Artificial Intelligence - Page 32

South Korea's memory chip exports to Taiwan surged over 225% year-over-year, thanks to the unstoppable demand of AI GPUs and AI accelerators and their respective HBM memory chips.

In a new report from the Korea Times, we're learning that outbound shipments of HBM memory chips to Taiwan reached $4.26 billion in the first 6 months of this year, up an incredible 225.7% from a year ago, "far outperforming" South Korea's overall increase of memory chip exports at 88.7%, according to data compiled by the industry ministry and the Korea International Trade Association.

Taiwan was the third-largest importer of South Korean memory chips in the period, pushing out both Vietnam and the United States. Market watchers said that the huge surge in memory chip exports to Taiwan is thanks to SK hynix's fantastic HBM supply to NVIDIA, which uses its HBM memory inside of its AI GPUs, and fabs its chips over at TSMC in Taiwan.

NEO Semiconductor has just unveiled the development of its new 3D X-AI chip technology, which aims to replace DRAM chips inside of HBM to solve data bus bottlenecks, by enabling AI processing through 3D DRAM.

The new 3D X-AI chip technology can reduce the huge amount of data transferred between HBM and GPUs during AI workloads, with NEO's innovative new 3D X-AI chip technology set to "revolutionize the performance, power consumption, and cost of AI chips for AI applications like generative AI".

NEO's new 3D X-AI chip technology has 100x the performance with 8000 neuron circuits to perform AI processing in 3D memory, a huge 99% power reduction that minimizes the requirement of transferring data to the GPU for calculation, reducing power consumption and heat generation by the data bus, and 8x the memory density with 300 memory layers, allowing HBM to store larger AI models.

When it comes to AI training and dealing with large data sets and increasingly complex large language models (LLMs) like Llama 70b, it's not simply a matter of being able to throw GPU horsepower at the problem until you find a solution. At least, it shouldn't be.

Phison's aiDAPTIV+ is a hybrid software and hardware solution for LLM training. It integrates Phison's Pascari A100 M.2 SSDs into a complete solution with linear scaling. Impressive! According to Phison, it unlocks access to run workloads previously reserved for data centers on a single workstation or server - supporting up to Llama-3 70B and Falcon 180B.

Phison's chart showcases the capabilities of aiDAPTIV+. Above, you can see a single system with the same configuration: four RTX 6000 Ada GPUs, 192GB of GDDR6 memory, and an additional 512GB of RAM. Phison's aiDAPTIVLink middleware extends this GPU memory capacity with two 2TB SSDs, paving the way for massive model support with low latency. This is impressive stuff, and it won the "Best of Show, Most Innovative AI Application" award at FMS: the Future of Memory and Storage.

It was only last month we heard about Apple, NVIDIA and many other big name players in the AI race being caught up in an investigative report that found they all used a public data set containing YouTube video transcripts to train their respective AI products, which is a violation of YouTube's terms-of-service (TOS).

YouTube has said in the past that any "unauthorized scraping or downloading of YouTube content" is strictly prohibited, and it's especially prohibited when that data is then used for commercial projects. Last month, a Proof News investigation found NVIDIA, Apple, and other AI companies used an academic data set containing subtitles from more than 170,000 YouTube videos to train AI models, and now NVIDIA has been caught in the spotlight again with a report from 404 Media.

According to the publication that spoke with a former NVIDIA employee about the company's internal processes, employees were instructed to scrape videos from Netflix, YouTube, and other sources to add to the data sets that are being used to an AI model for NVIDIA's Omniverse 3D world generator, self-driving car systems, a "digital human" AI avatar product, and the Cosmos deep learning model.

Dell announced its company is "getting leaner" with executives informing thousands of employees via a memo it has begun its effort into "streamlining layers of management."

President of global sales and customer operations Bill Scannell and global channels president John Byrne explained in the memo the company is attempting to grow faster in the market by concentrating its efforts on the development of "modern IT and AI" and that it can seemingly achieve that without 10% of it workforce, or approximately 12,500 employees. The layoffs are part of Dell's overall push to reduce its workforce to under 100,000 people.

These layoffs at Dell come only a matter of days after Intel announced it was laying off 15% of its workforce, amounting to 15,000 job cuts and a 26% stock drop. The massive stock drop is attributed to recently confirmed hardware issues with Raptor Lake chips, which is causing a fire internally for the CPU manufacturer. As for Dell, last year, the company planned on reducing its workforce by 5% but ended up reducing it by double that, or 13,000 people.

Since the explosion of AI-powered models, academics, teachers, and many other professions have been rightfully concerned about tools such as ChatGPT being used to assist in the writing of essays and research papers.

The easy accessibility to powerful AI models such as ChatGPT has given rise to a substantial increase in the amount of AI-generated content across the internet. Unfortunately, it appears some of this content has already made its way into scientific journals within submitted papers, and classrooms with students taking advantage of these free tools to write assignments and papers.

To prevent the abuse of this technology in the academic setting, companies rolled out AI detection tools. Unfortunately, these detection tools proved unreliable, but according to a new article by The Washington Post, OpenAI, the creators of ChatGPT, have developed a new method that can detect ChatGPT-generated content with a 99.9% accuracy rate. The new system issues a watermark to ChatGPT-generated content that cannot be seen by the human user but detected by the detection too.

