NVIDIA Tegra Architecture

[Deleted member 13524]

It's interesting that they specifically say Wolverine will have a Kepler GPU and not just a CUDA-enabled GeForce ULP.

 

[Ailuros]

It's interesting that they specifically say Wolverine will have a Kepler GPU and not just a CUDA-enabled GeForce ULP.

I'm not taking it literally, but that's just me.

 

[Ailuros]

No one else had 28nm products out when Tegra 3 was released so it's not like they they went against the industry.
And it's not like the first 28nm products were especially late vs expectations had for a good year beforehand. Expecting TSMC's 20nm to be ready for a mid to late 2014 release doesn't seem that unrealistic.

No it's not unrealistic but it does involve more risk.

You probably could make a good argument that nVidia would have been better off waiting for 28nm, but in that case nVidia needed something more suitable for phones out ASAP - Tegra 2 was a misstep in that direction since it couldn't power gate a single core and lack of NEON was quickly looking like a glaring mistake that needed prompt correction.

If I'd make a list with what I'd think they would had been better off some would prepare feathers to lynch me

The situation is a little different with Tegra 4. The die isn't nearly as small as Tegra 2's so they don't have loads of room to grow in. It also looks like the Cortex-A15s are going to use a lot of power. If they're going to substantially grow peak CPU power consumption with the next generation (as one would certainly expect) and if they're going to do it with Cortex-A15s or maybe Cortex-A57s (which is their only option if it's not using Denver yet) then they really need a lower power process. Right now they don't have the power budget to grow at all, not for any market they're remotely established in.

Why would they need more power for hypothetical A5x cores vs current A15 cores under 28HPM or whatever they might use in the future? The real burning question would be where that 50x times performance increase on a SoC level compared to Tegra2 is supposed to come from. For that - and you might excuse the sarcasm but it's not directed at you - not even 20nm by far is going to be enough.

 

[ninelven]

The real burning question would be where that 50x times performance increase on a SoC level compared to Tegra2 is supposed to come from.

A Kepler SMX at 600MHz is ~50x the Tegra 2 GPU, so I would imagine that is the source of the number.

Anyway, it seemed pretty clear from the presentation that:

Logan - 20nm
Parker - 16nm

 

[Ailuros]

A Kepler SMX at 600MHz is ~50x the Tegra 2 GPU, so I would imagine that is the source of the number.

That's not a ULP GeForce roadmap but obviously a Tegra roadmap. And yes I'm still with Arun's theory of a very "lose" Kepler derivate or realistically illustrated: a design based on Kepler but carefully adjusted for the SFF mobile market.

Anyway seemed pretty clear from the presentation that:

Logan - 20nm
Parker - 16nm

If they specifically mention it then fine.

 

[ninelven]

That's not a ULP GeForce roadmap but obviously a Tegra roadmap.

Eh, its a roadmap put together by marketing people. They are probably counting the fact that the cuda cores will be usable for compute purposes whereas the the GF ULP cores were not.

 

[Exophase]

No it's not unrealistic but it does involve more risk.

It is riskier but I don't think nVidia is going to be in a position to rest on their laurels and take it easy.

If I'd make a list with what I'd think they would had been better off some would prepare feathers to lynch me

Fair enough...

Why would they need more power for hypothetical A5x cores vs current A15 cores under 28HPM or whatever they might use in the future? The real burning question would be where that 50x times performance increase on a SoC level compared to Tegra2 is supposed to come from. For that - and you might excuse the sarcasm but it's not directed at you - not even 20nm by far is going to be enough.

