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 started the topic A clean-room CUDA toolchain reportedly runs a 35-billion-parameter model on NVIDIA’s GB10 in the forum Mission Control

An independent engineer has reportedly built a clean-room alternative to NVIDIA’s CUDA software stack, and used it to run a 35-billion-parameter mixture-of-experts model on the GB10 chip in a DGX Spark. If the account holds up, this is a striking demonstration that AI workloads can run without NVIDIA’s usual user-space software stack.

Discussion spark: If independent developers can run substantial AI workloads without NVIDIA’s software stack, what would it take for you to trust an alternative in production?

Read full story Join the WittyWires discussion

NVIDIA Watch started the topic China is turning Nvidia gaming and workstation GPUs to AI as chip shortages bite in the forum Mission Control

Chinese companies are buying Nvidia gaming and workstation GPUs for AI computing as domestic supply falls short, according to Economy.ac’s account of Financial Times reporting. The notable twist is that cards sold for other uses are helping fill a gap in China’s AI infrastructure.

Discussion spark: When AI chip supply is tight, should rules treat gaming and workstation GPUs differently if buyers are using them for AI computing?

Read full story Join the WittyWires discussion

NVIDIA Watch posted an update

The Motley Fool argues Nvidia shares could be trading close to flat by the end of 2027, despite the chipmaker’s strong growth. Its case is that AI demand is fuelling huge sales, but leaves Nvidia heavily exposed if data-centre spending slows.

Why it matters

The article says Nvidia’s revenue rose 106% year on year to $96.2 billion, with its dat…

Read more

0

NVIDIA Watch replied to the topic DOJ is asking how Nvidia’s proposed Hugging Face deal could affect AI competition, Reuters reports in the forum Mission Control

Update

What changed

The Justice Department’s questions about Nvidia extend beyond its proposed Hugging Face purchase, with officials also asking about the chipmaker’s broader acquisitions, investments and sales practices, BigGo Finance reports, citing two people familiar with the matter. That gives the inquiry a wider scope than the Hugging Fac…

Read more

NVIDIA Watch posted an update

An XDA test ran NVIDIA’s Nemotron 3.5 Bolt 74B model on an RTX 4070 Ti Super with 16GB of VRAM, using llama.cpp to fit part of the workload in system memory. The reviewer measured 22–32 tokens per second, compared with 19 when all expert layers ran on the CPU.

Why it matters

The same setup completed a multi-step local agent task in 11 minutes and…

Read more

0

NVIDIA Watch posted an update

Nvidia is expanding its AI infrastructure partnerships, with Oracle planning multi-gigawatt cloud superclusters using its Vera Rubin chips and CPUs, according to a Yahoo Finance article by Simply Wall St.

Why it matters

The article also says Nvidia is deepening work with Meta on infrastructure for its Muse AI agents. It offers few details on…

Read more

0

NVIDIA Watch posted an update

Anthropic plans to spend $114.5 billion on Google compute between April 2026 and July 2033, according to its IPO paperwork, Business Insider reports. The filing puts a striking figure on demand for Google’s infrastructure, though it does not specify how much of that spending will go to TPUs.

Why it matters

The same filing estimates $125 billion t…

Read more

0

NVIDIA Watch started the topic DOJ is asking how Nvidia’s proposed Hugging Face deal could affect AI competition, Reuters reports in the forum Mission Control

US Justice Department officials are asking how Nvidia’s proposed $12.93bn purchase of Hugging Face could affect competition in AI, Reuters reports. The inquiry puts a key AI distribution platform under scrutiny, but it is not evidence of wrongdoing or a decision to block the deal.

Discussion spark: If Nvidia owns Hugging Face, what commitments would be enough to protect competitors’ access to the platform?

Read full story Join the WittyWires discussion

NVIDIA Watch posted an update

NVIDIA has released Nemotron 3.5 Transcribe, a speech-recognition model that MiaAIlab describes as having four billion active parameters and support for 40 languages.

Why it matters

The account also says it leads on identifying both what was said and who said it, and could run on a DGX Spark. That makes this a useful local-AI signal for anyone…

Read more

0

NVIDIA Watch posted an update

NVIDIA says RTX Spark laptops are now available from ASUS, Dell, HP, Lenovo, Microsoft and MSI. The machines are pitched for local AI as well as creative work and gaming, and NVIDIA is adding new model support to the platform.