NVIDIA's next-generation Blackwell AI GPU has been reportedly experiencing some issues, which have caused some major delays that will see shipments kicking off in 2025... except for Foxconn.

Foxconn is the only company that has received enough of NVIDIA's new Blackwell GB200 Superchips, so it will be the only company pumping out new GB200 AI servers. According to UDN, they will be available in Q4 2024.

NVIDIA's new GB200 AI servers were meant to be shipping in large quantities in Q4 2024, but the yield issues with B200 forcing a redesign for B200A this time next year, is causing some major headaches for some of the biggest companies on the planet. UDN reports that it was found that the problem could stem from "unsatisfactory" yield rates of the advanced packaging required for GB200.

Elon Musk has said he wants to "give people superpowers" with his brain-computer interface (BCI) implants through Neuralink, which the SpaceX and Tesla founder said will let you "outperform a pro gamer".

Elon wants to increase humans' "output rate." With Neuralink's BCI, he said that there's the potential to have "three, maybe six, maybe more orders of magnitude" of how fast our brain processes signals to the chip.

The SpaceX and Tesla boss sat down for the latest episode of the Lex Fridman podcast, with Fridman saying there would be "hundreds of millions" of people getting Neuralink implants in the "next couple of decades" to which Elon agreed.

Stability AI's Stable Diffusion 3 is the latest version of its popular and powerful AI image-generation tool. It can deliver photorealistic results thanks to what it describes as the "most advanced text-to-image open model yet." For those following the rise of AI image generators, one thing that Stable Diffusion 3 brings to the table - or solves - is the ability to recreate repeating patterns and, yes, human hands.

It's impressive, and the model runs on hardware ranging from an Apple M2 Ultra to a GeForce RTX 4090-powered rig. Naturally, with its 24GB of GDDR6X memory and unlocked Ada Lovelace power, the latter is several magnitudes faster than any other "at-home" option. Still, with TensorRT acceleration and optimization, the GPU can render or generate images in real-time.

You can see this in action in NVIDIA's video, which it presented at SIGGRAPH 2024. The video showcases SDXL Turbo generating an image of "a hot rod, racing in the desert at sunset" in real-time, adding details to the image like adding a canyon to the background as it's being typed in. Impressive!

On August 8 the NFL will roll out new facial recognition technology across its thirty-two stadiums located in various states.

The new facial recognition software is called Wicket and it's powered by an artificial intelligence. The new system is designed with the goal of decreasing the time it takes people to get in and out of stadiums, verifying credentials, and enabling staff to move more easily around stadiums. Notably, the software is primarily designed to assist credentialed officials in verifying their identify at restricted areas, and to prevent any fraudulent use of counterfeit physical credientials.

As for ticketholders, some teams have extended the technology to customers, which will require customers to link their ticketing account to their facial image selfie. The Wicket software will create a unique digital code that represents a customer's facial features instead of storing an actual image of a person.

NVIDIA has hit some major roadblocks with its new Blackwell AI GPUs, with redesigns in the works, and a new B200A on the way... but in the middle of this mess, we get some new details on the specs of Blackwell Ultra.

NVIDIA first teased its mid-generation refresh "Blackwell Ultra" earlier this year, as well as its next-gen Rubin R100 AI GPU that will feature ultra-fast HBM4 memory in 2025 (if there's no delays for Rubin, as there are with Blackwell).

SemiAnalysis has the scoop here, with Blackwell Ultra acting as a mid-generaiton enhancement of Blackwell, with the standard CoWoS-L Blackwell Ultra known as "B210" or "B200 Ultra". Inside, NVIDIA's new Blackwell Ultra features both a memory refresh -- up to 288GB of 12-Hi HBM3E memory -- and performance enhancements of up to 50% in FLOPS.

NVIDIA is reportedly experiencing some major issues with its new Blackwell AI GPUs, with the new B100 and B200 AI GPU shipments heavily impacted, with design flaws causing major headaches not just for NVIDIA.

In a new report published by SemiAnalysis, the outlet reports that there are issues with the Blackwell die needing a possible redesign or that the bridge dies inside TSMC's new CoWoS-L advanced packaging needs a redesign.

The result? NVIDIA is now reportedly working on the B200A AI GPU, which will be released towards the second half of 2025 and won't have the issues that are now plaguing the fastest AI GPU before it could hit the market.

NVIDIA's upcoming Blackwell AI GPU family is "encountering major issues" according to a new detailed breakdown by SemiAnalysis, which will impact production targets through into the first half of 2025.

The main problems are from the NVIDIA Blackwell GPU architecture and the chip packaging used: TSMC's new CoWoS-L. SemiAnalysis explains that one of the rumors swirling is that one of the top few global routing metal layers and bump out of the Blackwell die needs a redesign, while another is that the bridge dies inside the CoWoS-L packaging need a redesign.