We don't know a lot about Cortex-A57 except that it's supposed to be ~30% faster, with no real description how or vs what exactly. They've shown a pipeline diagram that's the same as A15's, but that might just be a placeholder. It's not that A57 needs more power, it's that nVidia needs some substantial mix of more power and better perf/W to justify a new SoC release at all. A better GPU alone won't cut it. A57 would bring 64-bit which could be necessary if there's substantial pressure for it, but I'm not sure this will be the case even in 2014 (A15's PAE-style extensions may provide a tiny amount of breathing room w/4GB, probably not worth using w/8GB). It'll also probably bring larger area requirements. I'm sure a Kepler-based GPU will all the more so. I don't think nVidia wants to commit to 100+mm^2 GPUs but I'm not really basing this on anything.

Personally I think Logan's more likely to stay Cortex-A15 and hit 20nm. I would not hold my breath for Logan being on a newer node, even if the hybrid 20/14nm node is really being brought in substantially.

As for 20nm not being enough for 50x "total" power vs Tegra 2, that I could agree with.

 

[rpg.314]

Interesting that Project Denver is taking so long to make it nVidia's SoCs. Many figured Logan would have it. Some even thought Wayne would..

I wonder if Logan will have A15s again, or if it'll have A57s. That kind of raises the more general question of when we'll first see tablet manufacturers strongly embrace 64-bit. I wonder if Google even mentioned anything about 64-bit migration for Android..

Isnt java supposed to be 32/64b agnostic?

 

[Exophase]

Isnt java supposed to be 32/64b agnostic?

Android is not Java OS... the VM is Dalvik anyway, not JVM, but a bunch of stuff runs that isn't written in Java. Including the kernel, lots of userspace stuff, builtin apps and of course NDK which a lot of apps use. Getting this migrated to 64-bit is an actual real thing.

 

[Ailuros]

Eh, its a roadmap put together by marketing people. They are probably counting the fact that the cuda cores will be usable for compute purposes whereas the the GF ULP cores were not.

Under that reasoning T4 down to T2 should be all at zero. Of course is it a marketing diagram.

 

[ninelven]

Under that reasoning T4 down to T2 should be all at zero.

No they shouldn't. CPUs do compute work as well...

 

[Ailuros]

It is riskier but I don't think nVidia is going to be in a position to rest on their laurels and take it easy.

Agreed; however they're in no position also to afford any bigger delays than they have faced so far. Some minor hickups here and there are perfectly normal especially since NV is fairly "new" with SoC development.

We don't know a lot about Cortex-A57 except that it's supposed to be ~30% faster, with no real description how or vs what exactly. They've shown a pipeline diagram that's the same as A15's, but that might just be a placeholder. It's not that A57 needs more power, it's that nVidia needs some substantial mix of more power and better perf/W to justify a new SoC release at all. A better GPU alone won't cut it. A57 would bring 64-bit which could be necessary if there's substantial pressure for it, but I'm not sure this will be the case even in 2014 (A15's PAE-style extensions may provide a tiny amount of breathing room w/4GB, probably not worth using w/8GB). It'll also probably bring larger area requirements.

Can in your opinion a Tegra4 max out its real performance today? That roadmap isn't new they just re-adjusted it slightly recently. I was always breaking my head in the past years WTF T4 is placed so close to T3, while the distance between T3 and T2 is bigger. If you level the placement in that diagram (yes I know marketing yadda yadda...) to quite simple tasks with a very specific perf/W ratio then of course power consumption between T4 and T3 for those use cases is not going to change significantly but so won't performance despite the T4 carrying a by several times faster GPU than the T3.

I haven't digged too much into A5x to be honest, but it was my impression so far that besides 64bit one of the important changes were much higher perf/W.

I'm sure a Kepler-based GPU will all the more so. I don't think nVidia wants to commit to 100+mm^2 GPUs but I'm not really basing this on anything.

I'm confident that the result might resemble a lot from the outside to a reduced Kepler cluster; in reality it won't be anything else IMHO then a SoC GPU block fine tuned for SFF markets with all the lessons they learned with Kepler included.

Personally I think Logan's more likely to stay Cortex-A15 and hit 20nm. I would not hold my breath for Logan being on a newer node, even if the hybrid 20/14nm node is really being brought in substantially.

