Nvidia reveals the 64-bit Denver core... with a fake die shot

[Particle]

Official support or not, nVidia's part is going to be 64-bit and use an ARM instruction set.

[Macadamia]

Eagle A-15 should be 64bit compliant (memoryspace wise). But nVidia is developing their own core logic instead of licensing ARM's stock config- kinda the case of Snapdragon and their scorpion cores.

It can go both ways though- Tegra2 had more custom logic than OMAP4, but after a small driver update and clock boost the OMAP4 more than kept up in both CPU and GPU power. Sometimes it's actually not worth it as ARM's own implementation is quite decent.

[-Boris-]

why all the sesationalism here? amd freely admits to doing the same thing... no reason to give your competators a jump on what you're doing. :shurg:

the fact that nvidia is making a cpu should come as no supprise to anyone here, considering the fact that nvidia has been making mobile cpu's for years.... now they're just bringing the same tech to hpc

You are confusing AMD with Global Foundries. And no. I can't recall AMD or GF published a faked die shot claiming it was something else. GF published a die shot showing they were on track with 32nm. They did however mask it using not to reveal details AMD didn't want to reveal.

I seem to remember this being done with bulldozer?

Nope. The die shot you are thinking of was a real die shot masked a bit since AMD didn't want to make it official yet. GF would be in deep if they published real die shots without AMDs aproval. And it was never a secret that the die shot was altered, it was official. See the difference?

GF publishing a real dieshot of 32nm altered to conceal information and is open about it is not the same thing as nVidia faking a die shot.

[bamtan2]

Eagle A-15 should be 64bit compliant (memoryspace wise). But nVidia is developing their own core logic instead of licensing ARM's stock config

in other words, nvidia is extending the architecture just like amd did with amd64, and if nvidia does a good job, just like amd did, arm will just copy it, just like intel did

[Sonic-Boom]

Why fake? It's hard to see any similarity with these colors:




On the other hand we know really nothing about Denver... it might use Fermi GPU cores... who knows !

This is clearly google earth of some fields and a few trees.

[xVeinx]

in other words, nvidia is extending the architecture just like amd did with amd64, and if nvidia does a good job, just like amd did, arm will just copy it, just like intel did

I wouldn't be surprised if the two were partnering in the effort.

[Chumbucket843]

Why fake? It's hard to see any similarity with these colors:

there are several things i noted. several parts clearly look like a cut and paste of gf100. i would do a crappy job with paint to show you this but i dont have the time. secondly, the diffraction looks very strange and unnatural. look at any die shot and you will see clear transitions from color to color. also the caches look synthesized which is highly unlikely.

[[XC] hipno650]

if it's fake or not who cares. it's not like being fake means project Denver is not coming out eventually. To be honest should be exciting for most people especially with Windows 8 shipping with an ARM build... this could "revolutionize" the PC market if done well. or could fall flat. I would peg it at 50/50 chance of either happening haha.

it is almost hard to imagine what a modern PC would look like though without x86....

[cegras]

It's been proven to be fake, in any case. See: first page.

[saaya]

Can you explain how the die shot is "fake"? You can really tell the difference between a GF100 transistor and a Project Denver transistor? I suggest you stop sensationalizing.

As far as I know the different colors are a result of the angle light hits the silicon, I've never seen a die shot with rainbow colors like this, jumping from red to green to red makes no sense at all...

[JumpingJack]

Charlie has sunk his teeth into this:
http://semiaccurate.com/2011/03/11/n...dscrew-moment/

Even references this thread.... pat on the head to DarthShader

[Macadamia]

in other words, nvidia is extending the architecture just like amd did with amd64, and if nvidia does a good job, just like amd did, arm will just copy it, just like intel did

Nope. It STILL runs ARMv7.


And you'll probably be waiting a long, long time before ARM licenses anything nVidia's making.
In fact, if ARM finds nVidia to actually invade their turf, they'll just stop any future licensing. But I suppose Denver is probably more standard design than what nVidia is touting; unlike TI they're not even one of the premier licensees.

[bamtan2]

Nope. It STILL runs ARMv7.


And you'll probably be waiting a long, long time before ARM licenses anything nVidia's making.
In fact, if ARM finds nVidia to actually invade their turf, they'll just stop any future licensing. But I suppose Denver is probably more standard design than what nVidia is touting; unlike TI they're not even one of the premier licensees.

this post is not accurate

[STEvil]

this post is not accurate

care to provide details?

[scottsche]

I don't understand how a company as large as nVidia can release such poor marketing materials.

[sutyi]

I don't understand how a company as large as nVidia can release such poor marketing materials.

If you cant show anything real, show something that resembles your future product from a distance.

[saaya]

i thought ARM wasnt going to support 64bit processors anytime soon?

support? they wont make one, but if nvidia does it, im sure they will support them and not BLOCK it...

this is like nvidia not doing a dual 560 vga, but partners can still do it... and if it makes sense nvidia will support them...

[JkS]

If there is still anyone not convinced (you never know!), heres a 5 second photoshop color overlay of DarthShaders image

[davidzo]

Nope. It STILL runs ARMv7.


And you'll probably be waiting a long, long time before ARM licenses anything nVidia's making.
In fact, if ARM finds nVidia to actually invade their turf, they'll just stop any future licensing. But I suppose Denver is probably more standard design than what nVidia is touting; unlike TI they're not even one of the premier licensees.

ure right about that. CPUcores are gonna be armv7.

