It's that time again! 3DExperience World 2025 kicks off on February 23 and runs through February 26 at the George R. Brown Convention Center in Houston, Texas. The show is hosted by Dassault Systèmes and highlights annual advances and improvements throughout its product ecosystem. It's a great opportunity to meet the engineers, students, and industry professionals who use SolidWorks and other Dassault Systèmes applications across browsers, local workstations, and the cloud.

One of the best parts of the event for me is showcasing how advances in silicon engineering can lead to transformational products - systems that offer performance, features, and efficiency that wasn't possible before. In 2024, the AMD Ryzen Threadripper PRO 7000 WX-Series processor stole the proverbial show with its excellent single-thread performance, support for multi-GPU configurations for AI training, and up to 96 cores and 2T B of memory for the largest and most demanding projects. This year, AMD has complemented these full-size tower systems with compact and mobile workstations based on the new AMD Ryzen AI Max PRO Series processors. Drop by booth #919 and see the array of systems and demos on exhibit.

Senao Networks Inc. (SNI), a global leader in AI computing and networking solutions, will be exhibiting at 2025 Mobile World Congress (MWC) in Barcelona. At the event, SNI will showcase its latest AI-driven innovations, including AI Servers, AI Cameras, AIPCs, Cloud Solutions, and Titanium Power Supply, reinforcing its vision of "AI Everywhere."

Senao Networks continues to advance AI computing with new products designed to enhance security, efficiency, and connectivity.

Advantech, a leader in edge computing solutions, is excited to announce the launch of the UBX-330M, a compact NUC-sized edge computer designed to power intelligent city services with advanced AI capabilities.

The UBX-330M is powered by Intel Core Ultra Processors (Meteor Lake H and U), which integrate a CPU, GPU, and NPU to deliver exceptional AI performance in a small form factor. Built as an AI-ready platform, the UBX-330M is perfect for applications requiring robust processing power across industries such as smart retail, hospitality, public spaces, education, entertainment, enterprise, and smart buildings.

ASUS Republic of Gamers (ROG) announced that the 2025 ROG Flow Z13 is now available for pre-order. This versatile gaming 2-in-1 can feature up to AMD's newest AMD Ryzen AI Max+ 395 Processor with Radeon 8060S Graphics and a unified memory structure, allowing for incredible performance and power efficiency. A new stainless steel and copper vapor chamber, larger intake vents, and 2nd Gen Arc Flow Fans offer 70% more airflow for quiet and efficient cooling.

This 13-inch tablet boasts a stunning ROG Nebula Display, a 2.5K resolution 180 Hz touchscreen with 500 nits of peak brightness, and Corning Gorilla Glass 5 protection. The Flow Z13 now also features a larger 70Wh battery, a larger touchpad and keycaps, and a convenient Command Center button for quick access to vital system functions. With dual USB-C ports, both of which support USB4 and power delivery, as well as a dedicated HDMI 2.1 port, the Flow Z13 lets gamers leave their dongles at home.

The global semiconductor manufacturing industry closed 2024 with strong fourth quarter results and solid year-on-year (YoY) growth across most of the key industry segments, SEMI announced today in its Q4 2024 publication of the Semiconductor Manufacturing Monitor (SMM) Report, prepared in partnership with TechInsights. The industry outlook is cautiously optimistic at the start of 2025 as seasonality and macroeconomic uncertainty may impede near-term growth despite momentum from strong investments related to AI applications.

After declining in the first half of 2024, electronics sales bounced back later in the year resulting in a 2% annual increase. Electronics sales grew 4% YoY in Q4 2024 and are expected to see a 1% YoY increase in Q1 2025 impacted by seasonality. Integrated circuit (IC) sales rose by 29% YoY in Q4 2024 and continued growth is expected in Q1 2025 with a 23% increase YoY as AI-fueled demand continues boosting shipments of high-performance computing (HPC) and datacenter memory chips.

Nexa AI, today, announced NexaQuants of two DeepSeek R1 Distills: The DeepSeek R1 Distill Qwen 1.5B and DeepSeek R1 Distill Llama 8B. Popular quantization methods like the llama.cpp based Q4 K M allow large language models to significantly reduce their memory footprint and typically offer low perplexity loss for dense models as a tradeoff. However, even low perplexity loss can result in a reasoning capability hit for (dense or MoE) models that use Chain of Thought traces. Nexa AI has stated that NexaQuants are able to recover this reasoning capability loss (compared to the full 16-bit precision) while keeping the 4-bit quantization and all the while retaining the performance advantage. Benchmarks provided by Nexa AI can be seen below.