Trick question then: is it likelier that first desktop Maxwell chips will arrive on 28 or on 20HP and why?

As for 20nm not being enough for 50x "total" power vs Tegra 2, that I could agree with.

Well that's one of those typical marketing oxymorons you hit on all of those kind of slides whereever they come from. Someone in another forum tried even a funkier explanation and claimed the scale is for GFLOPs with the Parker GPU ending up at a glorified 1 TFLOP. The unfortunate thing is that its nonsense to compare FP20 with FP32 ALUs as the primary point and the next best being that the ULP GF in T2 delivered just 5.33 GFLOPs, so no that scale won't work just according to some folks convenience. Despite it being a marketing slide there is a reasoning behind it however twisted it might be due to its marketing nature.

 

[Ailuros]

No they shouldn't. CPUs do compute work as well...

Well thank you for admitting yourself in the end (even unwillingly) that the scale is based on SoC level performance.

 

[RecessionCone]

I'm confident that the result might resemble a lot from the outside to a reduced Kepler cluster; in reality it won't be anything else IMHO then a SoC GPU block fine tuned for SFF markets with all the lessons they learned with Kepler included.

I'm curious why you're so confident about this, when Nvidia said repeatedly and unambiguously that it would have a Kepler GPU? What could they say that would convince you?

 

[Ailuros]

I'm curious why you're so confident about this, when Nvidia said repeatedly and unambiguously that it would have a Kepler GPU? What could they say that would convince you?

Let's leave it at that then and if after the release it's won't be a strict 1:1 Kepler SMX copy in a SFF SoC I'll leave it up to you to come along with all the necessary explanations. Just as a completely unncessary sidenote, did it ever strike you that some folks here aren't exactly completely "alienated" to one or more IHVs and might know a couple of details more?

 

[RecessionCone]

Let's leave it at that then and if after the release it's won't be a strict 1:1 Kepler SMX copy in a SFF SoC I'll leave it up to you to come along with all the necessary explanations. Just as a completely unncessary sidenote, did it ever strike you that some folks here aren't exactly completely "alienated" to one or more IHVs and might know a couple of details more?

Deal on. After the launch, I'll be quoting this post at you.

 

[Ailuros]

Deal on. After the launch, I'll be quoting this post at you.

Watch it; Kayla is supposed to consist of not 1 but 2 SMXs, with the claim that it'll approach capability of the Logan SoC (not mentioning whether the approach is from top to bottom or vice versa) and that the integrated solution will be more "power efficient". So amongst others we'd need something like 32 TMUs to show up too....

 

[DSC]

http://www.youtube.com/watch?v=8kIQWWJs_po

Jen-Hsun announcing Logan and Parker at GTC 2013.

 

[mboeller]

Anyway, it seemed pretty clear from the presentation that:

Logan - 20nm
Parker - 16nm

How should that be possible.
According to heise-news they plan to announce Logan already 2013 and have it in phones or tablets in 2014.
Link (german) http://www.heise.de/newsticker/meldung/GTC-2013-Erste-Details-zu-Tegra-5-und-Tegra-6-1826134.html
Oh, and if this is true then Tegra4 seems to be death in the water . .

[edit]

Logan will be the successor to Wayne, more commonly known as the Tegra 4 chipset, and will be appearing in devices in early 2014

Link: http://blog.gsmarena.com/nvidia-reveals-their-tegra-roadmap/#more-46498

So only 6-8 months between Tegra4 and Logan. Tegra4 really seems death before arrival.

damn...firefox is really bad for writing comments. The format is completely wrong.[edit again]: javascript helps

 

[ltcommander.data]

I love how nVidia just finished trying to convince everyone that Tegra 4 being OES2.0 based is a big win due to being more die area and power efficient compared to an OES3.0 GPU. And now a year later with Logan they are saying the exact opposite by jumping over OES3.0 by adding full desktop OGL4.3 support with all its associated cruft and promoting it as a great design decision for mobile.