NVIDIA Geforce GTX 660M Release Information + CUDA Core Count

I am interested to buy a notebook with GTX 660M.
NVIDIA reference core count is 384.
However, Anandtech states up to 384. Will there be a repeat of GT 555M with castrated versions?

Some pricing information would be good.
Also, any performance figures and benchmarks for the GTX 660M in any configuration please post here.


AnandTech - NVIDIA's GeForce 600M Series: Mobile Kepler and Fermi Die Shrinks

GeForce GTX 660M - GeForce

Nvidia GeForce GT 640M, 650M, and GTX 660M debut, bring 28nm Kepler to mobile GPUs | The Verge

NVIDIA GeForce GTX 660M - Notebookcheck.net Tech

 

Excellent thread. Please post upcoming notebooks that will have the 660m.

I'll start with Lenovo y580 and Gigabyte P2542G.

 

from what I can see their entire line up is either rebrands or gt 555m all over again.

 

There's already a thread talking about this....

http://forum.notebookreview.com/gaming-software-graphics-cards/649052-hurray-nvidia-600-series-not-just-fermi.html

 

I want to specifically get GTX 660M info only, as I am looking for a notebook with one. Finally the waiting continues...

 

Gk107 384 shaders, 128bit gddr5.

 

This one is one possible configuration, but there seems to be more?

 

I think it will feature 384 shaders, and this is no surprise as Fermi shader is equivalent (somewhat) to 2 Kepler shaders (for example gtx 560m shaders are 1550mhz and 660m shaders are 835mhz, definitely not exactly but you get the point), so yes it will feature 384 shaders, but 650m and 640m may feature less, depending on the oem as I understand..

 

Won't that bandwidth be limiting the card if it had true 384 shaders

Sent from my samsung galaxy s2 using tapatalk

 

I would love to get this graphics card with the gddr5 memory in a 13 or 14 inch laptop.

 

It was probably a tradeoff for better efficiency. You can point to pretty much any card and say that it could use more bandwidth but it's not always worth the extra power consumption for that extra 5% of performance.

Maybe Kepler is more bandwidth efficient and can hit most of it's potential without a bigger memory bus. Also keep in mind that the memory clocks are so high that even with a 128bit bus it has more memory bandwidth than the 192bit 560m.

 

from what ive read ill tend to agree more with grazzt reply that 2kepler shaders are equal to 1 fermi shader. so this card is equal to a 192bit shader fermi card and for that a 128bit bus would be more suitable.

 

it supports a max digital resolution of 3840x2160, or QHD

 

I highly doubt that.. Probably even the 650m with GDDR5 wont be found in notebooks under 15 inches, or very fat 14 inches.. so the 650m GDDR5 probably needs good cooling

The upcoming MSI GE60 and GE70 also feature the 650m GDDR5, which are huge and fat 15 and 17 inch notebooks..

Notebookjournal Upcoming MSI notebooks

So your best bet is probably the 650m GDDR3 version, which can be found for example in the upcoming clevo 11.6 inch W110ER and the Samsung Q470.. The lenovo Y480 14 inch has the 640m I think

 

Because the 660M uses very fast 128-bit RAM and controller, the memory bandwidth is actually higher than the 192-bit of the 560m or 570m.

We really have to educate people about not being so worried about the "bit-ness" and concentrate on actual bandwidth.

 

Looks like stock memory clocks are going to be 1ghz.

1Ghz 128bit > 650mhz 192bit.

 

Here is what I wrote in the Hurray thread

 

Articles : GeForce 600M Notebooks: Powerful and Efficient - GeForce

Official article, only showcase up to 650M. So 640M is supposed to be equal to 460M or 285M.

 

I'm really hoping that we get some good 12-14 inch gaming machines out of the new crop of Kepler machines. It seems that most of the announced machines are going for ultrabook thinness in this size range instead of keeping to the older sizes but delivering more performance.