We can see that the Q4 K M quantized DeepSeek R1 distills score slightly less (except for the AIME24 bench on Llama 3 8b distill, which scores significantly lower) in LLM benchmarks like GPQA and AIME24 compared to their full 16-bit counter parts. Moving to a Q6 or Q8 quantization would be one way to fix this problem - but would result in the model becoming slightly slower to run and requiring more memory. Nexa AI has stated that NexaQuants use a proprietary quantization method to recover the loss while keeping the quantization at 4-bits. This means users can theoretically get the best of both worlds: accuracy and speed.

Yesterday, the ASUS Chinese office announced a special event—on February 25—dedicated to launching a next-gen AMD APU-powered premium notebook model: "ROG Magic X (or Illusion X) is the first to be equipped with the Ryzen AI MAX+ three-in-one chip, which can efficiently coordinate multiple modes, provide combat power and computing power on demand, and can handle e-sports, creation and AI with one chip!" The manufacturer's Weibo post has generated plenty of buzz; industry insiders reckon that reviews could be published today (February 18)—HXL/9550pro informed VideoCardz with a not so cryptic message: "STX-Halo NDA: Feb 18th 2025."

Western press outlets point out that the ROG Magic X is a local variant of the familiar ROG Flow Z13 design; a 2025 refresh brings in AMD's much anticipated "Strix Halo" APU design. Team Red-authored marketing material and pre-release evaluation leaks have hinted about impressive integrated graphics solution performance; equalling or even exceeding that of previous-gen dGPUs. Well-known North American hardware review outlets have dropped hints (NDA permitting) about AMD's Ryzen Al Max+ 395 and Max 390 processors. Hardware Canucks could barely contain their excitement regarding the potent Zen 4 and RDNA 3.5 combo package; to the point of wish listing a potential direct successor: "Strix Halo is one of the most exciting things launched into the PC space in the last half decade. Full stop...AMD can't keep this as a one-off. If it's followed up with Zen 6 and RDNA 4 next year...watch out." Naturally, Team Red's cutting-edge mobile CPU technology is arriving in devices with high asking prices. The aforementioned ROG Flow Z13 2025 model—configured with top specs—is priced at $2699. Notebookcheck reckons that ASUS has tacked on an extra $500, since an announcement of initial pricing at CES 2025.

TrendForce's latest findings reveal that the NAND Flash market continues to be plagued by oversupply in the first quarter of 2025, leading to sustained price declines and financial strain for suppliers. However, TrendForce anticipates a significant improvement in the market's supply-demand balance in the second half of the year.

Key factors contributing to this shift include proactive production cuts by manufacturers, inventory reductions in the smartphone sector, and growing demand driven by AI and DeepSeek applications. These elements are expected to alleviate oversupply and support a price rebound for NAND Flash.

ASUS's Republic of Gamers China account on Weibo has teased the ROG Magic X mobile device that combines laptop/tablet form with a detachable keyboard. Inside, the device is powered by AMD Ryzen AI MAX+ SoC, which ASUS called a "three-in-one" chip, mainly due to its CPU, NPU, and iGPU combination. One possible SKU for ROG Magic X is AMD's top-end Ryzen AI MAX+ 395 processor. Carrying 16 "Zen 5" cores and 32 threads, the chip is designed for AI-enhanced laptops with 126 combined TOPS of AI processing power. The Ryzen AI MAX+ 395 features 64 MB of L3 cache and operates at a base clock of 3 GHz, with boost capabilities up to 5.1 GHz, depending on workload conditions. Manufactured on TSMC's 4 nm process node, the processor maintains a modest 55 W TDP, suitable for high-performance mobile systems. The chip includes support for ECC memory and PCIe Gen 5. It integrates the Radeon 8060S solution based on RDNA 3.5 architecture for graphics.

We are yet to see more details about the ROG Magic X, but with the arrival of AMD Ryzen AI MAX+, we assume this machine will result in a powerful gaming device for users on the go. More details are expected on February 25, when ASUS plans to showcase it. Pricing and availability are also expected to follow soon after.

