Dell Unveils Groundbreaking 20-PetaFLOPS Desktop Powered by Nvidia's GB300 Superchip

Dell delivers 20-petaFLOPS desktop built on Nvidia's GB300

HPE and Lenovo also have plans to put Jensen's latest hardware to work GTC The age of the 20-petaFLOPS desktop is upon us as Dell announced a machine capable of achieving that massive processing power today at Nvidia's GPU Technology Conference in Silicon Valley. The Pro Max with GB300 takes its name from the GB300 Grace Blackwell Ultra Desktop Superchip Nvidia announced earlier in the day. The GB300 features an Nvidia Blackwell Ultra GPU with next-generation Tensor Cores and FP4 precision. It includes 784 GB of unified system memory - 288 GB of HBM3e in the GPU and 496 GB of LPDDR5X for the CPU - optimized for large-scale training and inference workloads. The GB300 connects via NVLink-C2C, ensuring high-speed, CPU-GPU coherent memory access. The workstation also includes the Nvidia ConnectX-8 SuperNIC, optimized for AI computing workloads. It supports networking speeds of up to 800 Gbps, enabling ultra-fast interconnects for distributed AI training and data transfers. Kevin Terwilliger, Dell VP and general manager for commercial, consumer, and gaming PCs, told The Register the machine "will make it much easier for developers to prototype, to test, and even scale their models into production environments." The VP thinks developers will use it to run large-scale training and inferencing workloads as they want to work on AI models at a scale typically deployed to servers. 20-petaFLOPS desktops are therefore needed. Before you ask, the Pro Max with GB300 should be able to run Doom as it can ship with either Ubuntu or Nvidia's Base OS. Dell also delivered a mini-workstation powered by the petaFLOPS-packing GB10 Grace-Blackwell desktop processor. We await news of pricing for both models, and details of their probably formidable I/O and storage options. As you'd expect, Dell has also made sure its servers can use Nvidia's newly announced Blackwell Ultra silicon. The firm's PowerEdge servers can house the B300 NVL16, GB300 NVL72, and RTX PRO 6000 Blackwell Server Edition units. The B300 NVL16 and GB300 NVL72 feature Nvidia's Blackwell Ultra GPUs, using NVLink-C2C for high-bandwidth interconnectivity. The RTX PRO 6000 Blackwell Server Edition is designed for professional visualization and AI workloads, offering enhanced memory bandwidth and performance for enterprise AI applications. Perhaps more interesting is the PowerEdge XE8712, 36 of which can be stacked in an Open Rack v3, accommodating up to 144 GB200 NVL4 accelerators. Dell says its Direct Liquid Cooling (DLC) will keep them all cool and support up to 264 kW of power per rack. We're told the machines will come in handy for scientific research, modeling financial markets, conducting genomic sequencing, or inclusion in a Dell/Nvidia AI Factory - the term Jensen Huang's gang uses to describe hardware dedicated to AI workloads. HPE has also made sure its ProLiant servers can handle Nvidia's latest. The company's Compute XD servers will support the HGX B300, while the DL384b Gen 12 machines handle the GB200 Grace Blackwell NVL4 Superchip. HPE ProLiant Compute DL380a Gen 12 hardware will run the RTX PRO 6000 Blackwell Server Edition silicon. Lenovo's pre-show announcements mention hardware that can run Nvidia's latest chips, but didn't specify any new models or which of its existing systems can handle Blackwell Ultra hardware. The Chinese company instead emphasized how it has bundled its hardware into solutions for workloads including AI reasoning, agentic AI, and real-time video generation. It's not alone in having built solutions and services around Nvidia's new toys. Dell and HPE have also assembled bundles and teams to help their customers build agents and implement other AI-infused applications. Dell has even created its own AI coding assistant that runs on-prem for orgs that think their developers need a hand but don't want their code to get anywhere near a cloud or SaaS platform. All three vendors' services are available now, even though we're told hardware hosting Nvidia products announced at GTC won't debut until the second quarter of 2025 - or even later in the year. ®

