Arm Total Compute Archives

Arm globally unveiled its Total Compute Solutions 2023 (TCS23) platform at the Computex 2023 expo in Taipei.
TCS23 will power the next-generation secure, intelligent (AI) and immersive (3D visual) experiences.
Arm continues to innovate across the stack to power visual computing applications designed for 3nm nodes and supported by advanced software and toolsets.
The next two years are going to be super-crucial for Arm as it pivots to a newer licensing model.

Leading mobile computing SoC architecture company Arm globally unveiled its Total Compute Solutions 2023 (TCS23) platform at the Computex 2023 expo, being held in Taipei from May 30 to June 2. The latest generation of TCS will power the next-generation secure, intelligent (AI) and immersive (3D visual) experiences.

Arm commenced the TCS journey in 2021 with a focus on three key aspects of SoC design – Compute Performance, Security and Software (Developer Access). This complete package helps an SoC designer to integrate Arm IP offerings across CPU, GPU and System IP (Interconnect, SLCs, MMUs) to reduce complexity, costs and time to market.

TCS23 brings optimized physical IP, new GPU and advanced software tools

Arm continues to innovate across the stack, from its Physical/Pop IP, CoreLink System IP, ARMv9 architecture CPU cluster (Cortex-X4, -A720 and -A520) to new architecture-driven 5^th Gen GPU (Immortalis-G720), to power visual computing applications designed for 3nm nodes and supported by advanced software and toolsets.

Source: Arm

CPU advancements

The ARM v9.2 Cortex cluster aims to deliver a double-digit performance boost with Cortex-X4 (+15%) at less power (40% lower) compared to the previous generation on the same process node.
Many of the future premium Android solutions will be powered by the advanced 3nm process node in the coming years, which will further benefit from the TCS23 and beyond cluster advancements.
According to Counterpoint’s foundry tracker, almost a third of the foundry smartphone AP/SoC wafer revenues in 2024 will come from the 3nm process node

See here:

Arm also announced the taping out of the upcoming Cortex-X4 on the TSMC N3E process, an industry first.

The new DynamicIQ Shared Unt (DSU-120) overall drives the demanding multi-threaded performance scaling across multiple computing devices, from smartphones and wearables to STBs to PCs, and for all premium, performance and efficiency use cases.
Further, TCS23 is also designed to accelerate machine learning (ML) and AI workload performance ranging from 12%+ depending on the CPU core type.

System IP and GPU advancements

The new innovations across the CoreLink CI-700 help reduce the silicon areas with a significant reduction in system interconnect latency.
Arm is also promising optimizations to drive improved latency and bandwidth reduction. Almost 30% DRAM bandwidth/frame traffic reduction is expected, which translates to lower power, great for gaming workloads.
These efficiency optimizations are possible with the new 5^th gen GPU architecture-based Immortalis-G720 featuring deferred vertex shading (DVS).
Almost 4x AI and ML performance boost can be achieved as well by leveraging the new GPU.

Security and software advancements

Arm continues to innovate on security across firmware, process and TEE.
The Memory Tagging Extension (MTE) capability increases memory safety from unsafe coded apps which normally comprise almost 75% of Android vulnerabilities.
Arm extends support for the Android Virtualization Framework (AVF) for secure execution of codes for 64-bit-based devices.
Newer algorithms such as QARMA3 and enhancements for Pointer Authentication (PAC) and Branch Target Identification (BTI) reduce the overheads for maintaining security performance across the cluster.
The 9 million app developers Arm enables continue to build applications for the 64-bit architecture. For example, the China ecosystem of app developers is 100% compliant with 64-bit.
Arm NN and Arm Compute Library offer optimization for ML workloads on Armv9 architecture, which is being used by Google apps on Android with over 100 million active users already.

Roadmap, ecosystem traction and outlook

Arm’s TCS24 will sport a Blackhawk premium core, Chaberton high-performance core, Hayes high-efficiency core and Krake GPU.