According to a recent report, Apple has apparently finalized on a partner to bring Apple Intelligence features to its second-largest market, China. Of course, China has strict regulations in place, which have clearly turned out to be rather difficult to navigate for Apple. Previous leaks have revealed that a Deepseek partnership was on the cards for the Cupertino giant, although that did not lead to fruition, with Apple settling for Alibaba as its primary partner. As MacRumors notes, Alibaba will function as a censorship layer of sorts, thereby complying with government-directed filtering and information control. Basically, if the Chinese government deems a certain kind of information inappropriate in any way, it can simply direct Alibaba to have it removed and not served again.

Baidu, Safari's default search engine in China, will function as a secondary partner, aiding Apple Intelligence's 'Visual Intelligence' features. If things go to plan, the AI features should roll out in China sometime towards the middle of this year. Apple Intelligence has faced quite a contentious launch, having struggled with multiple delays and widespread criticism regarding not only the real-world efficacy and usefulness of its features, but also the multiple times that its AI-generated notifications have wildly misrepresented real-life events. The report further states that iPhones sold outside of China will not have access to the China-specific AI system, which is to be expected. It is yet to be seen how the Apple Intelligence launch plays out in China, although we should be able to find out soon enough.

The notion of designing tailor-made AI-crunching chips is nothing new; several major organizations—with access to big coffers—are engaged in the formulation of proprietary hardware. A new DigiTimes Asia report suggests that DeepSeek is the latest company to jump on the in-house design bandwagon. The publication's insider network believes that the open-source large language model development house has: "initiated a major recruitment drive for semiconductor design talent, signaling potential plans to develop its proprietary processors." The recent news cycle has highlighted DeepSeek's deep reliance on an NVIDIA ecosystem, despite alternative options emerging from local sources.

Industry watchdogs believe that DeepSeek has access to 10,000 of sanction-approved Team Green "Hopper" H800 AI chips, and (now banned) 10,000 H100 AI GPUs. Around late January, DeepSeek's Scale AI CEO—Alexandr Wang—claimed that the organization could utilize up to 50,000 H100 chips for model training purposes. This unsubstantiated declaration raised eyebrows; given current global political tensions. Press outlets have speculated that DeepSeek is in no rush to reveal its full deck of cards, but they appear to have a competitive volume of resources; when lined up against with Western competitors. The DigiTimes news article did not provide any detailed insight into the rumored in-house chip design. DeepSeek faces a major challenge; the Chinese semiconductor industry trails behind market leading regions. Will local foundries be able to provide an advanced enough node process for required purposes?

AMD and Nutanix have jointly developed virtualization/HCI solutions since 2019, working with major OEMs including Dell, HP and Lenovo, systems integrators and other resellers and partners. You can learn more about AMD-Nutanix solutions here.

AMD EPYC Processors
The EPYC 9004 family of high performance processors provide up to 128 cores per processor to help meet the demands of a wide range of workloads and use cases. High density core counts allow you to reduce the number of servers you need by as much as a five to one ratio when looking at retiring older, inefficient servers and replacing with a new one. Systems based on AMD processors can also be more energy efficient than many competitive processor based systems. For example, running 2000 VMs on 11 2P AMD EPYC 9654 processor-powered servers will use up to 29% less power annually than the 17 2P Intel Xeon Platinum 8490H processor-based servers required to deliver the same performance, while helping reduce CAPEX up to 46%.

Retailers are set to open up pre-order floodgates for upcoming GeForce RTX 50-series laptops on February 25, as we learned earlier this week. According to a new DigiTimes report, the launch of mobile devices—sporting Team Green "Blackwell" GPUs—is expected to be "significantly delayed." A loose March launch window was teased during Jensen Huang's keynote presentation at CES 2025, but supply chain insiders have claimed that high-end RTX 50 laptops were "originally planned to be launched in January 2025." Additionally, they surmise that mid-range and low-end offerings are postponed to April. DigiTimes believes that the rumored postponements have surprised supply chain moles; Team Green is not known to delay product launches. Extenuating circumstances are cited as the reason behind alleged deferred release windows, but insiders have not yet determined the extent of lengthened launch parameters.

