Brace yourself: NEW MAXWELL CARDS INCOMING!

Well I might as well post it again since Wccftech thinks it is real plus there is a guy in the comment field who knows a guy from AIB partners who also says its real.

GTX 980
3DMark Firestrike: X5500 (GTX 780 Ti - X5000, GTX 780 - X4500, GTX 880M - X1900(ish))
3DMark11 Performance: P15000 (GTX 780 Ti - P13500, GTX 780 - P12100, GTX 880M - P8300)
TDP: 170W (GTX 780 Ti - 250W, GTX 780 - 230W (ish))

GTX 680M was based on GTX 670 which had a TDP of 170W

I think this means there is a really good chance of a GTX 780 performance from the GTX 980M.

 

They all look like sensible numbers to me, I think there's good chance that this is real. (Based on what we know about efficiency & performance of Maxwell combined with their need to offer a reasonable credible performance upgrade especially given changing the naming scheme from 7xx to 9xx, and also to allow for the fact that their 970 is gonna need to be as fast or faster than their 780). I reckon there's gonna be some cool running cards when Maxwell comes by (170W that you quoted)! Good news for small form factors.

 

I changed the Firestrike score for the GTX 880M btw. The one you are quoting is the 3DMark11 Extreme. My bad.

Yeah the above makes a lot of sense.
I can see this happening:

GTX 980: 170W
GTX 980M: 100W

GTX 970: 130W (ish)
GTX 970M: 75W

GTX 960: 110W maybe
GTX 960M: 60W maybe

I think GTX 960M will also be a huge jump over GTX 860M

 

If GTX 980 is ~10% better than 780 Ti at a TDP of 170W, how do you reckon that a 980M with only 60% of the desktop card's TDP will come close to a desktop 780? Not implying that TDP and performance scale linearly, but for such a steep drop in TDP I would expect a serious ding in performance, especially since both are supposed to be based on GM204.

Btw 780 is also 250W, it is the 770 that is 230W. <del>If I had to make an uneducated random guess I'm going to say 980M would be on par with desktop 770.</del> Nope scratch that, probably halfway between a 770 and 780.

Final prediction: 980M scores X3900 in Fire Strike

 

That TDP number is not restricted only to the core. Efficiency gains can be made on RAM chips too. Comparison: gtx780m uses the same Samsung chips that uses the Titan. Their operating frequency and voltage however differs.
Also undervolting and underclocking helps a lot. Lastly maybe some cores were disabled in the proccess and we get close to gtx970 performance, which per se wouldn't be bad too

 

Yup. Lots of stuff that comes in to play here. Downvolting, binning, using a 28nm process thats been here for 3 years vs 680M that used a brand new 28nm process, type of VRAM etc

 

GT72 with GTX 880M is now selling for $600 less.
Amazon.com : MSI GT72 DOMINATOR PRO-010 17.3-Inch Laptop : Computers & Accessories

Asus G750JZ with GTX 880M is selling for $300 less
Amazon.com : ASUS ROG G750JZ-DS71 17.3-inch Gaming Laptop, GeForce GTX 880M Graphics : Computers & Accessories

Trying to remove inventory before the 900M series are here....?

 

I'm confused about when we're hearing news/announcements on Maxwell - am I right in thinking there was a Sept 8th date floating around, as well as a Sept 19th date? If anyone can clarify, I'll make a note of the dates so I can try to catch the live feeds, etc.

 

I have tried asking around about the 9th-10th event, but havent gotten a reply yet.
For now, we should stick to September 18th until we hear more
Turn On, Tune In, Geek Out at GAME24 | NVIDIA Blog

I doubt they will announce GM204 on that event since its a 24 hour event. Yeah good luck watching all of that lol. Makes more sense to announce it on a Maxwell event, but who the hell knows what they are up to.

 

Except that we know nothing about 960m, could be a really cutted down GM204 or a GM206, or even a 860m rebrand, who knows

 

If they bump the TDP of 960m to 60W they should be able to boost clock speeds of 860m by quite a bit. If still 28nm I wouldn't say more than 15-20% though.

