Today, at Linley Fall Processor Conference 2022, Andes Technology, a leading provider of high efficiency, low power 32/64-bit RISC-V processor cores and founding premier member of RISC-V International, reveals its top-of-the-line AndesCore AX60 series of power and area efficient out-of-order 64-bit processors. The family of processors are intended to run heavy-duty OS and applications with compute intensive requirements such as advanced driver-assistance systems (ADAS), artificial intelligence (AI), augmented/virtual reality (AR/VR), datacenter accelerators, 5G infrastructure, high-speed networking, and enterprise storage.

The first member of the AX60 series, the AX65, supports the latest RISC-V architecture extensions such as the scalar cryptography extension and bit manipulation extension. It is a 4-way superscalar with Out-of-Order (OoO) execution in a 13-stage pipeline. It fetches 4 to 8 instructions per cycle guided by highly accurate TAGE branch predictor with loop prediction to ensure fetch efficiency. It then decodes, renames and dispatches up to 4 instructions into 8 execution units, including 4 integer units, 2 full load/store units, and 2 floating-point units. Besides the load/store units, the AX65's aggressive memory subsystem also includes split 2-level TLBs with multiple concurrent table walkers and up to 64 outstanding load/store instructions.

Tachyum today announced that it has responded to a U.S. Department of Energy Request for Information soliciting Advanced Computing Ecosystems for DOE national laboratories engaged in scientific and national security research. Tachyum has submitted a proposal to create a 20-exaflop supercomputer based on Tachyum's Prodigy, the world's first universal processor.

The DOE's request calls for computing systems that are five to 10 times faster than those currently available and/or that can perform more complex applications in "data science, artificial intelligence, edge deployments at facilities, and science ecosystem problems, in addition to the traditional modeling and simulation applications."

When designing integrated circuits, engineers aim to produce an efficient design that is easier to manufacture. If they manage to keep the circuit size down, the economics of manufacturing that circuit is also going down. NVIDIA has posted on its technical blog a technique where the company uses an artificial intelligence model called PrefixRL. Using deep reinforcement learning, NVIDIA uses the PrefixRL model to outperform traditional EDA (Electronics Design Automation) tools from major vendors such as Cadence, Synopsys, or Siemens/Mentor. EDA vendors usually implement their in-house AI solution to silicon placement and routing (PnR); however, NVIDIA's PrefixRL solution seems to be doing wonders in the company's workflow.

Creating a deep reinforcement learning model that aims to keep the latency the same as the EDA PnR attempt while achieving a smaller die area is the goal of PrefixRL. According to the technical blog, the latest Hopper H100 GPU architecture uses 13,000 instances of arithmetic circuits that the PrefixRL AI model designed. NVIDIA produced a model that outputs a 25% smaller circuit than comparable EDA output. This is all while achieving similar or better latency. Below, you can compare a 64-bit adder design made by PrefixRL and the same design made by an industry-leading EDA tool.

Researchers from Università di Bologna and CINECA, the largest supercomputing center in Italy, have been playing with the concept of developing a RISC-V supercomputer. The team has laid the grounds for the first-ever implementation that demonstrates the capability of the relatively novel ISA to run high-performance computing. To create a supercomputer, you need pieces of hardware that seem like Lego building blocks. Those are called clusters, made from a motherboard, processor, memory, and storage. Italian researchers decided to try and use something different than Intel/AMD solution to the problem and use a processor based on RISC-V ISA. Using SiFive's Freedom U740 SoC as the base, researchers named their RISC-V cluster "Monte Cimone."

Monte Cimone features four dual-board servers, each in a 1U form factor. Each board has a SiFive's Freedom U740 SoC with four U74 cores running up to 1.4 GHz and one S7 management core. In total, eight nodes combine for a total of 32 RISC-V cores. Paired with 16 GB of 64-bit DDR4 memory operating at 1866s MT/s, PCIe Gen 3 x8 bus running at 7.8 GB/s, one gigabit Ethernet port, USB 3.2 Gen 1 interfaces, the system is powered by two 250 Watt PSUs to support future expansion and addition of accelerator cards.

A few days ago, Intel announced its first official discrete graphics card efforts, designed for laptops. Called the Arc Alchemist lineup, Intel has designed these SKUs to provide entry-level to high-end options covering a wide range of use cases. Today, AMD has responded with a rather exciting Tweet made by the company's @Radeon Twitter account. The company compared Intel's Arc Alchemist A370M GPU with AMD's Radeon RX 6500M mobile SKUs in the post. These GPUs are made on TSMC's N6 node, feature 4 GB GDDR6 64-bit memory, 1024 FP32 cores, and have the same configurable TDP range of 35-50 Watts.

