HURRAY: Nvidia 600 series not just Fermi!! (Kepler)

The 2350M scores 130 more than 370M (look at the "CPU score"), so they are close. 2350M clocked at 2.3GHz, 370M clocked at 2.4GHz. Both missing turbo boost.

 

For some reason I thought the 370M was clocked lower. My bad.

 

I suspect the 680m to be full 560 Ti performance at least.

 

Man imagine if GT 650M is close to the GTX 460M. A freaking GT card, TDP with I don`t know, 35-40W maybe, performing like a 75W GTX card? Where does it put GTX 660M or GTX 680M?

No way

 

it would be sweet if gtx 680m had 768 shaders

 

The guy from post #10 said this (translated):

 

How do they look photoshopped :O and why would anyone want to.

 

I think the 3DMark scores look photoshopped. Note the out-of-place black pixel or two above the letter 'D' in 3DMarks in the screenie on the first post.

As for why anyone would want to, your guess is as good as mine.

 

I have a reasonable explanation for that: Most people use the prtsc button to take a picture of the screen after doing the benches, paste it in to paint and select whatever portion they want from the screen dump and copy it. The rest are cut out. It isn't rare that people accidentally click on somewhere in the pic in this process and accidentally make a little dot with paint. And another thing, what is it to edit over the "3dmark"? If anything have been edited, shouldn't it be the score and we should have seen a dot around the score? Just thinking out loud of course

 

The chinese guy says the Q470 is owned by his friend, that is why he would only post what his friend posts on Weibo.

 

Easiest way to find out GT650M's performance is to get a Quadro 2000M which should be OCed to core810MHz/memory900/shader1620MHz..

BTW: I repeat
Kepler 384sp = Fermi 192sp.

 

Compare to the nerfed 128 bit GTX560M in ASUS G53/74 will do.

 

Then you must downclock the vram from 1250MHz(625MHz) to 900MHz(450MHz
) as GTX560M uses GD5 other than D3.

 

Should not compare Quadro to Geforce cards in reality. There will be GDDR5 version of GT 650M, don't worry.

 

GD5 version may be named as 660M with same 384sp.

 

supposely the 640m in the acer aspire m3 5800 TG is just a rebadge of the 525m

Acer Aspire M3 5800 TG

This wouldnt speak for the 650m being kepler

Also the 635m (555m rebadge) in the asus n45 jay chou edition has an even higher score in the 3dmark 06 benchmark (at 1280x768, so probably the same at 1366x768)



source http://notebook.pconline.com.cn/testing/cn/asus/1201/2650855_3.html

Also I have searched for the Device ID of the GPU Z Screenshot, it seems to be part of an older PUBLIC driver (290.18)..

Source: http://forums.laptopvideo2go.com/topic/29043-asus-mobile-29518-released/
How some people have already noticed, GPU-Z reads out driver information.. Now if someone for example owns a 555m and knows a bit of reverse engineering, he can change the device ID to a different on in the driver, so GPU-Z reads out info from a hex edited ID.. But in this case the ID refers to N13P-GT, so GPU-Z wouldnt read out geforce 650m GT but rather the codename N13P-GT

This is just an educated guess, I might be completely wrong

Much more then believing this is a fake, I just cant understand how NVIDIA could rebadge their high end cards 670m and 675m (which are much less produced then low-to-midend cards) and mass produce the midend 640m and 650m with 28nm kepler architecture

 

640m is a rebadged 525m
while
the 630m is a rebadged 540m?

that woudl be so weird

 

I don't trust all this stuff anymore, 3 more days till CeBIT and maybe it will all be revealed

 

lol I won`t even try to guess what is or what isn`t anymore.

Yes gamba, the GT 650M could be a fake. It could also not be a fake. Nobody knows. I just don`t see how it would be a fake, especially when we have the screenshots that match the early article where someone was looking at a samsung laptop and it fits the specs he saw. See yknoyoung`s earlier post. The 650M beats the 555M, even more since the guy who tested the 635M used an i5 processor while 650M was done with an i3.

