News about the upcoming GTX 680M

I was talking to an Alienware tech regarding the 680M and the R3 and he said that at this point in time Dell recommends the 680M for the R4 only. The reason being that their engineering dept. hasn't tested the 680M in the R3 yet. There may be a chance that it will work but they won't recommend the 680M for the R3 until they've been able to test it in the R3 and update the drivers.

 

The real reason is that they want you to buy an R4 in order to get a 680M. You can bet the ranch that they are never going to recommend use of a 680M in the R3. As usual, we'll know about 680M compatibility only when a few brave R3 owners serve as pioneers to try it out and explore its real world capabilities and limitations in the R3.

 

they guys.. i got something for you.

this benchmark is run at default setting.

Is this the first member 680m benchmark result?

 

@crazyzxf -- Excellent. Thanks. Where did that come from? +Rep

 

This is real dell alienware part leaked from TSMC taiwan, I have a friend who is a senior engineer at Dell factory in China.
Getting my own soon, he promised to supply me some!!

 

Hmm, not too much higher than 7970 ...

 

Yeah it is the first one. Excellent work
BTW, whats up with the 2011 date?

Can you ask him to do "Extreme preset" in 3DMark 11 since that is what matters most? And then ask him to do 3DMark Vantage? Thank you
+rep

 

Ok here's a big favour. Ask him for a vbios dump.

 

LOL, you do really want to be ready for your new hardware. .

 

I have to say, he sounded like a happy man when i was talking to him this morning. He will re-test Vantage and Extreme tomorrow.

 

Waiting for the 680m to drop so I can order my M17x R4.

 

Its not higher at all, weve had members pulling 5900 scores on stock systems

 

Performance preset aka 720p counts very little with these powerful GPUs.
If they perform different, we should be able to separate them with "Extreme preset" aka 1080p with 4xMSAA+16AF which is what people will be gaming with

680M score 200 points higher on GPU score with this performance preset

 

^ This. Would. Be. Awesome!!!

 

Just curious, what do you guys need the vbios for?

 

I need it to improve this: Vbios Patcher - releases & changelog

And widezu... I don't know, but maybe to ask me to improve the thing above

 

All I want with the vbios was to ask my kind Swiss friend above to take a peak and find a potential overvolt. Just imagine...true GTX 670 clocks...

 

Cool. Awesome project

If we can overclock this GPU closely to 670, then woooow. That would be so sick

 

If there is an overvolt possible then yes we may see the core reaching 670 speeds but the memory clock is dreadfully slow @3.6Ghz effective where as the 670 has 6Ghz effective

 

I don't know about overvolting... Kepler works completely differently than Fermi. The voltage seems to get adjusted dynamically with the boost clocks. Did some first modifications on a 650m vbios, increased the boost clocks and it seems to work, thing is I don't have a Kepler card to do the testing myself.

 

you guys able to extract so much potential out of those high performance GPU. Sounds to me, stock setting benchmark is almost irrelevant at this moment other than satisfy curiosity.

On the other hand, the cost is the marketing deciding factor. My understanding is that the out of factory price between 7970m and 680m is not so different.
680m is higher but less than 25% percent. However, alienware asking price is a different matter.

 

Wow, svl7. If RBE can't handle the 7970M, you'll just roll your own vbios editor. Major props and multiple reps. Congratulations and thanks from the o/c community.

@svl7 and widezu69 -- Did you guys notice the variance between the Graphics score and the Combined Test score on the earlier 680M 3dMark11 screenshot? Wondering if that suggests a power issue. Sure hope not, but those two scores are ordinarily pretty close to each other. Your thoughts?

​

@crazyzxf -- I expect the upgrade price for the 680M will be much the same as the 675M. It's just the way these things usually work at Alienware.

 

Thank you very much @The Rev, I'm glad you to hear you like it!! Just want to add that it's not only my work, my buddies chmod1337 and geischtli are behind this as well.

Good catch with the combined score... I failed to see this. It's incredibly low, in fact lower than my M15x with 7970m and 920xm at stock.
I got 4.9k for the combined test, with no overclock. This almost seems like a power issue or throttling (CPU or GPU).

