Nvidia, Oracle, Google, Dell and 13 other companies reported how long it takes their computers to train the key neural networks in use today. Among those results were the first glimpse of Nvidia’s next generation GPU, the B200, and Google’s upcoming accelerator, called Trillium. The B200 posted a…

"Giving up on Nvidia here after its hit — Hopper — is like giving up on Apple at iPhone 1 or 2," Melius Research said.

Nvidia, the AI-chip giant, petitioned the nation's highest court after a lower court permitted a 2018 class action lawsuit to move ahead.

The U.S. Supreme Court will hear arguments in Nvidia’s bid to appeal a securities fraud lawsuit today, just days after… Continue reading US Supreme Court to hear bid by Nvidia to appeal securities fraud lawsuit

The post US Supreme Court to hear bid by Nvidia to appeal securities fraud lawsuit appeared first on ReadWrite.

Nov. 13, 2024 — Generative AI applications that use text, computer code, protein chains, summaries, video and even 3D graphics require data-center-scale accelerated computing to efficiently train the large language models […]

The post NVIDIA: Blackwell Delivers Next-Level MLPerf Training Performance appeared first on HPCwire.

TOKYO, Nov. 13, 2024 — NVIDIA has announced a series of collaborations with SoftBank Corp. designed to accelerate Japan’s sovereign AI initiatives and further its global technology leadership while also unlocking […]

The post NVIDIA Supports SoftBank in Building AI Supercomputer, Unveils AI-Driven Telecom Network appeared first on HPCwire.

Nvidia, Oracle, Google, Dell and 13 other companies reported how long it takes their computers to train the key neural networks in use today. Among those results were the first glimpse of Nvidia’s next generation GPU, the B200, and Google’s upcoming accelerator, called Trillium. The B200 posted a doubling of performance on some tests versus today’s workhorse Nvidia chip, the H100. And Trillium delivered nearly a four-fold boost over the chip Google tested in 2023.

The benchmark tests, called MLPerf v4.1, consist of six tasks: recommendation, the pre-training of the large language models (LLM) GPT-3 and BERT-large, the fine tuning of the Llama 2 70B large language model, object detection, graph node classification, and image generation.

Training GPT-3 is such a mammoth task that it’d be impractical to do the whole thing just to deliver a benchmark. Instead, the test is to train it to a point that experts have determined means it is likely to reach the goal if you kept going. For Llama 2 70B, the goal is not to train the LLM from scratch, but to take an already trained model and fine-tune it so it’s specialized in a particular expertise—in this case, government documents. Graph node classification is a type of machine learning used in fraud detection and drug discovery.

As what’s important in AI has evolved, mostly toward using generative AI, the set of tests has changed. This latest version of MLPerf marks a complete changeover in what’s being tested since the benchmark effort began. “At this point all of the original benchmarks have been phased out,” says David Kanter, who leads the benchmark effort at MLCommons. In the previous round it was taking mere seconds to perform some of the benchmarks.

Performance of the best machine learning systems on various benchmarks has outpaced what would be expected if gains were solely from Moore’s Law [blue line]. Solid line represent current benchmarks. Dashed lines represent benchmarks that have now been retired, because they are no longer industrially relevant.MLCommons

According to MLPerf’s calculations, AI training on the new suite of benchmarks is improving at about twice the rate one would expect from Moore’s Law. As the years have gone on, results have plateaued more quickly than they did at the start of MLPerf’s reign. Kanter attributes this mostly to the fact that companies have figured out how to do the benchmark tests on very large systems. Over time, Nvidia, Google, and others have developed software and network technology that allows for near linear scaling—doubling the processors cuts training time roughly in half.

First Nvidia Blackwell training results

This round marked the first training tests for Nvidia’s next GPU architecture, called Blackwell. For the GPT-3 training and LLM fine-tuning, the Blackwell (B200) roughly doubled the performance of the H100 on a per-GPU basis. The gains were a little less robust but still substantial for recommender systems and image generation—64 percent and 62 percent, respectively.

The Blackwell architecture, embodied in the Nvidia B200 GPU, continues an ongoing trend toward using less and less precise numbers to speed up AI. For certain parts of transformer neural networks such as ChatGPT, Llama2, and Stable Diffusion, the Nvidia H100 and H200 use 8-bit floating point numbers. The B200 brings that down to just 4 bits.

Google debuts 6th gen hardware

Google showed the first results for its 6^th generation of TPU, called Trillium—which it unveiled only last month—and a second round of results for its 5^th generation variant, the Cloud TPU v5p. In the 2023 edition, the search giant entered a different variant of the 5^th generation TPU, v5e, designed more for efficiency than performance. Versus the latter, Trillium delivers as much as a 3.8-fold performance boost on the GPT-3 training task.

But versus everyone’s arch-rival Nvidia, things weren’t as rosy. A system made up of 6,144 TPU v5ps reached the GPT-3 training checkpoint in 11.77 minutes, placing a distant second to an 11,616-Nvidia H100 system, which accomplished the task in about 3.44 minutes. That top TPU system was only about 25 seconds faster than an H100 computer half its size.