 

This GTX 660M should not exist. The GK 107 is a small chip for mid-range mobile GPUs, and was intended to succeed to the GT540/GT555M.

That means there will be a GTX 760M with 768 CC that will be a worthy successor to the GTX 560M in ~5 months, maybe less if it's released along with the mobile GK 104. In the meantime prepare for a market of overpriced gaming laptops that will feature a GK 107 that was never meant to target gamers or rebranded 40nm chips and will perform way under the potential of this generation.

 

This thread is confusing me even more on whether the 570M is stronger or the 660M.

 

700 series? Keep dreaming, long time until that happen

 

Meh, just wait till mid 2013

 

I think I might go for 680M but I need a date/month so that I don`t wait for nothing

 

The 570m/670m are intended to be more potent than the 660m which is supposed to be a successor to the 560m.

The 570m/670m simply has more hardware (albeit Fermi instead of Kepler).


I'll bet the 660m performs admirably and truly succeeds the 560m, but doesn't quite exceed the 570m with BIOS fix or 670m. It might give unpatched 570m's a run for the money though.

The 660m will likely seem to suck when it is first released as Nvidia drivers will probably need an official support release or two before ther 660m really shows its teeth.

 

the 570m will be stronger

EDIT: Kernal beat me to it lol

 

Even if it is meant to be for mid range, if it performs, it doesn't matter. Based on the benches of the 640M, which trades blows with the 555M, the 660M w/GDDR5 should be 50-60%, conservatively. This puts it well above the 560M, and more than suitable as a replacement.

 

Please be so kind as to remind me, how many months between 400M & 500M ? All evidence point to it. The GTX 670/675M will have to make way for kepler chips. And if the GK 104 is intended for the top of the range, another chip has to fill the huge gap between it & GK107.

More like above, no need for "well". It's placed under the GTX 570M (= GTX 675M) which is reasonably faster than a GTX 560M, not tremendously. So the GTX 660M can't be too far from its predecessor.

 

Kepler chips are all active and clocking well, I doubt we will see a decent refresh.

 

dude, 400-500 had a year between them, 485m and 580m had 6 months between them as they are pretty much the same thing.. don't confuse the two, at earliest we will see 700m around 2013 summer, they will squeeze the market as much as possible with 600m (that is probably why we are waiting for 680m, as they are still squeezing the 580m )

 

Wrong.

Here, and here. Then 4 months later, add a few mhz, and voilà, you have the shiny (not) new 500m series. Don't underestimate nvidia's ability to quickly release a new series in a few months if they feel like it and need to make room in their nomenclature for the Kepler chips that come later.

Should I add that on the desktop side, they have a bigger and faster, compute oriented chip coming later this year. What are they gonna name their top performing GPUs based on it if not "GTX 780" ?

 

With the 500 series we had a new chip revision which was more tuned to the TSMC process giving lower power consumption and higher average clocking.

Mostly it was correcting a mistake by Nvidia so, while yes they can rebrand, the 500 series actually helped.

This gen there is no mistake.

 

If anything, the limited capacity at TSMC will not help for any new series.

 

They didn't do nothing about the aforementioned mobile parts. The chips that got modified were GF100 & to a lesser extent GF 104. If you look at the links I posted you'll see it's still 96 & 144cc, plus to my knowledge 4 months isn't enough to do a respin or any serious design modifications. The few mhz that were added could already be obtained via OCing.

They were maybe better chips built on a more mature process, so what. That's a more legit motive to switch from 400 to 500 than launching a mobile GK 104 & GK 106 would be to switch from 600 to 700 ? Since when exactly ?

I don't even see any good reason why Nvidia couldn't be doing both at the same time (raising the clocks on existing parts & releasing new chips). You've seen the clocking potential of 28nm, compare with the specifications of the mobile lineup, factor in the fact that the process is in its early stages.

yknyong however makes a valid argument.

