American Sign Language is the third most prevalent language in the United States — but there are vastly fewer AI tools developed with ASL data than data representing the country’s most common languages, English and Spanish.
NVIDIA, the American Society for Deaf Children and creative agency Hello Monday are helping close this gap with Signs, an interactive web platform built to support ASL learning and the development of accessible AI applications.
Sign language learners can access the platform’s validated library of ASL signs to expand their vocabulary with the help of a 3D avatar that demonstrates signs — and use an AI tool that analyzes webcam footage to receive real-time feedback on their signing. Signers of any skill level can contribute by signing specific words to help build an open-source video dataset for ASL.
The dataset — which NVIDIA aims to grow to 400,000 video clips representing 1,000 signed words — is being validated by fluent ASL users and interpreters to ensure the accuracy of each sign, resulting in a high-quality visual dictionary and teaching tool.
“Most deaf children are born to hearing parents. Giving family members accessible tools like Signs to start learning ASL early enables them to open an effective communication channel with children as young as six to eight months old,” said Cheri Dowling, executive director of the American Society for Deaf Children. “And knowing that professional ASL teachers have validated all the vocabulary on the platform, users can be confident in what they’re learning.”
NVIDIA teams plan to use this dataset to further develop AI applications that break down communication barriers between the deaf and hearing communities. The data is slated to be available to the public as a resource for building accessible technologies including AI agents, digital human applications and video conferencing tools. It could also be used to enhance Signs and enable ASL platforms across the ecosystem with real-time, AI-powered support and feedback.
Whether novice or expert, volunteers can record themselves signing to contribute to the ASL dataset.
Supporting ASL Education, Exploring Language Nuance
During the data collection phase, Signs already provides a powerful platform for ASL language acquisition, offering opportunities for individuals to learn and practice an initial set of 100 signs so they can more effectively communicate with friends or family members who use ASL.
“The Signs learning platform could help families with deaf children quickly search for a specific word and see how to make the corresponding sign. It’s a tool that can help support their everyday use of ASL outside of a more formal class,” Dowling said. “I see both kids and parents exploring it — and I think they could play with it together.”
While Signs currently focuses on hand movements and finger positions for each sign, ASL also incorporates facial expressions and head movements to convey meaning. The team behind Signs is exploring how these non-manual signals can be tracked and integrated in future versions of the platform.
They’re also investigating how other nuances, like regional variations and slang terms, can be represented in Signs to enrich its ASL database — and working with researchers at the Rochester Institute of Technology’s Center for Accessibility and Inclusion Research to evaluate and further improve the user experience of the Signs platform for deaf and hard-of-hearing users.
“Improving ASL accessibility is an ongoing effort,” said Anders Jessen, founding partner of Hello Monday/DEPT, which built the Signs web platform and previously worked with the American Society for Deaf Children on Fingerspelling.xyz, an application that taught users the ASL alphabet. “Signs can serve the need for advanced AI tools that help transcend communication barriers between the deaf and hearing communities.”
The dataset behind Signs is planned for release later this year.
Start learning or contributing with Signs at signs-ai.com, and learn more about NVIDIA’s trustworthy AI initiatives. Attendees of NVIDIA GTC, a global AI conference taking place March 17-21 in San Jose, will be able to participate in Signs live at the event.
Wield magic and steel as GeForce NOW’s fifth-anniversary celebration summons Obsidian Entertainment’s highly anticipated Avowed to the cloud.
This first-person fantasy role-playing game is ready to enchant cloud gamers, leading the charge of six titles joining the over 2,000 games in the cloud gaming library.
GeForce NOW day passes are available to purchase again, in limited quantities each day. Members can currently purchase one day at a time, based on available capacity. Day pass users get 24-hour access to powerful cloud gaming with all the benefits of a GeForce NOW Ultimate or Performance membership. Stay tuned for updates as more membership options become available.
Choose Your Own Adventure
Cloudy with a chance of dragons.
Embark on a thrilling adventure in Avowed, set in the captivating world of Eora. As an envoy of Aedyr, explore the mysterious Living Lands, an island teeming with ancient magic and shifting secrets, as a dire threat looms over the realm: a mysterious plague that defies nature and reason, spreading chaos across the sprawling wilderness.
The Living Lands offer a diverse array of environments to explore, each with a unique ecosystem. Engage in visceral combat by mixing and matching swords, spells, guns and shields. Companions of various species, each with their own abilities and quests, will join the adventure, their fates intertwined with the players’ choices. As the story unfolds, every decision will ripple across the Living Lands, shaping the future of its inhabitants and testing the players’ resolve in the face of intrigue, power and danger.