An anonymous source stated: "NVIDIA, which has never been late in the past, also encountered this situation. It is probably related to NVIDIA's full sprint to AI servers. Even though there are differences in server and PC chip design and manufacturing processes, the company's resource allocation may still affect the debugging efficiency of new products." Other insiders have murmured about GeForce RTX 50-series mobile GPU performance not meeting expectations. Rumors have swirled about problems with early sample units; most notably the encountering of major screen issues when the "hardware is turned on." Laptop/notebook supply chain insiders reckon that manufacturers have anticipated a healthy level of growth in 2025—thanks to the emergence of new NVIDIA graphics cards—but targets have been reduced, due to anticipated delays. Optimistic industry chatter predicts higher education students and esport enthusiasts driving unit sales upward, following a stagnant 2024 market.

Thanks to a well-known industry leaker, Jaykihn, Intel's Panther Lake-H processor family, built on the 18A process node, features three distinct configurations, with several power profiles for each case. The flagship model combines four "Cougar Cove" P-cores, eight "Skymont" E-cores, and four LPE cores alongside a 12-core Xe3 "Celestial" GPU. This variant supports LPDDR5X memory exclusively and delivers 180 TOPS of computational power for local AI workloads, operating at 25 W PL1 (base) and 64 W PL2 (turbo) power levels. Secondary configurations include a 4P+8E+4LP+4Xe3 model and a 4P+0E+4LP+4Xe3 variant, both rated at 100 TOPS and supporting both LPDDR5X and DDR5 memory.

The entry-level model operates at 15 W PL1 with 44 W PL2 in baseline mode, scaling to 25 W PL1 and 55 W PL2 in performance mode. All variants feature Thunderbolt 4 connectivity, with high-end models incorporating expanded PCIe 5.0 lane configurations. The integrated Xe3 graphics architecture scales from four to 12 cores across the product stack, with the top SKU eliminating DDR5 compatibility in favor of LPDDR5X optimization. The Panther Lake processor family is slated for launch within the latter half of 2025, and it will be the company's leading 18A product. As a "Lunar Lake" successor, Panther Lake will deliver improved IPC and optimization for new cores in the same device form-factor like laptops, ultrabooks, and handhelds.

AMD's Radeon RX 9000 series GPUs are expected to come with up to 16 GB of GDDR6 memory. However, AMD is reportedly expanding its RX 9070 lineup with a new 32 GB variant, according to sources on Chiphell. The card, speculatively called the RX 9070 XT 32 GB, is slated for release at the end of Q2 2025. The current GDDR6 memory modules used in GPUs carry a capacity of 2 GB per module only, meaning that a design with 32 GB of VRAM would require as many as 16 memory modules on a single card. No 2 GB+ GDDR6 memory modules are available, meaning that the design would require memory module installation on both the front and back of the PCB. Consumers GPUs are not known for this, but it is a possibility with workstation/prosumer grade GPUs employing this engineering tactic to boost capacity,

While we don't have information on the GPU architecture, discussions point to potential modifications of the existing Navi 48 silicon. This release is positioned as a gaming card rather than a workstation-class Radeon PRO 9000 series product. AMD appears to be targeting gamers interested in running AI workloads, which typically require massive VRAM amounts to run locally. Additionally, investing in a GPU with a big VRAM capacity is essentially "future-proofing" for gamers who plan to keep their cards for longer, as recent games have been spiking VRAM usage by a large margin. The combination of gaming and AI workloads may have made AMD reconsider some of its product offerings, potentially giving us the Radeon RX 9070 XT 32 GB SKU. We have to wait for the Q2 to start, and we can expect more details by then.

Update 20:55 UTC: AMD's Frank Azor on X debunked rumors of the 32 GB SKU coming to gamers. So, this will not happen. Instead, we could be looking at prosumer oriented AMD Radeon Pro GPU with 32 GB of memory instead.

Hewlett Packard Enterprise announced today that it has shipped its first NVIDIA Blackwell family-based solution, the NVIDIA GB200 NVL72. This rack-scale system by HPE is designed to help service providers and large enterprises quickly deploy very large, complex AI clusters with advanced, direct liquid cooling solutions to optimize efficiency and performance. "AI service providers and large enterprise model builders are under tremendous pressure to offer scalability, extreme performance, and fast time-to-deployment," said Trish Damkroger, senior vice president and general manager of HPC & AI Infrastructure Solutions, HPE. "As builders of the world's top three fastest systems with direct liquid cooling, HPE offers customers lower cost per token training and best-in-class performance with industry-leading services expertise."