 

I think the GTX 960M will be based on a different chip than 860M.
960M should atleast have 8SMM (1024 cores). GTX 860M have 5SMM (640 cores)

GT 900M will probably continue with the GM107 chip

 

I am not so sure about that, although is tru that 660m was GK107 and 760m was GK106, but that doesn´t mean that Nvidia want to do that again, specially when AMD has nothing to compete with

 

There are 3 known desktop chips coming.
GTX 980
GTX 970
GTX 960 (launching in late October according to rumors)

Pretty safe to assume neither of them will be GM107 and we will undoubtly have mobile chips based on them...

The jump from 800M series to 900M series will be a great leap if
GTX 980 GM204 >>> GTX 980M GM204
GTX 970 GM204 >>> GTX 970M GM204
GTX 960 GM204/GM206 >>> GTX 960M/GTX 965M GM204/GM206

Speculation:

Spoiler

800M/900M cards:
GT 840M GM107
GTX 850M GM107
GTX 860M GM107
GTX 870M GK104
GTX 880M GK104

GT 940M GM107
GTX 960M GM204/GM206
GTX 970M GM204
GTX 980M GM204

 

Dammit Cloudfire, you know how i am with these whole new gen graphic cards subject, you are making my hype go sky high with a 960m performing like a 780m with a lot less TDP, and then after they launch you will speak about 20nm Maxwell and will make me wait again XD

At least they are quite near

 

Hype is what I do

 

Nice to see the speculation train is still in service.

 

You really think that many cores?

 

GK107: 384 cores - 2SMX
GK106: 960 cores - 5SMX
GK104: 1536 cores - 8SMX

GM107: 640 cores - 5SMM
GM206:
GM204:

 

Hah right, he's got me chomping at the bit to buy a new notebook once the new cards come out.

 

with gpus focusing more on being power efficient I hope the gap between mobile vs desktop GPUs get lower. Would be awesome if a 125w flagship mobile GPU can keep up with a X70 desktop version.

 

I wasn't going to replace my GTX 860m until they have a brand new card in it's place vs just a clock bumped, re-branded original 860m. But if they replace it with a chip with +/-40% more cores and equivalent performance bump, I just might replace it early.

I'm guessing they won't replace the 860m with brand new silicon after just 8 months on the market when they can easily just bump the clocks and say here's the new GTX 960m with it's 10% improvement over the 860m. Then presto, they can sell it for another 9 months. But, it is totally just a personal guess as I have nothing to back it up with. If they do put a vastly improved graphics card in that segment so soon I will be looking a new 13-14" lightweight laptop in a few months to replace my 'aging' Clevo W230SS.

 

Even assuming Cloudfire is right on the 960m though, going up from 45w to 60w TDP could cause a major problem in Ultrabooks. Should be a great upgrade for anything over 1" with decent cooling, but for the under 20mm crowd the old 860m maxwell might still be a better option.

 

Honestly, ultrabooks should be glad they even GET that kind of power at 45W. I'm starting to assume the ultrabook crowd is like the current-gen console loving crowd. They assume their stuff must cost little, perform close to a high-end PC and have no other problems, then bash anyone wanting to use a laptop weighing 5 pounds or more. Because they can't walk around with a 9 pound backpack to go to some classes sometimes. Hmph. Some people.

 

Prior to the 860m, most laptops carried a 765m which was more along the lines of 60W and managed it just fine. I think it would be a good fit for existing laptop configs as a direct swap 860m to 960m.

 

There were only 4-5 truly ultra thin laptops with the 765m though, and they all ran very hot. Most were over 25mm (1"). Not saying 60w is impossible at the 17-21mm most ultra thins go for, but its really pushing the heat envelope far on them.

 

that's assuming 960m will have increased TDP, afterall 860m is really a 50w chip that can be easily cooled. if 960m is 60w then it won't be as mainstream as 860ms

 

Oh yup. Did'n see the efficiency in context.

So according to you, 970M mobile will probably trump which current desktop card by performance numbers?

 

GTX 770 is my bet

 

My 13.3" clevo is not a super thin and light ultrabook. It is well proportioned for it's size but in no way meets the criteria that makes it an ultrabook. Nor do I expect this laptop to cost little and perform nearly as well as the machines that are 2 to 3 times it's size.