Source: Arm

Arm’s overall ecosystem is the broadest and deepest considering its leading position in low-power architecture.
Arm’s partnerships for TCS23 have mainly targeted Android-based computing devices and future Windows on Arm PCs

Source: Arm, Counterpoint Analysis

Arm’s tighter Google (Android) partnership is key to the development and success of TCS spanning from overall SoC-Android optimizations to building robust security features to AI/ML optimizations and more.
MediaTek and TSMC remain the key fabless and fab partners for Arm and key drivers for TCS.
MediaTek’s entry and growth in the premium smartphone segment is great news for Arm from the TCS adoption perspective.

See here:

MediaTek’s Dimensity 9200/9200+ was a great showcase for TCS22. The upcoming Dimensity 9300 is touted to sport four Cortex-X4 and four Cortex-A720 cores leveraging the TCS23 platform.
Arm has also seen great traction for TCS in the Chinese ecosystem of OEMs and app developers leveraging premium MediaTek solutions.
Samsung is another great customer/partner for Arm. It continues to grapple with its overall Exynos SoC strategy and performance and it remains to be seen how it leverages TCS23 or TCS24 for its future roadmap.
Unlike TSMC, the Samsung LSI is not on top of the list of announcements as a key Arm partner for Cortex-X4 tapeouts on its 3nm nodes. This casts some shadow on the adoption of TCS23 by Samsung Exynos or Google Tensor in near-to- mid term.
While Qualcomm and Apple will benefit from the new Armv9.2 cluster and system IP enhancements, they have tons of proprietary CPU customizations and in-house GPU designs and AI engines to completely leverage TCS23.
This is one of the reasons why there has been no direct endorsement from Qualcomm or Apple for the new clusters or overall TCS23 at launch.
Arm-based PCs is another area that has potential for growth mainly driven by Apple, Qualcomm and MediaTek.

See here:

The next two years are going to be super-crucial for Arm as it pivots to a newer licensing model. Licensing opportunities for chipsets, especially in smartphones and PCs, have been shrinking with more vertical integration at its licensees, so the pivot makes sense but will have to wait and see.

Overall, the end-to-end system optimization provided by TCS23 with enhancements across hardware and software unlocks performance and efficiencies to power future mobile computing experiences. Arm remains a linchpin and TCS a foundational platform for the industry to build on low-power but high-performance computing experiences.

Arm, a leading intellectual property (IP) company enabling cutting edge technologies across all levels of power-efficient compute from the edge to the cloud, held its inaugural three-day “virtual” developer conference last week. Arm renamed its in-person attended developer conference ‘Arm TechCon’ last year to a fully virtual ‘Arm DevSummit’ this year. The pandemic-hit 2020 has driven all announcements, developer conversations and engineer workshops to a virtual environment. But still, it was a well-executed conference, more than doubling the reach, attendance and engagement compared to the last year!

Counterpoint Research analysts attended this three-day conference and its keynotes, workshops, and, above all, discussions with key Arm executives on the announcements, trends and outlook.

NVIDIA-Arm Deal – Elephant in the Room

Arm CEO Simon Segars kicked off the conference addressing the elephant in the room – Nvidia-Arm M&A. Simon was joined by NVIDIA CEO Jensen Huang for a fireside chat. Following were the key takeaways from the discussion:

NVIDIA is excited about the acquisition and aims to build the “computing company of the age of AI”. This allows it to bring AI prowess from the cloud to different edge and endpoints.
NVIDIA “loves the Arm business model” for foreseeing the tremendous rise of super-energy efficient computing architecture, powering billions of devices over the three decades and married with a highly scalable licensing model prevalent across every vertical and ecosystem of partners.
NVIDIA aims to bring these four pillars to the Arm portfolio:
1. GPU/DPU/Acceleration SW
2. Acceleration Libraries – Domain-specific Arch, NVIDIA AI, Analytics
3. Computing Platforms – HPC/Cloud/Edge/Robotics/PC
4. Ecosystem – Dev and Partners
To offer, sell and proliferate NVIDIA’s “Soft IP” across Arm network to drive the accelerated computing vision.
NVIDIA will continue to offer Arm products, giving the ecosystem more choices in the overlapping areas (like NPUs vs GPUs).
Combine the vibrant Arm and NVIDIA ecosystems to offer one holistic platform, richer tools to software developers, and drive open source development.
The big trend to capitalize together is the wave of “Autonomous Intelligent Computers” driven by software (AI) which is written by software (machine). So, marrying NVIDIA’s AI capabilities and Arm’s reach is the key.
Regulatory process would be something to keep an eye on in the current geopolitical climate where the US and China are locked in a trade dispute.