GeForce NOW members can dive into this immersive fantasy world with the power of GeForce RTX-powered gaming rigs in the cloud. Ultimate members can stream the game at up to 4K resolution and 60 frames per second with high dynamic range on supported devices. These members enjoy additional benefits like NVIDIA DLSS 3 technology for enhanced frame rates and NVIDIA Reflex for ultra-low latency, delivering a seamless and visually stunning adventure through the Living Lands.
Time to Play
Some mixtapes are better left unplayed.
Lost Records: Bloom & Rage is the recently released narrative-adventure game by Don’t Nod, the creators of Life Is Strange. Set in the fictional Michigan town of Velvet Cove, the game follows four friends — Swann, Nora, Autumn and Kat — during the summer of 1995, as well as 27 years later in 2022.
Explore Swann’s world through a nostalgic 90s lens, complete with a camcorder for capturing and reliving memories. The story unfolds across two timelines, delving into themes of friendship and identity, as well as a mysterious secret that tore the group apart. With its immersive storytelling, interactive environments and choice-driven gameplay, Lost Records: Bloom & Rage promises a captivating journey through time, nostalgia and the complexities of lifelong friendships.
Look for the following games available to stream in the cloud this week:
Avowed (New release on Steam, Battle.net and Xbox, available on PC Game Pass, Feb. 18)
Warhammer 40,000: Rogue Trader (New release on Xbox, available on PC Game Pass, Feb. 20)
Lost Records: Bloom & Rage (New release on Steam, Feb. 18)
Editor’s note: This post is part of Into the Omniverse, a series focused on how developers, 3D practitioners and enterprises can transform their workflows using the latest advances in OpenUSD and NVIDIA Omniverse.
Humanoid robots are rapidly becoming a reality. Those built on NVIDIA Isaac GR00T are already learning to walk, manipulate objects and otherwise interact with the real world.
Gathering diverse and large datasets to train these sophisticated machines can be time-consuming and costly. Using synthetic data (SDG), generated from physically-accurate digital twins, researchers and developers can train and validate their AI models in simulation before deployment in the real world.
Universal Scene Description, aka OpenUSD, is a powerful framework that makes it easy to build these physically accurate virtual environments. Once 3D environments are built, OpenUSD allows teams to develop detailed, scalable simulations along with lifelike scenarios where robots can practice, learn and improve their skills.
This synthetic data is essential for humanoid robots to learn humanlike behaviors such as walking, grasping objects and navigating complex environments. OpenUSD is enhancing the development of humanoid robots and paving the way for a future where these machines can seamlessly integrate into people’s daily lives.
The NVIDIA Omniverse platform, powered by OpenUSD, provides developers a way to unify 3D assets from disparate sources such as 3DCAD and digital content creation (DCC) tools. This allows them to build large-scale 3D virtual environments and run complex simulations to train their robots, streamlining the entire process and delivering faster, more cost-effective ways to collaborate and develop physical AI.
Advancing Robot Training With Synthetic Motion Data
At CES last month, NVIDIA announced the Isaac GR00T Blueprint for synthetic motion generation to help developers generate exponentially larger synthetic motion datasets to train humanoids using imitation learning.
Highlights of the release include:
Large-Scale Motion Data Generation: Uses simulation as well generative AI techniques to generate exponentially large and diverse datasets of humanlike movements, speeding up the data collection process.
Faster Data Augmentation:NVIDIA Cosmos world foundation models generate photorealistic videos at scale using the ground-truth simulation from Omniverse. This equips developers to augment synthetic datasets faster, for training physical AI models, reducing the simulation-to-real gap.
Simulation-First Training: Instead of relying solely on real-world testing, developers can train robots in virtual environments, making the process faster and more cost-effective.
Bridging Virtual to Reality: The combination of real and synthetic data along with simulation-based training and testing allows developers to transfer the robots skills learned in the virtual world to the real-world seamlessly.
Simulating the Future of Robotics
Humanoid robots are enhancing efficiency, safety and adaptability across industries like manufacturing, warehouse and logistics, and healthcare by automating complex tasks and increasing safety conditions for human workers.
Major robotics companies including Boston Dynamics and Figure have already started adopting and demonstrating results with Isaac GR00T.
Get Plugged Into the World of OpenUSD
Learn more about OpenUSD, humanoid robots and the latest AI advancements at NVIDIA GTC, a global AI conference running March 17-21 in San Jose, California.
Don’t miss NVIDIA founder and CEO Jensen Huang’s GTC keynote on Tuesday, March 18 — in person at the SAP Center or online. He’ll share the latest technologies driving the next wave in AI, digital twins, cloud technologies and sustainable computing.
The inaugural GTC Humanoid Developer Day will take place on Wednesday, March 18. Following the sessions, join the Physical AI Developer Meetup to network with developers and researchers at NVIDIA GTC. Discuss the latest breakthroughs in OpenUSD and generative AI-powered simulation and digital twins, as well as innovations in generalist robotics for the next frontier of industries.