I simply prefer as much performance that a mid level class of laptop can reliably produce...which for me and my needs, is plenty. That means when my OCed 860m can no longer compete against another mid-level card in the same weight class...i.e. NOT a GTX x80 card, I will replace my pathetic POS under sized laptop with the NEW pathetic wanna be gamer laptop. Simply because I don't want to lug around an 11lb, 17.3" laptop is my own preference. But not once have I ever 'bashed' anyone because their laptop is too big and I don't like.

Yeah, some people.

 

About that GTX 960/960M:

The only chips Nvidia is working on is the following:
GM204
GM200
( GM206 is not found = They havent worked on it yet)

Forget about GM200 for mobile.

We know 3 desktop cards will arrive soon (They were recently renamed to 900 series):
GTX 980
GTX 970
GTX 960

So we know all those three will be GM204

Which means that Nvidia will disable a lot of cores with the GTX 960 since GM206 doesnt exist.
How insane have they sliced off cores before and essentially wasted silicon?

The answer is GTX 675MX and it was GK104.
http://forum.notebookreview.com/attachments/alienware-17-m17x/91259d1359743252-alienware-m17x-r2-running-gtx-675mx-akc.png

A full GK104 is 1536 cores. Or 8SMX.
GTX 675MX have 960 cores. 5SMX. Which means they disabled 38% of the cores.

So lets assume a full GM204 is 1920 cores (15SMM).
If Nvidia disable just as many cores here for the GTX 960M/960 as they did with GTX 675MX, we are looking at a 1200 core GPU, or the nearest, 9SMM, 1152 cores.

So I assume GTX 960M will be around the area of 8-10SMM somewhere. Or 1024 cores - 1280 cores.

 

Never mind...I totally misread something in the above post.

 

I think the GTX 670MX was around 10% faster than GTX 580M, so P8500 (GTX 880M) +10% is P9400. GTX 770 scores P11000 and GTX 880M scores P8500, so yeah, maybe somewhat close to GTX 770 perhaps? Somewhere in the middle? I have no idea.

 

Since the current 860m is based off the GTX 750ti, wouldn't it make sense the new 960m be based on the GTX 950ti...whenever that is released. Which I still have hunch both of those will be just clock bumped rebrands of the current 860m/750ti.

On edit:
That would give nvidia an easy out using that GM107 chip for one more go around before they do the offical 20nm die shrink. They can just do rebadges on the 860m, 850m and 840m and just up the clocks 10-15% and call them 960m, 950m and 940m. It's surely something nvidia has done before and it sort of makes sense.

I like your theory better though. I hope they do squeeze a hugely beefed up chip in its spot.

 

Your 13.3" clevo sounds to me like the W230SS... that machine is considered "too heavy" (at 4.6 pounds) for people who want thin gaming laptops. They want usually something like 3.5 pounds or less + gaming performance + low price point + don't overheat. This is from most of them that I've spoken to OUTSIDE of NBR. On places like PC gamer comments or Linus Tech Tips. Everyone is clamouring for sub 4 pound babies that can run 4K res and game decently and stuff. It's... not really going to happen. Not now, anyway. In fact, I'll go so far as to say it'll never happen, because as soon as video cards get stronger, games improve, or get more unoptimized and the new cards have to pick up the slack. The 580 and 6970 desktop cards were super good and people used 1-2 of them for Ultra settings back in the day. GTX 680 comes out; 1 x 680 = 3 x 580s... what happens? Within a few months, 1 x 680 wasn't even enough for 60fps 1080p maxed graphics on some games, and people still needed to lower settings for 120fps 1080p on lesser-demanding games. It felt like as strong as it was, it was hardly an improvement. Even the 780Ti feels to me like it'd be "just" enough as a single card, considering all the new (not always highest fidelity) games.

 

The 860M has a 4GB version that's kepler based off the GTX 760. The 960M would likely be based off of the entry level high end maxwell desktop card. The 860M 2GB maxwell could have been called the 855M or 850M Ti (but mobile cards don't get Ti versions), but nVidia kind of did a thing that made us wonder why they didn't make a 865M to distinguish kepler/maxwell variants or something.