But here I go again, BAH
Agreed, lets just wait and see what happens at CeBit

 

I hope 600m family will be better than rebadge, otherwise, we will wait another year for good GPUs

 

Agreed Cloudfire!

Since im from germany and got 2 free tickets for the CeBit ill maybe going to check out the new gpus myself (with a usb stick with 3dmark and such on it )

 

BTW, where does it say that the GT 640M is a Fermi (rebrand) gamba?
Man you got tickets to CeBit? Lucky lucky you. So much to see

 

Here is another source saying that the GT 650M is Kepler and is real and not a fake
Here is a picture of the GPU



GeForce GT 650M (Kepler GK107) pictured, tested ... ~ OBR-HARDWARE

 

why on earth this picture is proving it is Kepler??

 

No I just posted it now since I didnt earlier. But someone commented on the picture:

 

OBR is the most unreliable source ever if you check out their track record.. also the die size looks about the same as the 550m(540m) die size on pictures

the link you posted uses the same source of information as we do

 

The die size of 650M is smaller than 555M (140mm^2 compared to 228mm^2) and yet have better performance. It must be a Kepler. You had the same discussion here, but didn`t listen

Yes, OBR is not very reliable

 

What's the point of hooking 384 shaders to 16 Rops 128bit DDR3? You could hook up 192 and get virtually the same performance and a much smaller core.

It's like hooking a V12 to the wheels on a trolly.

Also the horrendously low clock rate only makes the Rop problem worse. 16 Rops at 400mhz is pretty much the same ROP potential of the 6770M which is just silly.

 

yes but why compare 555m to 650m? the first number is the generation

its not a a fair comparison.. the 550m should be compared to the 650m.. also the perfomance isnt proven at all

anyhow im on your side hoping that the 650m will be kepler!

EDIT:

check this out!

Read the post and then the discussion

Seems like its fake!

The post:

http://semiaccurate.com/forums/showpost.php?p=155107&postcount=1451


Discussion:

http://semiaccurate.com/forums/showthread.php?t=4849&page=145

 

I don't like it either. Nvidia is brute forcing for raw performance, and in a massively inefficient way.

Seems like they were desperate to get something Kepler out, thus was birthed this monstrosity.

The GT 650M is the GF110, GTX 480M of what we should expect from mobile Kepler, if it's even Kepler at all.

 

Just wait and see.

 

hmm, seems like they are still having a discussion. He is stating that it is a photoshopped GPU-Z or that he have fiddled with the GPU-Z, but portions of GPU-Z screeny actually match a GT 540M. But what riddles me is what GPU did they use to get the 3DMark06 score. An overclocked 555M? Photoshop there too? Are they really that sneaky? So that means the picture of the chip is fake?

Can you change the ini-file of the GPU to get GPU-Z to see 384 shaders and is it possible to get those "unknown" fields with a simple editing? Or do you have to photoshop?

Man this is all very confusing.

 

Wait a minute, why an overclocked 555m? As I said before even the 635m (slightly revised 555m with higher clocks) gets a bigger 3dmark 06 score then the one of the 650m screenshot..

650m


635m


Remember its 3dmark 06 @768p (which is only directx 9) so it could possibly be a standard 555m such as in the clevo or maybe the m14x 192 bit 555m.. Doesnt really matter though, I think its all fake.. I mean if the gpu z is fake the 3dmark score too since its all from one user or source

 

I was thinking they actually used the i3 that they used to test 650M in 3DMark06. If they did that 635M would have been lower than that. But then I remembered that it could all be fake

 

I'm not sure how GPU-Z reads the device ids, but when you install the drivers, you can fiddle around with the entries in the inf file to make Windows think you're using a GTX790 (for example) rather than, say, a GT555m. This method could screw GPU-Z up if it's reading from the driver rather than the hardware. You do not see any device manager screenshots since it reads from hardware, so the images you see could very well be manipulated and/or photoshopped.

EDIT: see this.

 

It makes total sense what he is saying, and sadly that leaves us in the dark again.

