• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Nvidia's next-generation GPUs coming sooner than expected?

Nickolp1974

Nickolp1974

Nickolp1974

Soldato
Joined
15 May 2012
Posts
5,830
Location
Louth, lincs
source: http://www.itworld.com/hardware/397985/nvidias-next-generation-gpus-coming-sooner-expected


The company has been saying April, but a vendor let slip it could be sooner than that.
December 30, 2013, 8:00 AM — Nvidia has been as consistent as Intel when it comes to new chip architectures: every two years, usually early in the year. Tesla came in 2008, Fermi came in 2010 and Kepler hit the market in 2012. The next generation, codenamed Maxwell, was supposed to be issued in March or April of next year. However, a leaked slide from an ODM suggests a sooner release.

Clevo, an original design manufacturer of high-performance laptops, let a roadmap slide out that showed Nvidia GeForce GTX and GT 8-series mobile graphics processing units (GPUs) in its product designs coming in February.

The leaked slides first showed up on ThinkComputers, a hobbyist/review site. The slides show three mobile parts using the MXM design, which means they are replaceable, interchangeable and in some cases support SLI, which means two GPU cards in one laptop running together in parallel: the GeForce GTX 880M 8GB, the GeForce GTX 870M 6GB and GeForce GTX 860M 4GB.

MXM is a mobile PCI Express spec that's been around almost a decade. It was meant to make laptops upgradeable, since for the longest time the GPUs were soldered to the motherboard. Clevo and Dell's Alienware subsidiary are the two biggest supporters of MXM because they target the performance/gaming market, but it's available from other ODMs and major vendors as well.


The slide leaves a lot of unanswered questions, though. First, Nvidia doesn't introduce mobile chips first, it introduces desktop add-in boards and mobile comes later. Now granted, this slide comes from a laptop ODM so they probably don't know Nvidia's board plans, but it would be a bit surprising to see Nvidia come out with new graphics boards so close to the passed Christmas.

Second, the GTX880M with 8GB of GDDR5 memory is definitely an eye opener, because Nvidia's top of the line add-in board, the GTX Titan, has 6GB of GDDR5 and that card costs $999. Memory is not getting cheaper, either. It makes me wonder what kind of monstrosity of a laptop will use these chips. It certainly won't be cheap. The GTX 870M comes with 6GB of memory, which is also a lot.

Then there's the question of why the chips are being pushed up two months. The first thing that comes to mind is pressure from AMD. Let's face it, AMD has been slapping Nvidia around lately. It won all three game consoles; Xbox One, PlayStation 4 and Wii-U. And the new R290 card for gamers can match Nvidia's top-end gaming card, the GTX 780, for $100 less. So Nvidia had to make a move.

Beyond the MXM chips there are three entry-level chips, the GeForce GTX 860M, the GeForce GTX 850M, and the GeForce GT 840M. These may or may not be rebranded Keplers. In the past, when Nvidia shipped a new GPU processor generation, it rebranded the prior generation of processors with a new name and made them the low-end/entry level. That said, it's not certain if the new chips are indeed old Keplers or new Maxwells.

So what's the story with the Maxwell architecture? Well, like the last three GPU generations, it's named after a famous scientist, in this case James Clerk Maxwell, the father of the theory of electromagnetic fields. This will be the first GPU core with an integrated CPU, an ARMv8-compatible core that's designed to make the GPU less reliant on the CPU. It will support unified virtual memory technology with CPUs, which will simplify programming apps that use both the CPU and GPU, and it is expected to be much more energy efficient than the two prior architecture generations.
 
Though not impossible I have my doubts that these are Maxwell - nvidia has always rebadged a mobile line using the next gen model numbers and previous gen core i.e. my GTX675m has a Fermi core and likewise mobile cards often have larger amounts but cheaper lower clocked VRAM.

It does usually happen very close to the next gen desktop release however.
 
I think people are rather silly, most likely the +4gb +8gb + 2gb is referring to the laptops themselves. If you get a p17 something xhshah, then if you have the 880gtx in it, the LAPTOP comes with 8gb system memory, if we put the 870gtx in it, the laptop comes with 4gb memory, etc, etc.

This is how OEM's work, Dell sell a box at $400, $500, $600, etc. But each box has a $10 more expensive cpu, gpu, memory, hdd, etc. Someone who wants a 880gtx mobile chip won't want a laptop with 4gb memory, a midrange user buying a 860gtx probably won't want the price of 8gb memory.

AS Rroff and a million other people have said and the article itself states, desktop first... mobile takes a decade(maybe not literally) for validating products, designing the laptops, testing, ignoring bumpgate failures, then selling to the public. Desktop chips are a quick check they process some data then out to the customer(maybe a little validation in there somewhere). These will both almost certainly be Kepler rebadged AND the memory amount will be referring to what comes with the system, not the gpu.

