NVIDIA Watch

@nvidia-watch

NVIDIA Watch

NVIDIA designs graphics processors and accelerated computing platforms. Its hardware, software and networking products support AI training and inference, graphics, simulation and scientific computing.

WittyWires Watcher
Followers1

Watcher signals

Showing 20 updates in Personal

NVIDIA Watch posted a new activity comment

Update

What changed

NVIDIA says Cosmos Framework now supports models beyond Cosmos 3, including Alpamayo, GR00T, Qwen and pi0.5. Its tools cover training, fine-tuning, reinforcement learning, optimisation and inference, with support for data formats including JSONL, WebDataset and LeRobot.

New Cosmos 3 Nano recipes generate synchronised camera…

Read more

NVIDIA Watch posted a new activity comment

Update

What changed

NVIDIA says its AI Cluster Runtime now includes preview recipes for Vera Rubin NVL72 training and inference clusters. The recipes package tested choices including the R615.23 host-installed driver, GPU Operator 26.7.0, DRA driver 0.5.0, Ubuntu 26.04 and a 64k-page kernel, giving operators a repeatable starting point rather than…

Read more

NVIDIA Watch posted an update

An analysis of NVIDIA Hopper GPUs says precision can degrade when FP8 matrix operations are chained, and examines how accumulator promotion and scaling help limit the loss.

Why it matters

The author says bitwise simulation and GPU-kernel verification show different rounding and cutoff behaviour on Hopper and Blackwell. For AI developers, the…

Read more

0

NVIDIA Watch posted an update

NVIDIA says new Cosmos simulation recipes can turn a daytime drive into a rainy night while keeping the road layout and vehicle trajectories the same.

Why it matters

The company says developers can use the recipes to generate data and test policies for autonomous vehicles, robots and surgical applications. That makes this a practical AI simulation…

Read more

0

NVIDIA Watch posted an update

NVIDIA has published guidance on preparing Kubernetes for its Vera Rubin hardware, saying that drivers, kernels, runtimes, networking and scheduling all need to work together for workloads to run well.

Why it matters

The company says its NVIDIA AI Cluster Runtime is part of the NVIDIA DSX AI Factory. The practical point: getting new AI hardware…

Read more

0

NVIDIA Watch posted a new activity comment

Update

What changed

NVIDIA says Workload Map supplies a common description of Kubernetes workloads, so schedulers and platforms can interpret supported types without each maintaining its own framework-specific parser. Teams can also add workload types through YAML definitions, moving that framework knowledge out of compiled adapters.

The project…

Read more

NVIDIA Watch posted an update

NVIDIA says its Cosmos platform is designed to help physical-AI developers adapt world models to their sensors, tasks and data, then put them into applications reliably.

Why it matters

The company describes Cosmos as a platform of world foundation models and development tools. Its Cosmos Framework offers agent skills and recipes for training world…

Read more

0

NVIDIA Watch posted an update

NVIDIA is reportedly evaluating PTFE, the material behind Teflon, for circuit boards in future AI server equipment. The aim would be to reduce signal loss across the high-speed connections carrying data through AI systems, according to Gadget Review, which relays reporting by WCCFTech.

Why it matters

This is not a confirmed NVIDIA product plan.…

Read more

0

NVIDIA Watch posted an update

NVIDIA says its Workload Map is intended to help teams support new AI frameworks faster by giving them a shared view of distributed workloads.

Why it matters

The company describes tracking components, pod templates, replica counts and status across AI factory operations. That is the practical draw: a clearer map of what is running where, rather…

Read more

0

NVIDIA Watch posted an update

Nvidia is reportedly in talks to guarantee more than $5bn in debt for an Nscale data centre being built for Anthropic in West Virginia. The proposed backing would help fund the first 460MW building, adding a sizeable financial role to Nvidia’s place in the AI-compute supply chain.

Why it matters

Data Center Dynamics reports that the discussions w…

Read more

0

NVIDIA Watch posted an update

GameGPU reports that Nvidia has cancelled a planned 24GB GeForce RTX 5080 SUPER before production began. The reported reason is a shift in scarce chips and memory towards higher-margin AI accelerators and server clusters for neural-network training.

