Broadcom Inc. (NASDAQ:AVGO), a global leader in semiconductor and infrastructure software solutions, today announced the shipment of its breakthrough Ethernet switch—the Tomahawk Ultra. Engineered to transform the Ethernet switch for high-performance computing (HPC) and AI workloads, Tomahawk Ultra delivers industry-leading ultra-low latency, massive throughput, and lossless networking.

"Tomahawk Ultra is a testament to innovation, involving a multi-year effort by hundreds of engineers who reimagined every aspect of the Ethernet switch," said Ram Velaga, senior vice president and general manager of Broadcom's Core Switching Group. "This highlights Broadcom's commitment to invest in advancing Ethernet for high-performance networking and AI scale-up."

European Processor Initiative (EPI) is nearing completion of its first goal. SiPearl, the leading developer behind the Rhea1 processor, has finally reached the tapeout stage after a string of delays, but it will not be ready for delivery until 2026 at the earliest. When the project launched in 2020, SiPearl planned to begin production in 2023; however, the 61 billion-transistor chip only entered tapeout this summer. The design, built on TSMC's N6 process, features 80 Arm Neoverse V1 cores alongside 64 GB of HBM2E memory and a DDR5 interface. While these specifications once looked cutting‑edge, the industry has already moved on, and Rhea1's raw performance may seem dated by the time samples are available. SiPearl initially explored a RISC‑V architecture back in 2019 but abandoned it after early feedback and comments highlighted the instruction set's immaturity for exascale computing.

Development was further interrupted by shifting core‑count debates, with teams alternately considering 72 cores, then 64, before finally settling on 80 cores by 2022. Those back‑and‑forth decisions, combined with evolving performance expectations, helped push the timeline back by years. Despite missing its original schedule, Rhea1 remains vital to European ambitions for high‑performance computing sovereignty and serves as the intended CPU for the Jupiter supercomputer. Thanks to Jupiter's modular design, the system was not left idle; its GPU booster module, running NVIDIA Grace Hopper accelerators, is already operational and approximately 80 percent complete. With the CPU clusters slated for mid-2026 deployment, full system readiness is expected by the end of 2026. To support this effort, SiPearl has recently secured €130 million in new financing from the French government, industry partners, and Taiwan's Cathay Venture. As Rhea1 finishes its goal, work on Rhea2 is already underway, and we can expect more updates about Rhea2 in a year or two.

Canonical, the company behind Ubuntu, has announced that its next-generation release will require RISC‑V processors to meet the newly ratified RVA23 profile. This specification, approved back in April, includes full support for Vector Instructions 1.0 and a Hypervisor extension. As Laurine Kirk, security researcher at Google, notes, by setting this higher baseline, Ubuntu 26.04 will no longer run on roughly 90% of existing RISC-V single-board computers, including the popular Raspberry Pi-style boards, unless their hardware is upgraded. Canonical's move puts it in step with Google and Microsoft, both of which have already targeted RVA23 for their RISC‑V builds. This change will drive manufacturers to ship more secure, future-proof silicon, helping to guard against exploits like GhostWrite, a memory-access vulnerability discovered last year in T-Head's XuanTie C910 CPUs.

We discussed RISC-V ecosystem fragmentation with Andrea Gallo, then the CTO and now the CEO of the RISC-V Foundation, regarding the fragmentation within the RISC-V ecosystem. "If you want to claim that you are RISC-V compatible, then there's an architecture compatibility test suite that verifies that you are complying with the ISA. We run the same tests on a golden reference model and compare the signatures of the tests to ensure alignment with the specification." He added that "We just ratified the RVA23 Profile. The newly ratified RVA23 Profile is a major release for the RISC-V software ecosystem and will help accelerate widespread implementation among toolchains and operating systems." For anyone who wants to ship a working RISC-V processor, be it data center or mobile, the RVA23 profile is the one that guarantees no fragmentation and compatibility.