Below, you can see AMD's benchmarks of the following select games: Hitman 3, Total War Saga: Troy, F1 2021, Strange Brigade (High), and Final Fantasy XIV. The Radeon RX 6500M GPU manages to win in all of these games, thus explaining AMD's "FTW" hashtag on Twitter. Remember that these are vendor-supplied benchmarks runs, so we have to wait for some media results to surface.

Memtest86+, the spiritual successor of MemTest86, has been somewhat stuck in development hell, but now the developer behind the memory testing software has promised to deliver version 6.00 sometime this summer. Version 5.31b was released in April 2020, some six years after the previous release and still not released as a final build, largely due to the pandemic. Version 6.00 promises a host of new features and the developers are already now asking for people that want to pitch in and help out with the project, since it's an open-source project.

The goal of version 6.00 is to deliver 64-bit support, UEFI and DDR5 with XMP support, as well as support for up to 256 CPU cores. Other features that are said to come include PXE and native USB boot. The goal is to provide a beta build sometime in April, but for those that don't mind compiling their own version of the code, can give it a try now, although the developers are warning that the code is going to be buggy at this point. The base code is said to have been re-written from scratch compared to prior versions and a lot of features are still missing and some features from the older versions are said to have been dropped, at least for the time being.

Intel's upcoming Arc Alchemist lineup of discrete graphics cards generates a lot of attention from consumers. Leaks of these cards' performance and detailed specifications appear more and more as we enter the countdown to the launch day, which is sometime in Q1 of this year. Today, we managed to see a slide from @9950pro on Twitter that shows the laptop memory configuration of Intel's DG2 GPU. As the picture suggests, we can see that the top-end SKU1 with 512 EUs supports a 16 GB capacity of GDDR6 memory that runs at 16 Gbps speeds. The memory runs on a 256-bit bus and generates 512 GB/s bandwidth while having eight VRAM modules present.

When it comes to SKU2, which is a variant with 384 EUs, this configuration supports six VRAM modules on a 192-bit bus, running at 16 Gbps speeds. They generate a total capacity of 12 GBs and a bandwidth of 384 GB/s. We have SKU3 DG2 GPU going down the stack, featuring 256 EUs, four VRAM modules on a 128-bit bus, 8 GB capacity, and a 256 GB/s bandwidth. And last but not least, the smallest DG2 variants come in the form of SKU4 and SKU5, feating 128 EUs and 96 EUs, respectively. Intel envisions these lower-end SKUs with two VRAM modules on a 64-bit bus, and this time slower GDDR6 memory running at 14 Gbps. They are paired with 4 GB of total capacity, and the total bandwidth comes down to 112 GB/s.

NVIDIA has just announced three new mobile GPUs, although the question is how new any of them really are, as the model names suggest they're anything but. First up is the GeForce RTX 2050, which should be based on the Turing architecture. The other two GPUs are the GeForce MX550 and MX570, both presumably based on the Ampere architecture, although NVIDIA hasn't confirmed the specifics.

The GeForce RTX 2050 features 2048 CUDA cores, which is more than the mobile RTX 2060, but it has lower clock speeds and a vastly lower power draw at 30-45 Watts depending on the notebook design choices and cooling. It's also limited to 4 GB of 64-bit GDDR6 memory, which puts this in GeForce MX territory when it comes to memory bandwidth, as NVIDIA quotes an up to memory bandwidth of a mere 112 GB/s.

Imagination Technologies announces Catapult, a RISC-V CPU product line designed from the ground-up for next-generation heterogeneous compute needs. Based on RISC-V, the open-source CPU architecture, which is transforming processor design, Imagination's Catapult CPUs can be configured for performance, efficiency, or balanced profiles, making them suitable for a wide range of markets.

Leveraging Imagination's 20 years of experience in delivering complex IP solutions, the new CPUs are supported by the rapidly expanding open-standard RISC-V ecosystem, which continues to shake up the embedded CPU industry by offering greater choice. Imagination's entry will enable the rapidly expanding RISC-V ecosystem to add a greater range of product offerings, especially for heterogeneous systems. Now customers have an even wider choice of solutions built on the open RISC-V ISA, avoiding lock-in with proprietary architectures.

NVIDIA today announced NVIDIA Quantum-2, the next generation of its InfiniBand networking platform, which offers the extreme performance, broad accessibility and strong security needed by cloud computing providers and supercomputing centers.

