti Using Memory Swap
A modder has successfully created a working RTX 3070 Ti with 16GB of GDDR6 VRAM by modifying a PCB and swapping memory chips.
Tom
AI
Running Qwen3.8-Flash-Next 125B on Consumer GPUs
The Qwen3.8-Flash-Next 125B model can now be run locally on GPUs with 8 GB of VRAM using AirLLM's layer-wise inference.
Хабр
That is everything