IBM and RIKEN, a national research laboratory in Japan, today unveiled the first IBM Quantum System Two ever to be deployed outside of the United States and beyond an IBM Quantum Data Center. The availability of this system also marks a milestone as the first quantum computer to be co-located with RIKEN's supercomputer Fugaku—one of the most powerful classical systems on Earth. This effort is supported by the New Energy and Industrial Technology Development Organization (NEDO), an organization under the jurisdiction of Japan's Ministry of Economy, Trade and Industry (METI)'s "Development of Integrated Utilization Technology for Quantum and Supercomputers" as part of the "Project for Research and Development of Enhanced Infrastructures for Post 5G Information and Communications Systems."

IBM Quantum System Two at RIKEN is powered by IBM's 156-qubit IBM Quantum Heron, the company's best performing quantum processor to-date. IBM Heron's quality as measured by the two-qubit error rate, across a 100-qubit layered circuit, is 3x10-3 (with the best two-qubit error being 1x10-3)—which is 10 times better than the previous generation 127-qubit IBM Quantum Eagle. IBM Heron's speed, as measured by the CLOPS (circuit layer operations per second) metric is 250,000, which reflects another 10x improvement in the past year, over IBM Eagle.

KIOXIA America, Inc. today announced the development and prototype demonstration of its new KIOXIA CD9P Series PCIe 5.0 NVMe SSDs. These next-generation drives are the latest SSDs built with KIOXIA's 8th generation BiCS FLASH TLC-based 3D flash memory. KIOXIA BiCS FLASH features CBA (CMOS directly Bonded to Array) technology, a breakthrough architecture that significantly boosts power efficiency, performance, and storage density, while doubling the capacity available per SSD compared with the previous generation model.

As GPU-accelerated AI servers drive up the demands on storage infrastructure, maintaining high throughput, low latency, and consistent performance is critical - including keeping valuable GPUs highly utilized. The KIOXIA CD9P Series is purpose-built for these next-generation environments, delivering the speed and responsiveness required by AI, machine learning, and high-performance computing (HPC) workloads to ensure GPUs stay fed with data and operating at maximum efficiency.

In a market reshaped by the compute demands of AI, Arteris, Inc. (Nasdaq: AIP), a leading provider of system IP for accelerating semiconductor creation, today announced an expansion of its multi-die solution, delivering a foundational technology for rapid chiplet-based innovation. "In the chiplet era, the need for computational power increasingly exceeds what is available by traditional monolithic die designs," said K. Charles Janac, president and CEO of Arteris. "Arteris is leading the transition into the chiplet era with standards-based, automated and silicon-proven solutions that enable seamless integration across IP cores, chiplets, and SoCs."

Moore's Law, predicting the doubling of transistor count on a chip every two years, is slowing down. As the semiconductor industry accelerates efforts to increase performance and efficiency, especially driven by AI workloads, architectural innovation through multi-die systems has become critical. Arteris' expanded multi-die solution addresses this shift with a suite of enhanced technologies that are purpose-built for scalable and faster time-to-silicon, high-performance computing, and automotive-grade mission-critical designs.

As AI computing accelerates toward higher density and greater energy efficiency, Compal Electronics (Compal; Stock Ticker: 2324.TW), a global leader in IT and computing solutions, unveiled its latest high-performance server platform: SG720-2A/ OG720-2A at both AMD Advancing AI 2025 in the U.S. and the International Supercomputing Conference (ISC) 2025 in Europe. It features the AMD Instinct MI355X GPU architecture and offers both single-phase and two-phase liquid cooling configurations, showcasing Compal's leadership in thermal innovation and system integration. Tailored for next-generation generative AI and large language model (LLM) training, the SG720-2A/OG720-2A delivers exceptional flexibility and scalability for modern data center operations, drawing significant attention across the industry.

With generative AI and LLMs driving increasingly intensive compute demands, enterprises are placing greater emphasis on infrastructure that offers both performance and adaptability. The SG720-2A/OG720-2A emerges as a robust solution, combining high-density GPU integration and flexible liquid cooling options, positioning itself as an ideal platform for next-generation AI training and inference workloads.

