10:58PM EDT – We’re again for day two of our reside weblog protection of Computex
10:59PM EDT – Following AMD’s massive, CPU-centric showcase, NVIDIA is on deck this night to ship what guarantees to be a various keynote touching upon a number of facets of NVIDIA’s enterprise
11:00PM EDT – The untitled keynote is slated to run for an hour and incorporates a who’s who of NVIDIA VPs, together with Ian Buck (VP Accelerated Computing), Michael Kagan (CTO), and Jeff Fisher (SVP GeForce)
11:00PM EDT – So count on the keynote to cowl all the things from datacenters to gaming
11:01PM EDT – And right here we go
11:02PM EDT – Starting with a fast promo video of the NVIDIA and the expansion of their GPU enterprise from gaming to AI and rather more
11:03PM EDT – First up, Ian Buck
11:03PM EDT – Starting with AI
11:04PM EDT – This requires reimagining the datacenter itself
11:04PM EDT – AI can be enabling a brand new marketplace for digital twins
11:04PM EDT – A possible $150B market
11:04PM EDT – And an extra $100B marketplace for cloud sport streaming
11:05PM EDT – All of which NVIDIA’s companions can faucet into by utilizing NVIDIA’s {hardware}
11:05PM EDT – These duties eat a whole lot of computing energy – so vitality effectivity is a giant deal
11:06PM EDT – Of course, NVIDIA can be joyful to argue that a whole lot of energy may be saved by transferring from utilizing CPUs for compute duties to GPUs
11:07PM EDT – And NVIDIA has a large assortment of different {hardware}, together with Grace CPUs and ex-Mellanox networking {hardware}
11:07PM EDT – Now recapping the H100 accelerator
11:08PM EDT – NVLink 4, Multi-Instance GPU, Confidential Computing, transformer accelerator, and so on
11:08PM EDT – And NVLink switches for forming bigger clusters of computer systems
11:10PM EDT – And a shout-out to NVIDIA’s Taiwanese companions for his or her half in constructing these new servers
11:10PM EDT – Current NVIDIA roadmap
11:11PM EDT – Two 12 months rhythm on {hardware}. Interleaving between x86 and Arm
11:11PM EDT – Now on to Grace
11:11PM EDT – Grace is on observe to ship subsequent 12 months
11:11PM EDT – Grace and Grace Hopper
11:12PM EDT – For CPU and CPU+GPU workloads respectively
11:12PM EDT – The Grace Superchip (twin Graces) has 144 cores. It has 1TB of LPDDR5X reminiscence and consumes 500 Watts
11:13PM EDT – And reiterating NVIDIA’s plans to place NVLink in all of their silicon, in order that it may be linked as much as different NVIDIA and third occasion {hardware} with a excessive pace cache coherent interconnect
11:14PM EDT – NVIDIA has developed numerous reference programs over time, and they’re now producing a brand new technology of them primarily based on Grace
11:15PM EDT – New CGX, OVX, and GFX refrence designs utilizing Grace and/or Grace Hopper
11:15PM EDT – Announcing HGX Grace and HGX Grace Hopper programs
11:15PM EDT – Both are particularly designed for OEM 2U chassis
11:16PM EDT – Can leverage their current system architectures
11:16PM EDT – Grace programs will begin transport within the first half of 2023
11:17PM EDT – Now on to speaking about Bluefield-3 NPUs
11:18PM EDT – NVIDIA is betting closely on Bluefield and DPUs changing into important for datacenters
11:18PM EDT – It is silicon-heavy and can have the ability to leverage NVIDIA’s GPUs for processing
11:19PM EDT – Or as NVIDIA likes to place it, this networking gear would be the spine of AI programs
11:19PM EDT – Now on to NVIDIA’s enterprise software program stack
11:20PM EDT – As a reminder, NVIDIA has extra software program engineers than {hardware} engineers. So on the enterprise stage, software program is much more of a differentiating issue for NV than {hardware} is
11:20PM EDT – The subsequent wave of AI: robotics
11:21PM EDT – “We see a transparent demand for automation”
11:22PM EDT – Recapping NVIDIA’s Isaac robotics platform
11:23PM EDT – And how Isaac interfaces with omniverse for simulation and robotic coaching