The most advanced end-to-end networking platform ever built, NVIDIA Quantum-2 is a 400 Gbps InfiniBand networking platform that consists of the NVIDIA Quantum-2 switch, the ConnectX-7 network adapter, the BlueField-3 data processing unit (DPU) and all the software that supports the new architecture.

It's been nearly six years since we unleashed the $5 Raspberry Pi Zero on an unsuspecting world. Of all the products we've launched, Zero is still the one I'm proudest of: it most perfectly embodies our mission to give people access to tools, and to eliminate cost as a barrier. We've sold nearly four million units of Zero, and its $10 wireless-enabled big brother Zero W, and they've made their way into everything from smart speakers to hospital ventilators. But where our larger products have grown steadily more powerful over the years, we've never found a way to pack more performance into the Zero form factor. Until today.

Priced at $15, Raspberry Pi Zero 2 W uses the same Broadcom BCM2710A1 SoC die as the launch version of Raspberry Pi 3, with Arm cores slightly down-clocked to 1GHz, bundled into a single space-saving package alongside 512 MB of LPDDR2 SDRAM. The exact performance uplift over Zero varies across workloads, but for multi-threaded sysbench it is almost exactly five times faster.

SiFive, Inc., the industry leader in RISC-V processors and silicon solutions, today announced launched the new SiFive Performance family of processors. The SiFive Performance family debuts with two new processor cores, the P270, SiFive's first Linux capable processor with full support for the RISC-V vector extension v1.0 rc, and the SiFive Performance P550 core, SiFive's highest performance processor to date. The new SiFive Performance P550 delivers a SPECInt 2006 score of 8.65/GHz, making it the highest performance RISC-V processor available today, and comparable to existing proprietary solutions in the application processor space.

"SiFive Performance is a significant milestone in our commitment to deliver a complete, scalable portfolio of RISC-V cores to customers in all markets who are at the vanguard of SOC design and are dissatisfied with the status quo," said Dr. Yunsup Lee, Co-Founder and CTO of SiFive. "These two new products cover new performance points and a wide range of application areas, from efficient vector processors that easily displace yesterday's SIMD architectures, to the bleeding edge that the P550 represents. SiFive is proud to set the standard for RISC-V processing and is ready to deliver these products to customers today."

SiFive, Inc., the industry leader in RISC-V processors and silicon solutions, today announced that Tenstorrent, an AI semiconductor and software start-up developing next-generation computers, will license the new SiFive Intelligence X280 processor in its AI training and inference processor. SiFive will deliver more details of its SiFive Intelligence initiative including the SiFive Intelligence X280 processor at the Linley Spring Processor Conference on April 23rd.

Tenstorrent's novel approach to inference and training effectively and efficiently accommodates the exponential growth in the size of machine learning models while offering best-in-class performance.

Editor's Note: This is written by Arm vice president and general manager Paul Williamson.

Over the last year, I have been inspired by the innovators who are dreaming up solutions to improve and enrich our daily lives. Tomorrow's mobile applications will be even more imaginative, immersive, and intelligent. To that point, the industry has come such a long way in making this happen. Take app stores for instance - we had the choice of roughly 500 apps when smartphones first began shipping in volume in 2007 and today there are 8.9 million apps available to choose from.

Mobile has transformed from a simple utility to the most powerful, pervasive device we engage with daily, much like Arm-based chips have progressed to more powerful but still energy-efficient SoCs. Although the chip-level innovation has already evolved significantly, more is still required as use cases become more complex, with more AI and ML workloads being processed locally on our devices.

There is expected to be more than 79 zettabytes of IoT data in 2025, but the real value of this data is found in the insights it generates. The closer to the data source we can produce these insights the better, because of the improved security, latency and energy efficiency enabled. Computational storage is emerging as a critical piece of the data storage puzzle because it puts processing power directly on the storage device, giving companies secure, quick and easy access to vital information.

Our expertise and legacy in storage puts Arm in a strong position to address the changing needs of this market - with around 85% of hard disk drive controllers and solid-state drive controllers based on Arm, we are already a trusted partner for billions of storage devices. Today, we're announcing Arm Cortex-R82, our first 64-bit, Linux-capable Cortex-R processor designed to accelerate the development and deployment of next-generation enterprise and computational storage solutions.