Supermicro, Inc., a Total IT Solution Provider for AI, Cloud, Storage, and 5G/Edge, is announcing that both liquid-cooled and air-cooled GPU solutions will be available with the new AMD Instinct MI350 series GPUs, optimized for unparalleled performance, maximum scalability, and efficiency. The Supermicro H14 generation of GPU optimized solutions featuring dual AMD EPYC 9005 CPUs along with the AMD Instinct MI350 series GPUs, are designed for organizations seeking maximum performance at scale, while reducing the total cost of ownership for their AI-driven data centers.

"Supermicro continues to lead the industry with the most experience in delivering high-performance systems designed for AI and HPC applications," said Charles Liang, president and CEO of Supermicro. "Our Data Center Building Block Solutions enable us to quickly deploy end-to-end data center solutions to market, bringing the latest technologies for the most demanding applications. The addition of the new AMD Instinct MI350 series GPUs to our GPU server lineup strengthens and expands our industry-leading AI solutions and gives customers greater choice and better performance as they design and build the next generation of data centers."

PCI-SIG today announced the official release of the PCI Express (PCIe) 7.0 specification, reaching 128.0 GT/s, to members. PCIe 7.0 specification targets data-driven applications like AI/ML, 800G Ethernet, cloud, and Quantum computing. Pathfinding for the PCIe 8.0 specification is already in progress to continue supporting the industry's investments and product roadmaps in the PCIe technology ecosystem.

PCIe 7.0 Specification Features

• Delivers 128.0 GT/s raw bit rate and up to 512 GB/s bi-directionally via x16 configuration
• Utilizes PAM4 (Pulse Amplitude Modulation with 4 levels) signaling and Flit-based encoding
• Provides improved power efficiency
• Maintains backwards compatibility with previous generations of PCIe technology

The Ultra Ethernet Consortium (UEC) today announces the release of UEC Specification 1.0, a comprehensive, Ethernet-based communication stack engineered to meet the demanding needs of modern Artificial Intelligence (AI) and High-Performance Computing (HPC) workloads. This release marks a critical step forward in redefining Ethernet for next-generation, data-intensive infrastructure.

UEC Specification 1.0 delivers a high-performance, scalable, and interoperable solution across all layers of the networking stack—including NICs, switches, optics, and cables—enabling seamless multivendor integration and accelerating innovation throughout the ecosystem.

Synopsys, Inc. today announced that its collaboration with Broadcom has achieved interoperability between Synopsys' PCIe 6.x IP solution and Broadcom's PEX90000 series switch. As a cornerstone of next-generation AI infrastructures, PCIe switches play a critical role in enabling the scalability required to meet the demands of modern AI workloads. This milestone demonstrates that future products integrating PCIe 6.x solutions from Synopsys and Broadcom will operate seamlessly within the ecosystem, reducing design risk and accelerating time-to-market for high-performance computing and AI data center systems.

The interoperability demonstration with Broadcom features a Synopsys PCIe 6.x IP solution, including PHY and controller, operating as a root complex and an endpoint running at 64 GT/s with Broadcom's PEX90000 switch. Synopsys will showcase this interoperability demonstration at PCI-SIG DevCon 2025 at booth #13, taking place June 11 and 12, where attendees can see a variety of successful Synopsys PCIe 7.0 and PCIe 6.x IP interoperability demonstrations in both the Synopsys booth and partners' booths.

ASUS today joined GTC Paris at VivaTech 2025 as a Gold Sponsor, highlighting its latest portfolio of AI infrastructure solutions and reinforcing its commitment to advancing the AI Factory vision with a full range of NVIDIA Blackwell Ultra solutions, delivering breakthrough performance from large-scale datacenter to personal desktop.

ASUS is also excited to announce a transformative partnership milestone in its partnership with Nebius. Together, the two companies are enabling a new era of AI innovation built on NVIDIA's advanced platforms. Building on the success of the NVIDIA GB200 NVL72 platform deployment, ASUS and Nebius are now moving forward with strategic collaborations featuring the next-generation NVIDIA GB300 NVL72 platform. This ongoing initiative underscores ASUS's role as a key enabler in AI infrastructure, committed to delivering scalable, high-performance solutions that help enterprises accelerate AI adoption and innovation.