Nvidia wants to put a Grace Blackwell Ultra on your desk

GTC After a Hopper hiatus, Nvidia's DGX Station returns, now armed with an all-new desktop-tuned Grace-Blackwell Ultra Superchip capable of churning out 20 petaFLOPS of AI performance. The system marks the first time Nvidia has updated its DGX Station lineup since the Ampere GPU generation. Its last DGX Station was an A100-based system with quad GPUs and a single AMD Epyc processor that was cooled by a custom refrigerant loop. By comparison, Nvidia's new Blackwell-based systems are far simpler, powered by a single Blackwell Ultra GPU and, as its GB300 codename would suggest, are backed by a Grace CPU. In total, we're told the system will feature 784 GB of "unified memory between the CPU's LPDDR5x DRAM and the GPUs' HBM3e." However, the system doesn't just cram the CPU and GPU onto the same mainboard. It also features an 800 Gbps ConnectX-8 NIC onboard in case you want to create a mini cluster of DGX Stations. In addition to its own gold-clad boxen, Nvidia says the GB300 Desktop board will also be available from OEMs including ASUS, BOXX, Dell, HP, Lambda, and Supermicro later this year. Alongside the DGX Station, Nvidia has started taking reservations for its Project Digits box. Now called the DGX Spark, the $3,000 system, first teased at CES earlier this year, is powered by a GB10 Grace Blackwell system-on-chip. In terms of performance, the system boasts up to 1,000 trillion operations per second of AI compute and 128 GB of unified system memory in addition to integrated ConnectX-7 networking. The L40S gets an RTX PRO replacement Alongside its new workstations, Nvidia also unveiled a series of new PCIe-based GPUs aimed at professional workstations and server applications. At the top of the stack are Nvidia's 96 GB RTX PRO 6000-series server and workstation chips. The parts are designed to replace Nvidia's aging L40S and RTX 6000 Ada graphics cards and boast between 2.5x and 2.7x higher floating-point performance at 3,753 and 4,030 teraFLOPS respectively. Of course, this doesn't tell the whole story, as the RTX PRO 6000 cards support the new FP4 datatype, while the Ada Lovelace GPUs do not. Normalized for FP8 performance, the gains aren't nearly so impressive, up roughly 28 percent on the server card and 38 percent for the workstation card. With that said, FP4 support is good for more than just juicing Nvidia's FLOPS claims. It also has the benefit of shrinking model sizes from 1-2 bytes per parameter at FP8 or FP16 to just 4 bits, which means enterprises can cram larger more capable models into fewer GPUs than before. Today, massive 600 billion-plus parameter models from the likes of DeepSeek, OpenAI, Anthropic, or Google often end up stealing the limelight. However, these models are usually overkill for the kinds of retrieval-augmented generation or customer service chatbots enterprises are interested in deploying. Because of this, models in the 24-70 billion parameter range have become quite popular as they're large enough to get the job done and are relatively easy to fine-tune, but are so big as to be impractical to deploy on more modest GPUs, like the L40S. Meta's Llama 3.3 70B, Mistral-Small V3, Alibaba's QwQ, and Google's Gemma 3 are just a handful of the mid-sized models aimed at enterprise deployments. With 96 GB of capacity, Nvidia's RTX PRO 6000-series cards can now run models that previously would have required two L40Ses on a single card, or run models twice that size if you're willing to take advantage of 4-bit floating-point quants. In addition to offering twice the capacity of their predecessors, the cards also offer between 66 and 88 percent higher memory bandwidth at 1.6-1.8 TBps for the server and workstation parts respectively. If you're wondering why there's such a large variance between the server and workstation card, we're told the main difference between the two is cooling. The workstation card is equipped with a larger active cooler reminiscent of the one found on the RTX 5090, while the server part features a more traditional passive heatsink. Speaking of thermals, this higher floating-point performance and memory capacity does come at the cost of higher power draw. According to Nvidia, the workstation card is rated for up to 600 W of power draw, twice that of the RTX 6000 Ada it replaces. The good news is that if you're looking for something more power frugal, Nvidia will also offer a Max-Q variant of the RTX PRO 6000 aimed at high-end mobile workstations. The mobile GPU promises the same 1.8 TBps of memory bandwidth, though it tops out at 24 GB and only manages 3.75 petaFLOPS of FP4 performance. The RTX PRO 6000 Blackwell Workstation Edition will be available in April, with server versions arriving in May and mobile variants launching in June. Nvidia also plans to release RTX PRO 5000, 4500, and 4000 across the workstation and mobile segments beginning later this summer. ®