 

you cannot overvolt desktop 670's via software...so i don't think the 680m will be any different.

This is the only way to overvolt desktop keplers beyond nvidia's control...you have to bypass via hardware mod. There are some retail cards you can buy that have already done the hardware mod similar to this (like the MSI's 670 with overvoltage). :

GTX 680 OC - Result-Thread --- Post your results here - Page 8

 

That picture looks a lot like svl7's modified M15x motherboard that he used to increase voltage to support o/c'ing his 920xm through the roof. Not for the faint of heart.

 

LOL...there is only one person that would even attempt that here...and you've already said his name lol (i.e., svl7)

I was looking for a way to go beyond Nvidia voltages on my SLI 670's but when i saw that pic, i gave up lol

 

Hahahaha, thanks guys! Lol, yeah... nice mod there!
Yes, it seems that at least the reference design Keplers can't be overvolted per software, well they actually can (per VBIOS) but it gets overwritten immediately by the dynamic overclocking thing when the boost clocks change.
It might be slightly different on the mobile side, but looking at some GPU-Z logs of mobile Kepler cards it seems the voltages behaves very dynamically as well.

 

All of this new technology brings performance not dreamed of even two years ago, but the increased complexity and locked down options have removed a lot of the fun, at least for me. With ThrottleStop and RBE or Nibitor, we could do almost anything to these systems. XTU and private utilities like svl7 or Svet has developed still allow some customization for some systems, but you have to wonder how long serious tweaking will still be possible in the PC world. Or maybe I'm just an old fart that can't keep up.

 

No, unfortunately your right. It certainly doesn't get easier to tweak and improve the system performance. True, there are a lot of sophisticated overclocking tools which are even co-developed with the card manufacturers, but they always come with limits.
For a private user it's really hard to develop something because there's pretty much no publicly available documentation when it comes to VBIOS tweaking etc.

Also talented people like the developer of RBE stop further developing because of the new difficulties, but also because it is an incredibly time-consuming hobby creating such a tool... I can totally understand this.

Unclewebb is still developing TS and trying to add an improve proper IB support, also for XM CPUs The problem is that it's really hard to do such things when you don't have access to hardware for testing... I'd love to get a Kepler GPU to test some mods, but they're quite expensive for a student, besides I only just got my 7970m, hahaha.

The 680m or Kepler generally is definitely very intriguing and a new challenge, but somehow I feel the same as you... back in the "5850m times" things were so much different, and a lot of fun!! Hardware-wise and community-wise Things change fast in the world of technology.

 

Well Schenker notebooks managed to get 6024 (almost 300 points more) in GPU score with their XMG P702 at stock speed. So perhaps they got it to work without throttling the GPU/CPU while the recent test had some issues?

GTX 680M

 

Here is the mine. Stock clocks, and I cannot run GPUZ, always return me an error. Saludos.

 

well, if you look at Schenkers results, their combined score is also quite low (i.e., 4822). it could be nvidia power control messing up the numbers. I'll stay positive and hope that this doesn't affect overclocking at all

i know my desktop 670's clock easily on the nvidia power scheme

 

Hm yeah thats true.

But question, isn`t "Physics" CPU test only? It looks like the 680M setup is scoring allright there right?

I see from your score that its scoring 700 points less in the "Combined score" though. Is that a lot?

 

Are we certain that the combined score of 4822 posted is not just a typo? - seems odd to me, given that the graphics score was 6024 and the physics score was 9267. Surely the combined would be higher than 4822????

 

Always a little careful with those Schenker numbers because they don't stay at default settings as we use the term. Those Vantage results clearly show that PhysX was being used, which completely skews the CPU results and therefore the 3D marks result. 3DMark06 appears to have been run with post processing disabled, which also inflates the results. Not much you can do to rig results with 3DMark11, but the Physics score suggests that the CPU is overclocked, even though CPU has very little influence on the final result. Still, I'd like to see some honest testing to give us an idea of the 680M's potential and how it compares to the 7970M. If the test purports to be at stock settings, then use stock settings and stop fudging the results. (This is not unique to Schenker; there's plenty of fudged "stock 7970M" results that are anything but posted in this forum). I suspect the two cards are going to be very close, probably with a slight advantage to the 680M. The Combined Test score remains a mystery.