The NVIDIA GB200 NVL72 features shared-memory, low-latency architecture with the latest GPU technology designed for extremely large AI models of over a trillion parameters, in one memory space. GB200 NVL72 offers seamless integration of NVIDIA CPUs, GPUs, compute and switch trays, networking, and software, bringing together extreme performance to address heavily parallelizable workloads, like generative AI (GenAI) model training and inferencing, along with NVIDIA software applications. "Engineers, scientists and researchers need cutting-edge liquid cooling technology to keep up with increasing power and compute requirements," said Bob Pette, vice president of enterprise platforms at NVIDIA. "Building on continued collaboration between HPE and NVIDIA, HPE's first shipment of NVIDIA GB200 NVL72 will help service providers and large enterprises efficiently build, deploy and scale large AI clusters."

OnLogic, a leading provider of edge computing solutions, has launched the Axial AX300, a highly customizable and powerful edge server. The AX300 is engineered to help businesses of any size better leverage their on-site data and unlock the potential of AI by placing powerful computing capabilities on-site.

The Axial AX300 empowers organizations to seamlessly move computing resources closer to the data source, providing significant advantages in performance, latency, operational efficiency, and total cost of ownership over cloud-based data management. With its robust design, flexible configuration options, and advanced security features, the Axial AX300 is the ideal platform for a wide range of highly-impactful edge computing applications, including:

• AI/ML inference and training: Leveraging the power of AI/ML at the edge for real-time insights, predictive maintenance, and improved decision-making.
• Data analytics: Processing and analyzing data generated by IoT devices and sensors in real-time to improve operational efficiency.
• Virtualization: Consolidating multiple workloads onto a single server, optimizing resource utilization and simplifying deployment and management.

People are using AI more and more at home, at work, in school, and everywhere in between. According to the most recent Microsoft Global Online Safety Survey, there has been a global increase in active generative AI users. Our findings showed that in 2024, 51% of people are users or experimenters of generative AI compared to 38% in 2023. Generation Z continues to drive this adoption with 64% of young adults reporting ever using the technology. That means it's up to us - especially those of us who work in technology and gaming - to make sure that young people have the support they need to navigate the world of AI safely while also fostering their curiosity and creativity in exploring these new technologies.

That's why, for Safer Internet Day 2025, Minecraft Education is releasing a new installment in the CyberSafe series where players can explore the risks and opportunities of AI use through fun, game-based challenges. In each instance, players are tasked with articulating guidelines for how to use AI safely and responsibly. Welcome to CyberSafe AI: Dig Deeper, available free on the Minecraft Marketplace and in Minecraft Education!

AI PC capabilities have evolved rapidly in the two years since AMD introduced the first x86 AI PC CPUs at CES 2023. New neural processing units have debuted, pushing available performance from a peak of 10 AI TOPS at the launch of the AMD Ryzen 7 7840U processor to peak 50+ TOPS on the latest AMD Ryzen AI Max PRO 300 Series processors. A wide range of software and hardware companies have announced various AI development plans or brought AI-infused products to market, while major operating system vendors like Microsoft are actively working to integrate AI into the operating system via its Copilot+ PC capabilities. AMD is on the forefront of those efforts and is working closely with Microsoft to deliver Copilot+ for Ryzen AI and Ryzen AI PRO PCs.

In the report "The Year of the AI PC is 2025," Forrester lays out its argument for why this year is likely to bring significant changes for AI PCs. Forrester defines the term "AI PC" to mean any system "embedded with an AI chip and algorithms specifically designed to improve the experience of AI workloads across the computer processing unit (CPU), graphics processing unit (GPU), and neural processing unit (NPU)." This includes AMD products, as well as competing products made by both x86 and non-x86 CPU manufacturers. 2025 represents a turning point for these efforts, both in terms of hardware and software, and this Forrester report is an excellent deep dive into why AI PCs represent the future for enterprise computing.

