NVIDIA Computex 2023 Recap: Lots of AI Talk, Powerful GH200 GPUs, Grace Hopper Superchips & New Servers

NVIDIA had loads of announcements &  product unveils during Computex 2023's opening keynote by CEO, Jensen Huang. As expected there was a lot of AI talk but there were some nice technologies introduced along with new hardware which is worth recapping for those who missed it out earlier.

AI, GPUs, Servers & "The More You Buy, The More You Save" Were They Key Topics By NVIDIA At Computex 2023

So there's no doubt that AI has been the leading force that has skyrocketed NVIDIA above and beyond everyone else in the technology sector. The company's & CEO, Jensen Huang's, bets on Artificial Intelligence (AI) that it had made several years ago are finally paying off. So let's have a recap of everything NVIDIA announced at the event.

Games With AI-Driven Characters Powered By NVIDIA ACE

It's odd to start off the recap with a software announcement but the truth is that NVIDIA's software is worth talking about more these days than the hardware. The company's event was loaded with several AI-powered software announcements and the one I personally liked the most was the ACE (Avatar Cloud Engine) demo which will allow game devs to utilize AI to create NPCs that are interactable and react naturally to player-based interactions.

NVIDIA Grace Hopper Superchips Enter Full Production With Arm CPU & GH200 GPU Cores

NVIDIA also revealed that its Grace Hopper Superchip featuring Arc GPU cores and GH200 GPU cores had entered full production and will be powering a range of solutions with the most impressive one being the DGX GH200 AI supercomputer, an Exaflop machine that will offer a total of 256 Grace Hopper Superchips and 144 TB of shared memory. Some numbers shared by NVIDIA reveal that the DGX GH200 AI supercomputer offers 500x more memory than the Ampere-based NVIDIA DGX A100 and 48x higher NVLINK bandwidth.

Faster & Efficient Generative AI Capabilities For Everyone With RTX

Moving back to software once again, NVIDIA announced that it is going to offer a new model for its RTX Windows 11 users that enables an efficient Generative AI mode known as Max-Q Low Power. With this mode, NVIDIA says that users and applications can access a complete RTX-accelerated development stack running on Windows 11. These AI applications can then be offered to an installed base of more than 100 million RTX users on PCs & workstations.

More Servers, More Buying, More Saving

In terms of the number of solutions on offer, NVIDIA introduced a new platform known as MGX which would allow users to mix and match from over 100 different server variations in any chip configuration. These servers will be designed according to the needs of customers and leading manufacturers will adopt MGX to significantly reduce their development costs. Jensen Huang also talked about the TCO and overall cost/performance advantages that come with buying NVIDIA's DGX/HGX/MGX solutions and you can read our detailed analysis over here.

Spectrum-X Is Here To Accelerate Network Platforms

Another key announcement in terms of hardware was the Spectrum-X Networking platform which combines NVIDIA's existing hardware such as Spectrum-4, and Bluefield-3 DPUs with class-leading software to enable faster networking speeds that can help improve generative AI content. The chip itself was a gargantuan design featuring 100 Billion transistors, delivers over 51 TB/s, and adopts a 400 GbE network that is fully optimized for clouds.

MediaTek & NVIDIA Joins Hands But Only For Autonomous Vehicles

One of the most anticipated announcements was the MediaTek and NVIDIA partnership. The announcement did happen but it turns out that both companies are only partnering for AI-driven automobiles and not for mobility SoCs. A new SoC platform known as MediaTek's Dimensity Auto will use NVIDIA's advanced driver assistance systems and a GPU chiplet that will be connected by an ultra-fast and coherent chiplet interconnect technology. This new partnership can lead to possible collaboration between the two companies on future mobile SOCs too.

These were the highlights of NVIDIA's Computex 2023 keynote. Oh and before I finish it all, I must also state that NVIDIA did reaffirm its Hopper Next GPU launch for 2024 so we are going to hear more about that by the next major GTC (2024) event.

Let us know your thoughts on the NVIDIA Computex 2023 keynote and what you think was the most interesting part of the CEO, Jensen Huang, keynote.

Written by Hassan Mujtaba