[Rumor] NVIDIA Maxwell, Denver CPU, Shield 2 and Tegra 5 Announcement 01/06 @ CES

[desprado]

NVIDIA Maxwell, Denver CPU, Shield 2 and More GeForce 700 Series Expected At CES 2014?

This is where the speculation begins and before starting, i would like to mention that last year at CES 2013 when NVIDIA announced Shield they mentioned that the handheld would see an annual update along with the TEGRA SOC. NVIDIA may announce their second generation Shield 2 at the CES 2014 which would feature their latest LOGAN SOC with Kepler cores and OpenGL 4.3 support. We may also see some updates to the design which would include smaller bezels, higher resolution and denser pixel screen with improved battery and controls. It may even launch at the same introductory price as the Shield for $349 which was reduced to $299 prior to launch.

The update regarding NVIDIA’s Maxwell generation of GPUs comes from Videocardz who have gathered some interesting details not only regarding Maxwell but also the Denver CPU and GeForce 700 series. According to the site, NVIDIA is expected to announce some details regarding the NVIDIA Maxwell GPU over at CES 2014 and we don’t know whether it would be make a public reveal or hold a post seminar briefing among the journalists.

Source and read more here: http://wccftech.com/nvidia-maxwell-...d-tomorrow-ces-2014-conference/#ixzz2pRDfuOjm
​

​

​

 

[Freddy1765]

28nm Maxwell, meh.. Wonder what kind of performance lead over 780 it's going to have.

 

[Kippa]

I am not that arsed about Maxwell's gaming capability, I'm more interested in how many CUDA cores it has compared to the Titan and how much ram it has. I do game but would more like to hammer the graphics card for 3d rendering in Lightwave 3d using the commercial Octane which specifically uses CUDA cores for rendering.

 

[ctk1981]

Wow...disappointing if true.

 

[sontin]

The last time nVidia mentioned something about specific graphic cards was 2010 with Fermi - "Fermi is ready and is going into production right now".

There will be no Maxwell news.

 

[Mondozei]

From what I've read the intitial Maxwell performance increase over Kepler will be more modest. The first Kepler GPUs performance over the GTX 500 series weren't that huge either and we will likely see a re-run of this in the beginning.

It is also almost surely going to be on 28 nm.

However, it could also be the case that they will simply deactivate a lot of SMX clusters on their cards, a lot more than they did with GTX 780 vs 780 Ti, and wait for 20 nm to roll around and then go big. This could happen in late 2014 or early 2015 and that is probably where the big performance increase will come.

 

[iMacmatician]

28nm Maxwell, meh.. Wonder what kind of performance lead over 780 it's going to have.

Since SA/Videocardz appears to be saying the GM107 is a 28 nm midrange chip, I suspect that the 780 is likely to have a performance lead over any GM107 parts.

Even if the GM204 is 28 nm (which Videocardz doesn't think is the case), I can see it being higher performing than the 780.

 

[blackened23]

I don't think 28nm is surprising at all. Keep in mind that nvidia has only done one line of chips on 28nm for Kepler - GK104 and GK110 have not been "refreshed" at any point and time. AMD has made 2 lines of chips on 28nm - Tahiti and Hawaii for the high end. Whereas nvidia has only made GK104/GK110 and they have existed since 2012 without being refreshed or overhauled.

Gains are also possible on 28nm still - AMD gained a sizable performance jump from Tahiti to Hawaii with both being 28nm. Apple's A7 is 28nm and is hands down the best/fastest SOC on the market. Gains are certainly still possible on 28nm, and since nvidia has not refreshed Kepler in any form or fashion, like I said...not surprising that it will debut on 28nm.

20nm isn't a real possibility until 2H 2014 anyway - i'd expect the high end Maxwell parts to be on 20nm, whenever 20nm is ready. 20nm certainly won't be ready by March, which is the rumored release date for the new Maxwell uarch.

 

[BallaTheFeared]

I think 20nm will be a pretty terrible node, and I don't think we'll see Nvidia's GF110->GK110->GM110 replacement on 20nm at all, but at 16nm so really 20nm, but the not terrible 20nm.

 

[Vesku]

Desktop GPUs are one of the more ideal products for benefiting from "just" density improvements. Although Nvidia has backed themselves into a bit of a marketing corner regarding power efficiency so the idea of leapfrogging from 28nm straight to 16nm FinFET would be attractive.

 

[blastingcap]

The update regarding NVIDIA’s Maxwell generation of GPUs comes from Videocardz who have gathered some interesting details not only regarding Maxwell but also the Denver CPU and GeForce 700 series. According to the site, NVIDIA is expected to announce some details regarding the NVIDIA Maxwell GPU over at CES 2014 and we don’t know whether it would be make a public reveal or hold a post seminar briefing among the journalists.

​

You cited to videocardz as a source without saying that videocardz cited SemiAccurate's Charlie D. as their source. A large number of people hate SA and Charlie because of their anti-NV stance, and many such people also claim SA is often inaccurate, so I thought I'd just throw that out there for people who did not read OP's links.

SA/Charlie also seem to be videocardz's source of info regarding GTX 790.

This is the videocardz article:

According to SemiAccurate’s sources NVIDIA might reveal some details of the upcoming Maxwell architecture during CES in Las Vegas. Whether this is going to be a public announcement or a special meeting behind closed doors, is yet to be confirmed.
As always take everything with a grain of salt.
Maxwell on 28nm process?