Intel experienced another leadership setback on Monday when Justin Hotard, who led its Data Center and AI (DCAI) division said he was leaving to become Nokia's CEO. Hotard joined Intel in early 2024 and worked there for just over a year. He will take over from Pekka Lundmark at Nokia on April 1. In his short time at Intel, Hotard oversaw the release of Intel's Sierra Forest E-core and Granite Rapids P-core Xeon 6 platforms. These helped Intel catch up to AMD in core count for the first time since 2017. Intel has temporarily appointed Karin Eibschitz Segal, an 18-year company veteran and co-CEO at Intel Israel, as the interim chief of DCAI.

However, Justin Hotard's exit comes as the DCAI division faces several problems. Not long ago, Intel said it would push back the launch of its next-generation Clearwater Forest Xeons to the first half of 2026 blaming low demand. The company also scrapped its Falcon Shores accelerators to focus on a future rack-scale platform called Jaguar Shores. These setbacks came after Intel fellow Sailesh Kottapalli left for Qualcomm last month. Kottapalli had worked at Intel for 28 years and played a key role in developing many Xeon server processors.

The future of AI at the edge has arrived, and ASRock Industrial is leading the way. With Super Mode support for the iEP-6010E Series and iEP-6010E Series Dev Kit, powered by NIVIDIA Jetson Orin NX and Jetson Orin Nano modules and optimized with NVIDIA JetPack 6.2 SDK, this upgrade supercharges AI acceleration, unlocking up to 2X faster generative AI performance. Delivering unprecedented inference speeds, real-time decision-making, and superior power efficiency, Super Mode is redefining edge AI computing - Powering everything from predictive maintenance and anomaly detection to AI-driven medical imaging, intelligent surveillance, and autonomous mobility - pushing performance beyond limits, faster, smarter, and more efficient than ever before. Whether driving Edge AI computing, real-time vision processing, intelligent autonomous vehicles, advanced robotics, industrial automation, or smart city infrastructure, ASRock Industrial's latest innovation empowers industries with cutting-edge AI capabilities that drive real-world transformation.

A Quantum Leap in Edge AI Performance
The iEP-6010E Series and iEP-6010E Series Dev Kit, now optimized with Super Mode, provides:

• Up to 2X Faster Gen AI Processing - Up to 157 TOPS (Sparse) on Orin NX 16 GB (Super) and 67 TOPS (Sparse) on Orin Nano 8 GB (Super), delivering blazing-fast generative AI performance for automation, robotics, and real-time vision systems.
• Optimized Power Boost and Efficiency - Supports power modes up to 40 W / MAXN Super on Orin NX and 25 W / MAXN Super on Orin Nano, while maintaining high-performance AI workloads in power-sensitive environments, making it ideal for industrial automation and smart city infrastructure.
• NVIDIA JetPack 6.2 Support - Fully compatible with the latest NVIDIA JetPack 6.2 SDK, offering enhanced libraries, frameworks, and tools to accelerate advanced AI development.

ASUS today announced the ASUS NUC 15 Pro mini PC, a compact powerhouse that delivers versatile performance for a wide range of development needs, from AI model training to data visualization. Featuring the latest Intel Core Ultra (Series 2) processors, DDR5 6400 MHz memory, and the Intel Arc GPU, NUC 15 Pro delivers outstanding speed and efficiency. Additionally, advanced Intel Wi-Fi 7 offers ultrafast connectivity for seamless collaborations. NUC 15 Pro offers an unbeatable combination of performance and innovative design to elevate workflows.

Optimized for AI
ASUS NUC 15 Pro unleashes the potential of AI thanks to the latest Intel Core Ultra processors (Series 2). Featuring up to 99 platform TOPS, this next-generation CPU is purpose-built for diverse workloads and unrivaled AI performance. These advanced processors are built on Intel 18A process and deliver remarkable compute performance. This innovative CPU redefines productivity and creativity to set a new benchmark for AI-powered computing.

IBM and Lenovo today announced at LEAP 2025 a planned expansion of their strategic technology partnership designed to help scale the impact of generative AI for clients in the Kingdom of Saudi Arabia. IDC expects annual worldwide spending on AI-centric systems to surpass $300 billion by 2026, with many leading organizations in Saudi Arabia exploring and investing in generative AI use cases as they prepare for the emergence of an "AI everywhere" world.