GeForce RTX 5070 Ti GPUs feature fifth-generation Tensor Cores with support for FP4, doubling performance and reducing VRAM requirements to run generative AI models.
In addition, the GPU comes equipped with two ninth-generation encoders and a sixth-generation decoder that add support for the 4:2:2 pro-grade color format and increase encoding quality for HEVC and AV1. This combo accelerates video editing workflows, reducing export times by 8x compared with single encoder GPUs without 4:2:2 support like the GeForce RTX 3090.
The GeForce RTX 5070 Ti GPU also includes 16GB of fast GDDR7 memory and 896 GB/sec of total memory bandwidth — a 78% increase over the GeForce RTX 4070 Ti GPU.
The GeForce RTX 5070 Ti GPU — a game changer.
NVIDIA DLSS 4, a suite of neural rendering technologies that uses AI to boost frames per second (fps) and improve image quality, is now available in professional-grade 3D apps like Chaos Vantage. D5 Render also adds DLSS 4 in beta with the new Multi Frame Generation feature to boost frame rates by 3x. 3D rendering software Maxon Redshift also added NVIDIA Blackwell support, providing a 30% performance increase.
The February NVIDIA Studio Driver, with support for the GeForce RTX 5070 Ti GPU, will be ready for download next week. For automatic Studio Driver notifications, download the NVIDIA app.
Use NVIDIA’s product finder to pick up a GeForce RTX 5070 Ti GPU or prebuilt system today. Check back regularly after 6 a.m. PT, as retail partners list their available models. Explore complete specifications.
Ready for the Generative AI Era
Black Forest Lab’s FP4-optimized FLUX.1 [dev] suite of image generation models is now available on Hugging Face.
FP4 is a lower quantization method, similar to file compression, that decreases model sizes. FLUX.1 [dev] at FP4 requires less than 10GB of VRAM, compared with over 23GB at FP16.
This means the state-of-the-art FLUX.1 [dev] model can run on the GeForce RTX 5070 Ti GPU as well as all GeForce RTX 50 Series GPUs. This is important because the FLUX.1 [dev] model wouldn’t be able to run at FP16, given memory constraints.
On the GeForce RTX 5070 Ti GPU, the FLUX.1 [dev] model can generate images in just over eight seconds on FP4, compared with 20 seconds on FP8 on a GeForce RTX 4070 Ti GPU.
Versatile Viewports
DLSS 4 is now available in Chaos Vantage and D5 Render in beta — popular professional-grade 3D apps for architects, animators and designers.
D5 Render also supports DLSS 4’s DLSS Multi Frame Generation to boost frame rates by using AI to generate up to three frames per traditionally rendered frame.
This enables animators to smoothly navigate a scene with multiplied frame rates and render 3D content, even with massive file sizes, at 60 fps or more.
Maxon Redshift — a 3D rendering software that uses GPU acceleration to visualize 3D models, scenes, animations and designs — has released an update to fully harness GeForce RTX 50 Series GPUs, accelerating performance by up to 30%.
In financial services, AI has traditionally been used primarily for fraud detection and risk modeling. With recent advancements in generative AI, the banking industry as a whole is becoming smarter and more intuitive, offering hyper-personalized services and real-time insights for customers.
In the latest episode of the NVIDIA AI Podcast, Barb Morgan, chief product and technology officer at banking and financial services technology company Temenos, shares how AI is reshaping the banking landscape, from enhancing customer experiences to ensuring robust data security.
Morgan explains that AI can tailor financial products and services to customer needs, making interactions more meaningful and relevant. Plus, AI-powered chatbots and digital interfaces can provide 24/7 support, addressing customer queries in real time.
AI adoption has grown significantly in financial services. Notably, the use of generative AI for customer experience, especially through chatbots and virtual assistants, has more than doubled, rising from 25% to 60% over the last year. Learn more in NVIDIA’s fifth annual “State of AI in Financial Services” report.
And see more of the latest technological advancements by registering for NVIDIA GTC, the conference for the era of AI, taking place March 17-21. Temenos will share more insights and examples in the session titled, “Generative AI for Core Banking.”
Time Stamps
08:30 – How AI can help banks process and analyze vast amounts of data to provide deeper insights and predictions.
11:56 – The importance of data management for effective AI implementation.
16:13 – Sustainability in the banking industry, and how AI can help banks and customers track and reduce their carbon footprints.
Learn how AI agents are transforming the retail landscape by personalizing customer journeys, converting marketing interactions into valuable research data and enhancing the customer experience with hyper-personalized insights and recommendations.
See how Snowflake’s AI Data Cloud platform helps enterprises unlock the value of data by transforming it into actionable insights and applications, using large language models.