Dell's new RTX Pro AI PC boasts an 'unlimited turbo' mode

Dell's renamed Precision workstation desktops and laptops are bursting with the latest technology. Dell's new Pro Max AI PCs boast the new Nvidia RTX Pro Blackwell GPUs, making them desirable to AI developers for that reason alone. But the new desktop and laptop lineup also include several nerdy new innovations that we haven't seen before, including an "infinite turbo" mode that most gamers would pine for and 18-inch displays. Dell's "Pro Max" lineup isn't for the everyday user. Instead, the Pro Max is what Dell traditionally called the Precision lineup: workstations, basically. Dell will ship its 14- and 16-inch laptops with the new Nvidia RTX Pro laptop GPUs in July, alongside AMD Ryzen and Intel Core Ultra Series 2 chips; versions with integrated graphics will ship on March 27. Dell's desktops -- the Slim, Micro, and Tower series -- will be available with "Ada"-class RTX 4000-series GPUs on March 27; versions with the Core Ultra Series 2 desktop or AMD Threadripper CPUs plus the RTX Pro Blackwell desktop GPUs will ship in July. Dell did not announce prices. Dell also plans to sell Nvidia's new GB300 Grace Blackwell Ultra Desktop Superchip inside a "desktop" design. That's 20,000 TOPS, Nvidia says, plus a whopping 784GB of memory (288GB to the GPU and the remaining 496GB to the CPU) to basically run all of the AI models that you can think of. That "desktop," though, will be likely out of reach for many. Dell's Tower, Slim, and Micro Pro Max PCs should be priced much more affordably, and include some cool tech that is making us drool already. Dell claims that its Dell Pro Max Tower T2 will be the world's fastest tower for single-threaded application performance, based on what Dell is calling its exclusive unlimited turbo duration technology. As many enthusiasts know, Intel and AMD cite both a base clock speed and a "turbo" clock speed at which one or all of its cores can run. Eventually, the CPU has to drop out of its turbo or boost speed to avoid overheating; the time in which the CPU runs in turbo mode is known as the "tau," and is prolonged as long as possible. It's one of the reasons that PCWorld uses the prolonged transcoding application Handbrake as part of our laptop testing, to try and see how long the laptop can run in turbo mode. Dell is apparently saying that the T2 never has to drop out of turbo mode, due to the tower's cooling. (I've asked for confirmation of this, but haven't heard back from Dell.) The idea, of course, is to allow the T2 to serve as an AI workstation, running prolonged AI applications. It would certainly be nice to see this feature appear in a gaming PC from Alienware someday, however. Interestingly, "select" Dell Pro Max desktops will ship with a modular USB-C port, designed to enhance durability. It's not clear what the purpose is on a desktop, although the poor mechanical reliability of USB-C ports on smartphones can lead to them wearing out. A modular USB-C port would be a great addition to a future laptop. Dell's AI Max laptops, meanwhile, ship in three product categories: the Base, Plus, and Premium tiers. The base 14-inch and 16-inch tier receives a boost just from its upgraded components, between 33 and 36 percent more performance from the Precision 3-series workstations. The Plus tier offers both 16- and 18-inch models, which I'd like to see fit inside my backpack. (At 6.9 pounds minimum, to boot!) Here, we're seeing Thunderbolt 5's power delivery, delivered: These laptops will boast the latest 48-volt EPR USB-C power adapters, delivering up to 240W directly to the laptop. Naturally, these laptops are designed for running code, and will include RTX Pro 5000 GPUs with 24GB of GPU memory and up to 256GB of system memory, cooled with three fans. Inside is up to 16 terabytes -- yep, terabytes -- storage, with RAID options. The Premium tier, somewhat surprisingly, doesn't offer an 18-inch model. Instead, Dell's premium AI PCs boast tandem OLEDs, which debuted with the M4 version of the Apple iPad but have since started appearing in laptop PCs. Tandem OLEDs combine two OLED panels one on top of another, potentially increasing the HDR light output (to TrueBlack 1000, in this case) without robbing them of their key advantage: "perfect," deep blacks. The laptops also include a zero-lattice keyboard. Dell is including up to 64GB of LPDDR5X memory and up to 8TB of dual-storage RAID.

Nvidia's DGX Station brings 800Gbps LAN, the most powerful chip ever launched in a desktop workstation PC