The 65th edition of the TOP500 showed that the El Capitan system retains the No. 1 position. With El Capitan, Frontier, and Aurora, there are now 3 Exascale systems leading the TOP500. All three are installed at Department of Energy (DOE) laboratories in the United States.

The El Capitan system at the Lawrence Livermore National Laboratory, California, remains the No. 1 system on the TOP500. The HPE Cray EX255a system was measured with 1.742 EFlop/s on the HPL benchmark. LLNL now also submitted a measurement for the HPCG benchmark, achieving 17.41 Petaflop/s, which makes the system the new No. 1 on this ranking as well.

IQM Quantum Computers, a global leader in superconducting quantum computers, announced today the signing of an agreement with VTT Technical Research Centre in Finland to deliver a 150-qubit and a 300-qubit quantum computer. The systems will be delivered in 2026 and 2027 and integrated with the Finnish HPC infrastructure. The 300-qubit quantum computer is purpose-built and designed to support quantum error correction experiments an essential step toward fault-tolerant quantum computing. The system is expected to enable algorithm research for techniques such as circuit knitting.

IQM has previously delivered 5-qubit, 20-qubit, and 50-qubit quantum computers to VTT, marking key milestones in the development of Finland's quantum ecosystem. Each system has played a crucial role in advancing research capabilities and supporting the growing needs of the quantum community in Finland.

MiTAC Computing Technology Corporation, a leading server platform designer, manufacturer, and a subsidiary of MiTAC Holdings Corporation, has launched its full suite of next-generation servers for data centers and enterprises at COMPUTEX 2025 (Booth M1110). Powered by Intel Xeon 6 processors, including those with Performance-cores (P-cores), MiTAC's new platforms are purpose-built for AI, HPC, cloud, and enterprise applications.

"For over five decades, MiTAC and Intel have built a close, collaborative relationship that continues to push innovation forward. Our latest server lineup reflects this legacy—combining Intel's cutting-edge processing power with MiTAC Computing's deep expertise in system design to deliver scalable, high-efficiency solutions for modern data centers." - Rick Hwang, President of MiTAC Computing.

The University of Tokyo (UTokyo) and IBM have announced plans to deploy the latest 156-qubit IBM Heron quantum processing unit (QPU), which will be operational in the IBM Quantum System One administered by UTokyo for the members of the Quantum Innovation Initiative (QII) Consortium. The IBM Heron QPU, which features a tunable-coupler architecture, delivers a significantly higher performance than the processor previously installed in 2023.

This is the second update of the IBM Quantum System One as part of the collaboration between UTokyo and IBM. It was first deployed with a 27-qubit IBM Falcon QPU, before being updated to a 127-qubit IBM Eagle QPU in 2023. It is now transitioning to the latest generation highly performant IBM Heron later this year. IBM has deployed four Heron-based systems worldwide and their performance shows significant improvement over the previous Eagle QPU, with a 3-4x improvement in two-qubit error rates; an order of magnitude improvement in device-wide performance benchmarked by errors across 100-qubit long layers; continued improvement in speed, with a 60 percent increase in CLOPS expected; and a system uptime of more than 95%. The latest IBM Heron processor has continued to demonstrate immense value in orchestrating utility-level workloads, to date, with multiple published studies leveraging these systems' capability of achieving more than 5,000 gate operations.

ASUS today announced that its flagship high-density AI server, ESC A8A-E12U, now supports the latest AMD Instinct MI350 series GPUs. This enhancement empowers enterprises, research institutions, and cloud providers to accelerate their AI and HPC workloads with next-generation performance and efficiency—while preserving compatibility with existing infrastructure.