I assume that by mistake you mean "not being able to squeeze a few mhz more off your chips because of the lacks of process & a more polished design", and then how would you know. Because I personally don't see the "mistakes" of the 400M series. There was nothing wrong with the GTX 485M / GTX 460M. I wouldn't even be so sure that their successors were based on chips that bore any significant changes.

 

after looking at some old anandtech articles i found that actually what happened was the 485m, and all of the lower end 500m series were launched at the same time. initially they were all rebadges of the 400m series(gf10x), except the 485m of course. then later on a few of the 500m chips were coming out with the newer gf11x type chips. so basically all of these new 600m chips which are rebadges will be rebadges of gf11x and not gf10x. it may not be an upgrade but it is not just a rebadge of 400m chips either.

so infact this 600m series is more impressive since we get cards of all ranges that are the new gk10x aka kepler.

 

dude, sorry to break it down to you but that date is wrong for example:
NVIDIA GeForce GTX 485M - Notebookcheck.net Tech

I know that 485m released on a January.. AND on 2011

next time don't trust notebookcheck too quick and yeah, on February-March-April 20 11 Alienware was still selling with 460m...

Here is the date of first Verde driver for 500m series (yes not whql):

 

AnandTech - NVIDIA GeForce 500M: Refreshing the 400M

notice "anandtech" not notebookcheck.net

 

thanks for the link, my bad then, but last summer I was in the lappy market and I remember how long I waited for 560m-570m-580m, the first lappy with them was around july and august.. still cannot believe that 555m was in a laptop in january, one of the first m14x was released around april..

at least one thing I know is there is about a year between gtx 460m - 560m, maybe because I don't pay a lot of attention to mainstream chips (well, until now 640m and 650m looks darn sexy )

 

Kepler have just been launched. And the high end of the 600 series have not even been spoken off. Lets say GTX 680M arrive in July, 4 months from now, do you really expect the 700 series to launch the month after the high end 600 series is launched? Shouldn`t the replacement for 570M be 665M or something instead instead of jumping up to 700 series and having to deal with 600 and 700 series at the same time? I atleast think we won`t see 700 series until very late this year

Since GTX 680M is GK104, maybe GTX 665M is GK106 which on the desktop side is supposed to replace GTX 550Ti? It kinda makes sense too since GTX570M was = GTX 550 Ti,
AND GK106 is one of the first GK cores that is due to launch VERY soon...
After all GTX 570M and 580M was also different Fermi cores which very well might be the case here too

Another prediction from Cloudfire
Who agrees?

 

What makes you think GTX 680M is GK104? The high end mobile card is never based off the high end desktop card.

 

A) Because GTX 680 is not high end desktop, but mid range
B) Because GK104 have been released already and Nvidia is probably working on fitting it inside a mobile card as we speak
C) Because GK104 is 256 bit
D) And most importantly because GK104 is GTX 670Ti which is the successor of GTX 560Ti which GTX 580M was based on

But like always, I don`t know this. Just me guessing based on what is available now. I think GTX 665M will be GK106 which on the desktop side is replacing GTX 550Ti which 570M was based on which 665M will be the successor to. And GTX 680M is GK104 which is GTX 670Ti which 680M will be based on.

I was right about GT 640M, GT 650M and GTX 660M being GK107, lets hope I get 2/2

 

Errr, why are you repeating what I said? However I did not say the 5xx series was impressive, I said it was a minor tweak to help it overcome the limitations of the process.

You're at it as well, where did I say they did nothing exactly? Also I can tell by taking even a quick glance at the clock rates and power consumption :/

 

Well I guess the 7970 is a mid range chip too because it's about the same size as the 680.

And if DK104 is mid-range, what does that make DK107? Looks like ultra-low end to me. And you have the 660M based on that low end part, but you think Nvidia is going to make the 680M off of a 4x stronger part? Keep dreaming.

