The Nvidia GPU Technology Conference 2020, also known as GTC, keynote was released to the public today. The annual conference is aimed at providing training, insights and direct access to experts regarding the current biggest topics in computing. This year's keynote was mostly focused on data centers and AI technology, especially in the pursuit of helping organizations with the COVID-19 pandemic.
We were greeted with all sorts of interesting and wonderful news from the CEO of Nvidia Jensen Huang’s very own kitchen, after the company announced they will be transitioning to an online conference instead due to COVID-19. The GTC 2020 “kitchen keynote” brought us the new Ampere GPU architecture and the new A100 GPU for data centers. So here's our GTC 2020 overview and the biggest news from this year's online conference.
Nvidia Ampere A100 GPU
Huang unveiled the world’s largest graphics card, the A100, fresh from out of the oven... literally. Okay, I have to admit that last bit was teased yesterday with the video shown below, but it was still a big part of the conference and we found out a lot more info about it today including that it uses the company’s latest 7nm Ampere GPU architecture.
Ampere is the latest GPU architecture from Nvidia and is what the A100 is based on. Even though it's only a data center GPU and not at all intended for home use or gaming, Hunag has reportedly mentioned that the Ampere architecture will replace the Turing and Volta architecture for all next-gen graphics cards for the next year or two.
It was most certainly the biggest announcement of the keynote, as it’s the first GPU to use the Ampere architecture from Nvidia. However It’s mostly designed for Artificial Intelligence, but is already in full production and shipping to customers now. Boasting over 54 billion transistors, it’s dubbed as the world’s largest 7nm processor, which provides the biggest generational leap in AI performance with a 20 times increase in power compared to Nvidia’s previous Volta V100.
Imagine hooking that up to run some modern games! What FPS would it even achieve? And can it run Crysis Remastered?
Working off the A100 announcement, the DGX A100 is a deep learning system that includes 8 of the aforementioned A100’s and a whopping 5 petaflops of performance for a complete and flexible data center package for the very first time. This even allows for cheaper and more energy efficient systems as the performance is equivalent to 50 DGX-1 systems and 600 CPU systems, offering around 1/10th of the cost and 1/20th of the power usage overall.
HGX A100 Server Building Block
Along with the launch of the A100, the HGX A100 was also announced as a way to help OEMs build their own systems based on the A100. Using an integrated baseboard and Nvidia’s NVLink and NVSwitch, it allows for up to 8 A100’s to communicate with each other and even serve as one single GPU capable of 10 petaflops of performance.
The EGX A100 combines the A100 GPU with a Mellanox ConnectX-6 Dx SmartNIC is aimed for 5G applications and allows for the protection and real-time processing of edge sensor streaming data. This makes it the ultimate platform for real-time decision making in places like hospitals or retail stores as it can route data directly to the GPU memory for AI or 5G signal processing with up to 200Gbps of data.
One of the big announcements Huang made at the GTC 2020 conference was the acquisition of Mellanox Technologies which has made most of the technology mentioned possible, thanks to their high performance networking capabilities. Huang mentioned that high speed networking and high speed computing “go hand in hand” which was a main reason as to why Nvidia bought Mellanox.
Is it just me, or does this kind of sound like that Silicon Valley TV show?
So that was all the big news that came out of GTC 2020 today, tell us what you found the most interesting (although we can have a pretty good guess) and let us know if we missed anything else!