Huawei has been readying the entire new breed of desktop PCs with a custom motherboard, custom processor, and even a custom operating system. Being that Huawei plans to supply Chinese government institutions with these PCs, it is logical to break away from US-made technology due to security reasons. And now, thanks to the YouTube channel called "二斤自制" we have the first look at the new PC system. Powered by Huawei D920S10 desktop motherboard equipped with Kunpeng 920 7 nm Arm v8 processor with 8 cores, the PC was running the 64-bit UOS operating system, which is a Chinese modification of Linux. In the test, the PC was assembled by a third-party provider and it featured 16 GB of 2666 MHz DDR4 memory and 256 GB SSD.

The YouTube channel put it to test and in the Blender BMW render test, it has finished in 11 minutes and 47 seconds, which is quite slow. The system reportedly managed to stream 4K content well but has struggled with local playback thanks to poor encoding. Being that it runs a custom OS with a custom processor, app selection is quite narrow. The app store for the PC is accessible only if you pay an extra 800 Yuan (~$115), while the mentioned system will set you back 7,500 Yuan (~$1,060). At the heart of this system is eight-core, eight threaded Kunpeng 920 2249K processor. It features a clock speed of 2.6 GHz, has 128K of L1 cache (64K instruction cache and 64K data cache), 512K of L2, and 32 MB of L3 cache.

Intel today released and updated version of its "Architecture Instruction Set Extensions and Future Features Programming" Reference document with the latest advanced matrix extension (AMX) programming reference. This gives us some insights into AMX and how it works. While we will not go in too much depth here, the AMX is pretty simple. Intel describes it as the following: "Intel Advanced Matrix Extensions (Intel AMX) is a new 64-bit programming paradigm consisting of two components: a set of 2-dimensional registers (tiles) representing sub-arrays from a larger 2-dimensional memory image, and an accelerator able to operate on tiles, the first implementation is called TMUL (tile matrix multiply unit)." In other words, this represents another matrix processing extension that can be used for a wide variety of workload, mainly machine learning processing. The first microarchitecture that will implement the new extension is Sapphire Rapids Xeon processor. You can find more about AMX here.

Raspberry Pi 4 is almost a year old, and it's been a busy year. We've sold nearly 3 million units, shipped a couple of minor board revisions, and reduced the price of the 2 GB variant from $45 to $35. On the software side, we've done enormous amounts of work to reduce the idle and loaded power consumption of the device, passed OpenGL ES 3.1 conformance, started work on a Vulkan driver, and shipped PXE network boot mode and a prototype of USB mass storage boot mode - all this alongside the usual round of bug fixes, feature additions, and kernel version bumps.

While we launched with 1 GB, 2 GB and 4 GB variants, even at that point we had our eye on the possibility of an 8 GB Raspberry Pi 4. We were so enthusiastic about the idea that the non-existent product made its way into both the Beginner's Guide and the compliance leaflet. The BCM2711 chip that we use on Raspberry Pi 4 can address up to 16 GB of LPDDR4 SDRAM, so the real barrier to our offering a larger-memory variant was the lack of an 8 GB LPDDR4 package. These didn't exist (at least in a form that we could address) in 2019, but happily our partners at Micron stepped up earlier this year with a suitable part. And so, today, we're delighted to announce the immediate availability of the 8 GB Raspberry Pi 4, priced at just $75.

It seems Microsoft has begun the long process of phasing out 32-bit support for Windows 10 beginning with version 2004, all new OEM Windows 10 systems will be required to use 64-bit builds and Microsoft will no longer release 32-bit builds for OEM distribution. This will not affect those of you running 32-bit versions of Windows 10 who will continue to receive updates and Microsoft plans to continue to sell 32-bit versions of Windows 10 through retail channels for the foreseeable future. This is likely just the first step in what will probably be a multi-year project to gradually phase out 32-bit support as more consumers and businesses switch to 64-bit systems.

Zhaoxin is a brand that makes multi-core 64-bit x86 processors primarily for use in Chinese state IT infrastructure. It's part of the Chinese Government's ambitious plan to make its IT hardware completely indigenous. Zhaoxin's x86-64 CPU cores are co-developed by licensee VIA, specifically its CenTaur subsidiary that's making NCORE AI-enabled x86 processors. The company's KaiXian KX-6780A processor is now commercially available in China to the DIY market in the form of motherboards with embedded processors.

The KaiXian KX-6780A features an 8-core/8-thread x86-64 CPU clocked up to 2.70 GHz, 8 MB of last-level cache, a dual-channel DDR4-3200 integrated memory controller, a PCI-Express gen 3.0 root-complex, and an iGPU possibly designed by VIA's S3 Graphics division, which supports basic display and DirectX 11.1 readiness. The CPU features modern ISA, with instruction sets that include AVX, AES-NI, SHA-NI, and VT-x comparable virtualization extensions. The chip has been fabricated on TSMC 16 nm FinFET process.

