Components & Peripherals News
Nvidia CEO Jensen Huang’s 10 Biggest Statements At GTC 2020
Among his biggest statements at GTC 2020, Huang believes the chipmaker’s EGX edge AI platform is ’the iPhone moment for the world’s industries’ and that its new partnerships with VMware are ’as exciting as when VMware first virtualized the data center.’
Making The Case For Nvidia
See the latest entry: Nvidia CEO Jensen Huang’s Top 5 Remarks From GTC 2022
Nvidia CEO Jensen Huang is making a case that the chipmaker is becoming an essential company for modern data centers and computing at GTC 2020 this week.
The Santa Clara, Calif.-based company kicked off the virtual conference on Monday with a variety of big announcements, including the reveal of a new family of BlueField DPUs that the company believes will benefit every server — in the data center, in the cloud and at the edge — regardless of workload.
While Nvidia’s traditional business has been in GPUs, first for PC gaming and now AI applications, the company has expanded the view of itself to a “data-center-scale company” that also provides high-speed and intelligent network technology, systems and software. And with the company’s pending $40 billion acquisition of British chip designer Arm, Nvidia could own an even bigger part of the pie.
“AI requires a whole reinvention of computing, full-stack rethinking, from chips to systems, algorithms, tools, the ecosystem,” Huang said in his keynote for the fall GTC 2020 event. “Nvidia is a full-stack computing company. We love working on extremely hard computing problems that have great impact on the world. This is right in our wheelhouse.”
What follows are Huang’s 10 biggest statements he made at GTC 2020 regarding the future of computing, AI and data centers as well as what role Nvidia will play in those areas.
‘AI Requires A Whole Reinvention Of Computing’
The software written by AI is very different than that written by a human. It is vastly more parallel, and thousands-to-millions-of-times more compute intensive. The method of developing the software is different, the computing infrastructure is different, the tools are different, the software runs differently, and even the method of deployment is different.
AI requires a whole reinvention of computing, full-stack rethinking, from chips to systems, algorithms, tools, the ecosystem. Nvidia is a full-stack computing company. We love working on extremely hard computing problems that have great impact on the world. This is right in our wheelhouse. We’re all in to advance and democratize this new form of computing for the age of AI. Nvidia’s dedicated to advancing accelerated computing. Enabling developer success on our platform is core to everything we do.
‘Ampere Is The Fastest Ramp In Our History’
Twenty years ago, Nvidia introduced the programmable shading GPU. The GPU revolutionized computer graphics. Hundreds of millions of insatiable gamers made the GPU the most advanced chip in the world. Nvidia GPU processing has increased a stunning 100,000-fold.
We recently launched Ampere [GPUs for PC gaming]. It is the greatest generational leap ever. We expected and prepared for huge demand, and the response was still staggering. I know many of you are anxiously waiting for your new Ampere GPU. It’s totally understandable and I assure you that it is worth the wait. Ampere is the fastest ramp in our history. Everything we ship is instantly sold out. We’re working around the clock. Your new GPU is coming. You’re going to love the Ampere GPU.
‘Nvidia GPUs Will Represent 90% Of Total Cloud Inference’ In A Few Years
Processing the AI on Nvidia is the most performant and cost effective. This year is the 10th anniversary of the first Nvidia GPU in the AWS cloud. Since AWS, every cloud now offers Nvidia GPU, and the aggregate compute throughput has increased 10x every two years. This year, we will ship over 166 exaops of inference compute to CSPs more than 6x that we shipped last year.
We estimate that the aggregate Nvidia GPU-inference-compute in the cloud now exceeds that of all cloud CPUs. With this trend, in a couple of two-three years, Nvidia GPUs will represent 90 percent of the total cloud inference compute. Any AI application and service can now rely on Nvidia Inference. We are past the tipping point.
Nvidia Has The ‘Opportunity To Revolution Video Conferencing Of Today’
Live video calls are one of the highest traffic on the internet today. For work, social, school, virtual events, doctor visits video conferencing is now the most important application for many people.
Today we are announcing Nvidia Maxine. a cloud-native streaming video AI platform for applications like video calls. AI can do magic for video calls. Using AI, we can perceive the important features of a face, send only the changes of the features, and re-animate your face at the receiver. This reduces bandwidth by a factor of 10. AI can reorient your face so you’re making eye contact with each person on the call, individually. Your face is regenerated to make eye contact with each person.
AI can realistically animate an avatar based on just the words you are speaking. AI can remove background noise, super-res, see better in low light, replace your background, and even re-light your face. And with Jarvis conversational AI, we can do real-time language translation and closed captioning: one person is heard but everyone can simultaneously talk. What they say is closed captioned or texted.
With Jarvis and Maxine, we have an opportunity to revolutionize video conferencing of today and invent the virtual presence of tomorrow.
‘Data Center Is The New Unit Of Computing’
Data center is the new unit of computing. Modern data center workloads are diverse and not limited to monolithic applications running on single nodes. AI and data analytics applications are distributed, running on multi-GPU and multi-node. Cloud services are disaggregated and composed of microservices. These new workloads put tremendous pressure on the network. East-west traffic has skyrocketed.
Nvidia created the Magnum IO SDK to offload moving data from the CPU with Mellanox RDMA and accelerate the networking, storage, and security processing on Mellanox NICs. Magnum IO SDK includes acceleration for each computing domain.
Modern data centers are also software-defined. The compute virtualization trend that enabled easier resource pooling and management has now extended to networking, storage, and security. What used to be dedicated hardware appliances are now software services running on CPUs. The entire data center is software programmable and can be provisioned as a service.
