NVIDIA Announces Tegra 'Parker'

[HKZeroFive]

Quote

For Hot Chips 28, NVIDIA has the first details on the new Tegra chip, codenamed Parker, in their luggage. This is based on two Denver-2 ARM cores plus four Cortex-A57 cores, coupled with a 256 shader-strong graphics solution from the Pascal-generation. With a focus on the automotive market, Nvidia has introduced the new Tegra solution more closely. This is part of the overall package of the Drive PX 2 revealed at CES 2016 and combined it with the latest technologies from Nvidia in a modern 16-nm FinFET process from TSMC.

 

The heart of the new Tegra solution are the revised Denver-2-cores that will compete generously equipped with large L1I caches of 128 kb and 64 kb L1D cache per core and 2MB L2 cache for two cores. They are also supported by four classic Cortex-A57 and should as a CPU Complex form. No less important is the 256 shader strong graphics solution from the current Pascal generation. This is supplemented with a doubling of LPDDR4 memory including ECC support and a bandwidth of 50 gigabytes per second now instead of the previous 25 GB / s.

 

Quote

NVIDIA now marketed Parker along with the Drive PX2 as  "the world's first supercomputer AI for self-propelled cars ". The cores are aggregated around the kit, as in marketing typically called as a first: 12 CPU cores and Pascal graphics solutions for a total of 8 TFLOPS performance. However, Nvidia surprised at the presentation and set a comprehensive block diagram of the entire system before.

 

No statements for other use purposes.

 

NVIDIA is indeed going hard with marketing the new Tegra chip with Drive PX 2, aligning their intentions to further expand into the automotive industry. Whether or not NVIDIA reworks it for other uses remains ambiguous at the moment but I can say with certainty that this will not be the chip that will power the upcoming the Nintendo NX. Don't get your hopes up just yet.

 

Source: https://www.computerbase.de/2016-08/nvidia-tegra-parker-denver-2-arm-pascal-16-nm/

[Djole123]

Oh nice!

[Duduix]

Sweet

[TheRandomness]

Parker and drive. 

 

Nvidia. 

I think you should go die. 

[Trixanity]

A shame about those A57 cores. A72 is a much better design. I know I know "long development time" but it was after all announced like 18 months ago with the successor A73 right around the corner.

[AlTech]

OH NIce! Now what are the GPU specs?

1 hour ago, Trixanity said:

A shame about those A57 cores. A72 is a much better design. I know I know "long development time" but it was after all announced like 18 months ago with the successor A73 right around the corner.

It uses 2 Denver and 2 A57.

[Trixanity]

7 minutes ago, AluminiumTech said:

OH NIce! Now what are the GPU specs?

It uses 2 Denver and 2 A57.

Actually it uses 2 Denver cores and 4 A47 cores but that's beside the point. The A57 was a pretty bad product. I guess it was all they (Nvidia) had to work with at the time of development.

[Castdeath97]

Wait didn't Nvidia state that they will stop making SoC? I'm I missing something?

[AlTech]

10 minutes ago, Trixanity said:

Actually it uses 2 Denver cores and 4 A47 cores but that's beside the point. The A57 was a pretty bad product. I guess it was all they (Nvidia) had to work with at the time of development.

ohhhhhhhhhh. I thought it was 2 A57.

 

10 minutes ago, Castdeath97 said:

Wait didn't Nvidia state that they will stop making SoC? I'm I missing something?

Nope. They just said they'll stop making tablets. Not SOCs

[Nineshadow]

Is it a...Parker Square?

I don't know how many of you know the joke.

[ScratchCat]

NVIDIA - Now water cooling ARM chips! 

[LAwLz]

I wonder if they included those four A57 cores because Denver is bad, or if they did it for some other purpose.

[Trixanity]

1 minute ago, LAwLz said:

I wonder if they included those four A57 cores because Denver is bad, or if they did it for some other purpose.

If I recall, Denver had excellent performance in some work loads and awful performance in others, so it was very inconsistent. I guess they're trying to account for that by splitting up the work to favor each core's forte. I'm sure our local expert robot overlords will jump at the chance to tell you all about how the Denver core works.

[patrickjp93]

5 hours ago, Castdeath97 said:

Wait didn't Nvidia state that they will stop making SoC? I'm I missing something?

No. Seriously where do you people get this crap info?

[Unhelpful]

5 minutes ago, patrickjp93 said:

No. Seriously where do you people get this crap info?