A Dell Technologies computer fine-tuned the Llama 2 70B large language model using about 75 cents worth of electricity.

In the closest head-to-head comparison between v5p and Trillium, with each system made up of 2048 TPUs, the upcoming Trillium shaved a solid 2 minutes off of the GPT-3 training time, nearly an 8 percent improvement on v5p’s 29.6 minutes. Another difference between the Trillium and v5p entries is that Trillium is paired with AMD Epyc CPUs instead of the v5p’s Intel Xeons.

Google also trained the image generator, Stable Diffusion, with the Cloud TPU v5p. At 2.6 billion parameters, Stable Diffusion is a light enough lift that MLPerf contestants are asked to train it to convergence instead of just to a checkpoint, as with GPT-3. A 1024 TPU system ranked second, finishing the job in 2 minutes 26 seconds, about a minute behind the same size system made up of Nvidia H100s.

Training power is still opaque

The steep energy cost of training neural networks has long been a source of concern. MLPerf is only beginning to measure this. Dell Technologies was the sole entrant in the energy category, with an eight-server system containing 64 Nvidia H100 GPUs and 16 Intel Xeon Platinum CPUs. The only measurement made was in the LLM fine-tuning task (Llama2 70B). The system consumed 16.4 megajoules during its 5-minute run, for an average power of 5.4 kilowatts. That means about 75 cents of electricity at the average cost in the United States.

While it doesn’t say much on its own, the result does potentially provide a ballpark for the power consumption of similar systems. Oracle, for example, reported a close performance result—4 minutes 45 seconds—using the same number and types of CPUs and GPUs.

Huang said SoftBank was the first to receive its new Blackwell chip designs

Huang elaborated on his philosophy by expressing that he is not driven by traditional ambition. Instead of striving for more, he focuses on doing better with what he already has.

Published at LXer: Have an NVIDIA GPU? Use GE-Proton? You should probably make sure you're on GE-Proton 9-15 which was just released. This is a "hotfix" build to clear up some problems from the...

NVIDIA nous annonce un nouveau bundle pour ses cartes graphiques RTX 4000. Ainsi, vous obtiendrez l'édition numérique premium d' Indiana Jones et le Cercle Ancien™ pour l'achat d'une carte graphique éligible NVIDIA GeForce RTX™ 4090, RTX 4080 SUPER, RTX 4080, RTX 4070 Ti SUPER, RTX 4070 Ti, RTX 4070 SUPER, RTX 4070 (ou d'un PC de bureau doté de l'un de ces modèles) ou d'un PC portable éligible doté d'un GPU RTX™ 4090, RTX 4080 ou RTX 4070. Tous les détails se trouvent dans la page dédiée à ce nouveau bundle. […]

De nouvelles rumeurs font écho d'un focus de la part de NVIDIA se concentrerait sur la production des futures RTX 5000, cela se traduirait par l'arrêt des chaines de production de la plupart des RTX 4000. Concrètement, NVIDIA aurait cessé de produire des puces AD102, 103, 104 et 106, seules les AD107 seraient encore en production, qui équipent classiquement les RTX 4060 desktop et les RTX 4050 laptop. […]

De nouveaux drivers sont disponibles du côté de NVIDIA, les GeForce Game Ready 566.14 WHQL. Outre la signature numérique, les nouveaux drivers promettent des performances optimales dans les jeux S.T.A.L.K.E.R 2: Heart of Chornobyl et Microsoft Flight Simulator 2024, ainsi que la prise en charge de la technologie DLSS3 dans ces titres. Le téléchargement est possible ici. […]

NVIDIA APP, ce nom ne vous parle peut-être pas encore et pourtant il s'agit du futur pour les cartes graphiques vertes, NVIDIA a, en effet, engagé une refonte de ses divers logiciels. Le but est de proposer une interface modernisée et d'unifier le panneau de configuration avec les applications GeForce Experience et RTX Experience. NVIDIA APP promet de devenir le centre névralgique de votre GPU, avec les profils de jeux, les nouveaux drivers… L'overlay (télémétrie de votre système de jeu) est toujours accessible grâce à la combinaison de touches ALT + Z, mais l'affichage est repensé avec une multitude d'informations (fréquences, tensions, latences, températures...) et le joueur pourra personnaliser pleinement les informations affichées : […]

Analogous to Marc Andreessen’s “software is eating the world”, Nvidia’s CEO Jensen Huang on the impact of AI: “AI is eating software,” Huang continued. “The way to think about it is that AI is just the modern way of doing software. In the future, we’re not going to see software that is not going to continue […]

The post Nvidia on AI everywhere first appeared on Tom Markiewicz.

(Telecompaper) Nvidia CEO Jensen Huang said SoftBank, GMO Internet Group, Highreso, KDDI, Rutilea and Sakura Internet are building AI infrastructure with Nvidia...

(Telecompaper) SoftBank is slated to receive the world's first Nvidia DGX B200 systems, which will serve as the building blocks for its new Nvidia DGX SuperPOD supercomputer...

(Telecompaper) SoftBank announced a series of collaborations with Nvidia to deploy what they claim is a new kind of telecommunications network that can run AI and...