‘A New Type Of Processor Is Needed’ For Data Movement, Security Processing
The software-defined data center is great for manageability, scalability, and security, but all the data center infrastructure processing in software is a huge tax on CPUs. As more users load the hyperscalers, each microservice comes with it the associated virtualization, networking, storage and security processing — all of it consuming CPU resources.
A new type of processor is needed that is designed for data movement and security processing. We call it the data processing unit. The DPU consists of accelerators for networking, storage, and security, and programmable Arm CPUs to offload the hypervisor. The DPU is a data center infrastructure processing chip. It is estimated that at least 30 percent of the CPU cores can be consumed running the data center infrastructure.
Today we are announcing the BlueField-2 DPU. It is a programmable processor with accelerators and engines for at-line-speed processing for networking, storage, and security. The BlueField DPU is a data center infrastructure on a chip. BlueField-2 has Arm CPUs and a whole host of state-of-the-art accelerators and hardware engines.
BlueField-2 does the security processing for private, public, and hybrid clouds. BlueField air-gaps the application domain from the infrastructure domain, stores and accelerates encryption keys, does SHA-256 authentication and encryption protocol processing. It does regular expression and deep packet inspection processing for app recognition, intrusion protection, web application firewall, and out-of-band malware detection. BlueField-2 includes traffic-rate-controlled packet pacing for video streaming, SMPT-2110 broadcast, and 5G networks.
BlueField-2 is also an amazing NVMe storage processor. It’s great for both compute as well as the storage server. It does elastic block storage, block storage encryption, de-dup, and compression. These are some of the examples of advanced functions that BlueField-2 offloads from the CPU and accelerates.
New VMware Partnership For BlueField ‘Will Redefine The Data Center’
Today we are thrilled to announce a major partnership with VMware. VMware runs the world’s enterprise. They’re the OS platform in 70 percent of the world’s companies. VMware pioneered the virtualized enterprise and now driving the software-defined data center revolution.
Together, we are porting VMware onto BlueField. BlueField is a data center infrastructure processor and VMware is the data center infrastructure OS. Our partnership will redefine the data center. We will offload virtualization, networking, storage, and security onto the BlueField and enable distributed, zero-trust security. Our work will give the 30-40 million enterprise servers around the world a big boost in performance and security.
VMware AI Computing Tie-Up ‘As Exciting As When VMware First Virtualized The Data Center’
There are two types of computing environments, bare metal for scale-out and virtualized multi-tenant. Bare metal is essential to achieve multi-GPU, multi-node scale-out, needed for large data processing or training. Scale-out and virtualized computing cannot share one infrastructure today.
This creates two distinct infrastructures that are managed and operated differently, not shareable and more difficult to secure. There is a solution. Nvidia and VMware are announcing a second partnership. We are going to do some serious computer science to create a data center platform that can support GPU acceleration for all three major domains of computing today — virtualized, distributed scale-out and composable micro-services — all Nvidia-accelerated.
Enterprises running VMware will be able enjoy Nvidia GPU and AI computing in any computing mode. This is a massively exciting project and every bit as exciting as when VMware first virtualized the data center. VMware and Nvidia are going to bring AI to the world’s enterprises.
EGX Is ‘The iPhone Moment For The World’s Industries’
Nvidia EGX is designed to make it easy for the world’s enterprises to quickly stand up a state-of-the-art edge AI server. Nvidia EGX can control factories of robots, perform automatic checkout at retail, orchestrate a fleet of inventory movers or help nurses monitor patients. EGX is a full-stack platform consisting of the AI computer, system software, AI frameworks, and fleet management cloud service. Deploying and provisioning services on EGX is simple with one-touch authentication to setup a new node. No Linux admins are required.
One of the most important pillars of EGX is the rich ecosystem of partners OEMs, software partners, and industry-focused solution makers. The EGX AI computer integrates a Mellanox BlueField-2 DPU and Ampere GPU into a single PCI express card, turning any standard OEM server into a secure, accelerated AI data center. Security is designed from the ground-up on EGX. OS is booted securely and measured, the applications and AI models are signed and protected in the AI enclave, data in-motion and data at-rest are encrypted and secure.
This is the Nvidia EGX Edge AI Platform. And with our network of partners, EGX can reach companies all over the world in manufacturing, healthcare, retail, logistics, to transportation. Nvidia EGX will make it easy to create, deploy and operate industrial AI services. This is the iPhone moment for the world’s industries. Over the past few months, we’ve been building the platform, preparing our partners, and working with lighthouse partners to refine EGX.
New Initiative Will Make Arm ‘Leading-Edge At Accelerated And AI Computing’
A few weeks ago, we announced our intention to acquire Arm. Arm is the most popular CPU in the world. Together, we can offer Nvidia accelerated and AI computing technologies to the Arm ecosystem, reaching computers everywhere. Last year, we announced porting CUDA and our scientific computing stack to Arm. The performance was excellent and the response from the HPC community was fantastic.
Today we are announcing a major initiative to advance the Arm platform. We are making investments across three dimensions. First, we’re complementing Arm partners with GPU, networking, storage, and security technologies to create complete accelerated platforms. Second, we’re working with Arm partners to create platforms for HPC, cloud, edge, and PC. This requires chips, systems, and system software. And third, we are porting the Nvidia AI and Nvidia RTX engines to Arm.
Today, these capabilities are available only on x86. With this initiative, Arm platforms will also be leading-edge at accelerated and AI computing.