Built on the 4th Gen AMD CDNA architecture, AMD Instinct MI350 series GPUs deliver powerful new capabilities, including 288 GB of HBM3E memory and up to 8 TB/s of bandwidth—enabling faster, more energy-efficient execution of large AI models and complex simulations. With expanded support for low-precision compute formats such as FP4 and FP6, the Instinct MI350 series significantly accelerates generative AI, inference, and machine-learning workloads. Importantly, Instinct MI350 series GPUs maintain drop-in compatibility with existing AMD Instinct MI300 series-based systems, such as those running Instinct MI325X—offering customers a cost-effective and seamless upgrade path. These innovations reduce server resource requirements and simplify scaling and workload management, making Instinct MI350 series GPUs an ideal choice for efficient, large-scale AI deployments.

Super Micro Computer, Inc. (SMCI), a Total IT Solution Provider for AI/ML, HPC, Cloud, Storage, and 5G/Edge, is announcing several improvements to its Direct Liquid Cooling (DLC) solution that incorporate new technologies for cooling various server components, accommodate warmer liquid inflow temperatures, and introduce innovative mechanical designs that enhance AI per watt. The Supermicro DLC-2 solution reduces data center power consumption by up to 40% compared to air-cooled installations. These advanced technologies enable faster deployment and reduced time-to-online for cutting-edge liquid-cooled AI infrastructure. Additionally, the total cost of ownership decreases by up to 20%. The comprehensive cold plate coverage of components allows for lower fan speeds and fewer required fans, significantly reducing data center noise levels to approximately 50dB.

"With the expected demand for liquid-cooled data centers rising to 30% of all installations, we realized that current technologies were insufficient to cool these new AI-optimized systems," said Charles Liang, president and CEO of Supermicro. "Supermicro continues to remain committed to innovation, green computing, and improving the future of AI, by significantly reducing data center power and water consumption, noise, and space. Our latest liquid-cooling innovation, DLC-2, saves data center electricity costs by up to 40%."

AMD is gearing up to expand its Instinct MI family in the latter half of 2026 with two purpose‑built UDNA‑based accelerators. These new models, MI430X UL4 and MI450X, will cater respectively to high‑precision HPC tasks and large‑scale AI workloads. The MI430X UL4 is designed for applications that rely heavily on double‑precision FP64 floating‑point performance, such as scientific simulations, climate modeling, and others. It features a large array of FP64 tensor cores, ensuring consistent throughput for tightly coupled compute jobs. Because a dedicated UALink switch from partners like Astera Labs or Broadcom won't be available at launch, AMD has chosen a four‑GPU point‑to‑point mesh configuration for the MI430X UL4. This limited scale‑up approach keeps latency low and synchronization tight, making it ideal for small cluster deployments. On the AI side, the MI450X embraces Ethernet‑based Ultra Ethernet for scale‑out.

With UEC‑ready switches already on the market, this open‑standard networking solution lets users build expansive AI farms across dozens or even hundreds of nodes. By relying on established Ethernet technology instead of the new UALink ecosystem, AMD provides customers with immediate, hardware‑accelerated networking for both inference and training. While UALink promises an open, vendor‑neutral accelerator-to-accelerator fabric, its progress has been slow. Committee reviews and limited investment in switch silicon have delayed switch availability. Broadcom, viewed as the primary UALink switch supplier, has been cautious about its commitment, given expected lower volumes compared to Ethernet alternatives. AMD's acceleration segmentation addresses these realities. The MI430X UL4 delivers high‑precision compute in compact GPU meshes. The MI450X offers large‑scale AI connectivity via Ethernet. Should UALink switch development catch up in the future, AMD could revisit native GPU‑to‑GPU fabrics.

Vultr, the world's largest privately-held cloud infrastructure company, today announced that it is one of the first cloud providers to offer the new AMD EPYC 4005 Series processors. The AMD EPYC 4005 Series processors will be available on the Vultr platform, enabling enterprise-class features and leading performance for businesses and hosted IT service providers. The AMD EPYC 4005 Series processors extend the broad AMD EPYC processor family, powering a new line of cost-effective systems designed for growing businesses and hosted IT services providers that demand performance, advanced technologies, energy efficiency, and affordability. Servers featuring the high-performance AMD EPYC 4005 Series CPUs with streamlined memory and I/O feature sets are designed to deliver compelling system price-to-performance metrics on key customer workloads. Meanwhile, the combination of up to 16 SMT-capable cores and DDR5 memory in the AMD EPYC 4005 Series processors enables smooth execution of business-critical workloads, while maintaining the thermal and power efficiency characteristics crucial for affordable compute environments.