It is possible, don't get me wrong, 20nm is in production now at AFAIK both TSMC and glofo, it's just not high volume, lower yields and expensive unless you can get a pay per chip deal, I'm fairly sure Jensen has extremely comprimising photo's of the head honcho's at TSMC to have gotten such deals previously, it's saved Nvidia pretty much hundreds of millions. They could do some midrange smaller parts on 20nm to start with and it's possible they wouldn't be that far away but I wouldn't put money on that being the case.
 
drunkenmaster said:
I think people are rather silly, most likely the +4gb +8gb + 2gb is referring to the laptops themselves. If you get a p17 something xhshah, then if you have the 880gtx in it, the LAPTOP comes with 8gb system memory, if we put the 870gtx in it, the laptop comes with 4gb memory, etc, etc.
Click to expand...

Current GTX780M is 16x RAM modules IIRC for a total of 4096MB @ 256bit its possible they'd do some kind of ugly duckling 8GB for the highest end refresh. I'd be unsuprised to see a ~2048 core config mobile kepler with 8GB VRAM masquerading as the 800 series - though its not totally impossible we'll see like the 600 series mobile where theres a mix of Fermi and slightly later released MX parts with Kepler cores (but in this case Maxwell).

RichardH754 said:
Think we will see lower end Maxwell April time then higher end September.

It will be the 680/780 all over again!
Click to expand...

Suspect the "full fat" core will be delayed until atleast 16 possibly 14nm hybrid processes are viable which is looking like atleast a year at the moment.
 
Last edited:
RichardH754 said:
Think we will see lower end Maxwell April time then higher end September.

It will be the 680/780 all over again!
Click to expand...

Which was no bad thing, especially with the 690 thrown in which still exceeds the performance of the single card 780 solutions. Wonder if they'll do that again. 2* 890's with power requirements of 1.5*780 would be nice
 
Last edited:
Rroff said:
Current GTX780M is 16x RAM modules IIRC for a total of 4096MB @ 256bit its possible they'd do some kind of ugly duckling 8GB for the highest end refresh. I'd be unsuprised to see a ~2048 core config mobile kepler with 8GB VRAM masquerading as the 800 series - though its not totally impossible we'll see like the 600 series mobile where theres a mix of Fermi and slightly later released MX parts with Kepler cores (but in this case Maxwell).



Suspect the "full fat" core will be delayed until atleast 16 possibly 14nm hybrid processes are viable which is looking like atleast a year at the moment.
Click to expand...

There is looking to be not a whole lot of die size space saving from 16/14nm(I think its 16nm at TSMC, 14 at glofo, it's all advertising, they both have a tighter gate pitch than Intel for a given process. Intel's 22nm is closer to TSMC/glofo's 28nm than their 20nm processes.

The finfet style 14/16nm upgrades for TSMC/Glofo are going to offer anything up to 40% power saving........ needless to say the cost and expense of a low volume product(high end gpu's aren't terrible volume but they aren't high for sure) taped out at 20nm when a 40% power saving is due under a year later. I can't see it at all, also while certain things can't shrink at 14/16nm, some of it certainly is and it likely means for a given die size their will be more empty space between transistors, meaning almost certainly better thermal properties, better heat spreading, lower temp running.

AS for 8gb, it is possible, just pointless. If it's a re-badged kepler then we're talking about a GK104... there just isn't any reason for a midrange kepler to have that much memory, it doesn't have 1/4 of the horsepower required for any kind of display set up or game that could push that much memory usage.

Of course, that hasn't stopped AMD, Nvidia or OEM's from putting in crap people don't need almost purely to make themselves seem better to the uninformed buyer.
 
^^ Could be GK110 on the new stepping and lower clock speeds/higher SP count, they are a fair bit better power/heat wise than the titan, etc. is on the original stepping on the new stepping. (My GHZ on the B1 stepping gets around 100MHz higher boost clock than the A1 stepping).
 
Last edited:
I'm happy with a late Summer release of the desktop Maxwell GPUs, that will allow me to see if my 780 can handle the early games of next year at ultra settings - to be honest, I'll find any excuse I can to buy an 880 probably! Hope it's going to be a great year for graphics cards :)
 
Nvidia are in no rush to launch Maxwell, the arrival of the 780Ti and price cuts to the 780 and 770 have them positioned where they want against the 290X/290/280X respectively, Nvidia have always aimed to charge a bit more for the same performance as ATi/AMD.

The thing that interests me is, what do AMD have planned for when Maxwell launches, is the 290X the real top end or do they have something bigger/bader waiting in the wing? considering how close R290X/780Ti are one would assume AMD held back on the 200 series or at least hope so.
 
You must log in or register to reply here.
Back
Top Bottom