 

The physics test in 3dmark11 is done with CPU only. Only the CPU renders that particular test so a highly clocked Xm processor will score very high in this test...the Schenker test scored over 9k and they had a 3920XM so results are consistent with an OC'ed Xm processor

Stevie, that 4822 combined score looks right since we now have two 680m 3dmark11 benches showing a low combined score

In the Schenker test, the combined score should be much higher but I'm currently researching what this means...will edit post if I find anything

The Schenker numbers are definitely "optimized" in vantage and 3dmark06. i don't know about the 3dmark11 though...it could be legit, but seeing how the 9k physics score points to an overclocked CPU, it is possible that BLCK bumping could have influenced the 6k GPU score, but who knows

I believe the other almost 5900 screenshot though. It would be cool if the tester posted a link the official 3dmark11 link so i can see the rest of his system

 

When I was running 3DMark 11 with my 7970m with a very OC'd CPU, the Combined score was lower which suggests that the GPU dials back when the mobo is supplying a huge amount of power both ways.

 

^^ You see that with a 580M and a heavily overclocked XM CPU too (as you know well), but not nearly to the same degree as we're seeing with the 680M. Could be the 680M is making huge power demands on the system, but that much at stock clocks? That would not auger well for o/c efforts.

 

Could it be due to not having any official drivers optimized for GTX 680M? Per now, no drivers support GTX 680M

 

Sure, raw drivers can cause all sorts of issues. The fact is it's all speculation until these 680M's get into the hands of people that can and will test them under a variety of conditions and setups with no axe to grind and no systems to sell. Then we'll know more.

 

That variance almost shows that the 680m is demanding more than 100w from the system (if that's even possible)...i think the MXM 3b slots can give more but not much more

EDIT: from my investigation, the low combined score is a result of the 680m having weak direct compute...explanation below, 4 posts down

 

...But the result we got on this board by our fellow NBR member, he didn`t overclock anything
He also got low "Combined score"

 

Here is the mine 3d mark11 with stock clocks AMD Radeon HD 7970M video card benchmark result - Intel Core i7-3720QM Processor,Alienware M17xR4 score: P5848 3DMarks

 

GTX 680M scores like 700 points below in combined score although it score slightly better in GPU score...

LOL

 

Ok, i think i know why the 680m scores such a low combined score. The combined score is the last test in the benchmark suite and it runs with both CPU and GPU rendering (i.e., they each have a role in rendering the last scene)

The explanantion of "Combined Score" is this:

Combined Test

* Runs rigid body physics simulation on a moderate number of objects (CPU)
* Soft body physics using DirectCompute and the Bullet physics library running on the GPU
* Medium tessellation
* Medium lighting

The keplers are known for their lack of directCompute, hence a very low score

EDIT: I reread what i wrote and i think i should clarify...the keplers are capable of DirectCompute, but they are just weak at it. Nvidia, has instead pushed off directcompute functionality to the Quadros. The 680m is still good at tessellation, shading and general rendering, just not DirectCompute

 

I think you are right Slickdude

 

Wow, I just read this and the 680 desktop really does suck, on the whole, when it comes to directcompute....explains alot regards to the 680m poor combined score...+rep Slick....

 

any closer to a release yet ? i need to order

Why all the fuss about directe compute guys , 3dmark doesnt relate to gaming performance , you cant play 3d mark all that matters for me is the 680m is fast

 

this is the test that says it all...pro apps will suck with the 680m, even the last gen gtx 570 kills the gtx 680. ATI is head and shoulders better at gpgpu. This says that the 680m is a pure gaming card (which is fine for the target audience). Anyone looking for a gaming/workstation combo is better served looking at amd

you can't play 3dmark, but it is an indicator of what to expect from the card. It puts it in the ballpark without predicting individual game bias (i.e., some games like nvidia and some amd)

And from all initial indications, the card should be fine in games, but we were investigating the low combined 680m score mystery. DirectCompute should have little bearing on a game unless the game uses gpgpu to calculate something...like direct physics.

 

Amazing insight. Makes a lot of sense