ReportWire

Tag: supercomputer

  • Why Your Business Should Consider Nvidia’s New $4,000 Desktop Supercomputer

    [ad_1]

    Nvidia’s desktop supercomputer went on sale Wednesday, evoking FOMO from some tech enthusiasts who weren’t able to get their hands on one and raves from those who did. But the DGX Spark may not be a must-have for everyone in the AI world.

    First announced in January at CES, the DGX Spark was originally priced at $3,000 and was scheduled to be released in May. Things change in the tech world, though.

    The Spark was originally known as “Project DIGITS,” but got a name upgrade. The launch was pushed back another five months also—and during that time Nvidia increased the price to $3,999. (That’s still far less than the $129,000 Nvidia charged for the DGX-1 supercomputer in 2016.)

    That hasn’t reduced demand, however, as the device sold out almost immediately on Wednesday. Described as powerful enough to build complex AI models but small enough to fit on your desk, it’s a computer that could democratize AI beyond the corporate giants, broadening innovation.

    Curious about the DGX Spark or thinking about getting one when supplies are restocked? Here’s what you need to know.

    How can I order a DGX Spark?

    The Spark can be ordered online at nvidia.com, as well as from select partners and stores, including Micro Center and PNY. (The system is sold out on Nvidia’s website and the partner sites appear to be out of stock as well.)

    Who is the DGX Spark designed for?

    With its eye-popping specifications, the DGX Spark might seem like a dream machine for any power-user of PCs, but it’s best suited for researchers. (It won’t play video games and is entirely overpowered for web browsing.) Nvidia CEO Jensen Huang, when introducing the device, said that “placing an AI supercomputer on the desks of every data scientist, AI researcher and student empowers them to engage and shape the age of AI.”

    The company envisions the initial target audience will be twofold: Developers at large tech companies, who can create a working use case (in other words, a practical application of AI technology), which can then be scaled via data centers or cloud computing—or smaller developers who don’t have the finances to access those data centers, but still have ideas for new applications.

    Nvidia has its eyes on a broader audience in the long-term, though. At that same CES keynote, Huang noted that in the near future, anyone “who uses computers as a tool” will need their own personal AI supercomputer.

    What are the system specs of the DGX Spark?

    If you’re not super fluent in computer-speak, brace yourself. The Spark uses Nvidia’s GB10 Grace Blackwell (or, if you prefer, just Blackwell) GPU chip and has 128 GB of GPU memory. It boasts up to 4TB of NVMe SSD (solid state) storage. And Nvidia says it can deliver a petaflop of AI performance, which works out to a quadrillion calculations each second (technically, these calculations are called FLOPS, for “floating point operations per second).

    For comparison, the fastest supercomputer in the world is El Capitan at the Lawrence Livermore National Laboratory. It is rated at 2.79 exaflops and is designed to “help researchers ensure the safety, security, and reliability of the nation’s nuclear stockpile in the absence of underground testing.” However, it’s not for sale. 

    What can all of the DGX Spark’s horsepower actually do?

    The Spark is capable of handling AI models that have as many as 200 billion parameters. That’s something that used to require access to data centers that were far, far beyond the budget of smaller developers. Making a more affordable system will let developers prototype, fine-tune and test complex AI models. And if a model is too big for a Spark to handle, the computer can be linked to another Spark to let researchers move forward with testing.

    Will there be other versions of the DGX Spark?

    Yes. Several Nvidia partners will make their own desktop supercomputers using the Nvidia GB10 Grace Blackwell Superchip, which powers the Spark. Among the PC companies that will offer these are Dell, Asus, Acer, HP, Lenovo, MSI and Gigabyte. None are currently available. However, the majority are expected later this year (with the caveat that some could slip to early 2026). Expect to pay largely the same as the Spark’s $4,000 for these.

    [ad_2]

    Chris Morris

    Source link

  • Nvidia

    Nvidia

    [ad_1]

    Workers install cooling fans on a supercomputer that will train Tesla’s new Autopilot. The supercomputer will consist of 50 thousand Nvidia H100 accelerators. Such a data center requires approximately 75 megawatts of electricity. Located in a gigafactory in Texas.