Scientists everywhere can now access Evo 2, a powerful new foundation model that understands the genetic code for all domains of life. Unveiled today as the largest publicly available AI model for genomic data, it was built on the NVIDIA DGX Cloud platform in a collaboration led by nonprofit biomedical research organization Arc Institute and Stanford University.
Evo 2 is available to global developers on the NVIDIA BioNeMo platform, including as an NVIDIA NIM microservice for easy, secure AI deployment.
Trained on an enormous dataset of nearly 9 trillion nucleotides — the building blocks of DNA and RNA — Evo 2 can be applied to biomolecular research applications including predicting the form and function of proteins based on their genetic sequence, identifying novel molecules for healthcare and industrial applications, and evaluating how gene mutations affect their function.
“Evo 2 represents a major milestone for generative genomics,” said Patrick Hsu, Arc Institute cofounder and core investigator, and an assistant professor of bioengineering at the University of California, Berkeley. “By advancing our understanding of these fundamental building blocks of life, we can pursue solutions in healthcare and environmental science that are unimaginable today.”
The NVIDIA NIM microservice for Evo 2 enables users to generate a variety of biological sequences, with settings to adjust model parameters. Developers interested in fine-tuning Evo 2 on their proprietary datasets can download the model through the open-source NVIDIA BioNeMo Framework, a collection of accelerated computing tools for biomolecular research.
“Designing new biology has traditionally been a laborious, unpredictable and artisanal process,” said Brian Hie, assistant professor of chemical engineering at Stanford University, the Dieter Schwarz Foundation Stanford Data Science Faculty Fellow and an Arc Institute innovation investigator. “With Evo 2, we make biological design of complex systems more accessible to researchers, enabling the creation of new and beneficial advances in a fraction of the time it would previously have taken.”
Enabling Complex Scientific Research
Established in 2021 with $650 million from its founding donors, Arc Institute empowers researchers to tackle long-term scientific challenges by providing scientists with multiyear funding — letting scientists focus on innovative research instead of grant writing.
Its core investigators receive state-of-the-art lab space and funding for eight-year, renewable terms that can be held concurrently with faculty appointments with one of the institute’s university partners, which include Stanford University, the University of California, Berkeley, and the University of California, San Francisco.
By combining this unique research environment with accelerated computing expertise and resources from NVIDIA, Arc Institute’s researchers can pursue more complex projects, analyze larger datasets and more quickly achieve results. Its scientists are focused on disease areas including cancer, immune dysfunction and neurodegeneration.
NVIDIA accelerated the Evo 2 project by giving scientists access to 2,000 NVIDIA H100 GPUs via NVIDIA DGX Cloud on AWS. DGX Cloud provides short-term access to large compute clusters, giving researchers the flexibility to innovate. The fully managed AI platform includes NVIDIA BioNeMo, which features optimized software in the form of NVIDIA NIM microservices and NVIDIA BioNeMo Blueprints.
NVIDIA researchers and engineers also collaborated closely on AI scaling and optimization.
Applications Across Biomolecular Sciences
Evo 2 can provide insights into DNA, RNA and proteins. Trained on a wide array of species across domains of life — including plants, animals and bacteria — the model can be applied to scientific fields such as healthcare, agricultural biotechnology and materials science.
Evo 2 uses a novel model architecture that can process lengthy sequences of genetic information, up to 1 million tokens. This widened view into the genome could unlock scientists’ understanding of the connection between distant parts of an organism’s genetic code and the mechanics of cell function, gene expression and disease.
“A single human gene contains thousands of nucleotides — so for an AI model to analyze how such complex biological systems work, it needs to process the largest possible portion of a genetic sequence at once,” said Hsu.
In healthcare and drug discovery, Evo 2 could help researchers understand which gene variants are tied to a specific disease — and design novel molecules that precisely target those areas to treat the disease. For example, researchers from Stanford and the Arc Institute found that in tests with BRCA1, a gene associated with breast cancer, Evo 2 could predict with 90% accuracy whether previously unrecognized mutations would affect gene function.
In agriculture, the model could help tackle global food shortages by providing insights into plant biology and helping scientists develop varieties of crops that are more climate-resilient or more nutrient-dense. And in other scientific fields, Evo 2 could be applied to design biofuels or engineer proteins that break down oil or plastic.
“Deploying a model like Evo 2 is like sending a powerful new telescope out to the farthest reaches of the universe,” said Dave Burke, Arc’s chief technology officer. “We know there’s immense opportunity for exploration, but we don’t yet know what we’re going to discover.”
The telecom industry’s efforts to drive efficiencies with AI are beginning to show fruit.
An increasing focus on deploying AI into radio access networks (RANs) was among the key findings of NVIDIA’s third annual “State of AI in Telecommunications” survey, as more than a third of respondents indicated they’re investing or planning to invest in AI-RAN. The survey polled more than 450 telecommunications professionals worldwide, revealing continued momentum for AI adoption — including growth in generative AI use cases — and how the technology is helping optimize customer experiences and increase employee productivity.