That one didn't have a Nvidia Arm CPU and needed separate PCIe AI accelerators (A100); the 2025 iteration doesn't. It also carried a price of more than $100,000 at launch. The GB300 Grace Blackwell Ultra Desktop Superchip that powers it delivers up to "20 PFlops of AI performance" which is likely to be measured using FP4 with sparsity. That would also infer that it is half the performance of GB200 Grace Blackwell Superchip, so something's not clear here and I wonder whether there's more than one version of the GB300. Nvidia hasn't said how many or what type of (Arm) CPU cores the GB300 uses; ditto for the GPU subsystem. Its predecessor, GH200, had 72 Arm Neoverse 2 CPU cores clocked at 3.1GHz, up to 144GB HBMe memory and 480GB LPDDR5x memory. What we do know is that it has 784GB of unified system memory, which one can assume means HBM (288GB HBM3e) plus what Nvidia calls Fast Memory (496GB LPDDR5x most probably). Fastest NIC in a computer Nvidia also disclosed the DGX Station will use its proprietary ConnectX-8 SuperNIC, a network technology that can deliver up to a staggering data center-class 800Gb/s connectivity. A close-up of the opened chassis shows the workstation has three forward-facing 120mm fans, a motherboard with three PCIe slots, an Nvidia-branded soldered chip (perhaps the SuperNic), and two large uncovered dies. One of which is the Grace GPU and the other with eight distinct tiles, the Blackwell GPU. No details about expansion or storage capabilities, the PSU capacity, the cooling solution used or the price have been revealed. Additionally, we do not know whether you will be able to plug in accelerator cards like the H200 NVL (or a theoretical B300 NVL) to significantly improve the performance of the DGX Station. The DGX station is expected to compete with the likes of the Camino Grando, an EPYC-powered tower workstation that packs two AMD CPUs and up to eight GPUs. We'll strive to update this article when further details of this workstation PC (including pricing and availability) are published. Nvidia's GTC Keynote also saw the formal launch of DGX Spark, formerly known as Project Digits, 12 new professional GPUs and Blackwell Ultra (or GB300), Nvidia's most powerful GPU ever.

Dell unveils levelling 20-petaFLOPS desktop: NVIDIA GB300 Superchip, 784GB of unified memory

TL;DR: NVIDIA's GTC features Dell's new Pro Max with GB300, a powerful workstation with 20-petaFLOPs processing capability. It includes NVIDIA's GB300 Grace Blackwell Ultra Superchip, 784 GB unified memory, and 800Gbps networking. Designed for AI training and engineering workloads, pricing details are yet to be announced. NVIDIA's GPU Technology Conference (GTC) has kicked into high gear with a slew of various announcements, and one that stands out is Dell's new monstrous workstation powered by NVIDIA's latest Superchip. Dell has unveiled the Pro Max with GB300 at GTC, and the system is nothing short of shocking when it comes to processing power as it is capable of 20-petaFLOPs. What's inside? Almost every bell and whistle you can think of when it comes to a workstation desktop, with the system featuring NVIDIA's latest GB300 Grace Blackwell Ultra Desktop Superchip that features an NVIDIA Blackwell Ultra GPU with next-generation Tensor Cores and FP4 precision. Additionally, the system features 784 GB of unified system memory - 288GB of HBM3e within the GPU and 496GB of LPDDR5X for the CPU. As you can probably imagine, this system has been optimized for large-scale AI training workloads, hence the inclusion of the NVIDIA ConnectX-8 SuperNIC, which ushers in support for an insane networking speed of 800Gbps. Dell intends the system to be used by developers who are looking to test prototypes and scale their AI models into production environments. Moreover, the new system is also aimed at engineers looking to run large-scale training and inferencing workloads, which would typically be run on servers. At the moment pricing isn't available for the Dell Pro Max with GB300, but given the hardware that's inside we can expect it to cost a pretty penny.

Dell Expands AI Factory With Next-Gen AI PCs & Data Center Solutions

As the global leader in workstations that feature NVIDIA's most powerful professional graphics, Dell expands and innovates the Dell Pro Max high-performance AI PC portfolio to meet the needs of today's AI developers, power users and specialty users. The portfolio offers a versatile range of powerful AI PCs designed for demanding tasks - from light AI development, data analysis and design simulation to training, inferencing and fine-tuning the most complex LLMs, before deploying at scale. The new Dell Pro Max with GB10 packs exceptional performance in a compact and power-efficient form factor. This AI developer workstation features the NVIDIA GB10 Grace Blackwell Superchip, a system-on-a-chip based on the NVIDIA Blackwell architecture. Delivering up to one petaflop (1000 TFLOPs) of AI computing performance and 128GB of unified memory, AI developers can develop, train and test models before deploying on other Dell infrastructure offerings. The new Dell Pro Max with GB300, at the top end of the high-performance PC range, gives AI developers and data scientists a computing class of its own - bringing server-level compute to a desktop. With the new NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip, the system delivers up to 20 petaflops of AI computing performance, 784GB unified system memory (up to 288GB HBME3e GPU memory and 496GB of LPDDR5X CPU memory) and the fastest networking solution with NVIDIA ConnectX-8 SuperNIC to power the most intensive and largest AI workloads, training up to 460billion parameter models. New Dell Pro Max notebooks and desktops offer outstanding power, reliability and scalability. Equipped with NVIDIA RTX PRO Blackwell Generation GPUs and Intel® Core™ Ultra (Series 2), AMD Ryzen-powered Copilot+ PCs with AI experiences and AMD Thread ripper processor options, along with a new bold and elevated design, users can drive productivity across every intensive workload.