The key themes which revolved around the Arm DevSummit this year can be summarized in three areas as highlighted by Segars:

🔋 Power

Project Triffid: Extreme Low Power Computing

Arm’s positioning revolves around pushing the computing envelope with a focus on lower per unit energy, be it from a chip powering a microcontroller or the one powering a supercomputer.
Arm’s Project Triffid, an experimental core taped out in 28 nm, driving ultra-low-power computing with just a single flash from RFID reader to billions of edge endpoints from logistics.
This should open up intelligence across trillions of potential sensor-based nodes or IoT endpoints and nicely align with the rise of low-power connectivity such as NB-IoT or LoRa.

Infrastructure

Arm’s SVP/GM of Infrastructure Group Chris Bergey shared Arm’s traction, growth prospects and roadmap in infrastructure space from cloud powering data centers to HPC (High-Performance Computing) to evolving cloudified telco infrastructure and focus on bringing out that performance per watt, a big differentiator for Arm!

Another area where Arm sees the importance of power economics is infrastructure.
With growing workloads and internet traffic, the power requirements for the next decade could rise exponentially. This drives the need for highly efficient power vs cost vs performance metrics which can reduce the OPEX for the entire ecosystem.
With Arm’s relentless focus on power performance, Arm’s Neoverse platform for servers, networking and other infra equipment will deliver more performance per watt, sq foot and other resources.
In one of the panel discussions between AWS executive and its customers, it shared price vs performance metrics for workloads of an impressive 30-40% performance gains on the cloud servers powered by ARM Neoverse-based AWS Graviton 2 processors vs traditional servers. Further, moving workloads from Intel-based traditional servers to Graviton-based servers took less than a week and in many cases days.
Arm is promising an upward of 40% in terms of more threads capacity, double the computing performance and almost 50% gain in performance/thread for a server such as Ampere Altra (80 Arm Neoverse N1 cores, 2.6GHz) vs a traditional Intel-based 28 cores, 2.2Ghz server in the same 42U rack space.
So basically, an Ampere Altra can increase the capacity to serve 900 more customers with some amount of power and space vs a traditional server.
Arm is seeing greater adoption of its Neoverse-based partner solutions (like Ampere Altra and Marvell ThunderX3/X2) from marquee infrastructure players such as AWS, Microsoft, Tencent Cloud, Alibaba Cloud, Oracle and Baidu.
AWS customers such as Netflix, Redbox, Flickr, SmugMug and Crowdstrike have already considerable workloads running on Arm-based AWS Graviton servers.
Project Cassini, which drives standardization efforts.

🛠️ Platform

Arm’s President of IP Products Group Rene Haas talked about two key Arm approaches – Arm Total Compute, a design framework for developers to realize solution-driven optimizations across the entire Arm IP portfolio, from Cortex-A/R/M to Mali GPU to Neoverse. Haas also talked about Arm Flexible Access to offer startups and institutes greater freedom to innovate using the Arm IP.

