Nvidia GTC 2020 – TLDW Highlights!
Mike Sanders / 12 months ago
Jensen Huang is today delivering his keynote speech ahead of Nvidia’s digital GTC 2020 conference. With the event kicking off with a pre-recorded video, there’s a good chance that you want all the latest news, but may not be in a position to watch it hot off their YouTube page. Well, either that or you can’t be bothered to.
Let us, therefore, do the hard work for you by presenting this TLDW version of it. – If you do, incidentally, actually want to watch it, you can check out the link here!
Rest assured, we’ll be watching the event as it happens and updating this article as it goes along to give you live updates. In addition to this, we’ll also give you a quick ‘bullet-point’ on the key reveals made! So, hit that refresh button every couple of minutes because we’re going to be more than a little busy here!
The key highlights of the video will be listed below. For the live updates (and our initial thoughts) as they happen, check out further down!
NB. This has been written live while watching the videos, so please forgive any typographical or grammatical errors. There’s over an hour of video footage to digest here!
- DPU – Reveals a DPU board. A circuitry essentially designed to manage and move data. Jensen Huang believes that will, in the near future, be as important as a CPU or GPU.
- Nvidia Omniverse – A networking system allowing for 3D designers to work on a project in real-time! Server-based stuff, but this could be very exciting stuff for developers!
- Nvidia technology will be based in Spark 3.0 – A data center that can handle 163GBps. 1/5 the cost, 1/3 the power (consumption)!
- Nvidia Merlin – A new system designed to vastly improve AI-generated ‘recommendation’ systems. Interesting stuff for retailers and companies who rely on automated systems.
- Nvidia Jarvis – A new AI designed to respond accurately to voice. Speech synthesis on a new and much better level! Better still, it can associate the audio to a graphics generated model that will listen and ‘talk’ back to you. This could be an amazing new development in AI models you can actually have a legitimate conversation with, with merely a few hundred-millisecond of delay! – This could allow also for REAL-TIME language translation in audio or video conversations. You could talk to someone speaking Chinese perfectly with this feature… in theory.
- Nvidia A100 – 7nm node design Ampere processor. 1TBps bandwidth! – “The greatest generational leap” – 8 GPUs connected by NVLink. – Workstation stuff mostly, so sorry, not much in terms of gaming graphics cards. The A100 will allow 1-7 graphics cards which can each be used individually or as a whole. MIG – Multi-Instance GPU. A fully-integrated AI system. – $199,999 – A lot of money, but damn that’s a lot of performance!
- NVIDIA EGX A100 – A new AI system that will take production line automation to the next level. BMW is already signed up to install this in their factories.
- Nvidia Autonomous Vehicles – Nvidia has revealed details on its latest self-driving car technology.
So, a lot of very interesting tech there, but the biggest key ommission is that while Ampere was discussed, there was literally no news about any gaming graphics cards. With the event set to run for another week, however, there’s still time. I can’t help but feel though that if they were going to do it, they’d have done it here! Oh well…
Live Reactions As We Watch – Latest at the Top With Time Stamps!
15:53 – Nope, an AI demo video… welp, that’s about it! – That’s all folks – Catch you next time!
15:52 – One more treat?…
15:51 – It is literally a conclusion so far. He’s simply going over everything we’ve already seen and covered. I’m not sure this ‘conclusion’ was necessary and if you truly want a TLDW, this might be the video for you! It’s quick and to the point going over everything revealed here today!
15:50 – If gaming graphics cards are going to be revealed (which seems increasingly unlikely) it’s got to happen now!
VIDEO 9 – THE CONCLUSION
15:44 – A virtual reality presentation of their automated AI driving system. Looks pretty smart, albeit, I’d prefer to see it in the real world. Who knows, maybe coming to a car near you?…
15:41 – “The NVIDIA Ampere architecture is coming to the NVIDIA DRIVE platform for autonomous vehicles. NVIDIA CEO Jensen Huang describes how the scalable, efficient and high-performance architecture will power NVIDIA Orin devices spanning the whole range of AV use cases, from ADAS to Level 2+ autopilots to fully autonomous robotaxis.”
15:41 – The chances of a 30XX reveal look ever remote…
VIDEO 8 – LAST ONE FOLKS! – NVIDIA Ampere Architecture Comes to Orin for Autonomous Vehicles
15:40 – A lot of other companies are apparently on board with this! No mention of Amazon… now that IS curious!