No, that's not a typo. AsRock has actually launched an AMD Radeon 550 graphics card this late into the game. There is some sense behind the business decision, though; AMD's Radeon 550 is the company's entry-line offering, which aims only to improve upon the performance of integrated graphics solutions, and no more. In that sense, the Radeon 550 certainly delivers, though it does so in an underwhelming way for any tech enthusiasts. The Radeon 550 2G features 2 GB of GDDR5 memory clocked at 1,750 MHz (7,000 MHz effective) across a 64-bit bus, which feeds a die powered by 512 Stream Processors.

There are three operating modes on offer: Silent, for HTPC environments, which clocks the graphics card at 1,183 MHz boost clock and 7,000 MHz memory clocks. Default mode runs the card at AMD's defaults (1,183 MHz boost, 7,000 MHz memory). The OC mode overclocks the graphics card's boost clock to 1,230 MHz and the memory to 7,038 MHz. I/O stands at 1x dual-link DVI-D connector, 1x HDMI 2.0b port and 1x DisplayPort 1.4. A 50 W TDP should make it easy to cool in cramped spaces, and the graphics card doesn't require any power pins. No pricing was available at time of writing.

Primate Labs, developers of the ubiquitous benchmarking application GeekBench, have announced the release of version 5 of the software. The new version brings numerous changes, and one of the most important (since if affects compatibility) is that it will only be distributed in a 64-bit version. Some under the hood changes include additions to the CPU benchmark tests (including machine learning, augmented reality, and computational photography) as well as increases in the memory footprint for tests so as to better gauge impacts of your memory subsystem on your system's performance. Also introduced are different threading models for CPU benchmarking, allowing for changes in workload attribution and the corresponding impact on CPU performance.

On the Compute side of things, GeekBench 5 now supports the Vulkan API, which joins CUDA, Metal, and OpenCL. GPU-accelerated compute for computer vision tasks such as Stereo Matching, and augmented reality tasks such as Feature Matching are also available. For iOS users, there is now a Dark Mode for the results interface. GeekBench 5 is available now, 50% off, on Primate Labs' store.

Today Intel announced ModernFW - an experimental approach to building a minimum viable platform firmware for machines such as cloud server platforms. The reason for this software is that, while traditional PC Firmware has evolved over time and retained its backward compatibility, it has become very big and often inefficient.

So to meet the requirements of new platforms that need to be built quickly and adapted easily, Intel decided to offer a new software package that will help with that. The new firmware package targets x86_64 from ISA standpoint and Linux kernel based OSes.

Intel has unceremoniously, through a product change notification (PCN), discontinued the Itanium family of microprocessors. The Itanium 9700 "Kittson," which was released in 2017, is the final generation of Itanium, and its sales to new customers have stopped according to the PCN. The series has been marked "end of life" (EOL). Existing customers of Itanium who already have their IT infrastructure built around Itanium 9700 series, have an opportunity to determine their remaining demand of these processors, and place their "Last Product Discontinuance" order with Intel. The final LPD shipments would go out mid-2021.

With this move, Intel has cast uncertainty over the future of the IA-64 microarchitecture. IA-64 was originally conceived by Intel to replace 32-bit x86 at the turn of the century, as an industry-standard 64-bit processor architecture. AMD laid the foundation for its rival standard AMD64, which could go on to become x86-64. AMD64 won the battle for popularity over IA-64, as it maintained complete backwards-compatibility with x86, and could seamlessly run 32-bit software, saving enterprises and clients billions in transition costs. Intel cross-licensed it as EM64T (extended memory 64-bit technology), before standardizing the name x86-64. Itanium dragged on for close to two decades serving certain enterprise and HPC customers.

One of the runaway features for NVIDIA's latest RTX-20 series of graphics cards was the introduction of support for the OC Scanner feature - a program that automagically tests a range of frequencies on your NVIDIA graphics card and overclocks it to a deemed "stable" sweet-spot. This practically obviates the need for manual fine-tuning, though of course, the best results should always be found down that road - provided there's enough tinkering.

The latest version of MSI's Afterburner (v4.6.0 beta 10, available in the source link) now brings this functionality to Pascal-based graphics cards (besides some other features, such as voltage control, for Turing; check the entire release notes after the break). Have fun.