Of the telecommunications professionals surveyed, almost all stated that their company is actively deploying or assessing AI projects. Here are some top insights on impact and use cases:
84% said AI is helping to increase their company’s annual revenue
77% said AI helped reduce annual operating costs
60% said increased employee productivity was their biggest benefit from AI
44% said they’re investing in AI for customer experience optimization, which is the No. 1 area of investment for AI in telecommunications
40% said they’re deploying AI into their network planning and operations, including RAN
Business Impact on AI in Telecommunications
Survey results highlight that use of AI in the telecom industry has helped increase revenue and reduce costs. 84% of respondents said that the technology is helping increase their company’s annual revenue, with 21% saying that AI had contributed to a more than 10% revenue increase in specific business areas. In addition, 77% agreed that AI helped reduce annual operating costs.
The wide array of AI use cases and impact on the bottom line has led to greater confidence in the future: 80% of respondents believe that AI is crucial for their company’s future success, while two-thirds plan to increase spending on AI infrastructure this year.
The telecommunications industry is at the forefront of AI adoption, with a clear focus on enhancing employee productivity, customer experience and network operations. By continuing to invest in AI infrastructure and training, telecom companies can stay ahead of the curve and capitalize on the numerous benefits that AI offers.
AI Finds Its Way Into the Network Stack
AI in the telecommunications network is gaining momentum, with 37% of respondents saying they’re investing in AI to improve network planning and operations. Similarly, 33% said they invested in using AI for field-operations optimization in the last year.
Of the respondents investing in AI for 5G monetization and/or 6G research and development, 66% are aiming to deploy AI services on RAN for operational and user needs, 53% are aiming to enhance spectral efficiency for the RAN, and 50% are aiming to colocate AI and RAN applications on the same infrastructure.
Generative AI Goes Mainstream
Generative AI is gaining significant attention in telecoms. More than half of survey respondents who said they’re using generative AI have already deployed their first use case, while another third plan to do so this year.
Of those respondents adopting generative AI, 84% said that their companies plan to offer generative AI solutions externally to customers. 52% said they would offer generative AI as a software-as-a-service solution, while 35% will offer generative AI as a platform for developers, including for compute services.
There’s also a notable trend toward using multiple approaches for AI development, including a rise in in-house and open-source capabilities.
Asteroids were responsible for extinction events hundreds of millions of years ago on Earth, providing no shortage of doomsday film plots for Hollywood.
But researchers focused on asteroid tracking are on a mission to locate them for today’s real-world concerns: planetary defense.
The new and unexpected discovery tool applied in this research is NASA’s James Web Space Telescope (JWST), which was tapped for views of these asteroids from previous research and enabled by NVIDIA accelerated computing.
An international team of researchers, led by MIT physicists, reported on the cover of Nature this week how the new method was able to spot 10-meter asteroids within the main asteroid belt located between Jupiter and Mars.
These rocks in space can range from the size of a bus to several Costco stores in width and deliver destruction to cities on Earth.
The finding of more than 100 space rocks of this size marks the smallest asteroids ever detected in the main asteroid belt. Previously, the smallest asteroids spotted measured more than half a mile in diameter.
Researchers say the novel method — tapping into previous studies, asteroid synthetic movement tracking and infrared observations — will help identify and track orbital movements of asteroids likely to approach Earth, supporting asteroid defense efforts.
“We have been able to detect near-Earth objects down to 10 meters in size when they are really close to Earth,” Artem Burdanov, the study’s co-lead author and a research scientist in MIT’s Department of Earth, Atmospheric and Planetary Sciences, told MIT News. “We now have a way of spotting these small asteroids when they are much farther away, so we can do more precise orbital tracking, which is key for planetary defense.”
Capturing Asteroid Images With Infrared JWST Driven by NVIDIA GPUs
Observatories typically look at the reflected light off asteroids to determine their size, which can be inaccurate. Using a telescope with infrared, like the JWST, can help track the thermal signals of asteroids for a more precise way at gauging their size.
Asteroid hunters focused on planetary defense are looking out for near-Earth asteroids. These rocks have orbits around the Sun that are within 28 million miles of Earth’s orbit. And any asteroid around 450 feet long is capable of demolishing a sizable city.
The asteroid paper’s co-authors included MIT professors of planetary science co-lead Julien de Wit and Richard Binzel. Contributions from international institutions included the University of Liege in Belgium, Charles University in the Czech Republic, the European Space Agency, and institutions in Germany including the Max Planck Institute for Extraterrestrial Physics and the University of Oldenburg.
The 2024YR4 near-Earth asteroid — estimated as wide as 300 feet and capable of destroying a city the size of New York — has a 2.3% chance of striking Earth.