"Vultr is committed to delivering the most advanced cloud infrastructure with unrivaled price-to-performance," said J.J. Kardwell, CEO of Vultr. "The AMD EPYC 4005 Series provides straightforward deployment, scalability, high clock speed, energy efficiency, and best-in-class performance. Whether you are a business striving to scale reliably or a developer crafting the next groundbreaking innovation, these solutions are designed to deliver exceptional value and meet demanding requirements now and in the future." Vultr's launch of systems featuring the AMD EPYC 4245P and AMD EPYC 4345P processors will expand the company's robust line of Bare Metal solutions. Vultr will also feature the AMD EPYC 4345P as part of its High Frequency Compute (HFC) offerings for organizations requiring the highest clock speeds and access to locally-attached NVMe storage.

At today's Intel Foundry Direct Connect 2025 event, Synopsys, Inc. announced broad EDA and IP collaborations with Intel Foundry, including availability of its certified AI-driven digital and analog design flows for the Intel 18A process node and production-ready EDA flows for the Intel 18A-P process node with RibbonFET Gate-all-around transistor architecture and the industry's first commercial foundry implementation of PowerVia backside power delivery. To drive multi-die design innovation forward, Synopsys and Intel Foundry are collaborating to enable Intel's new Embedded Multi-die Interconnect Bridge-T (EMIB-T) advanced packaging technology with an EDA reference flow powered by Synopsys 3DIC Compiler. With its EDA flows, multi-die solution, and broad portfolio of Synopsys' foundation and interface IP on Intel 18A and Intel 18A-P, Synopsys is helping designers accelerate the development of highly optimized AI and HPC chip designs from silicon to systems.

In a keynote presentation at today's event, John Koeter, Senior Vice President, for the Synopsys IP Group, emphasized: "The successful collaboration between Synopsys and Intel Foundry is advancing the semiconductor industry with silicon to system design solutions to meet the evolving needs for AI and high-performance computing applications. Our production-ready EDA flows, IP, and multi-die solution, provides our mutual customers with comprehensive technologies to accelerate the development of chip designs that meet or exceed their requirements."

MiTAC Computing Technology, a global leader in high-performance and energy-efficient server solutions, is proud to announce its participation at the OCP EMEA Summit 2025, taking place April 29-30 at the Convention Centre Dublin. At Booth No. B13, MiTAC will showcase its latest innovations in server design, sustainable cooling, and open-source firmware development - empowering the future of data center infrastructure.

C2810Z5 & C2820Z5: Advancing Sustainable Thermal Design
MiTAC will debut two new OCP server platforms - C2810Z5 (air-cooled) and C2820Z5 (liquid-cooled), built to meet the demands of high-performance computing (HPC) and AI workloads. Designed around the latest AMD EPYC 9005 series processors, these multi-node servers support the latest EPYC 9005 processors and are engineered to deliver optimal compute density and power efficiency.

At this year's Technology Symposium, TSMC unveiled an engaging multi-year roadmap for its packaging technologies. TSMC's strategy splits into two main categories: Advanced Packaging and System-on-Wafer. Back in 2016, CoWoS-S debuted with four HBM stacks paired to N16 compute dies on a 1.5× reticle-limited interposer, which was an impressive feat at the time. Fast forward to 2025, and CoWoS-S now routinely supports eight HBM chips alongside N5 and N4 compute tiles within a 3.3× reticle budget. Its successor, CoWoS-R, increases interconnect bandwidth and brings N3-node compatibility without changing that reticle constraint. Looking toward 2027, TSMC will launch CoWoS-L. First up are large N3-node chiplets, followed by N2-node tiles, multiple I/O dies, and up to a dozen HBM3E or HBM4 stacks—all housed within a 5.5× reticle ceiling. It's hard to believe that eight HBM stacks once sounded ambitious—now they're just the starting point for next-gen AI accelerators inspired by AMD's Instinct MI450X and NVIDIA's Vera Rubin.

