Automotive parts take longer to qualify than parts for phones/tablets. A part for a car has to be around for 10 years unlike a phone part that gets a new model the next year.
Time to market probably. In a car they don't care about a few 100 mW saved by an A72, and they've got the Denver cores and GPUs for heavy number crunching.
I wish. But Nvidia hasn't actually followed up any of the Shield lineup. I wonder if there just isn't a market for it. The Shield TV is a fantastic device. It would be disappointing if sales didn't demand a v2 next year. Same goes for the Shield Portable and Shield Tablet.
There is a rumour that the new Nintendo NX console is using a Tegra SoC, and that this may be the cause of Nvidia not 'competing' with them with another Shield TV or Portable.
That's one weird chip CPU-wise at least. Nvidia had better not messed up how the CPUs work, as that could cause real-world crashes. big.Little was troublesome enough driver-wise.
That said, I'm glad they adopted Imagination's OmniShield-like (and Qubes OS-like) virtualization to keep things isolated, although I'd still feel safer if the car makers just used different systems altogether.
One of them is already strongly rumored to have a Pascal Tegra, but not from Sony or Microsoft who are now tied into GCN compatible architectures for a while (well, at least Microsoft with their whole beyond generations thing).
The Nintendo NX is rumored to be a Pascal Tegra handheld in a tablet form factor that docks to a TV.
A57 cores probably was already validated for automotive, and fast enough that the extra cost to validate the A72 was not worth, either to ARM, Nvidia or any other licensees. Denver cores make no sense to me. Seems like they're just included to try to recover some R&D by justifying a higher price? GPU, I am semi-dissapointed in. I expected a higher clockspeed, 1.7-1.8Ghz. But that's a non-issue.
Seems to look okay for the NX. Developers doing ports should be able to just downgrade from 1080/900p on PS4/XBO to 720p on Nintendo. And developers that want to push performance to the max can play with putting "unimportant" pieces through FP16. Boosting the performance.
Although, I may still be to optimistic in my targets. More realistically probably they move to 720p plus a settings downgrade.
Denver is good at straight line tests with lots of repetitive loops, where it chokes up is spaghetti code where things get unpredictable and it's binary translator can't keep it fed at full width. I can only assume, automotive would rely on a lot of repetitive processing.
It also makes sense to leverage the two core types together for the strengths of each, anything that gets more messy can go to the Cortex, while anything predictable can enjoy higher throughput.
Hopefully the connective fabric is good enough at picking and choosing.
Automotive is in large part detecting cars around you. For self driving. The last thing I would want is a core which is unpredictible to be handing ANY information of that sort.
A little disappointing to see EMMC being utilized. This is such a mish-mash, high speed DDR4 with ECC next to an A57. Pascal next to EMMC. The validation for automotive must be the reason for such an odd assembly, but I really would hope they would put something together that is far more current for any future tablet/consoles. Denver + A72 with DDR4, Volta, and UFS would be sweet.
AVB over ethernet sounds very interesting. I could see the use of a set of rear/side cameras, and maybe the rear speakers connected to a AD/DA hub in the trunk. A single ethernet cable gets run from the hub to the the front of the vehicle (maybe two for redundancy). Should cut down on complexity and weight of all the wires in a harness. Its low latency and already technology that exists today.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
29 Comments
Back to Article
Eden-K121D - Thursday, August 25, 2016 - link
Interestingp1esk - Thursday, August 25, 2016 - link
When?lolipopman - Wednesday, October 12, 2016 - link
What?mmrezaie - Thursday, August 25, 2016 - link
Why did they use A57 and not the a72!? There must be a reason since A72 is faster and more power efficient. It is indeed interesting.extide - Thursday, August 25, 2016 - link
Probably the A72 design came out too late for them to put it into this chip.syxbit - Thursday, August 25, 2016 - link
Believable, but not a good excuse to be shipping A57 in late 2016.Amazon actually shipped their Fire TV 4k with A72 last year!!!
mmonnin03 - Thursday, August 25, 2016 - link
Automotive parts take longer to qualify than parts for phones/tablets. A part for a car has to be around for 10 years unlike a phone part that gets a new model the next year.MrSpadge - Thursday, August 25, 2016 - link
Time to market probably. In a car they don't care about a few 100 mW saved by an A72, and they've got the Denver cores and GPUs for heavy number crunching.DCstewieG - Thursday, August 25, 2016 - link
With the focus on automotive, would this still be a candidate for a Shield TV 2?syxbit - Thursday, August 25, 2016 - link
I wish.But Nvidia hasn't actually followed up any of the Shield lineup.
I wonder if there just isn't a market for it. The Shield TV is a fantastic device. It would be disappointing if sales didn't demand a v2 next year. Same goes for the Shield Portable and Shield Tablet.
