NVIDIA Tegra Architecture

ROFL since it's nothing more than a reworded CPU whitepaper (source NVIDIA).

David Kanter gave some input on this new article
I worked with Linley a bit on this article to make sure we got as much uarch details as possible - so thanks : )

The Denver design is quite interesting, and I hope they'll be a bit more open going forward.

David

http://www.realworldtech.com/forum/?threadid=145674&curpostid=146346

I'm not sure why you think it is funny or has nothing added as the article clearly has much more details on the inner workings of Denver.
 
Yes it does. Instead of "I brushed though it quickly" try reading it in whole as there is much more detail.

It still doesn't state anything exciting I haven't read before.

David Kanter gave some input on this new article

And?

I'm not sure why you think it is funny or has nothing added as the article clearly has much more details on the inner workings of Denver.

Excuse me if I would had prefered an analysis based on real time measurings of the CPU instead of any theoretical data that they got from an as subjective source as NVIDIA. I'd like too to see a full review from a site like Anandtech if you don't mind.
 
Last edited:
The article does lose a bit of credibility when it claims that apple A7 has a "SGX5 MP4" GPU. The A7 was released over a year ago and this article is 4 months old.
 
The fact that it was under subscription does not change that there are several mistakes in it and that there isn't any data in it about Denver from an independent source. Yes it highlights details of the architecture but Nexus9 is a real product and we should have a few serious reviews by now about it. It's almost like no one cares anymore, NV included.
 
This article originally required a PAID subscription. It is now released as a FREE article. It does have new information on the inner workings of Denver that were not available in any public Denver article before including the Nvidia Denver white paper.

How old is the article then? Did they change the date after the subscription requirement was lifted?
All I see is one article release date - August 18 2014 - where the author should have the necessary information to put the correct name in the A7's GPU.
As such, my comment still stands - the article does lose credibility for showing the wrong GPU in A7.
 
It's almost like no one cares anymore, NV included.

Touché! That is really really awkward isn't it? Being a marketing force to be reckoned with, its very strange that nVIDIA did not continue to rumble the war drums... Plus, one would expect them to continue waving "design win numbers" like on past Tegra launches. Well, they are probably busy sorting their legal action against Samsung :p
 
Well it was Erinyes that first mentioned that Erista has according to his info a 4*A57+4*A53 big.LITTLE config. No one outside of NV knows yet if there's also going to be an Erista variant with Denver cores, but I find the standard ARM cores at least peculiar and that's where I wonder if that defeaning silence around Denver lately is just a coincidence.

Drums will roll for the dual cluster GPU in Erista and it is truly the real highlight of the SoC, as is GK20A in K1.
 
Last edited:
CES 2015 is a little over three weeks away.

What should we expect?

20nm?
Maxwell GPU?
Denver CPU?

Check - it is 20nm
2 SMM Maxwell GPU
4*A57 + 4*A53 big.LITTLE. No idea about the Denver variant (or
Well it was Erinyes that first mentioned that Erista has according to his info a 4*A57+4*A53 big.LITTLE config. No one outside of NV knows yet if there's also going to be an Erista variant with Denver cores, but I find the standard ARM cores at least peculiar and that's where I wonder if that defeaning silence around Denver lately is just a coincidence.

Drums will roll for the dual cluster GPU in Erista and it is truly the real highlight of the SoC, as is GK20A in K1.

Why is it peculiar that it has standard ARM cores? NV have been using standard ARM cores since the first Tegra. The only surprise is big.LITTLE as till now they were committed to their Penta core config. I think the silence is due to the fact that there is just one product using it at the moment. I think Denver just ended up being very late..much later than they hoped. As it turns out, Samsung beat them to the 64 bit game with a 20nm Exynos with A57 cores.

True..IMHO the GPU will be the major differentiator as the CPU wont be any different to Qualcomm/Samsung.
 
Yes they have been using standard ARM cores since the beginning, but they haven't had so far a custom core before Denver either. Having standard 64bit ARM cores in parallel with custom 64bit Denver cores is peculiar. If there is only one Erista variant in the end, it's a completely different chapter then.
 
Why is it peculiar that it has standard ARM cores?

Erm... because nVidia spent millions developing their own ARMv8 core, which is already shipping in a final product.
One would assume they'd at least leverage that investment by not purchasing off-the-shelf cores from ARM.

This could mean that nVidia just isn't confident enough on the Denver's standby consumption.
 
OT: but is anyone even measuring standby consumption? I'm just asking since you recently bought an Intel SoC tablet recently from what I recall. The one I'm playing with at the moment has a 4*A15+4*A7 config for which I'm estimating a rough =/>22hours standy time (full loaded battery doing nothing until it drains).
 
This could mean that nVidia just isn't confident enough on the Denver's standby consumption.
That's to put it lightly. Denver has basically no reason to exist in its current shape. Going back to vanilla cores is a wise decision.