I think theyre talking about the graphics or general ALUs part of the chip. Currently tegra T20s Geforce GPU is at best a Geforce6 derivative, even that is a little overstating.
Its difficult to process any general code on such an old fixed function gpu not even talking about double precision what is the industry asks for.

As we all know 64bits is double precision Floating Point Arithmetic. A feature on GPUs exclusive to GT200 and later GF1x0 GPUs and several ATIs.

The mentioning of 64bit is probably a hint that the future Tegra ALU part will support these gpgpu features.
Thus the Chip can calculate 64bits, albeit not on its CPU, lol.

What nvidia will definitely implement to the CPU is a Physical Adress Extension to allow more than 4gig of RAM (PAE-like).
Oh wait, wasn't it ARM that anounced this feature for the Cortex A15 already? So thats not even nvidas feat.
Dont expect too much from a Company that has not even half the workforce of the weakest competitor (AMD) and most of them in marketing, the others playing games most of the time.


- So there u have it. 64bit computing on Tegra Codenamed Denver!
RAM support and some double precision ALU, thats it.

[AbortRetryFail?]

ure right about that. CPUcores are gonna be armv7.

I think theyre talking about the graphics or general ALUs part of the chip. Currently tegra T20s Geforce GPU is at best a Geforce6 derivative, even that is a little overstating.
Its difficult to process any general code on such an old fixed function gpu not even talking about double precision what is the industry asks for.

As we all know 64bits is double precision Floating Point Arithmetic. A feature on GPUs exclusive to GT200 and later GF1x0 GPUs and several ATIs.

The mentioning of 64bit is probably a hint that the future Tegra ALU part will support these gpgpu features.
Thus the Chip can calculate 64bits, albeit not on its CPU, lol.

What nvidia will definitely implement to the CPU is a Physical Adress Extension to allow more than 4gig of RAM (PAE-like).
Oh wait, wasn't it ARM that anounced this feature for the Cortex A15 already? So thats not even nvidas feat.
Dont expect too much from a Company that has not even half the workforce of the weakest competitor (AMD) and most of them in marketing, the others playing games most of the time.


- So there u have it. 64bit computing on Tegra Codenamed Denver!
RAM support and some double precision ALU, thats it.

IIRC, Bobcat (and Llano, and BD 'flex FP') all do DP-64 (with AVX instruction).

[xVeinx]

While I can't speak to the degree to which the SEC would care about such matters, I think the important portion of the diagram is the layout of where things are. It was never said it was a die shot, just that this was the layout of the die, which it is. They aren't intending this to be a complete release of details . I think people are making far too much of this. As far as concerns the design of the processor, I'd wait awhile before making too many judgments.

[Russian]

Pretty sure they weren't necessarily trying to show it off as a real image, I really think that using this as a rallying cry against nvidia is a little excessive considering that this was an analyst day and nowhere does it say that its actually finished silicon or that the image isn't just a representation.

[davidzo]

While I can't speak to the degree to which the SEC would care about such matters, I think the important portion of the diagram is the layout of where things are. It was never said it was a die shot, just that this was the layout of the die, which it is. They aren't intending this to be a complete release of details . I think people are making far too much of this. As far as concerns the design of the processor, I'd wait awhile before making too many judgments.

well, wouldn't showing a Block diagramm show exactly what they could tell us about the architecture now?
Showing a dieshot really is a bit misleading because that assumes that they actually have taped out the chip. Showing coloured blocks like AMD did with Bobcat is not in that Ballpark. Faking Dieshots is trying to deceive Investors about the actual Progress of the Product. In the History of Chips a lot of Designs have ben scrapped even before tapeout, aka in the planning stage, especially from graphics chip companies. Showing a Dieshot concludes that it is taped out and will not be scrapped. But before Tapeout the Product may even be canceled completely.
I wonder if this will have some legal consequences for nvidia. If i was an Investor, i would feel seriously deceived.

[Russian]

well, wouldn't showing a Block diagramm show exactly what they could tell us about the architecture now?
Showing a dieshot really is a bit misleading because that assumes that they actually have taped out the chip. Showing coloured blocks like AMD did with Bobcat is not in that Ballpark. Faking Dieshots is trying to deceive Investors about the actual Progress of the Product. In the History of Chips a lot of Designs have ben scrapped even before tapeout, aka in the planning stage, especially from graphics chip companies. Showing a Dieshot concludes that it is taped out and will not be scrapped. But before Tapeout the Product may even be canceled completely.
I wonder if this will have some legal consequences for nvidia. If i was an Investor, i would feel seriously deceived.

You are reading way too much into this, like most people.

This is a slide, there are no claims... its a mock-up just like the wood-screws card. In that, charlie is right... but in neither case nvidia made any claims regarding the product shown. It was more of a simple mock-up, Nvidia probably would be more tasteful if they actually had a disclaimer saying it is merely a mockup, but if they make no claims they aren't liable to anything.

[davidzo]

sure its a representation. just like the woodscrews card was a representation.

but for what?
Its definitely not a representation for an Architecture. Taht would be much better shown with Diagrammes and even the Layout would be shown with blocks like in the bobcat sheets.

Its a representation for an actual chip, not for an Architecture. Its meant to give the impression they finished the Designstage already and are working on actual silicon. Thats what they suggest. Its not what they write, but they dont deny it either.