Coverage
GPU news: NVIDIA and AMD accelerator launches, supply and allocation, cloud availability, pricing moves and deployment at scale.
OpenAI has introduced GPT-6 Astra Ultrafast, utilising Nvidia's Blackwell graphics processors across developer API access, ChatGPT Work, and Codex. Inference optimisations tailored directly to the Blackwell silicon allow the model to deliver up to an eightfold acceleration in token production over the standard edition.
Sony has introduced a fresh artificial intelligence graphics scaling feature for the baseline PlayStation 5 console, created through its Project Amethyst collaboration with AMD. Known as Quick Spectral Super Resolution, the software adds a new tier of AI-driven upscaling alongside PlayStation Spectral Super Resolution.
Retailers have discounted an MSI gaming laptop equipped with an Nvidia RTX 5090 mobile graphics processor and an AMD Ryzen AI CPU. The high-performance portable machine also features 64GB of memory alongside high-capacity solid-state storage. Such promotions highlight pricing movements within the premium consumer and mobile computing hardware segment.
A call centre consultancy leased four Nvidia H200 accelerators to assess whether running DeepSeek V4.1 Flash for Claude Code justified claims of substantial cost savings. The testing demonstrated that querying DeepSeek's hosted API directly was more economical than renting dedicated hardware.
Nvidia has opened submissions for the 2027–2028 cycle of its Graduate Fellowship Program, providing selected doctoral candidates with grants of up to $60,000. The scheme offers financial support, mentorship and technical resources to researchers developing projects relevant to accelerated computing architectures.
A developer has trained an artificial intelligence world model using an individual RTX 3080 Ti graphics card to play Pokémon Red. The project implements a JEPA architecture derived from LeWorldModel research co-authored by Yann LeCun.
A former product manager at EVGA has detailed the deteriorating commercial ties that culminated in the 2022 dissolution of the firm's long-standing collaboration with Nvidia. The account highlights how the debut of Nvidia's own Founders Edition graphics cards alarmed the board partner, with mounting pricing friction and forward-looking product risks ultimately fracturing the tie-up. The reflection offers rare insight into the operational pressures faced by third-party graphics card manufacturers.
A new E-Series GPU IP architecture has been introduced to merge graphics, general compute, and machine learning workloads onto a single design and programmable software layer. Running at 1GHz, the core provides up to 32 TOPS of Int8 throughput and supports simultaneous execution of visual and artificial intelligence tasks.
The SC26 supercomputing conference has honoured Nathan Bell and Michael Garland with its Test of Time Award for their seminal research into sparse matrix-vector multiplication. Originally unveiled at SC09 during the initial era of CUDA, the study provided key breakthroughs in harnessing parallel GPU architectures for complex mathematical workloads.
AMD has set the baseline pricing for its 192 GB Gorgon Halo offering at $6,799. The steep price tag underscores the substantial costs associated with deploying hardware configured with such high memory volumes.
Graphics hardware vendor Zotac has refused warranty service for an RTX 3060 graphics card in India after 12 months, despite promoting three years of standard coverage. The manufacturer claimed that the warranty duration is calculated from the product's 2023 import date rather than its 2025 retail purchase.
Nvidia is preparing to allocate hundreds of billions of dollars toward repurchasing its own equity. The move follows a compression in the chip designer's valuation when measured against projected forward earnings. Chief executive Jensen Huang is capitalising on the comparatively subdued share price to execute the record programme.
Eric Gullichsen, an early adviser to Nvidia, claims that an administrative error dating to 1993 has deprived him of shares currently valued near $1bn. Nvidia has turned down settlement overtures aimed at resolving the decades-old dispute. The disagreement centres on the vesting and execution of an initial stock option award.
Hardware manufacturer GMKtec has introduced the Evo-X5 Pro, a compact system powered by AMD's high-end Ryzen AI Max+ Pro 495 processor. The compact machine incorporates up to 192 gigabytes of system memory aimed at localized artificial intelligence workloads, commanding a top price of over seven thousand dollars. The launch demonstrates rising hardware specifications in small-form-factor devices targeting on-device model deployment.
Page 1 · 14 stories
GPU Data Hub Daily
The most important GPU, AI, data-center, semiconductor and cloud developments delivered directly to your inbox.