Arm Total Compute:

Arm Total Compute design paradigm focuses on three main pillars – performance, security and developer access, which promises a level of performance independent of hardware or software. This ensures exposure of various Arm features to developers across the software ecosystem to develop unparalleled experiences on Arm-based devices.
This framework will help accelerate compute performance and will be well-optimized according to the end application or workloads.
Further, it will enable a secure foundation for hardware at the platform and compute levels (like memory tagging extensions (MTE) and Pointer Authentication Codes (PAC)).
For software, with application/services-layer security with isolation architecture, secure interconnectivity, and third-party data protection.
Along with compute and security optimizations, access to the right software and toolsets for developers, such as Arm Mobile Studio, Performance Advisor and Arm NN (for ML), is key to help developers understand and visualize the performance vectors.
Arm is working with Google, Facebook, Unity and others to help drive the Total Compute framework to a broader set of developers developing apps using their SDKs. For example, The Unity game engine powers more than 50% of games and 80% of all XR experiences.
Arm announced that the number of partners signed up for this program has doubled to more than 60 from a year ago.
Some of the startups benefitting from the program include Hailu, Atmosic, Femtosense, MemryX and ZhorTech. Bigger companies include Socionext, Nordic Semi and Faraday.
This is a very interesting move to lower the entry barriers to building cutting-edge solutions without any IP bottleneck, thus making the ecosystem vibrant and competitive.

Project Cassini: Arm SystemReady

Arm has been at the forefront of taking a platform approach to enable software and hardware development. At the summit, Arm reinforced its focus on standardization of technology, ensuring easy access to tools, systems and development, and driving AI across several different applications and markets in a scalable way.
With this in mind, the company launched an enhancement to its platform Arm SystemReady, a common framework to help a variety of software work seamlessly on Arm architecture.
It ensures consistency and stability starting from boot sequences across different hardware (Platform Security Architecture (PSA)), multiple elements of the software stack to accelerate time-to-market, and reduced development costs without reinventing the wheel.
Already a dozen of Arm customers have adopted Arm SystemReady.

Devices: Moving to 64-bit World and CPU Roadmaps

Arm’s VP & GM of Client Line of Business, Paul Williamson, talked about transitioning to a 64-bit world.

Crux of the mobile experience development has been the smartphone platform, which is evolving into multiple form-factors, from foldables to wearables.
These experiences built upon the base smartphone platform are driven by key technologies such as AI, XR, 5G, and Security. This will require powerful but power-efficient 64-bit computing cores to drive immersive and intelligent experiences.
Wearables such as wrist might remain at 32-bit, whereas XR wearable devices will be 64-bit by default.
So, from 2022, Arm’s big cores will be 64-bit only, boosting mobile app experiences, be it performance (20% better) or frame-rate uplift in games (17% higher) across smart devices from smartphones to laptops to even cars.
Arm’s Total Compute design approach will be key to optimizing these experiences across different form-factors and drive efficiencies.
Arm showcased examples of its collaborations with partners, design frameworks and toolkits on what is possible for developers to build unique, differentiated and scalable software and hardware experiences, whether it is developing a gaming app using Unity or an Augmented Reality Lens from startups such as Mojo Vision.

Arm also shared details on its future CPU roadmap called Arm Matterhorn & Makalu for the next two years. It will drive up to 30% uplift in peak performance vs the current generation.
Further, Arm is tying up with Microsoft and partners such as Fujitsu and Marvell to bring Arm-based compute using the Arm ServerReady architecture for Microsoft Azure Cloud for internal storage and VM hosting services.
It will be interesting to see how Arm’s future roadmap aligns to take on Intel and AMD in the PC space for Windows, similar to what Apple is trying to do with Apple Silicon.
Obviously, with Qualcomm’s Snapdragon 8cx powered ACPC devices, we have already seen significant battery life, silicon-level benefits (advanced process nodes 7nm), thinner and fanless designs, and easy integration of advanced connectivity capabilities (e.g. 4G, 5G), which will be the key moving forward in the PC segment.
Current partners pushing the “Windows on Arm” include Acer, Samsung, Microsoft Surface, and HP.
If Arm, Qualcomm and Microsoft can successfully attract and convince smartphone OEMs to build Windows on Arm PCs, this segment will see faster growth for Arm.
With the move to 64-bit accelerating, Microsoft is increasingly investing in tools for developers to develop/easily port majority of the apps to Arm architecture.
Possibly, NVIDIA partnership could accelerate this development further for the Microsoft PC ecosystem.
Arm is already well-positioned in wearables space in smartwatch and XR space via partners such as Qualcomm.
In the IoT space, while there are billions of IoT devices powered by the Cortex-R/M-based processors, partnership with Qualcomm and NXP (iMX8) processors can run Windows 10 IoT Core OS platform in advanced IoT devices.