15:38 – And Jensen confirms that BMW has already signed up to the technology. I have a friend who works there that is undoubtedly going to be thrilled (or terrified) about this.
15:34 – “State of the art, cloud nature, data center” – Capable of managing a fleet of centers and sensors. It sounds mostly like a really advanced form of factory security/production surveillance. Something that could be used to run automated production lines. I guess this is where BMW is entering the picture…
15:33 – Talking a lot about the specifications, but I’m still not entirely certain what he’s ‘revealing’ here.
15:29 – “NVIDIA CEO Jensen Huang introduces the NVIDIA EGX A100 converged accelerator, which combines an NVIDIA Ampere GPU and NVIDIA Mellanox ConnectX-6 DX SmartNIC to bring secure computing to the edge. Huang also announces that BMW has selected the NVIDIA Isaac robotics platform for its factory logistics.”
VIDEO 7 – NVIDIA EGX A100 Converged Accelerator and Isaac Robotics Platform
15:36 – Data center comparisons in terms of space and power. Before and after! $11m data center into a $1m rack!
15:25 – Up to 150x faster than existing server systems. It’s in full production and available now for $199,999.
15:34 – 2D CAD designs turned into this demo video. A demo of the A100 in action.
15:22 – The worlds first integrated AI system. This is, pretty much, supercomputer territory! PCI-4 SSD’s with 4 petaflops of total capability!
15:20 – 70% more transistors (presumably compared to Volta). We get a demo of it working on noise recognition. Specifically, it can identify a bird by its call. I think we saw this demoed around 6 months ago. He’s specifically using the demo to show one ‘MIG’ working on the task, which in itself is faster than Volta. 500 comparisons on MIG compared to 80 from Volta.
15:18 – The A100 will allow 1-7 graphics cards which can each be used individually or as a whole. MIG – Multi Instance GPU
15:15 – And this was what Jensen brought out of the oven yesterday… The A100 – The world’s largest graphics card – 8 GPUs connected by NVLink,
15:14 – Speaking how Ampere is going to be significantly faster than Volta. If the demo images is legitimate, that might be an understatement! – “The greatest generational leap”
15:13 – Nvidia A100 – 7nm node design Ampere processor. 1TBps bandwidth!
15:12 – Cloud computing – If you were hoping for a Nvidia 30XX announcement, I think you’re going to be disappointed. I stand to be corrected, but I think this is mostly a ‘wow, look at us’ presentation.
15:10 – “NVIDIA CEO Jensen Huang introduces the NVIDIA A100 data center GPU, which dramatically increases throughput of scale-out and scale-up applications whether for data analytics, training or inference. The A100 delivers 1.5 terabytes per second of bandwidth thanks to its third-generation Tensor Cores, multi-instance GPU technology, sparsity acceleration, and third-generation NVLink and NVSwitch interconnect technology. It’s based on new NVIDIA Ampere architecture, offering the greatest generational leap in performance in the company’s history. The A100 is offered in the DGX A100, an integrated AI system with 5 petaflops of performance on a single node that comes with fully accelerated software stacks. It’s also available in HGX A100 hyperscale data center accelerators from the world’s leading systems makers.”
15:09 – Following a brief mention earlier, it sounds like we’re finally getting to Ampere. Likely that Nvidia is looking to deploy this for both their commercial and gaming graphics cards.
VIDEO 6 – NVIDIA A100 Data Center GPU Based on NVIDIA Ampere Architecture
15:03 – Jensen is talking to ‘Misty’ a demo of the Jarvis AI. It’s pretty amazing as the graphic model responds not only to Jensen, but also gives a visual representation as to the topic of the conversation. When he talked about cold, she froze. When he talked about thunder, she took a virtual bolt! – Conversational AI!
15:02 – The Jarvis AI utilizes a graphics generated face to respond to the voice output. A demo video displays showing a head ‘singing’ along to a hip hop track.
15:01 – Does this mean I don’t have to shout my postcode at the phone in an annoying high level of volume?…
15:00 – “NVIDIA Jarvis is a new application framework that recognizes vision and sound to power conversational AI workflows. Watch NVIDIA CEO Jensen Huang interact with Misty, a conversational AI that uses inference to intelligently respond in just milliseconds. With Jarvis, companies can create real-time translations, closed captioning of everyone speaking, or transcriptions of video calls in real time or power a number of other applications such as smart speakers, call centers, and retail services.”