There is a rumour that the new Nintendo NX console is using a Tegra SoC, and that this may be the cause of Nvidia not 'competing' with them with another Shield TV or Portable.
klmccaughey - Thursday, August 25, 2016 - link
What is this machine for? I can't figure out what it would be used for.MrSpadge - Thursday, August 25, 2016 - link
Last slide: "Autopilot to self driving car".. and lesser tasks, if you wish.tipoo - Thursday, August 25, 2016 - link
Nintendo NX, I hope, for the Tegra itself. The PX Drive system with two dGPUs is for cars.TheinsanegamerN - Thursday, August 25, 2016 - link
And sadly there will be no shield tablet with parker in it.tipoo - Thursday, August 25, 2016 - link
Probably stipulated to differentiate the NX, which seemed very Shield-eyKrysto - Thursday, August 25, 2016 - link
That's one weird chip CPU-wise at least. Nvidia had better not messed up how the CPUs work, as that could cause real-world crashes. big.Little was troublesome enough driver-wise.That said, I'm glad they adopted Imagination's OmniShield-like (and Qubes OS-like) virtualization to keep things isolated, although I'd still feel safer if the car makers just used different systems altogether.
PVG - Thursday, August 25, 2016 - link
Am I the only one that sees something derived from Tegra, with a few ARM cores and a beefier GPU powering the future generations of consoles?tipoo - Thursday, August 25, 2016 - link
One of them is already strongly rumored to have a Pascal Tegra, but not from Sony or Microsoft who are now tied into GCN compatible architectures for a while (well, at least Microsoft with their whole beyond generations thing).The Nintendo NX is rumored to be a Pascal Tegra handheld in a tablet form factor that docks to a TV.
testbug00 - Thursday, August 25, 2016 - link
A57 cores probably was already validated for automotive, and fast enough that the extra cost to validate the A72 was not worth, either to ARM, Nvidia or any other licensees.Denver cores make no sense to me. Seems like they're just included to try to recover some R&D by justifying a higher price?
GPU, I am semi-dissapointed in. I expected a higher clockspeed, 1.7-1.8Ghz. But that's a non-issue.
Seems to look okay for the NX. Developers doing ports should be able to just downgrade from 1080/900p on PS4/XBO to 720p on Nintendo. And developers that want to push performance to the max can play with putting "unimportant" pieces through FP16. Boosting the performance.
Although, I may still be to optimistic in my targets. More realistically probably they move to 720p plus a settings downgrade.
tipoo - Thursday, August 25, 2016 - link
Denver is good at straight line tests with lots of repetitive loops, where it chokes up is spaghetti code where things get unpredictable and it's binary translator can't keep it fed at full width. I can only assume, automotive would rely on a lot of repetitive processing.It also makes sense to leverage the two core types together for the strengths of each, anything that gets more messy can go to the Cortex, while anything predictable can enjoy higher throughput.
Hopefully the connective fabric is good enough at picking and choosing.
testbug00 - Friday, August 26, 2016 - link
Automotive is in large part detecting cars around you. For self driving. The last thing I would want is a core which is unpredictible to be handing ANY information of that sort.fanofanand - Thursday, August 25, 2016 - link
A little disappointing to see EMMC being utilized. This is such a mish-mash, high speed DDR4 with ECC next to an A57. Pascal next to EMMC. The validation for automotive must be the reason for such an odd assembly, but I really would hope they would put something together that is far more current for any future tablet/consoles. Denver + A72 with DDR4, Volta, and UFS would be sweet.psychobriggsy - Friday, August 26, 2016 - link
To be fair, it does have SATA as well, and likely some spare PCIe lanes that could be used for an SSD if that's desirable.jwcalla - Thursday, August 25, 2016 - link
"At the same time, Parker's big use (at least within NVIDIA)..."Mmmm hmmm...
Joshua knows something.
HigherState - Friday, August 26, 2016 - link
AVB over ethernet sounds very interesting. I could see the use of a set of rear/side cameras, and maybe the rear speakers connected to a AD/DA hub in the trunk. A single ethernet cable gets run from the hub to the the front of the vehicle (maybe two for redundancy). Should cut down on complexity and weight of all the wires in a harness. Its low latency and already technology that exists today.Jumangi - Friday, August 26, 2016 - link
Great job Nvidia comparing a SoC that goes in a car vs a tiny smartphone case. Totally not Apples to Oranges.daku123 - Friday, August 26, 2016 - link
>> NVIDIA quotes an FP16 rate of 1.5 TFLOPs for Parker, which implies a GPU clockspeed of around 1.5GHz.TFLOPs depends largely on number of cores and how the SOC is designed. Statement above is not particularly geeks friendly :)
tipoo - Friday, August 26, 2016 - link
But we know the architecture and the core count...It's a perfectly reasonable statement.Shaders * 2 floating point operations per cycle * billion cycles per second = theoretical Gflops.
It does not denote performance, it's exactly what it says, theoretical peak Gflops.
darkich - Friday, August 26, 2016 - link
Nintendo NX