◉ The “vacuum tube” era of Quantum computing begins. While we are still many years away from a viable quantum computer, 2020 is the year we will see the first real quantum technology applied to solve small problems in a radically new way. Long- term we see three big conditions that must be true before we have viable impactful quantum technology:
1. A viable quantum computing architecture needs to be built. Today we have 25 or 53 qbit systems that are nowhere near the scale needed to run advanced algorithms or solve the theoretical problems quantum may address. We also do not have consensus on what a quantum computer is as different teams propose different models including trapped Ion, trapped photon, etc.
2. Quantum systems must be practical in real-world environments. Today the early systems are exotic to the extreme. Many need supercooled cryogenic systems to exist and they are incredibly fragile. We need quantum qbit capacity to be delivered much as we deliver compute capacity – via standardized chip level building blocks.
3. We need a standardized way to interact via software with quantum. Today there isn’t standard consensus API or even a broad agreement on how quantum sits in the rest of the IT stack. There is a shift to agree that it will not replace traditional compute but look much like an accelerator (GPU, FPGA, SmartNIC, etc), however, the actual way that happens is still not a reality.
Look for lots of quantum announcements and “breakthroughs” in these areas in 2020 but we predict we will still be in the vacuum tube era when 2020 ends.
◉ Domain Specific Architectures in Compute will become a reality. We have lived in a world of homogeneous compute for many years. X86 is the compute architecture powering the cloud era and most modern IT. While x86 is still critical to run broad general software, as we enter the AI/Machine Learning era, we need far greater compute capacity per watt and, with Moore’s Law winding down, we need alternative models. The one that seems to be the winner is X86 augmented by domain specific architectures to accelerate specific kinds of software and functions. We already have this for features such as encryption but in 2020 we will see a massive expansion of the available chipsets that accelerate specific domains. Some examples will be a next wave of SmartNICs to offload and accelerate not just networking but higher level functions in the communications stream, general purpose AI/ML chips that are optimized for 4 or 8 bit precision and only accelerate AI/ML tasks, chips that emulate neural networks in silicon, low power AI inferencing chips for the edge, and many more.
2020 will be the first year that we have a wide range of domain specific architectures and that will cause us to change system architecture to accommodate them. We will need dense acceleration servicers (Like Dell DSS8440 or 940XA); we will need an ecosystem approach to these accelerators to pre-integrate them into solutions and make consumption easy; and we will need to virtualize and pool them (VMWare Bitfusion as an example) and create APIs to interact with them such as OpenCL and CUDA. By the end of 2020, we predict most enterprises will begin the process of shifting to a heterogeneous compute model built with X86 plus domain specific architectures.
◉ 5G will change how we think about wireless network capabilities. Early 5G roll outs are happening now but in 2020 we will start to see the full potential of 5G. It clearly will give us higher bandwidth and lower latency than 4G and that’s good but what will change is that we will begin to think about how we use the new capabilities of 5G. New is the ability to program 5G to deliver network slices for specific enterprise applications and users to create one end to end experience via cloud orchestration (we showed this in 2019 at Mobile World Congress).
Addictingly, 5G is not one size fits all wireless. Beyond the first use case of Mobile Broad Band (mBB), 5G will add two entirely different capabilities in the wireless systems. First, is Ultra Reliable Low Latency Communication (URLLC) which will make real-time systems like drones and AR more effective. Second, 5G will add massive machine type communications (mMTC) that will optimize 5G for the world of billions of low power lightly connected sensors. However, the biggest new capability 5G will expose is an edge compute model that enterprises will begin to look at to deploy their real-time and data-intensive applications into the 5G network close to the users to get a faster response time for tasks like AI-driven control systems in factories or cars but also will push pre-processing of data to that edge to control the data flow back into data centers and clouds. By the end of 2020, we predict customers will begin to fully understand the significant capability change of 5G and start developing ways to take advantage of it to digitize their businesses.
There are many other emerging technologies that will show up in 2020 but these three – 5G, Domain Specific Architectures and Quantum represent the ones that are likely to change the trajectory of the industry over the long term. While Quantum will not do that for many years, what’s exciting about 2020 is that all three of them are becoming real enough to now enter the technical and business dialog for the first time broadly.
Dell Technologies is working in all of these areas as we see a data explosion coming that will require orders of magnitude more compute, storage, networking and applications capacity to keep up. We are, understandably, excited that 2020 will be a year where we not only continue to move existing technologies forward but also a year where many potential game-changing technologies become real enough to be part of the strategies our customers are developing to win in the digital transformed world.
Posts
0 comments:
Post a Comment