SemiAnalysis explains that there has a been a warpage in the CoWoS-L packaging that was caused by heat, with TSMC's new CoWoS-L advanced packaging being more complex than the now more mainstream CoWoS-S packaging.

Elon Musk has officially dubbed Tesla's new supercomputer cluster at Giga Texas "Cortex" which will be powered by 100,000 x NVIDIA H100/H200 AI GPUs with "massive storage" for video training its FSD and Optimus projects.

Elon Musk revealed the new name of the Tesla Gigafactory Texas supercomputer cluster in a post on X, which comes as no surprise to the SpaceX and Telsa (and X) boss.

Elon did say that the new Cortex supercomputing cluster will feature around 100,000 x NVIDIA H100 and H200 AI GPUs that will be used to video train the neural networks for FSD (Full Self-Driving) and Optimus robotics.

The US Department of Justice is being pushed to investigate NVIDIA over anti-competitive concerns in its dominance in the AI GPU business.

Progressive groups backed by Senator Elizabeth Warren, have written a letter to anti-trust boss Jonathan Kanter, to open an investigation into NVIDIA. The push is coming from Demand Progress and 9 other entities that push for government scrutiny against monopolies in business, and now NVIDIA is in their crosshairs.

A magnifying glass is being put over NVIDIA's bundling of hardware and software, which has created a tight-grip monopoly on the AI GPU market. The groups wrote: "This aggressively proprietary approach, which is strongly contrary to industry norms about collaboration and interoperability, acts to lock in customers and stifles innovation".

NVIDIA's next-gen Blackwell AI GPUs have reportedly been delayed, with The Information reporting that "design flaws" are stopping Blackwell from flooding the world of AI.

The Information reports on information from two unidentified people that helped produce the Blackwell AI GPU and its server hardware said that NVIDIA's new Blackwell AI GPUs could be delayed for 3 months or more, affecting major customers like Meta, Google, and Microsoft.

NVIDIA reportedly made Microsoft aware of the issue of Blackwell AI GPUs being delayed this week, at least its "most advanced" AI chip models, according to Bloomberg reporting on unidentified Microsoft staffers and "another person." The delays mean that big shipments aren't to occur until Q1 2025, according to The Information.

NVIDIA ChatRTX is an AI chatbot that is fully customizable, from the LLM used to the data. The company describes it as an AI that lets you "chat with your files" as you can customize the data set across different file formats, documents, images, and other data. The latest version of the application is available now, with a minimal and clean UI overhaul that highlights how powerful it can be.

As you can see in the screenshot, the new UI (built using the Electron + Material UI framework) highlights ChatRTX's two main components: the AI Model and the Data. Even better, users and developers can add their own UI elements and expand ChatRTX's functionality by using an open UI framework.

ChatRTx can tap into powerful AI models for customization and, as such, requires at least a GeForce RTX 30 or GeForce RTX 40 Series GPU with at least 8GB of VRAM to run.

At SIGGRAPH 2024, NVIDIA introduced and expanded on the future of digital humans by introducing its latest AI advancement - adding a path-traced realistic-looking human face (and voice) with emotion and detailed animation to the idea of interacting with an AI assistant. This is a massive leap forward compared to the usual text-based interface for AI chatbots and assistants, and it was brought to life by the latest in RTX technologies.

'James' is built using similar NVIDIA ACE technology that we went hands-on with at Computex 2024 as part of an interactive game demonstration where you interact with digital humans in real-time to solve a mystery. This time, however, it's being used in the context of a virtual assistant or customer service agent.

'James' is also built on NVIDIA NIM microservices; Jensen Huang explains how it works and confirms that the visual 'RTX On' front end can connect to existing AI chatbots like ChatGPT.

Google has been battling deep fakes of people who didn't consent to their creation for quite some time, but recently, the search engine giant has upped its efforts.

The increase in its efforts against AI-generated non-consensual deepfakes is in response to the rise of generative AI tools that are making it easier for people to create deepfakes of people without their consent. Google has decided to respond to the influx of AI-generated content with a new system that streamlines the request for removal process from the victims' perspective, along with improvements to the response to removal requests.

Additionally, when the new system confirms a request it is then added to the algorithm, which is designed to filter out any other explicit instances related to the individual from Google Search. Any duplicates of the content will automatically be removed, even if its posted on various social media platforms. Moreover, the new system works to proactively reduce the spread of deepfake content that is emerging.

On the heels of Reddit rolling out an update that booted Microsoft's Bing from accessing its data, Reddit's CEO has come out slammed Microsoft and other big tech companies for refusing to negotiate.

Reddit pivoted early this year and made major changes to its business, particularly how it handles its API with third parties. Seeing the massive rise in demand for data to train AI models, Reddit altered its API policies and began charging for access to its massive data pool. This change nuked popular third-party Reddit apps that were used by thousands of users. Reddit users caused the platform to go dark after that, but it has seemingly recovered.

These new API changes and Reddit now selling access to its data scored the company a deal with Google, which includes the $60 million a year paycheck for direct access to Reddit's wells of user data. Additionally, Reddit also signed on to OpenAI, which enables Reddit posts to appear in responses from ChatGPT and other AI tools. No specific financials were revealed about the deal.