Movies like Armageddon provide fictional solutions, like implanting a nuclear bomb, but it’s unclear how this could play out off screen.
The JWST technology will soon be the only telescope capable of tracking the space rock as it moves away from Earth before coming back.
The new study used the JWST, the best-ever telescope in the infrared, on images of TRAPPIST-1, a star studied to search for signs of atmospheres around its seven terrestrial planets and located about 40 light years from Earth. The data include more than 10,000 images of the star.
After processing the images from JWST to study TRAPPIST-1’s planets, the researchers considered whether they could do more with the datasets. They’re looking at if they can search for otherwise undetectable asteroids using JWST’s infrared capabilities and a new detection technique they had deployed on other datasets called synthetic tracking.
The researchers applied synthetic tracking methods, which doesn’t require previous information on an asteroid’s motion. Instead it does “fully blind” search by testing possible shifts, like velocity vectors.
Such techniques are computationally intense, and they created bottlenecks until NVIDIA GPUs were applied to such work in recent years. Harnessing GPU-based synthetic tracking increases the scientific return on resources when conducting exoplanet transit-search surveys by recovering serendipitous asteroid detections, the study said.
After applying their GPU-based framework for detecting asteroids in targeted exoplanet surveys, the researchers were able to detect eight known and 139 unknown asteroids, the paper’s authors noted.
“Today’s GPU technology was key to unlocking the scientific achievement of detecting the small-asteroid population of the main belt, but there is more to it in the form of planetary-defense efforts,” said de Wit. “Since our study, the potential Earth-impactor 2024YR4 has been detected, and we now know that JWST can observe such an asteroid all the way out to the main belt as they move away from Earth before coming back. And in fact, JWST will do just that soon.”
It’s a match made in heaven — GeForce NOW and Warner Bros. Games are collaborating to bring the beloved Batman: Arkham series to the cloud as part of GeForce NOW’s fifth-anniversary celebration. Just in time for Valentine’s Day, gamers can fall in love all over again with Gotham City’s Dark Knight, streaming his epic adventures from anywhere, on nearly any device.
The sweet treats don’t end there. GeForce NOW also brings the launch of the highly anticipated Sid Meier’s Civilization VII.
It’s all part of the lovable lineup of seven games joining the cloud this week.
A Match Made in Gotham City
GeForce NOW is welcoming Warner Bros. Games to the cloud with the Batman: Arkham series, including Arkham Asylum Game of the Year Edition, Batman: Arkham City Game of the Year Edition and Batman: Arkham Knight Premium. Don the cape and cowl of the world’s greatest detective, bringing justice to the streets of Gotham City with bone-crushing combat and ingenious gadgets.
Villains check in, but they don’t check out.
Experience the dark and gritty world of Gotham City’s infamous asylum in the critically acclaimed action-adventure game Batman: Arkham Asylum. The Game of the Year (GOTY) Edition enhances the original title with additional challenge maps, allowing players to test their skills as the Dark Knight. Unravel the Joker’s sinister plot, face off against iconic villains and harness Batman’s gadgets and detective abilities in this groundbreaking title.
Every street is a crime scene.
Members can expand their crimefighting horizons in the open-world sequel, Batman: Arkham City. The GOTY Edition includes the full game, plus all downloadable content, featuring Catwoman, Nightwing and Robin as playable characters. Explore the sprawling super-prison of Arkham City, confront a rogues’ gallery of villains and uncover the mysteries behind Hugo Strange’s Protocol 10. With enhanced gameplay and an even larger arsenal of gadgets, Batman: Arkham City elevates the Batman experience to new heights.
Just another Tuesday in the neighborhood.
Conclude the Batman: Arkham trilogy in style with Batman: Arkham Knight. The Premium Edition includes the base game and season pass, offering new story missions, additional DC Super-Villains, legendary Batmobiles, advanced challenge maps and alternative character skins. Take control of a fully realized Gotham City, master the iconic Batmobile and face off against the Scarecrow and the mysterious Arkham Knight in this epic finale. With stunning visuals and refined gameplay, Batman: Arkham Knight delivers the ultimate Batman experience.
It’s the ideal time for members to be swept off their feet by the Caped Crusader. Stream the Batman: Arkham series with a GeForce NOW Ultimate membership and experience these iconic titles in stunning 4K resolution at up to 120 frames per second. Feel the heartbeat of Gotham City, the rush of grappling between skyscrapers and the thrill of outsmarting Gotham’s most notorious villains — all from the cloud.
Build an Empire in the Cloud
GeForce NOW’s fifth-anniversary celebration continues this week with the gift of Sid Meier’s Civilization VII in the cloud at launch.
Grow your civilization with the power of the cloud.
