RCHeliguy
Premium
RCHeliguy
Premium
The AI chips they are talking about are serious hardware. They are definitely working on multi-chip technology. They are teasing Grace with 1Tb/s communication between chips.
The term industrial digital twin was coined where an entire environment is modeled photo realistically that can be viewed as a digital representation of a real environment in real time.
They showed a real time spoken language conversion where the speaker is heard speaking multiple foreign languages simultaneously by different people with digital lips matched to the foreign language so it looks more natural to the listener.
Interesting stuff.
Then they fit some of these pieces together using digital twin of the world around their self driving car concept. They claim that Mercedes will be the first to sell a car in 2024 with 100% self driving capability based on their Hyperion 8 auto driving system.
Doesn't look like any gaming information.
However they are now using talking interactive AI Avatar assistants to help on conference calls.
I did see this from over a week ago....
I wonder how accurate this is.
The term industrial digital twin was coined where an entire environment is modeled photo realistically that can be viewed as a digital representation of a real environment in real time.
They showed a real time spoken language conversion where the speaker is heard speaking multiple foreign languages simultaneously by different people with digital lips matched to the foreign language so it looks more natural to the listener.
Interesting stuff.
Then they fit some of these pieces together using digital twin of the world around their self driving car concept. They claim that Mercedes will be the first to sell a car in 2024 with 100% self driving capability based on their Hyperion 8 auto driving system.
Doesn't look like any gaming information.
However they are now using talking interactive AI Avatar assistants to help on conference calls.
I did see this from over a week ago....
I wonder how accurate this is.
Last edited:
Beyond Nvidia' s Ada Lovelace (RTX 4000 GPU) technology: Grace Hooper
H100 is the first PCIe Gen-5 and High-Bandwidth Memory 3 (HBM3) GPU
"Twenty H100s can sustain the equivalent of the entire world's internet traffic"
Grace is a 144-CPU-core component consisting of two Arm-based processors
interconnected within the same package
via Nvidia’s new NVLink chip-to-chip interconnect technology,
and it will support 1TB of LPDDR5x memory.
https://www.theregister.com/2022/03/22/nvidia_jensen_gtc/
An evolutionary step over LPDDR5,
LPDDR5x increases possible data rates from 6400Mbps to 8533Mbps.
https://www.micron.com/about/blog/2022/february/lpddr5x-memory-performance-that-pushes-the-limits
H100 is the first PCIe Gen-5 and High-Bandwidth Memory 3 (HBM3) GPU
"Twenty H100s can sustain the equivalent of the entire world's internet traffic"
Grace is a 144-CPU-core component consisting of two Arm-based processors
interconnected within the same package
via Nvidia’s new NVLink chip-to-chip interconnect technology,
and it will support 1TB of LPDDR5x memory.
https://www.theregister.com/2022/03/22/nvidia_jensen_gtc/
An evolutionary step over LPDDR5,
LPDDR5x increases possible data rates from 6400Mbps to 8533Mbps.
https://www.micron.com/about/blog/2022/february/lpddr5x-memory-performance-that-pushes-the-limits
RCHeliguy
Premium
Remember when the 10 series came out?
There was both a drop in Watts needed and an increase in power.
The 20 series had a small bump in power required.
The 30 series had a good bump in power required.
The 40 series will have a jump in power required.
Since they are running out of room to shrink the dies ( for now ), larger and multi-chip configurations are where they are feeling forced to go which is just a way to effectively create larger dies.
There are some promising technologies that will allow them to shrink the die sizes a good bit further, but we are approaching the limits of where Si based chips can go linearly. After that multi-chip configurations will allow them to continue to increase performance but with a linear increase in power consumption.
The interesting thing is that most people do not need any more performance than they have. I was having no problem doing software development on a 2nd gen i7 machine for a decade. My updated system renders video faster, but that is about the only time I notice a difference. It's mostly gaming systems that are driving this.
Given how efficient most console gaming systems are in terms of power vs performance, I wonder if that is where gaming will go entirely at some point.
There was both a drop in Watts needed and an increase in power.
The 20 series had a small bump in power required.
The 30 series had a good bump in power required.
The 40 series will have a jump in power required.
Since they are running out of room to shrink the dies ( for now ), larger and multi-chip configurations are where they are feeling forced to go which is just a way to effectively create larger dies.
There are some promising technologies that will allow them to shrink the die sizes a good bit further, but we are approaching the limits of where Si based chips can go linearly. After that multi-chip configurations will allow them to continue to increase performance but with a linear increase in power consumption.