"Every other telco will have to follow this new wave," SoftBank Group CEO Masayoshi Son said at an AI event where he was speaking alongside Nvidia CEO Jensen Huang.

AWS said researchers from Carnegie Mellon University and the University of California, Berkeley, are taking part in the program. The company plans to make 40,000 of the first-generation Trainium chips available for the program.

Nvidia CEO Jensen Huang has requested SK Hynix to expedite the delivery of HBM4 chips by six months. This request, revealed by SK Group Chairman Chey Tae-won, highlights the surging demand for Nvidia's AI accelerators, which heavily rely on HBM chips.

The round was led by venture capital firm IVP and valued the company at about $520 million, according to the company. NEA, NVIDIA, Databricks, and Bessemer Venture Partners also participated in the round.

Microsoft's clusters of Advanced Micro Devices' flagship MI300X AI chips will be sold through its Azure cloud computing service. AMD, which expects $4 billion in AI chip revenue this year, has said the chips are powerful enough to train and run large AI models.

MLPerf is considered the most reliable global AI semiconductor benchmark competition. Leading companies and research institutes such as Google, Microsoft, Facebook, Stanford University and Harvard University host the event annually.

Spyrosoft today announced it has joined NVIDIA Inception, a program designed to nurture startups revolutionising industries with advancements in AI and data science.

The future of AI goes far beyond individuals using ChatGPT. Companies are now integrating artificial intelligence into all aspects of their businesses. One key player in this transition is Nvidia, the AI-driven computing company, which makes both hardware and software for a range of industries. In this episode, HBR editor in chief Adi Ignatius speaks with Nvidia’s CEO and cofounder Jensen Huang at HBR’s Future of Business conference about how he keeps his company agile in the face of accelerating change and where he sees AI going next.

Atlantic.Net?empowers customers to advance offerings through market-leading AI cloud solutions, helping to speed up the adoption of AI compute for software developers and businesses

Mastek's next-gen solution dramatically reduced customer response time

We are proud to announce that we recently joined the NVIDIA Inception Program, which will help to speed up our deep learning development process and therefore offer the best possible audio processing tools to our users.

What is NVIDIA Inception

NVIDIA is a global leader in hardware and software for Artificial Intelligence (AI).
Their NVIDIA Inception Program will enable us to leverage NVIDIA's cutting-edge technology by accessing more diverse cloud and GPU (Graphics Processing Unit) product offerings, which are used in most Machine Learning and Deep Learning model training instances worldwide. This will allow us to streamline AI development and deployment and train bigger machine-learning models to test and evaluate algorithms faster. The program will also offer us the opportunity to collaborate with industry-leading experts and other AI-driven organizations, among other things.

Our Deep Learning Development Process

For our development process, more GPU capacity means a great saving of time and therewith of course a saving of costs. As an example, one training cycle of our dynamic denoiser model takes almost a week trained with GPUs, however the same training cycle trained with CPUs would take several months.

To illustrate, a CPU (Central Processing Unit) can be compared to a race car, which is very fast but can only transfer a small number of packages, while a GPU in this comparison is a big truck, which can transfer a huge number of packages more slowly. Deep learning algorithms require for training very large datasets consisting of thousands of files, therefore our 'trucks', the GPUs, are the best hardware to choose processing multiple computations simultaneously.

The more GPU capacity we can use, the faster we get results for our tested algorithms, and the faster we know which way we should follow to offer our users the best possible audio processing tools.
Unfortunately, the world is right in the middle of a Global Chip Shortage, so the latest GPUs are very hard to get and super expensive to purchase – unless you have a partnership with a GPU manufacturer.

Conclusion

We are happy to join such a renowned program and look forward to the updates to our product that we will be able to implement and potentially a greater industry transformation.

You can read our full press release here: AuphonicNVIDIAInceptionPressRelease (pdf)

Want to know the kind of performance you will see at 1440p on an NVIDIA GeForce RTX 2060 SUPER FE when it is overclocked? Check out our gaming review.... [PCSTATS]

When NVIDIA debuted its GeForce Experience app over a decade ago, the online login requirement made many PC hardware grognards (like this author) leery of the software despite its many useful functions. The green team made many updates to the GeForce Experience application over the years, but it has ever remained optional, to NVIDIA's credit. If

As all things must come to an end, NVIDIA's GeForce RTX GPUs follow in the cycle of life. With NVIDIA going full-throttle ahead with its GeForce RTX 50 series launch, it appears that its current RTX 40 series GPUs are about to be sunset (most of them, anyway). NVIDIA has been absolutely buzzing with productions for the AI craze sweeping the

The Italian competition watchdog requested the investigation for fear the acquisition will affect trade in the single market.

Read more: EU investigates Nvidia acquisition of Run:ai over competition risk

This all-in-one app makes it easier to maximize your Nvidia GPU.

NVIDIA saw blowout second-quarter results, surging margins, and incredible demand, which prompted one analyst from Constellation Insights to conclude that "it's clear the company has little competition and a lot of pricing power."

AI chipmaker Nvidia to join Dow Jones, replacing rival Intel