Building upon their 20-year partnership, IBM and Lenovo will collaborate to deliver AI solutions comprised of technology from the IBM watsonx portfolio of AI products, including the Saudi Data and Artificial Intelligence Authority (SDAIA) open-source Arabic Large Language Model (ALLaM), and Lenovo infrastructure. These solutions are expected to help government and business clients in the Kingdom to accelerate their use of AI to improve public services and make data-driven decisions in areas such as fraud detection, public safety, customer service, code modernization, and IT operations.

AI reasoning models and agents are set to transform industries, but delivering their full potential at scale requires massive compute and optimized software. The "reasoning" process involves multiple models, generating many additional tokens, and demands infrastructure with a combination of high-speed communication, memory and compute to ensure real-time, high-quality results. To meet this demand, CoreWeave has launched NVIDIA GB200 NVL72-based instances, becoming the first cloud service provider to make the NVIDIA Blackwell platform generally available. With rack-scale NVIDIA NVLink across 72 NVIDIA Blackwell GPUs and 36 NVIDIA Grace CPUs, scaling to up to 110,000 GPUs with NVIDIA Quantum-2 InfiniBand networking, these instances provide the scale and performance needed to build and deploy the next generation of AI reasoning models and agents.

NVIDIA GB200 NVL72 on CoreWeave
NVIDIA GB200 NVL72 is a liquid-cooled, rack-scale solution with a 72-GPU NVLink domain, which enables the six dozen GPUs to act as a single massive GPU. NVIDIA Blackwell features many technological breakthroughs that accelerate inference token generation, boosting performance while reducing service costs. For example, fifth-generation NVLink enables 130 TB/s of GPU bandwidth in one 72-GPU NVLink domain, and the second-generation Transformer Engine enables FP4 for faster AI performance while maintaining high accuracy. CoreWeave's portfolio of managed cloud services is purpose-built for Blackwell. CoreWeave Kubernetes Service optimizes workload orchestration by exposing NVLink domain IDs, ensuring efficient scheduling within the same rack. Slurm on Kubernetes (SUNK) supports the topology block plug-in, enabling intelligent workload distribution across GB200 NVL72 racks. In addition, CoreWeave's Observability Platform provides real-time insights into NVLink performance, GPU utilization and temperatures.

Moore Threads, a Chinese manufacturer of proprietary GPU designs is (reportedly) the latest company to jump onto the DeepSeek-R1 bandwagon. Since late January, NVIDIA, Microsoft and AMD have swooped in with their own interpretations/deployments. By global standards, Moore Threads GPUs trail behind Western-developed offerings—early 2024 evaluations presented the firm's MTT S80 dedicated desktop graphics card struggling against an AMD integrated solution: Radeon 760M. The recent emergence of DeepSeek's open source models has signalled a shift away from reliance on extremely powerful and expensive AI-crunching hardware (often accessed via the cloud)—widespread excitement has been generated by DeepSeek solutions being relatively frugal, in terms of processing requirements. Tom's Hardware has observed cases of open source AI models running (locally) on: "inexpensive hardware, like the Raspberry Pi."

According to recent Chinese press coverage, Moore Threads has announced a successful deployment of DeepSeek's R1-Distill-Qwen-7B distilled model on the aforementioned MTT S80 GPU. The company also revealed that it had taken similar steps with its MTT S4000 datacenter-oriented graphics hardware. On the subject of adaptation, a Moore Threads spokesperson stated: "based on the Ollama open source framework, Moore Threads completed the deployment of the DeepSeek-R1-Distill-Qwen-7B distillation model and demonstrated excellent performance in a variety of Chinese tasks, verifying the versatility and CUDA compatibility of Moore Threads' self-developed full-featured GPU." Exact performance figures, benchmark results and technical details were not disclosed to the Chinese public, so Moore Threads appears to be teasing the prowess of its MTT GPU designs. ITHome reported that: "users can also perform inference deployment of the DeepSeek-R1 distillation model based on MTT S80 and MTT S4000. Some users have previously completed the practice manually on MTT S80." Moore Threads believes that its: "self-developed high-performance inference engine, combined with software and hardware co-optimization technology, significantly improves the model's computing efficiency and resource utilization through customized operator acceleration and memory management. This engine not only supports the efficient operation of the DeepSeek distillation model, but also provides technical support for the deployment of more large-scale models in the future."