    [ad_2]

    Source link

  • This record number in Nvidia earnings is a scary sight

    This record number in Nvidia earnings is a scary sight

    [ad_1]

    Nvidia Corp.’s financial results had a bit of a surprise for investors, and not on the good side — product inventories doubled to a record high as the chip company gears up for a questionable holiday season.

    Nvidia reported fiscal third-quarter revenue that was slightly better than analysts’ reduced expectations Wednesday, but the numbers weren’t that great. Revenue fell 17% to $5.9 billion, while earnings were cut in half thanks to a $702 million inventory charge, largely relating to slower data-center demand in China.

    Gaming revenue in the quarter fell 51% to $1.57 billion. Nvidia said it is working with its retail partners to help move the currently high-channel inventories.

    While the company was writing off the inventory for China, its own new product inventory was growing. Nvidia
    NVDA,
    -4.54%
    reported that its overall product inventory nearly doubled to $4.45 billion in the fiscal third quarter, compared with $2.23 billion a year ago and $3.89 billion in the prior quarter. Executives cited its coming product launches, designed around its new Ada and Hopper architectures, when asked about the inventory gains.

    In the semiconductor industry, high inventories can make investors nervous, especially after the industry had so many supply constraints in recent years that quickly swung to a glut of chips in 2022. With doubts about demand for gaming cards and consumers’ willingness to spend amid sky-high inflation this holiday season, having all that product on hand just amps up the nerves.

    Full earnings coverage: Nvidia profit chopped in half, but tweaked servers to China offset earlier $400 million warning

    Chief Financial Officer Colette Kress told MarketWatch in a telephone interview Wednesday that the company’s high level of inventories were commensurate with its high levels of revenue.

    “I do believe….it is our highest level of inventory,” she said. “They go hand in hand.” Kress said she was confident in the success of Nvidia’s upcoming product launches.

    Nvidia’s revenue reached a peak in the April 2022 quarter with $8.3 billion, and in the past two quarters revenue has slowed, with gaming demand sluggish amid a transition to a new cycle, and a decline in China data-center demand due to COVID-19 lockdowns and U.S. government restrictions.

    For its data-center customers, the new architectures promise major advances in computing power and artificial-intelligence features, with Nvidia planning to ship the equivalent of a supercomputer in a box with its new products over the next year. Those types of advanced products weigh on inventory totals even more, Kress said, because of the price of the total package.

    “It’s about the complexity of the system we are building, that is what drives the inventory, the pieces of that together,” Kress said.

    Bernstein Research analyst Stacy Rasgon believes that products based on Hopper will begin shipping over the next several quarters, “at materially higher price points.” He said in a recent note that he believes Nvidia’s numbers were likely hitting a bottom in this quarter.

    “We remain positive on the Hopper ramp into next year, and believe numbers have at this point likely reached close to bottom, with new cycles brewing and an attractive secular story even without China potential,” Rasgon said in an earnings preview note Tuesday.

    Read also: Warren Buffett’s chip-stock purchase is a classic example of why you want to be ‘greedy only when others are fearful’

    Nvidia Chief Executive Jensen Huang reminded investors on a conference call that the company’s inventories are “never zero,” and said everyone is enthusiastic about the upcoming launches. But it doesn’t take too long of a memory to conjure up a time when Nvidia went into a holiday with an inventory backlog that included new architecture and greatly disappointed investors: Four years ago, Huang had to cut his forecast for holiday earnings twice amid a “crypto hangover” with similar dynamics to the current moment

    Investors need faith that this holiday season will not be the same, even as demand for some videogame products declines after a pandemic boom just as the market for cryptocurrency — some of which has been mined with Nvidia products — hits a rough patch. Huang said that Nvidia’s RTX 4080 and 4090 graphics cards based on the Ada Lovelace architecture had an “exceptional launch,” and sold out.

    Nvidia shares gained more than 2% in after-hours trading Wednesday, suggesting that some are betting that this time will be different. That enthusiasm needs to translate into revenue for Nvidia so that this big gain in inventories does not end up being part of another write-down at some point in the future.

    [ad_2]

    Source link