14:59 – I suspect that this video might be something to do with the generated voice and music shown in the first promotional video. What it may essentially boil down to is a huge improvement in automated phone lines.
VIDEO 5 – NVIDIA Jarvis for Conversational AI
14:54 – Nvidia Merlin
14:52 – I’m not 100% sure I’m grasping it, but it appears this system is being designed for automated retailers to help improve ‘recommendations’. This isn’t incidentally just things like YouTube. It can also be set in help centers and potentially hospitals to help recommend the next most logical step for an individual.
14:50 – The only thing really grabbing my attention at the moment is that graphics card casually sat on the side…
14:48 – “NVIDIA Merlin is a new application framework that simplifies and codifies recommendation systems for use by all industries. Recommender engines are one of the most important and complex machine learning pipelines, and they’re key to having a personalized internet experience.”
VIDEO 4 – This one caused the delay as it was released as one of the last on their channel – “NVIDIA Merlin for Recommendation Systems”
14:40 – Spark 3.0 will be an Nvidia based system. Making massive data handling fast! Benchmark figures showing Dell’s existing 16GB/s benchmark system. Nvidia’s new Spark 3.0 will can run as fast as 163GB/s. 1/5th the cost, 1/3 the power.
14:38 – Machine Learning Algorithms – Data processing and AI. “SPARKS turns an entire data center into a compute engine” – Discusses the issue of cache memory limitations in such large systems. Presumably, because Jensen’s about to tell us how much better they’ve made it.
14:37 – AMPERE is mentioned for the first time! Get Amped?…
14:36 – “GPU Accelerating HPC and Scientific Computing” – Sounds a little dry…
VIDEO 3 BEGINS
14:34 – Real-time ray-tracing tech graphical demo. It looks insanely impressive and it was all remotely designed! ‘Marbles RTX’.
14:33 – NVIDIA OMNIVERSE – Sounds fancy, but it essentially means a conglomeration of their technology for 3D design applications. Real-time editing by various users – Demo video starts
14:30 – Minecraft RTX – Discussing how DLSS helped bring the frame rates up in the game when ray tracing was turned on. Honestly nothing much new so far, this is merely a recap of already known information. – Video presentation of the game running with RTX on/off comparisons.
14:29 – Discussing DLSS 2.0 which was launched a month ago and the improvements it brought are honestly pretty stunning!
14:27 – Largely just talking about how ray tracing works – He’s now moved onto a recap of DLSS. 16K image AI upscaled from 720p!
14:24 – Well, we knew this was happening, but ray tracing is getting faster. A LOT FASTER!
14:21 – Ray Tracing is mentioned, this is already sounding like a good watch!
14:20 – Apologies for the delay, but Nvidia released the videos in the wrong order! AGHHHH!!!
14:18 – FOUR NEW SYSTEMS! – Moving onto video 2!
14:16 – Talking about the number of companies who are choosing to use Nvidia for their design and development.
14:15 – DPU – Data processing unit. Jensen thinks this will be as integral to new computing as CPUs and GPUs.
14:13 – Talks about their recent acquisition of Mellanox and reveals some of their technology. Clearly some very impressive networking technology here. Masses of ports on a router with equal data distribution
14:12 – Starts talking about data centres. In other words, AI and GPU technology looking to make things even faster!
14:11 – Gee, Jensen sure has a nice kitchen… Albeit, that black jacket and shirt is in danger of turning into a trademark 😀
14:10 – Ok, that’s impressive. The music and voice over in the video have been AI-generated. Really pushing AI in this presentation.
14:05 – Presentational video. Nothing specific discussed here except Nvidia’s technology in the various fields it researches and develops.
14:06 – Jensen starts by talking about how Nvidia has been working (with their AI systems) to help try and accelerate finding the cure for COVID-19.
14:01 – Ok, we didn’t expect 9 separate videos, but here we are! Let’s get into this with video 1!
What Do You Think?
So, there you have it, Nvidia’s keynote GTC 2020 has concluded and we certainly have more than a little to digest here, but what do you think?
Let us know what your highlights have been in the comments and, rest assured, we’ll be bring you all the big news from the event as this proceeds over the next week!