Integrated Fan-Out, or InFO, adds another dimension with flexible 3D assemblies. The original InFO bridge is already powering AMD's Instinct cards. Later this year, InFO-POP (package-on-package) and InFO-2.5D arrive, promising even denser chip stacking and unlocking new scaling potential on a single package, away from the 2D and 2.5D packaging we were used to, going into the third dimension. On the wafer scale, TSMC's System-on-Wafer lineup—SoW-P and SoW-X—has grown from specialized AI engines into a comprehensive roadmap mirroring logic-node progress. This year marks the first SoIC stacks from N3 to N4, with each tile up to 830 mm² and no hard limit on top-die size. That trajectory points to massive, ultra-dense packages, which is exactly what HPC and AI data centers will demand in the coming years.

MSI, a leading global provider of high-performance server solutions, is bringing AI-driven innovation to Japan IT Week Spring 2025 at Booth #21-2 with high-performance server platforms built for next-generation AI and cloud computing workloads. MSI's NVIDIA MGX AI Servers deliver modular GPU-accelerated computing to optimize AI training and inference, while the Core Compute line of Multi-Node Servers maximize compute density and efficiency for AI inference and cloud service provider workloads. MSI's Open Compute line of ORv3 Servers enhance scalability and thermal efficiency in hyperscale AI deployments. MSI's Enterprise Servers provide balanced compute, storage, and networking for seamless AI workloads across cloud and edge. With deep expertise in system integration and AI-driven infrastructure, MSI is advancing the next generation of intelligent computing solutions to power AI's next leap.

"AI's advancement hinges on performance efficiency, compute density, and workload scalability. MSI's server platforms are engineered to accelerate model training, optimize inference, and maximize resource utilization—ensuring enterprises have the processing power to turn AI potential into real-world impact," said Danny Hsu, General Manager of MSI Enterprise Platform Solutions.

AMD officially released ROCm 6.4, its latest open‑source GPU compute stack, bringing several under‑the‑hood improvements while still lacking official RDNA 4 support. The update improves compatibility between ROCm's user‑space libraries and the AMDKFD kernel driver, making it easier to run across a wider range of Linux kernels. AMD has also expanded its internal testing to cover more combinations of user and kernel versions, which should reduce integration headaches for HPC and AI workloads. On the framework side, ROCm 6.4 now supports PyTorch 2.5 and 2.6 out of the box, so developers can use the latest deep‑learning features without building from source. The Megatron‑LM integration adds three new fused kernels, Attention (QKV), Layer Norm, and ROPE, to speed up transformer model training by combining multiple operations into single GPU passes. Video decoding gets a boost, too, with VP9 support in both rocDecode and rocPyDecode, plus a new bitstream reader module to streamline media pipelines.

Oracle Linux 9 is now officially supported, and the Radeon PRO W7800 48 GB workstation card has been validated under ROCm. AMD also enabled CPX mode with NPS4 memory configurations, catering to advanced memory bandwidth scenarios on MI Instinct accelerators. Despite these updates, ROCm 6.4 still does not officially support RDNA 4 GPUs, such as the RX 9070 series. While community members report that the new release can run on those cards unofficially, the lack of formal enablement means RDNA 4's doubled FP16 throughput, eight times INT4 sparsity acceleration, and FP8 capabilities remain largely untapped in ROCm workflows. On Linux, consumer Radeon support is limited to just a few models, even though Windows coverage for RDNA 2 and 3 families has expanded since 2022. With AMD's "Advancing AI" event coming in June, many developers are hoping for an announcement about RDNA 4 integration. Until then, those who need guaranteed, day‑one GPU support may continue to look at alternative ecosystems.