 

Actually 7970 is a high end chip, beaten by a mid range GTX 680. GK110 is considered high end in the 600 series. And its rumored to be gigantic in comparison to the 680, 294mm2 vs 550mm2. And its a single GPU, not a Dual GPU. AMD didn`t have any huge dies with 6000 series (6970 was 389mm2, 7970 is around 350mm2) and it looks like they don`t plan on building bigger this generation (7000 series) either because the next GPU from AMD is the 7990 which is a dual GPU (2x7970). .

GT 635M and under is low end. GT 640M, GT 650M and GTX 660M is mid range. GTX 680M (and maybe GTX 665M if it comes out) is high end in the 600 series which is why GTX 680M is based on GK104 since mobile GPUs is not based on high end desktop GPUs. Maybe Nvidia call 660M for a high end since it have the amazing three letters "GTX" but if it is High end, then GT 650M must also be considered high end since they are so alike but I don`t agree with calling it high end unless it atleast get a 192 bit bus

 

I don't think we will see a 665M and in general I think the 600M series will lack high end performance. Nvidia wants the 660M in Ultrabooks and Kepler series in general is designed for use in such devices, with targets at tablets and phones. We may see their heavy compute architecture in the 700 series, but right now the market isn't geared toward high end performance. Nvidia has also mentioned they are more interested in compute technology then games, and we may start seeing a split in their chip designs.

 

The problem is that NVIDIA may use GK116 instead.

 

That is for the 700 series right?

 

It seems to me that Nvidia made a shady product.
While it's true that mid-range gpu's saw an increase in performance, their efficiency is lacking, all the while the high-end segment actually saw REDUCTION of performance.
There was another test done that showed the 640m generating enormous amounts of heat (93 to 97 deg C) coupled with a low end IB (especially when compared to the older generation).

What's the point of having an 8 hours battery life (pathetic as it is) if it can be done on IDLE only without any serious workload, or having a system that effectively overheats if you fire up the gpu and want to play games?

If this is what Nvidia touts as 'progress', I have to say it's moronic.

They could have merely reduced the existing FERMI's into a lower manuf. process and put the GTX 560M into the mid-range while bumping up it's clocks (or improving bandwidth) and probably still get away with a lower power consumption.

Instead, they come out with a 'new' architecture that seems to require twice as many shaders to accomplish same performance, consumes seemingly less power but heats up a lot more.

Am I the only one who thinks that this is idiotic?

Now, I agree that shoddy laptop cooling implemented by manufacturers is a definitive concern, but I'm not about to drop premium cash on a product that overheats to no end makes playing games a skin-burning experience.

 

I do certainly agree, although your point-of-view is kind of extreme!

I think the Acer Aspire M3 5800tg with the ULV i5 and the 640m already shows that the new kepler architecture gets relatively hot.. When tested with furmark (without prime 95) the GPU already reaches 93 degrees maximum, which is ok for a furmark test, but not so good in combination with CPU heavy games.. Although this doesnt create throttling, it certainly would create throttling in combination with a fullpowered non-ulv 35W or 45W TDP Cpu..

Thats why im really astonished that the clevo 11.6 inch netbook features a 650m (DDR3) with a fullpowered sandy bridge CPU, supposely even a 45W quad..

I guess we will just have to wait and see, but faster perfomance achieved through more transistors certainly creates more heat.. The question is how good the cooling will be on the upcoming notebooks, only time will tell

From this point on we can only make assumptions and guesses.. Of course NVIDIA claims that kepler runs more efficient and with less power/heat.. but do you really believe them when they say stuff like this?

NVIDIA launches 28nm graphics processor for ultrabooks - Liliputing

Common sense, laws of physics and slight marketing knowledge tells the informed customer, that these claims are simply made to create interest...

Another PR Stunt is this

Nvidia CEO wants to bring Kepler GPU to 'superphones' | The Verge

So think for yourselfs, rather then believing everything the company wants you too xD

 

The heat problem is caused by NVIDIA's new baby, the dear GPU Boost.

In fact Intel also suffered from the same problem with Nehalem and Sandy Bridge.