 

There are three Maxwell Mobile chips coming up for sure (See first post of this thread)
NVIDIA_DEV.1617 = "NVIDIA N16E-GX"
NVIDIA_DEV.1618 = "NVIDIA N16E-GT"
NVIDIA_DEV.1619 = "NVIDIA N16E-GX-B"

What is what I have no idea, but considering that there are 3 Maxwell desktop chips coming, all GM204, and I know "DEV_16xx" is GM204, I am preeeeetty sure those 3 will be based on the GTX 980, GTX 970 and GTX 960

 

I was forgetting about the Kepler 860m and I have never paid attention to which card in the desktop world it was spawned from.

 

Between different architectures indeed it is very difficult to gage memory bandwidth needs and its impact in performance. Higher numbers mean nothing across architectures because some are more efficient than others. GTX 580 had slightly higher memory badwidth than GTX660 but on extreme testing and settings, they basically performed the same. The bits is meaningless on its own. It was 144 GB/s vs 192 GB/s and GTX660 and GTX660ti still spanked the previous gens!

GTX8800 and such gen had a bit of trouble because they had low memory bandwidth compared to GTX260 (88 GB/s or so, GTX260 and 280 had over 110 GB/s) And the architecture, as you mention, was not as efficient using that memory bandwidth to it's advantage.

It is not only a manner of how much, but how you use it. R9 290x and 780 ti have very comparable numbers, so I don't know what you are talking about. 780 Ti has slightly higher texture fillrate, memory bandwidth etc. That is all in terms of raw numbers. Performance is similar and R9 was tweaked to from tahiti to a new distribution of the same architecture. Maxwell seems much better at using its memory bandwidth than kepler and GCN. Even tonga is able to compete despite much lower memory bandwidth.

All in all, Maxwell seems like a very good architecture, and I am very confident as cloudfire, hopeful, that we will get a powerful 980m

 

980M will be powerful no doubt, but I'm highly skeptical that 980M = desktop 780. Looking back, going from 580M to 680M, it took an arch change (Fermi --> Kepler) AND die shrink (40nm --> 28nm) for 680M to just barely catch up with desktop 580. So I don't see how an arch change alone would give 980M performance parity with desktop 780.

Don't mean to be raining down on your parade, but I think the hype has gotten a bit out of hand.

 

It will perform similarly to a GTX 780, slightly underclocked. I saw the future.

I hope one of those articles quote me. I am the Notebook Prophet after all. (See under my username.) It's fate.

Let us feast!

 

The 680MX came out at the same time as the 680M though; and was essentially a downclocked 780M. It wasn't available for laptops, only for iMac, but it still existed at the same time. If nVidia does things correctly, we could see mobile chips keeping up with the desktop chips from the get-go, with maybe pure downclocks being the only downside. Also, even though it's late, the 870M is a 660Ti and a mobile "midrange". They could have done that with the 680M from the start, but didn't (though the core count was the same).

 

Kinda annoying to be so close yet know so little. Damn the secrecy

 

What's another 2-4 weeks when we've waited 6 months already?

 

You make a good point. Allright, I`ll try to keep myself occupied in the meantime. It only gets worse with the rumors lol

 

Yeah, I know, right? Now we've got people claiming to have seen the future... Nice going, Cloud.

We've come so far.

 

You are forgetting that, despite Maxwell have only changed the architecture without a die shrink, there ir similar improvement on performance/watt that Fermi-->Kepler

 

Well at least now we can say with 100% confidence that SinOfLiberty was a total BSer

Umm, if performance/watt improvement is similar from Kepler --> Maxwell as it was from Fermi --> Kepler, then how do you propose a 580M --> 680M style improvement without a die shrink? Did you mean to say performance/watt improvement will be greater for Kepler --> Maxwell?

 

Wondering why no one's shared this yet: Only at VC: NVIDIA GeForce GTX 980, GTX 970, GTX 980M and GTX 970M 3DMark performance | VideoCardz.com

For once the source doesn't appear to be Cloud!