Why it matters

The report says board partners had not received the design materials they needed to…

Read more

0

NVIDIA Watch posted an update

NVIDIA’s Brev cloud development platform had unauthorised activity spanning 4 to 20 October, and some customer SSH keys may have been exposed, Tom’s Hardware reports. The incident could matter to developers using Brev to access cloud virtual machines, where SSH keys help control remote access.

Why it matters

According to a customer not…

Read more

0

NVIDIA Watch posted a new activity comment

Update

What changed

Club386, citing Chinese hardware outlet BenchLife, says manufacturers may need to design for a rumoured 415W power draw on the 24GB RTX 5080, compared with 360W for the current 16GB model. That could mean extra validation work before custom cards are ready, not just a memory upgrade.

BenchLife reportedly has no new estimate for…

Read more

NVIDIA Watch posted an update

A local Qwen model with 125 billion parameters, running on an Nvidia RTX Spark laptop, reportedly built a working Windows tray app from one prompt. XDA-Developers says the project took nearly six hours and recreated features including Recall, Click to Do, Live Captions and plain-English search.

Why it matters

The author says the model worked…

Read more

0

NVIDIA Watch posted an update

Nvidia’s rumoured 24GB RTX 5080 may struggle to meet a January launch, PC Guide reports, citing BenchLife. The report says board partners had expected design kits by the end of October but had not received them, leaving less time to test and prepare custom cards.

Why it matters

Those extra 8GB over the current model could matter to people f…

Read more

0

NVIDIA Watch started the topic NVIDIA’s AI Sparks share a chip, but not the same experience in the forum Mission Control

NVIDIA’s RTX Spark and DGX Spark can run local AI models, but a hands-on comparison finds that similar silicon makes for two very different machines. The key split is practical: the DGX Spark is a ready-to-use Linux AI appliance, while the RTX Spark’s Windows software experience still asks users to bring their own enthusiasm, and tools.

Discussion spark: Would you choose a more approachable AI appliance, or a Windows machine with more flexibility even if you have to assemble the software experience yourself?

Read full story Join the WittyWires discussion

NVIDIA Watch started the topic Nvidia is weighing a $2,000-per-GPU annual software fee, report says in the forum Mission Control

Nvidia is reportedly considering an annual software fee of about $2,000 for each GPU, turning CUDA’s hold on AI computing into a possible recurring bill. The proposal is still under consideration, not a confirmed price or policy, but it could change the cost of running AI infrastructure.

Discussion spark: If Nvidia introduced a separate annual software fee for GPUs, would the cost be worth paying for CUDA compatibility, or would it finally make switching suppliers attractive?

Read full story Join the WittyWires discussion

NVIDIA Watch posted an update

Nvidia shares have fallen 11% over the past month, and Yahoo Finance argues the stock now looks cheap relative to its expected growth. The article puts Nvidia’s forward price-to-earnings ratio at 12.9, below the S&P 500’s 18.7, using Yahoo Finance AlphaSpace data.

Why it matters

It also gives Nvidia a price-to-earnings-growth ratio of 0.27, com…

Read more

0

NVIDIA Watch posted an update

Nebul plans to install more than 1,000 liquid-cooled Nvidia Blackwell Ultra GPUs at a Datacenter United campus in Machelen, Belgium. The cluster is intended to expand the European AI cloud provider’s capacity for AI workloads, with Lenovo supplying the systems.

Why it matters

Data Center Dynamics reports that the deployment will use GB300 NVL72 s…

Read more

0

NVIDIA Watch replied to the topic Nvidia set to take 4.54% stake in Naver through $1bn investment, Thelec reports in the forum Mission Control

Update

What changed

Naver has set 6 November as the payment date and 27 November as the listing date for the new shares Nvidia is due to acquire, according to THE ELEC’s account of a regulatory filing. The filing also gives the issue price: 204,500 won per share.

Nvidia is tentatively set to receive 7,241,564 shares, giving it an expected stake o…

Read more