I’ve been hearing about Maxwell for few months. The rumor was that NVIDIA plans two major GPUs, the GM107 processor which would still be made in 28nm process and a new flagship (GM204) based on 20nm architecture. There were rumors about the GM104, but apparently it was canceled.
Charlie claims that NVIDIA will stick to 28nm process with the new architecture. He is basically saying there won’t be any 20nm GPUs from NVIDIA in 2014. I find this highly unlikely, it would deny all previous rumors and simply cause a huge disappointment.
One thing is certain though, there are no 20nm GPUs in first and second quarter of 2014. First Maxwell samples were made few months ago, they are almost ready for mass production and they are all in 28nm fabrication process.
I could be wrong but I have a feeling that GM100 (2014) are 28nm, while GM200 (late 2014/2015) are 20nm, we will verify this in the upcoming months.
NVIDIA Maxwell GM107 and GM204 release dates

According to S|A, the GM107 would be revealed in February. This is a mid-range GPU for cards like GTX 860, maybe even GTX 870. The Big Maxwell, which is codenamed GM204 would arrive later, possibly in fourth quarter. Of course this would be a GTX 780 (Ti) replacement.
What is new in Maxwell?

The Maxwell architecture will be a derivative of Kepler. The shaders will look similar to Kepler’s. The new feature is coherent memory, and here we should mention Project Denver, which apparently is not yet ready to be implemented into Maxwell. Instead of using a separate CPU NVIDIA will use a controller to take some steam off the GPU. How is this going to affect the performance? Let’s wait and see.
GeForce GTX 790?

Speaking of the new cards, we might soon welcome a new GeForce 700 graphics card. The GeForce GTX 790 is supposedly being prepared for February, so yes, this is going to be an interesting month. I heard it could feature 10GB memory (through 2x 320b interface) and 2x 2496 CUDAs. I could not verify this information, thus I decided not to make a separate post, but if these specs are even slightly true then I think you should know them.
Almost forgot, for those wondering where are 6GB 780 Tis, I think this has something to do with TITAN successor..

 

[Ajay]

Charlie claims that NVIDIA will stick to 28nm process with the new architecture. He is basically saying there won’t be any 20nm GPUs from NVIDIA in 2014. I find this highly unlikely, it would deny all previous rumors and simply cause a huge disappointment.
One thing is certain though, there are no 20nm GPUs in first and second quarter of 2014. First Maxwell samples were made few months ago, they are almost ready for mass production and they are all in 28nm fabrication process.
I could be wrong but I have a feeling that GM100 (2014) are 28nm, while GM200 (late 2014/2015) are 20nm, we will verify this in the upcoming months.

This makes more sense to me than anything else. Top Maxwell parts are supposed to have ~twice the number of shaders than GK100 - clearly that's not going to happen on 28nm. I'm guessing 20nm by Q3/Q4 2014, unless NV is really going for 16nm (for lower power) - then it will be 2015, IMHO.

 

[nenforcer]

They just released the 780 Ti a couple of months ago. I also think it's highly unlikely there will be any high end Maxwell cards right away like they've done that last couple of years.
(680 in 2012, and 690 / Titan / 780 in 2013).

 

[f1sherman]

yay its ON

http://blogs.nvidia.com/blog/2014/01/05/live-las-vegas/

Jensen has new cool jacket

 

[BallaTheFeared]

96 pre g80 cores to 192 kepler cores...

What's the performance increase on that?

 

[wasabiman123]

That is awesome for mobile GFX, nothing to sneeze at for Tegra 5 going forward. I want this in a Surface 2/3 NOW, modern instruction support and efficiency is a huge huge leap from Tegra 4.

 

[BallaTheFeared]

Wow this is on K1, awesome :thumbsup:

 

[sontin]

Project Denver comes in H2 - from the press release:

Tegra K1 is offered in two pin-to-pin compatible versions. The first version uses a 32-bit quad-core, 4-Plus-1™ ARM Cortex A15 CPU. The second version uses a custom, NVIDIA-designed 64-bit dual Super Core CPU. This CPU (codenamed "Denver") delivers very high single-thread and multi-thread performance. It is based on the ARMv8 architecture, which brings the energy-efficient heritage of ARM processor technology to 64-bit computing.
Both versions of Tegra K1 deliver stunning graphics and visual computing capabilities powered by the 192-core NVIDIA Kepler GPU. The 32-bit version is expected in devices in the first half of 2014, while the 64-bit version is expected in devices in the second half of the year.

http://phx.corporate-ir.net/phoenix.zhtml?c=116466&p=irol-newsArticle&ID=1887853&highlight=


Shield 2 with the power of a Xbox360? Bring it on!

/edit: And people believed that mobile will never get to the next gen consoles. Look at this jump from Tegra 4. And next year there will be 16nm with FinFet.

 

[BallaTheFeared]

 

[f1sherman]

$@#(* man all this runs on Tegra?!
Thats pretty crazy.

Dat room!! Was that real time ray tracing? Looks like Nvidia has its own close-to-metal tools.

 

[sontin]

Yes, it's unbelievable.
Nearly 3x times faster than A7 on 28nm.

 

[24601]

WTF Denver is at least double the size of A15

I wonder how crazy powerful it will be.

 

[blackened23]

Whoa. Denver is done.

 

[24601]

7 way superscaler :O

 

[DaZeeMan]

Just an observation, aren't we to XBox 1 and PS4 now? Comparing to the last generation of gaming consoles instead of the current generation seems a little silly, although I do get that NVidia was in those last gen consoles.

Also, I'm sort of getting the feeling that NVidia is starting to fall behind in the process department, similar to what has already happened on the Intel vs. AMD front. Hopefully they are able to keep up.