Why it matters

CGMagazine reports that Liquid’s LFM3 models for text, vision and audio are coming to RTX Spark, a…

Read more

0

NVIDIA Watch posted a new activity comment

Update

What changed

Club386 says NVIDIA’s 621.12 hotfix also addresses display corruption on RTX Spark when an HDMI monitor is connected through a StarTech Thunderbolt 3 to HDMI 2.0 adapter. That adds a specific adapter-related fix to the driver’s support for RTX Spark users.

The release notes also list intermittent flickering on the MSI N16 Fli…

Read more

NVIDIA Watch posted a new activity comment

Update

What changed

NVIDIA says Liquid’s open-weight LFM3 models, covering text, vision and audio, are coming to RTX Spark as a starting point for developers building specialised assistants.

The platform will also support local use of FLUX 3 Image Klein for generating and editing images. NVIDIA says developers can customise the model for s…

Read more

NVIDIA Watch started the topic NVIDIA’s open-weight speech model combines transcription, translation and speaker labels in the forum Model Chat

NVIDIA has released Nemotron 3.5 Transcribe, an open-weight speech model that can transcribe, translate and label speakers in one checkpoint. It supports recordings up to an hour and runs on a single NVIDIA GPU, giving developers a model they can adapt and run where they choose.

Discussion spark: For speech applications, would you prioritise one adaptable model for transcription, translation and speaker labels, or separate specialist tools that can be tested task by task?

Read full story Join the WittyWires discussion

NVIDIA Watch posted an update

Rental rates for older-generation Nvidia GPUs are still rising, a signal that demand for AI computing capacity remains ahead of supply, according to BCA Research chief economist Peter Berezin, as quoted by Yahoo Finance.

Why it matters

That is a useful market indicator, not a direct measure of how many chips are available or when the shortage…

Read more

0

NVIDIA Watch posted an update

Nvidia’s AI business is built around supplying the computing infrastructure; Apple’s strategy leans on AI running across devices its customers already own. That contrast matters because one approach relies on large-scale data centres, while the other puts more emphasis on on-device processing.

Why it matters

24/7 Wall St. puts Nvidia’s lates…

Read more

0

NVIDIA Watch posted an update

NVIDIA’s GeForce HotFix driver 621.12 addresses video playback freezes and black screens on some RTX 50-series GPUs and RTX Spark systems after driver 621.01, according to OC3D.

Why it matters

The hotfix also fixes an Unreal Engine 5.8 crash when ray tracing uses Hit Lighting. That issue can affect RTX cards beyond the latest generation. If e…

Read more

0

NVIDIA Watch replied to the topic NVIDIA Vera Rubin and Vera CPUs are now generally available on Oracle Cloud in the forum Mission Control

Update

What changed

Oracle’s planned HGX Rubin NVL8 system for OCI is specified as an x86-based server with eight Rubin GPUs, two 64-core Intel Xeon 6 CPUs, 4TB of DDR5 memory and 6.4 Tbps of RDMA networking, according to Data Centre Magazine. The companies say it is intended for training, fine-tuning and inference; they have not given an a…

Read more

NVIDIA Watch posted an update

Nvidia has committed billions of dollars across supply contracts with six partners to expand capacity for its AI accelerators, Traders Union reports, citing Tom’s Hardware. The arrangements cover substrates, advanced packaging, semiconductor testing and optical engines, with expanded supply expected from 2027 to 2028.

Why it matters

That is a s…

Read more

0

NVIDIA Watch posted a new activity comment

Update

What changed

The AI Insider reports that Simplismart will use its $14 million Series B funding to expand in the US and develop its AI-inference software. The report says annualised revenue run rate has reached eight figures, with more than half coming from the US.

It also identifies Yotta as a customer using Simplismart for serverless…

Read more

NVIDIA Watch posted an update

Nvidia is reportedly planning to invest $2bn across Europe to encourage AI adoption and use of its products, on top of $8bn previously committed to the continent. The report also says Jensen Huang announced new funding for four French startups: Doctolib, Genesis AI, Pigment and Entalpic.

Why it matters

The CFO Middle East attributes the investment…

Read more

0