2K Games’ highly anticipated sequel comes to the cloud with innovative gameplay mechanics. This latest installment introduces a dynamic three-Age structure — Antiquity, Exploration and Modern — allowing players to evolve their civilizations throughout history and transition between civilizations with the flexibility to shape empires’ destinies.
Explore unknown lands, expand territories, engage in diplomacy or battle with rival nations. Sid Meier’s Civilization VII introduces a crisis-event system at the end of each era, bringing challenges for players to navigate.
With its refined gameplay and bold new features, the title offers both longtime fans and newcomers a fresh and engaging take on the classic formula that has defined the series for decades.
Prepare for New Games
Steel and sorcery, meet cloud and FPS.
Legacy: Steel & Sorcery is an action-packed player vs. player vs. environment extraction role-playing game (RPG) set in the fantasy world of Mithrigarde by Notorious Studios, former World of Warcraft developers. Choose from distinctive classes like Warrior, Hunter, Rogue and Priest, each with unique abilities and environmental interactions. The game features a dynamic combat system emphasizing skill-based PvP, a full-loot system and RPG progression elements. Explore expansive outdoor zones solo or team up with friends to search for treasures, complete quests and battle both AI-controlled foes and rival players for an immersive fantasy RPG experience with a fresh twist on the extraction genre.
Look for the following games available to stream in the cloud this week:
Just as there are widely understood empirical laws of nature — for example, what goes up must come down, or every action has an equal and opposite reaction — the field of AI was long defined by a single idea: that more compute, more training data and more parameters makes a better AI model.
However, AI has since grown to need three distinct laws that describe how applying compute resources in different ways impacts model performance. Together, these AI scaling laws — pretraining scaling, post-training scaling and test-time scaling, also called long thinking — reflect how the field has evolved with techniques to use additional compute in a wide variety of increasingly complex AI use cases.
The recent rise of test-time scaling — applying more compute at inference time to improve accuracy — has enabled AI reasoning models, a new class of large language models (LLMs) that perform multiple inference passes to work through complex problems, while describing the steps required to solve a task. Test-time scaling requires intensive amounts of computational resources to support AI reasoning, which will drive further demand for accelerated computing.
What Is Pretraining Scaling?
Pretraining scaling is the original law of AI development. It demonstrated that by increasing training dataset size, model parameter count and computational resources, developers could expect predictable improvements in model intelligence and accuracy.
Each of these three elements — data, model size, compute — is interrelated. Per the pretraining scaling law, outlined in this research paper, when larger models are fed with more data, the overall performance of the models improves. To make this feasible, developers must scale up their compute — creating the need for powerful accelerated computing resources to run those larger training workloads.
This principle of pretraining scaling led to large models that achieved groundbreaking capabilities. It also spurred major innovations in model architecture, including the rise of billion- and trillion-parameter transformer models, mixture of experts models and new distributed training techniques — all demanding significant compute.
And the relevance of the pretraining scaling law continues — as humans continue to produce growing amounts of multimodal data, this trove of text, images, audio, video and sensor information will be used to train powerful future AI models.
Pretraining scaling is the foundational principle of AI development, linking the size of models, datasets and compute to AI gains. Mixture of experts, depicted above, is a popular model architecture for AI training.
What Is Post-Training Scaling?
Pretraining a large foundation model isn’t for everyone — it takes significant investment, skilled experts and datasets. But once an organization pretrains and releases a model, they lower the barrier to AI adoption by enabling others to use their pretrained model as a foundation to adapt for their own applications.
This post-training process drives additional cumulative demand for accelerated computing across enterprises and the broader developer community. Popular open-source models can have hundreds or thousands of derivative models, trained across numerous domains.
Developing this ecosystem of derivative models for a variety of use cases could take around 30x more compute than pretraining the original foundation model.
Developing this ecosystem of derivative models for a variety of use cases could take around 30x more compute than pretraining the original foundation model.
Post-training techniques can further improve a model’s specificity and relevance for an organization’s desired use case. While pretraining is like sending an AI model to school to learn foundational skills, post-training enhances the model with skills applicable to its intended job. An LLM, for example, could be post-trained to tackle a task like sentiment analysis or translation — or understand the jargon of a specific domain, like healthcare or law.
The post-training scaling law posits that a pretrained model’s performance can further improve — in computational efficiency, accuracy or domain specificity — using techniques including fine-tuning, pruning, quantization, distillation, reinforcement learning and synthetic data augmentation.
Fine-tuning uses additional training data to tailor an AI model for specific domains and applications. This can be done using an organization’s internal datasets, or with pairs of sample model input and outputs.
Distillation requires a pair of AI models: a large, complex teacher model and a lightweight student model. In the most common distillation technique, called offline distillation, the student model learns to mimic the outputs of a pretrained teacher model.