The interesting thing is that most people do not need any more performance than they have. I was having no problem doing software development on a 2nd gen i7 machine for a decade. My updated system renders video faster, but that is about the only time I notice a difference. It's mostly gaming systems that are driving this.
Given how efficient most console gaming systems are in terms of power vs performance, I wonder if that is where gaming will go entirely at some point.
Last edited:
Indeed, 40 series predictions reflect that, with prices increasing much faster than performance.
Bingo, the AMD+Intel+TSMC consortium may be in reaction to Nvidia's Grace.
I was peripherally involved in something comparable, when Fore delivered promising ATM LAN tech,
which was killed by IBM and others in the ATM Forum promoting an incompatible alternative,
then commodity ethernet switches delivered the coup de grâce.
One can hope that open source progress will facilitate Linux PC gaming.
RCHeliguy
Premium
The end of this shows the new NVidia AI GPU GH100 is using TMC's new 4nm process, but still went from the GA100 with TDP of 400W to the GH100 with TDP of 700W.
Based on what I saw of the presentation the NVidia CPU while impressive in it's own right has one thing that Intel and AMD don't have, an ultra high bandwidth direct to GPU connection that scales up to 8 GPU's. Obviously this is serious money and for commercial use only, but it could be a sign of things to come.
Based on what I saw of the presentation the NVidia CPU while impressive in it's own right has one thing that Intel and AMD don't have, an ultra high bandwidth direct to GPU connection that scales up to 8 GPU's. Obviously this is serious money and for commercial use only, but it could be a sign of things to come.
Steam alpha on select Chromebooks
Only Chromebooks with Intel XE graphics, an 11th-gen i5 or i7 processor, and at least 8GB of RAMhttps://www.theverge.com/2022/3/22/22990267/steam-alpha-chromeos-chromebook-supported-models
.. plus new external NVLink Switch for speeding multi-chip interconnections
.. plus Scalable Hierarchical Aggregation and Reduction Protocol reducing software overhead.
Nvidia's ability to optimize performance across so many areas – from the GPU, to the system, to the network, to the software – underlines why CEO Jensen Huang believes the Nvidia needs to be a "full-stack computing company" rather than just a provider of GPUs.
RCHeliguy
Premium
For gaming that would be disruptive AND $$$$$$$$$$$$
For AI, I get it.
By 2024, both AMD and Intel might announce for PCs
all-in-one enthusiast-grade CPU+GPU multi-chip modules
RCHeliguy
Premium
Great way to force people to buy all new hardware.
Imagine Chrome OS robustness with 3080 Ti level graphics and full SteamVR game support.
I like the sound of the last two bits but I wont be accepting google back in to my life again
Some of Chrome OS is proprietary, but much is open source Linux.
The real trick for open source VR sim is SteamVR or alternative interoperability
without having games' business models hacked.
How about a link?
Latest News
-
Need For Speed Unbound To Add Playable Motorbikes In Vol. 9Throughout the rich thirty-year history of Need For Speed, motorbikes have always evaded the...- Connor Minniss
- Updated:
- 3 min read
-
Revisiting Our Top 3 2024 FIA WEC Moments in Le Mans UltimateThe 2024 FIA World Endurance Championship may be over but its intense action rests heavy on the...- Angus Martin
- Updated:
- 6 min read
-
Community Question: Which Car Would You Like To See Brought To Assetto Corsa?Assetto Corsa is renowned for being the most versatile racing simulator on the market with...
- Connor Minniss
- Updated:
- 2 min read
-
"A Lifelong Dream": How OverTake Community Members Tackled The Mantorp 6 HoursSim racing can have a great sense of community. A handful of OverTake members took this one step...- Yannik Haustein
- Updated:
- 6 min read
-
F1 Manager 2024: Pit Lane Starts & More New Features In Latest UpdateAs the Formula One season starts to draw to a close, F1 Manager 2024 has been hit with a big...
- Connor Minniss
- Updated:
- 2 min read
-
Weekly Races & Twitch Broadcasts: Join The OverTake British GT4 League In ACC!Motorsport's real-world racing series are all winding down as 2024 enters the final stretch -...
- Yannik Haustein
- Updated:
- 2 min read
-
WATCH: Sensit! Aims To Simulate That Seat-Of-The-Pants FeelingOne thing sim racing cannot really replicate well is the sensory feedback drivers get from their...
- Yannik Haustein
- Updated:
- 1 min read