11:24PM EDT – An enormous a part of enabling this are NVIDIA’s pre-trained fashions obtainable on NGC
11:25PM EDT – Which is one other instance of the place NVIDIA is utilizing software program to construct up their ecosystem
11:26PM EDT – Isaac Sim utilizing Omniverse
11:26PM EDT – NVIDIA desires to shut the sim-to-real hole
11:26PM EDT – Announcing Isaac Sim 2022.1 launch
11:27PM EDT – Now rolling a demo of Isaac Sim
11:29PM EDT – Now on to NVIDIA’s {hardware} for robotics: Jetson
11:31PM EDT – NVIDIA’s Orin SoC is reportedly performing effectively in MLPerf benchmarks
11:31PM EDT – The Jetson AGX Orin dev equipment is out there now
11:32PM EDT – Production programs with Jetson Orin can be found now, and lots of are being introduced at Computex this week
11:33PM EDT – Isaac Nova Orin. A state-of-the-art structure for autonomous cell robots
11:34PM EDT – One thing more: Drive Hyperion
11:35PM EDT – Hyperion model 8 will begin transport in 2024
11:35PM EDT – And Hyperion 9 is in improvement for automobiles transport beginning in 2026
11:35PM EDT – Processes twice as a lot sensor information as Hyperion 8
11:36PM EDT – And NVIDIA has signed up further supplier-partners for Hyperion
11:36PM EDT – And now on to gaming with Jeff Fisher
11:37PM EDT – “Taiwan is the birthplace of the PC ecosystem”
11:37PM EDT – Recapping NVIDIA’s present stack of {hardware}, together with GeForce RTX sequence video playing cards, Max-Q laptops, and G-Sync screens
11:38PM EDT – PC sport {hardware} is anticipated to be a $67B market this 12 months, with double digit development over coming years
11:38PM EDT – Jeff is thanking NVIDIA’s companions for his or her merchandise
11:39PM EDT – “[Ampere] is the world’s quickest GPU”
11:39PM EDT – Recapping Ampere household options equivalent to RT cores, DLSS, and NVENC video encoder
11:40PM EDT – Now rolling a video on NVIDIA’s driver improvement course of
11:41PM EDT – This seems to be the identical (or comparable) Game Ready improvement video they launched a few weeks again
11:41PM EDT – “Our mission with drivers is to be invisible”
11:42PM EDT – There at the moment are over 250 RTX video games and functions
11:42PM EDT – Hitman 3 is getting an replace with ray tracing and DLSS
11:43PM EDT – Also asserting that F1 22, which is launching in July, may also have RT and DLSS
11:44PM EDT – Now on to NVIDIA Reflex
11:44PM EDT – “Low system latency helps all players”
11:44PM EDT – NVIDIA lately performed a research on how system latency impacts purpose accuracy
11:44PM EDT – Kovaak’s system latency problem
11:45PM EDT – Running one other promo video, this time on Reflex
11:45PM EDT – Talking about decreasing system latency with Reflex
11:46PM EDT – Reflex is even a plugin in Unreal Engine as of late
11:46PM EDT – Currently as much as 38 video games and 22 screens
11:47PM EDT – Introducing a brand new Reflex-capable show
11:47PM EDT – A brand new 500Hz 1080p 24-inch Asus monitor
11:47PM EDT – TN panel
11:47PM EDT – Includes the Reflex Analyzer
11:48PM EDT – Now on to GeForce laptops with Max-Q
11:48PM EDT – Recapping the Max-Q characteristic set
11:49PM EDT – New gaming laptops at Computex from MSI, Asus, Gigabyte, and others
11:49PM EDT – Now on to NVIDIA Studio
11:49PM EDT – NVIDIA’s creation-focused platform
11:50PM EDT – Now over 200 Studio accelerated functions
11:51PM EDT – And how Omniverse may also assist sport builders and content material creators
11:51PM EDT – Seen a 10x improve in Omniverse downloads
11:51PM EDT – Omniverse is being up to date with new options as effectively, equivalent to Omniverse Cloud and Audio2Face
11:53PM EDT – Now recapping the gaming section of this keynote and the state of NVIDIA’s gaming ecosystem
11:53PM EDT – And Jeff thanks NVIDIA’s companions one final time
11:54PM EDT – And by signing off with Toy Jensen, that is a wrap. Thanks for becoming a member of us.
https://www.anandtech.com/present/17404/computex-2022-nvidia-keynote-live-blog-11pm-et0300-utc
