Modders have released firmware exploits that remove Nvidia's artificial half-speed limit on FP32-accumulate matrix math for RTX 20, 30, and 40-series GeForce cards, according to VideoCardz. The unlock lets the RTX 4090 reach double its normal tensor rate and boosts the RTX 3090 Ti's matrix math by up to 80%, though gains vary by card and chatbot token speeds stay roughly the same since those depend on memory bandwidth rather than compute. Three separate projects—gpu-unlock, gpu-unlock-ada, and turing-unlock—provide unofficial Linux kernel modules covering different GPU generations, with experimental Windows builds also in development.
tomshardware.com
· 2026-10-26
Reports and rumors indicate Nvidia's RTX 5090 flagship GPU is being phased out, with no replacement expected until at least 2028. Retail prices have climbed from a $1,999 MSRP to over $7,500, and Micro Center is said to be ending its stock of the standalone card, leaving prebuilt PCs as one of the remaining purchase routes.
tomshardware.com
· 2026-10-16
A Russian repair specialist known as VIK-off examined an RTX 4090 bought as a bare PCB for about 30,000 rubles ($350) and discovered it contained a fake GPU die soldered on via BGA along with counterfeit memory chips mimicking GDDR6X markings. Closer inspection revealed the die's contact area didn't match a real processor and the solder mask degraded during cleaning, while the memory chips were actually GDDR6 rather than GDDR6X, with VIK-off suggesting the board may actually be a repurposed RTX 3090 Ti PCB.
tomshardware.com
· 2026-10-12
A Tom's Guide tech writer compiled a list of 15 affordable gadgets currently discounted to under $50, including items as cheap as under $25, that they personally own and use. Among the featured deals is Rosewill's NVMe M.2 SSD Cloner, now $36.09 (28% off its $49.99 price) on Amazon, which lets users clone or transfer drives via a 20 Gbps USB-C connection without extra enclosures or software.
tomshardware.com
· 2026-10-07
A free, open-source tool called Strata enables the 125-billion-parameter Qwen3.8-Flash-Next AI model to run locally on consumer gaming PCs with 12GB+ VRAM, rather than requiring server hardware. Testing on an RTX 5070 and RX 9070 XT showed response speeds ranging from roughly 44 to 94 tokens per second depending on model compression level, with all processing staying on the local machine.
github.com
· 2026-10-04
Ollaya is a new local server for running open-source 'decision models' like Laya, gliclass and NLI, designed to answer structured questions in a single forward pass rather than generating text token by token. It implements the same /v1/systemone and /v1/models endpoints as TypeSafe's hosted API, so the official TypeSafe Python SDK can point at a local Ollaya server without code changes.
ollaya.dev
· 2026-09-25
Reviews from Notebookcheck and PC Gamer found that external GPU enclosures connected via Thunderbolt suffer meaningful performance losses compared to desktop PCIe slots, ranging from 7 to 32 percent depending on resolution and card. PC Gamer found an RTX 4070 Ti ran about 25 percent slower over Thunderbolt than OCuLink, with worse stuttering and lower frame-rate consistency, and found that pairing an eGPU with a faster RTX 4090 yielded diminishing returns versus a full desktop connection.
engadget.com
· 2026-09-17
A Reddit user reported months of black screens and system instability on their AMD Radeon RX 7900 XTX before discovering burn marks on the card's 8-pin power connector after disassembling their PC. Notably, the connector was routed through a Thermal Grizzly WireView power monitor meant to prevent such damage, and while the monitor took the brunt of the melting, it kept the GPU itself from being harmed.
techspot.com
· 2026-09-15
A technical deep-dive follows the STG.E instruction—the global memory store operation in a simple vector-add CUDA kernel—through an RTX 4090's pipeline, from register read through the load/store unit, coalescer, L1 cache, crossbar, and L2 slice, down to DRAM. The author benchmarks how frequently a single warp can issue store instructions, finding a new STG.E can be pushed through the register file, LSU, and coalescer roughly every 6.1 cycles (about 2.3 nanoseconds at 2.6 GHz).
blog.doubleword.ai
· 2026-09-08
A benchmarking study tested several GGUF quantizations of the Qwen3.8 27B model against benchmarks including GPQA Diamond, IFBench, and Terminal-Bench 2.1. The 17GB Q4_K_M version performed on par with the full 55GB BF16 model on coding tasks and fits on a consumer 24GB GPU, while the smallest 6.2GB 1-bit version dropped to near-random accuracy on reasoning tasks, with longer reasoning chains making results worse.
quesma.com
· 2026-09-08
GoKawiil