He's thinking of tablets. Although there isn't anything stating they're done with tablets, just that nvidia file a dismissal letter with the FCC for the shield tablet 2 

[Castdeath97]

15 minutes ago, patrickjp93 said:

No. Seriously where do you people get this crap info?

Bingo found it:

 

[Trixanity]

7 minutes ago, Castdeath97 said:

Bingo found it:

 

Hmm, seems like it has been misinterpreted a bit. They have quit smartphone SoCs and probably tablet as well (including actual devices). Tegra will live on as automotive chips only; that's a very different market.

[Castdeath97]

3 minutes ago, Trixanity said:

Hmm, seems like it has been misinterpreted a bit. They have quit smartphone SoCs and probably tablet as well (including actual devices). Tegra will live on as automotive chips only; that's a very different market.

So no "mobile" SoCs, but we are getting car and possibly console SoCs if I understood correctly.

[Trixanity]

1 minute ago, Castdeath97 said:

So no "mobile" SoCs, but we are getting car and possibly console SoCs if I understood correctly.

I'd take the Nintendo rumor with a grain of salt. I mean it's not impossible - not at all - but no solid evidence has been shown, so let's see how that plays out.

 

And yeah, I suppose so on the "no mobile SoCs" although I'm not sure if the quote meant smartphones only (very different TDP and Nvidia couldn't make it work in the end) or if the tablet market along with their devices is abandoned as well. Besides the Shield Tablet, I think the Nexus was the only product to ship with an Nvidia chip, so that would explain why they would shift their focus if they abandon it. All depends on the sales of their Shield Tablet and if the FCC info is confirmation (probably is) or if they merely wanted to pull the product to put it back on the drawing board (unlikely).

[PRTI]

6 hours ago, AluminiumTech said:

ohhhhhhhhhh. I thought it was 2 A57.

 

Nope. They just said they'll stop making tablets. Not SOCs

Yea,that explains why i'm still alive  

[Exotath]

So is this unit water cooled?

[patrickjp93]

37 minutes ago, Castdeath97 said:

Bingo found it:

 

 

30 minutes ago, Trixanity said:

Hmm, seems like it has been misinterpreted a bit. They have quit smartphone SoCs and probably tablet as well (including actual devices). Tegra will live on as automotive chips only; that's a very different market.

Based o the language of the article, it seems to only be applying to phones and phablets. As long as the ARM tablet market is around, Nvidia would love to have a piece of it.

[LAwLz]

39 minutes ago, Trixanity said:

If I recall, Denver had excellent performance in some work loads and awful performance in others, so it was very inconsistent. I guess they're trying to account for that by splitting up the work to favor each core's forte. I'm sure our local expert robot overlords will jump at the chance to tell you all about how the Denver core works.

Yeah, but I don't get the reason for the hybrid design.

Denver was slow in two scenarios.

1) If the code morphing failed to create efficient code.

2) When the code morphing was completely skipped and it executed native ARM code.

 

 

I can see those scenarios happening a lot on phones and tablets, but in cars? You'd hope that the one developing the car OS which will control the AI want to spend a bit more time on optimizations than what a developer creating a 1 dollar app for Android would do. So scenario 1 should hopefully not happen at all (or at least not enough times for it to warrant this hybrid design), and couldn't you solve scenario 2 by "pre-morphing" the code (I just assume this is possible)?

 

If Denver is a lot better than A57 then shouldn't Nvidia have tried to use them for everything, and not spent a ton of space and silicon on the A57 cluster?

If Denver isn't that much better than A57 then why designing this complicated hybrid design to begin with?

 

 

Maybe they will be used for different purposes. Like maybe the Denver cores will be used for some AI thing, and the A57 cores will be used to run something like Android Auto or some more general car OS. Nvidia probably has some idea behind it but to me it seems weird.

[Trixanity]

2 minutes ago, LAwLz said:

snip

Well, if one set of cores runs Android Auto and the other runs AI/autonomous driving/auto pilot, that would be different work loads, so my comment isn't wrong, is it?  

 

And I imagine that your assumption is right on the money in that regard. However who knows what Nvidia is thinking? They might have gone bonkers when Tegra crashed and burned on phones after all their investments in the market. Seems like they're trying to find a place for Tegra and this time it will be in cars and this time they will make sure it succeeds.

[AlTech]

3 hours ago, LAwLz said:

I wonder if they included those four A57 cores because Denver is bad, or if they did it for some other purpose.

Denver performance is usually good but I think they were worried about power consumption.