OT: but is anyone even measuring standby consumption?
If by standby you mean not touching the device then it's quite good. As soon as anything happens then it goes downhill.
 
Yes that's what I meant. I don't know if Tottentranz meant the same with "standby consumption".

By the way do you have any idea when we can count on a full N9 review from Anandtech?
 
OT: but is anyone even measuring standby consumption?
Anandtech does (or at least they did with Snapdragon 600):
oKHKwyH.png



I'm just asking since you recently bought an Intel SoC tablet recently from what I recall. The one I'm playing with at the moment has a 4*A15+4*A7 config for which I'm estimating a rough =/>22hours standy time (full loaded battery doing nothing until it drains).
It arrived in my country almost a month ago, but unfortunately it's still stuck in the customs :(
22 hours standby is terrible! Maybe you have one of those trojans/malwares that are constantly using an internet connection to send advertisements into your notification area?
I actually ordered another windows tablet to give to my parents this Christmas, this time a 9.7" model with the ipad air's 2046*1536 screen and a BayTrail.
Dirt cheap, 179€ shipping included and it was actually bought directly from a portuguese retailer so no customs nightmare this time.


That's to put it lightly. Denver has basically no reason to exist in its current shape. Going back to vanilla cores is a wise decision.
If this is a preview of what we'll see in the Nexus 9 review, then Auch...
 
The one I'm playing with at the moment has a 4*A15+4*A7 config for which I'm estimating a rough =/>22hours standy time (full loaded battery doing nothing until it drains).

That is awfully bad! My Asus Transformer (Tegra 3) with Cyanogen Kitkat can stay on standby for days at a time. I do not use it that much at the moment (its basically my travel companion plus Skype machine (do not have Cam on desktop)) and sometimes I pick it up expecting the juice to have run out, but nope, still hanging there. Its true it has 2 batteries (one on keyboard dock) but I would estimate it at nearly 4 days standby time. For a 40nm quad core chip, I think its brilliant. The companion core (if cyanogen actually makes use of it, i'm not sure) must work great. I really wonder if Big Little is a better strategy...
 
It's getting way OT, but I need to think of a reliable method to estimate standby time. Having days of differences in so far results means that any so far layman's method just sucks ;)
 
Yes they have been using standard ARM cores since the beginning, but they haven't had so far a custom core before Denver either. Having standard 64bit ARM cores in parallel with custom 64bit Denver cores is peculiar. If there is only one Erista variant in the end, it's a completely different chapter then.

Erm... because nVidia spent millions developing their own ARMv8 core, which is already shipping in a final product.
One would assume they'd at least leverage that investment by not purchasing off-the-shelf cores from ARM.

This could mean that nVidia just isn't confident enough on the Denver's standby consumption.
I'm going to quote an old post of mine where I had the same discussion with Ailuros a while back. See below
Ailuros said:
In the meantime what still escapes that picture is the necessity of the big.LITTLE A5x config in Erista. If Denver is so efficient then why bother with standard ARM cores in the first place?

The Denver Soc may be on a different design cycle(~6 months behind) compared to the standard ARM variant. Could be a case of hedging their bets in case of any delays to Denver (the timescales involved here are pretty large..the design process for these chips was started sometime in 2013). And possibly big.LITTLE could be more power efficient than having only Denver cores..which matters a lot more in the mobile/tablet segment (see above on my views of where Denver is targeted) And the marketing aspect of a "octa core" is very real. Even Qualcomm, who had were very opposed to it, are now offering a Octa core A53 SoC for this very reason (I'd rather have two higher clocked A57s or a 2 A57 + 2 A53 config for this segment, but I doubt we will ever see such an SoC ever made).

As we've already seen with Tegra K1, the denver variant came out ~6 months after the A15 variant. Both Samsung and Qualcomm will be shipping 20nm 64 bit "Octa" cores in Q1'15. If Nvidia wants to compete at all..it will have to match that schedule, which they probably could not do with Denver cores. And if they are indeed working on a Denver version, I'm sure it will be an evolution of the cores which were in Tegra K1 so the development time could be a lot more.

That's to put it lightly. Denver has basically no reason to exist in its current shape. Going back to vanilla cores is a wise decision.

If by standby you mean not touching the device then it's quite good. As soon as anything happens then it goes downhill.

That is surprising news indeed..all the reviews so far have indicated that it isnt all that terrible. However, as others have stated..I'm waiting for the Anandtech review as well to see what you have to say about it.
 
Last edited:
Well the puish back of the Nexus 9 review by anandtech is really suspicious to me now, if it makes it after the buying season and the result of their in depth testing underwhelms I will sure know what to think... The buying season is almost over already...
That article about the HP Stream 11 which avoid any reference to competition like the Asus x205 or Acer products (and others) doesn't help my mixed feeling either.
 
Back
Top