PC Games Hardware benchmarked a modder-supplied RTX 2080 Ti with VRAM doubled to 22GB against the stock 11GB version across nine memory-intensive games and an AI image-generation task. The extra memory made little difference at 1440p but produced large gains at 4K with upscaling and frame generation in titles like Flight Simulator 2024, where the stock card stuttered or an 11GB card crashed outright in Indiana Jones and the Great Circle.
techspot.com
· 2026-11-02
Tom's Hardware compared NVIDIA's 8GB and 16GB RTX 5060 Ti cards, which share the same chip and clock speeds, across 21 games and found the 8GB model averaged 2.3% slower at 1080p medium settings but 12% slower at ultra, with minimum frame rates down 16%. Hardware Unboxed reported a bigger gap in The Last of Us Part II (67fps vs 109fps at 1080p very high) and said Indiana Jones and the Great Circle crashed on the 8GB card at 1080p ultra.
engadget.com
· 2026-09-29
GPU prices are climbing sharply, with NVIDIA's RTX 50-series up as much as 39 percent and AMD cards rising up to 10 percent, driven by demand from AI companies and individuals building local AI setups. Cards with large VRAM capacities, such as the older NVIDIA GeForce RTX 3090, are commanding strong resale value because they can handle AI workloads despite their age.
engadget.com
· 2026-09-26
AMD's newly announced RX 9050 GPU exists in two very different versions sharing the same name: an 8GB retail card sold standalone in some markets, and a 4GB variant that can only be bought bundled inside prebuilt desktops. Testing found the 4GB model, discovered inside a $750 CyberPowerPC system from Best Buy, has half the memory bandwidth and significantly worse performance than the retail card most buyers would research online.
techspot.com
· 2026-09-22
A hobbyist project called Mini-AGI has been shared on Hacker News as a byte-level language model that builds its own architecture and trains from scratch on a single consumer GPU with 8GB of VRAM. Rather than keeping all parameters in memory, it stores weights as disk files and loads them as needed, letting it expand or prune capacity while it continuously reads and learns from a text corpus without forgetting prior knowledge. The creator says it's currently a toy-scale demonstration, not a competitive model, and weights won't be released until the current training pass over its corpus finishes in a few weeks.
github.com
· 2026-09-21
Brazilian YouTuber Fabian (fmklab) built a working version of the RTX 3070 Ti 16GB that Nvidia never shipped, by fitting an RTX 3070 Ti GPU onto an RTX 3070 PCB and swapping in higher-capacity Samsung GDDR6 chips. After reconfiguring the board's memory straps using Nvidia's BIOS Reader tool, the card correctly recognized the full 16GB pool in GPU-Z, matching the specs of a prototype that had only ever surfaced in database listings and photos.
tomshardware.com
· 2026-09-09
Tom's Hardware benchmarked Alibaba's newly released Qwen3.8 27B open-weight AI model across discrete GPUs like the RTX 5090, RTX 4090, RTX 3090, and AMD/Intel cards, as well as unified-memory systems including the DGX Spark, Mac Studio, and Ryzen AI Halo. Despite the four-bit quantized model needing only about 17GB of VRAM, the tests found that software stacks and inference engines often bottleneck real-world performance, affecting time-to-first-token and throughput once context windows fill up.
tomshardware.com
· 2026-09-08
AMD's new Radeon RX 9050 will ship in two versions: an 8GB retail model and a 4GB variant reserved for OEM prebuilt PCs. Testers found the 4GB configuration echoes the widely criticized Radeon RX 6500 XT from 2022, struggling to run current games smoothly even years after that earlier card's flaws were exposed.
techspot.com
· 2026-09-08
GoKawiil