But what I don`t get from what he is writing is that GT 650M will be a lot slower than a 540M. Even if it was a rebrand shouldn`t it atleast be equal to a 550M? If it were the GPU-Z from a downclocked 540M, why is there 2GB of RAM? The 540M have maximum of 1.5GB according to notebookcheck

I personally thought that the low speed of the shaders and GPU made total sense with a new architecture where Kepler could be faster clock for clock, plus the amount of shaders. With 96 shaders at this low frequency it is lulzworthy

Here are the specs of both GPUs:

 

Here is a Q2000M 06 benchmark from here http://dell.benyouhui.it168.com/thread-1462713-1-1.html

Spoiler

Stock clock(550/900 D3/linked) resolution 1280x720

And he can OC it to 760/1060 D3/linked

while he was too embarassed to give out the OCed score as he left these words:

“超频的3Dmark06超频就不上了。基本没区别，日。 ”
Literally,
(I) won't post the OCed 3Dmark06 score, as there is no difference (between stock and OCed), (what the) fvck!


I guess low bandwidth provided by 128bit D3 vram becomes the bottle neck for its 06 benchmark.

Note: That GT650M 06 benchmark is with resolution 1366x768.

 

I think the point of him posting the second screenshot is to prove that it can be very easily photoshopped rather than to prove that the GT650m is slower than the GT540m. Either that or Nvidia is positioning the GT650m as a low powered GPU, which would be a big mistake imo since the speeds should improve generation upon generation rather than underpowering them.

 

Interesting. I didn`t imagine 128bit being that bad. Guess we can forget about overclocking a low clocked Kepler if it only have DDR3 and 128bit bus then.

 

But I can see a significant improvement in OCed Q2000M vantage and 11 benchmarks in that thread.

And a 760/1060/linked Q2000M can get 10266 GFX score in vantage under 1280x720 resolution.

Spoiler

Do you remember someone claimed that 650M on W110ER could achieve P10000 vantage score? (Which I think might be true for a similar resolution other than P set.)

 

If that holds true i can already see DXG reballing a low profile Kepler on a 580/675 Fermi board and OC the heck out of it. AKA homemade 680m

 

That would be a downgrade. 580M/675M can surely kick mobile GK107's a55.

I hope 680m could be made of mobile GK104
since someone has revealed that desktop GK104 only has 2-6pins power supply and similar TDP as GTX560TI.

 

Kepler or not, 128 bits + DDR3 = kthxforthelulzbye

 

Don't yall know never to trust pre-lauch GPUz or Benchmarks? 95% are fakes..

 

true! cebit in 2 days

 

WOW, that is a pretty good OC result. About the same score as GTX 570M got and almost 2x the performance at stock settings. This could mean that a 28nm with a lot of headroom can give an average Kepler pretty good performance, even if it is a 128bit. I`m curios about real game performance with this overclocked 2000M though since I don`t entirely trust benchmarks. Do you know if the 760/1060 was stable, or if the GPU was just barely hanging on?

If only this GT650M is real, that could give a good bang for the bucks.

 

Awesome thank you!

Looks to be the first real benchmarks of the 640m!

How can I rep someone?

Very interesting! They have only run a 3dmark 05 benchmark, cinebench 10, cinebench 11.5 and tested 3 Games.. Stalker, RE:5 and Lost Planet 2

Spoiler

Lets see how much the 540m in the m11x r3 gets then ( also ULV, but a bit better the 2657m)

Source

Now lets not forget that the acer uses an Ultra-low-Voltage i5-2467m processor

Can someone give a good indication of where the 640m stands in comparison to the 500m series???

 

WHAT?!! wth is 3dmark 2005, asus g74sx beats an m17x with 580m? crazy

 

Its probably too outdated.. but maybe the avg fps of stalker, resident evil 5 and lost planet 2 give a good indication? Also there are cinebench 10 and 11.5 benchmarks

Stalker

Spoiler

Resident Evil 5

Spoiler

Lost Planet 2

Spoiler

Cinebench 10

Spoiler

Cinebench 11.5

Spoiler

Okay its not the best benchmarks but its better than nothing right?

 

GT 650M = GK107-200
GTX 660M = GK107-300

NVIDIA GK107 Details by VR-Zone.com