facebook rss twitter

Nvidia Orin SoC melds next gen GPU and Arm Hercules CPU cores

by Mark Tyson on 18 December 2019, 10:11


Quick Link: HEXUS.net/qaegxl

Add to My Vault: x

Nvidia has revealed a powerful new SoC called Orin. The fruit of four years of research, development and investment, Orin packs Nvidia's next generation GPU architecture alongside Arm Hercules CPU cores in a 17 billion transistor SoC. The result sounds impressive; an Orin SoC is approx 7x faster that the Nvidia Xavier SoC, and can deliver 200TOPS.

The initial plan for Orin is that it will form an essential part of Nvidia's Drive AGX Orin, a highly advanced software-defined platform for autonomous vehicles and robots. Orin is highly suited to the tasks that are central to these automatons as it can adeptly multitask traditional apps as well as deep neural networks.

In Nvidia's development tests Drive AGX Orin could scale from a Level 2 to full self-driving Level 5 vehicle. Like Xavier it is programmable through open CUDA and TensorRT APIs and libraries.

CEO Jensen Huang said that Nvidia Drive AGX Orin addresses what he thinks might be "society's greatest computing challenge," - creating safe autonomous vehicles. Orin will help vehicle makers meet this incredibly complex challenge due to it being a scalable, programmable, software-defined AI platform, asserted Huang. Importantly the Orin SoC based system can meet systematic safety standards such as ISO 26262 ASIL-D.

Nvidia reckons vehicles packing its Drive AGX Orin system will deliver a significant step forward in autonomous driving tech for its customers. Expect vehicles featuring Nvidia Drive AGX Orin to start to become available from 2022.

Before we get too excited by Nvidia's PR, earlier this year analyst James Wang from ARKInvest noted that, even with Orin in production, Nvidia is still years behind Tesla in L5 driving computers. Today's news and official Nvidia Drive AGX Orin unveiling don't appear to change that.

HEXUS Forums :: 5 Comments

Login with Forum Account

Don't have an account? Register today!
Powering the New Switch XL Pro coming Winter 2020.
UNFORTUNATELY most of these other companies (Amazon,Google,Tesla, China* etc) are coming up with their own custom ASIC that outperforms these Nvidia products. Investors better watch out!
UNFORTUNATELY most of these other companies (Amazon,Google,Tesla, China* etc) are coming up with their own custom ASIC that outperforms these Nvidia products. Investors better watch out!

Far outstripped Nvidias, for sure!
Good luck beating NV tesla. The difference is the price. Tesla sells how many cars a year? LOL@James guy…Must own Tesla stock or something.

How many watts will it be at 5nm in 2021? Even currently, as OP in tweet noted, it's about 5.3% of your battery life even now. Dev watt/perf is not what it will be at release. When it's a chip for MANY more people (devices even) than just telsa cars, it will be cheap as chips to make, which is the whole point vs. something like a 100K tesla (and they shipped only about 100K in 2019…ROFLMAO). If you want it in your 25K Ford, ask Nvidia for it, not tesla. This guy acts as though there is some magic keeping NV from competing. Wake me when tesla sells L5 in a $20K-30k car. I'm guessing NV will beat them to it, probably by years…LOL.

Saying NV is 4yrs behind on something that isn't even AVAILABLE FROM ANYONE at retail, is really funny. Is it april 1st or Christmas? Just for giggles here's a reality check of where you are now:
I don't think it will take that long (die shrinks, new tech, etc is massively speeding things up), but the point is, you're can't win until it's out IMHO. ;) After that we'll see, but until then, it's anybody's game still and I'd bet on the leading gpu company (20yrs?) before a car company for anything like this. This requires software and hardware that car guys just aren't used to making. It's far worse than something like saying you think Intel will win the gpu wars, when they've already failed MULTIPLE times vs. NV/AMD. After years of dealing with them and their drivers, I just laugh at talk like this. Tesla has how much experience in software/hardware? I've got a $20 Intel fails again and bails on desktop in under 5yrs, ends up as a server chip or something.

Raja, who failed at AMD, will now make magic at Intel. It's not the money, it's the engineers. If he/she/they suck, so does your product. Worse, now he's working on a failing process currently and no end in sight IMHO. At this point I have to SEE 10nm in a 200mm^2+ chip or I don't believe it yet. Heck I might be tempted to say make a 16c cpu (or larger gpu) or bust on it as the other guy has huge 7nm chips out NOW. AMD's 8c chiplet in threadrippers are ~80mm^2. AMD 5700 XT chip is ~250mm^2. So basically Intel should be able to fit a 16c in that at 10nm right? Roughly speaking, but you should get the point. NV is going HUGE next, and is pointing devs at that, mostly I think due to a possible Intel competitor. AMD can't afford smack down perf, but Intel could, so better safe than sorry right? So NV says to devs, expect massive perf next gen. So how big does an Intel gpu have to be to get in this race for real? Pretty big even at 7nm.

But never mind, Tesla has this gpu/imaging/AI stuff licked. Yeah…Whatever. How many L5 cars do they have on the road?
Good luck beating NV tesla.

Tesla aren't an NV competitor, they don't seem to be making any effort at all on selling their silicon to others, they just fab for their own consumption.

That's the worry for Nvidia, if it's the wrong product then no-one want it at any price. Tesla don't waste any silicon on GPU functionality, it is purely a TPU, and are clearly finding that by making the chip themselves they save money for a faster product even with their current sales levels.

This could be the next Tegra, which at one point Nvidia wanted to be in every phone and tablet.