Building Autonomous Systems in IoT & Mobility:

SVP & GM, Automotive & IoT, Dipti Vachani’s keynote highlighted the availability of the IP, software tools to drive streamlining and scalability to the developers for building next generation intelligent and autonomous IoT and mobility systems.

Arm has been powering billions of IoT endpoints and millions of mobility solutions over the last three decades.
However, as we move to the AIoT (AI + IoT) or IIoE (Intelligent Internet of Everything) era, these connected endpoints or edge devices are being equipped with AI/ML capabilities.
This brings a host of complexities for developers in developing for different devices, OS, middleware, cloud, and so forth, in addition to the need for accelerated hardware capabilities, end-point security and more.
Arm via Project Cassini aims to standardize this development with an open-source, portable autonomous software reference stack addressing real-time autonomy and workload acceleration.
Further, Arm has dedicated architecture solutions for autonomy, featuring capability to handle ML workloads, richer OS and advanced safety – Ethos-U55, Cortex R-82 and AE line of CPUs, GPUs and ISPs for industrial and mobility applications.
Arm is combining its next-generation development solution for Endpoint AI innovations, combining Arm Keil MDK with Arm Development Studio.
Arm has also expanded its partnership with Microsoft to efficiently deploy AI/Ml models on the low-power ARM-based IoT endpoints out of the box.
This partnership is a great catalyst for Microsoft’s Azure IoT platform which has been the leading IoT platform from cloud to edge.

Arm Flexible Access:

Arm launched Arm Flexible Access a year ago in a very fundamental shift to stimulate the ecosystem growth by providing new and existing partners 75% of the Arm IP portfolio, tools and training support without any upfront licensing commitment.
This is great for universities, researchers and startups to design and develop their own unique SoCs across different applications from IoT, automotive and AI to autonomous systems.
The important step in this journey would be how these players scale and thrive within the Arm ecosystem, especially after the NVIDIA acquisition, as many of these startups could be potential NVIDIA competitors.

🌎 Pervasiveness:

More than 180 billion Arm-based chipsets shipped by Arm partners impacting societies, businesses and economies, with 100 billion coming from just the last five years.
Arm running programs such as GenArm2Z to drive sustainability and resource efficiencies to tackle different problems via Arm Community.
It is great to see the partner, developer and customer ecosystem thriving.
Arm has one of the broadest ecosystem reach and can make a considerable impact even with one small change in its approach, architecture and initiatives at a global scale

Conclusion:

NVIDIA-Arm deal is a match made in heaven to create a leading AI-driven compute company. However, with the current geopolitical climate, tech cold war and efforts needed in convincing the broader Arm ecosystem, it is not going to be straightforward or quick.
Arm’s relentless focus to design efficient architecture to achieve “higher performance per watt” is seen across everything from HPC to IoT endpoint and is Arm’s biggest differentiator.
Arm’s CPU roadmap for client and infrastructure business both looks promising in terms of performance gains to disrupt newer segments from PCs, XR and Servers to HPCs.
Total Compute framework drives home the advantage for Arm in bringing standardization, security, time-to-market advantage and scalability to the developers.
Project Cassini – Arm SystemReady builds on the message to offer an open, collaborative environment for developers to harmonize the software from the cloud to the intelligent edge.
With Flexible Access, Arm further stimulates the ecosystem growth by bringing IP, and thus innovation opportunities, at the grassroots levels.
Overall, Arm is well-positioned to ride the upcoming “fifth wave” of the intelligent computing era.

Top

Insights Tag: Arm Total Compute

Arm Platform TCS23 Sets Benchmark to Power Advanced, Holistic Mobile Computing Experiences

Related Posts

ARM Developer Summit 2020: A Pervasive Platform Approach to Drive Efficient & Intelligent Computing at Global Scale

NVIDIA-Arm Deal – Elephant in the Room

🔋 Power