Reinforcement learning, or RL, is a machine learning technique that uses a reward model to train an agent to make decisions that align with a specific use case. The agent aims to make decisions that maximize cumulative rewards over time as it interacts with an environment — for example, a chatbot LLM that is positively reinforced by “thumbs up” reactions from users. This technique is known as reinforcement learning from human feedback (RLHF). Another, newer technique, reinforcement learning from AI feedback (RLAIF), instead uses feedback from AI models to guide the learning process, streamlining post-training efforts.
Best-of-n sampling generates multiple outputs from a language model and selects the one with the highest reward score based on a reward model. It’s often used to improve an AI’s outputs without modifying model parameters, offering an alternative to fine-tuning with reinforcement learning.
Search methods explore a range of potential decision paths before selecting a final output. This post-training technique can iteratively improve the model’s responses.
To support post-training, developers can use synthetic data to augment or complement their fine-tuning dataset. Supplementing real-world datasets with AI-generated data can help models improve their ability to handle edge cases that are underrepresented or missing in the original training data.
Post-training scaling refines pretrained models using techniques like fine-tuning, pruning and distillation to enhance efficiency and task relevance.
What Is Test-Time Scaling?
LLMs generate quick responses to input prompts. While this process is well suited for getting the right answers to simple questions, it may not work as well when a user poses complex queries. Answering complex questions — an essential capability for agentic AI workloads — requires the LLM to reason through the question before coming up with an answer.
It’s similar to the way most humans think — when asked to add two plus two, they provide an instant answer, without needing to talk through the fundamentals of addition or integers. But if asked on the spot to develop a business plan that could grow a company’s profits by 10%, a person will likely reason through various options and provide a multistep answer.
Test-time scaling, also known as long thinking, takes place during inference. Instead of traditional AI models that rapidly generate a one-shot answer to a user prompt, models using this technique allocate extra computational effort during inference, allowing them to reason through multiple potential responses before arriving at the best answer.
On tasks like generating complex, customized code for developers, this AI reasoning process can take multiple minutes, or even hours — and can easily require over 100x compute for challenging queries compared to a single inference pass on a traditional LLM, which would be highly unlikely to produce a correct answer in response to a complex problem on the first try.
This AI reasoning process can take multiple minutes, or even hours — and can easily require over 100x compute for challenging queries compared to a single inference pass on a traditional LLM.
This test-time compute capability enables AI models to explore different solutions to a problem and break down complex requests into multiple steps — in many cases, showing their work to the user as they reason. Studies have found that test-time scaling results in higher-quality responses when AI models are given open-ended prompts that require several reasoning and planning steps.
The test-time compute methodology has many approaches, including:
Chain-of-thought prompting: Breaking down complex problems into a series of simpler steps.
Sampling with majority voting: Generating multiple responses to the same prompt, then selecting the most frequently recurring answer as the final output.
Search: Exploring and evaluating multiple paths present in a tree-like structure of responses.
Post-training methods like best-of-n sampling can also be used for long thinking during inference to optimize responses in alignment with human preferences or other objectives.
Test-time scaling enhances inference by allocating extra compute to improve AI reasoning, enabling models to tackle complex, multi-step problems effectively.
How Test-Time Scaling Enables AI Reasoning
The rise of test-time compute unlocks the ability for AI to offer well-reasoned, helpful and more accurate responses to complex, open-ended user queries. These capabilities will be critical for the detailed, multistep reasoning tasks expected of autonomous agentic AI and physical AI applications. Across industries, they could boost efficiency and productivity by providing users with highly capable assistants to accelerate their work.
In healthcare, models could use test-time scaling to analyze vast amounts of data and infer how a disease will progress, as well as predict potential complications that could stem from new treatments based on the chemical structure of a drug molecule. Or, it could comb through a database of clinical trials to suggest options that match an individual’s disease profile, sharing its reasoning process about the pros and cons of different studies.
In retail and supply chain logistics, long thinking can help with the complex decision-making required to address near-term operational challenges and long-term strategic goals. Reasoning techniques can help businesses reduce risk and address scalability challenges by predicting and evaluating multiple scenarios simultaneously — which could enable more accurate demand forecasting, streamlined supply chain travel routes, and sourcing decisions that align with an organization’s sustainability initiatives.
And for global enterprises, this technique could be applied to draft detailed business plans, generate complex code to debug software, or optimize travel routes for delivery trucks, warehouse robots and robotaxis.
AI reasoning models are rapidly evolving. OpenAI o1-mini and o3-mini, DeepSeek R1, and Google DeepMind’s Gemini 2.0 Flash Thinking were all introduced in the last few weeks, and additional new models are expected to follow soon.
Models like these require considerably more compute to reason during inference and generate correct answers to complex questions — which means that enterprises need to scale their accelerated computing resources to deliver the next generation of AI reasoning tools that can support complex problem-solving, coding and multistep planning.
NVIDIA GeForce NOW (@NVIDIAGFN) 
