nVidia GTX 780M Benchmarks and Overblocks (vBios, Afterburner, eVGA etc)

well that sucks...
I had the same problem with my 680m .
but now my 780m on stock volt(no bios flashed) I can get +135mhz+450mhz with no problems at all.
before, to clock my 680m at 980mhz-2250mhz I must have 1.05v bios installed
now only 0.88v to enjoy better performance.

 

I am running the 1050 mV BIOS at 1019/2350 MHz myself, been doing so for a long time. Max Temps (seldom reached) are quite low at 82-83 degrees, I'm using a laptop cooler though.

 

Are you using a modded AW PSU? I've found that if I apply a similar overclock I get stuttering during gaming and benchmarks. The overall experience is smoother on stock clocks. Looking to get an AW PSU now tho

 

No it doesn't and quit telling stories. LOL! If you have a dell 680M then i helped work on it to make it what it is. So I know exactly what it's capable of doing at at what voltages. That goes for the 780M as well. And I can tell just by what your posting that you didn't even figure out how to make that work right at stock voltage.
Plain and simple. Stock vs Stock on unlocked vbios the 780m is better. Not sure how your missing this but ok. Do not mistake you not knowing how to make it work for it being broken. Had a few members in the same spot your in now, but once they saw what they were doing wrong. it's like night and day. And once you figure it out, it will be the same.

Don't fall for that Joker. it's the same mistake a few others made when running the new vbios.
LOL

 

It is possible to get a bum GPU johnksss. I've had them in the past, and in some review units. They run fine at stock and even slight OC, but are just bummers for any kind of significant OC, which is why the cards are clocked how they are, to accommodate the lowest common denominator quality part. Otherwise why not clock them all 100MHz higher to start?

 

Is it possible, yes. In this case, no.

Why do you think we took a week testing this sort of thing? To find out when these sorts of problems happen and why they happen. Not. "Oh, it doesn't clock well so it stinks" To many people jumping the gun on what is and isn't. Like i stated before. Some have already had the same issue and once they figured out what they were doing wrong, the problem was fixed. With the same cards they called junk not a day before.

As to your 100 mhz theory.... The reason why is because the way they have the card set it will never go over well. All it will do is throttle, just like it does now on a factory vbios. But the difference between my answer and others is I know exactly why it will, not a bunch of guess work that is far from being in the ballpark.

I can take his card and make it do wonders...what would that then say to the person who had it before me?

 

yeap.....

 

I want to believe this card can do more as well, but am in the same position as a couple others: my 780ms (both of them) won't due a mhz over 100+ and 300+ on memory on stock vbios.

Pardon my naivite, but where should I begin unlocking the card's (theoretical) potential? Is the vbios here the way to go? What are the exact setting where you've found the card maxxed out (vbios version, vcore, core/mem)?

 

Your aggressive tone is quite unwarranted for. Have you had a new AW17 for testing? Do you even know that it throttles on stock vBIOS, which is why it doesn't score more than 7000 points in 3DM11?
I have applied the unlocked vbios posted by svl7 on Tech Inferno.

I have then increased temperature and power targets and
- incrementally increased gpu clock @stock voltage while leaving memory untouched ==> not even stable at + 100 MHz
- incrementally increased gpu clock while applying overvoltage while leaving memory untouched ==> somewhat stable at +140 MHz with 1062 mV applied, not tested for long for obvious reasons though
- tried combinations of core/memory overclock to see what works best ==> stable at +100/200 MHz with slight overvoltage.

I used both nVidia Inspector and EVGA Precision X for applying overclocks/overvoltages to check whether there is any difference.

Temps and voltages were monitored with the newest Version of EVGA Precision X. Overvoltage was reported to be applied correctly.

Instead of calling me a moron, why not just enlighten us all as to what would be the "obvious" right thing to do?

 

Does anyone know for sure why the 780m is not a good overclocker compared to 680m? Surely temperature throttle can't be the only reason.

 

1: I did not call you a moron nor refer to you as such.
2: There is nothing aggressive in my tones what so ever, just stating the obvious. Lets not wear our hearts on our sleeves here.
3: In directly, yes. I had one for testing.
4: You cant increase temperature past 92C, so that means you can not change that in the up direction only down. Which means it's fine where it is.
5: Why don't you try moving the power slider to say something like 140 then retest.
And you need your screen to look something like this



to understand what works and doesn't work. Where it throttles and why. How much power draw your pulling. And if your on a AW17 with a 240W psu, you can for get it. My single heavily over clocked gpu and over clocked cpu will draw 340W You psu maxes out at about 300W

The problem I have is people jump out there bad mouthing stuff without knowing how to fully test all avenues. If it wasn't for people like us not caring what we are lead to believe is true. You (Me or anyone else) would not have any performance gains on anything.

So my advice when "trying" to understand something is to move all variables out of the picture and concentrate on one thing only. So turn off cpu over clocking and work on gpu.
That gpu is not going to break so move the slider all the way to right and then do one single test at stock to see if you gain or lose. If you lose here, then i will go with what htwingnut said, but if you don't....then user error. That means compare it to what you already have ran on original stock vbios. then just up the core to a 100+ and run it again. And report that score. Reason is so i can compare it. And post it here. Have no clue what driver your using but it's probablly best to go with 320.49 whql or the new beta.

Also, you will need to make sure they are giving you the right amount of amps, because these shipping now...are pretty much set to wrong....

 

@johnksss: I'm sorry if I misjudged your tone.

I cannot heed yiur advice right now, as I've sent back my AW17 already and I'm waiting for an sli setup (no more AW though) with (again) 2 more 780m.
I had moved the power slider all the way to the right anyway. I had not overclocked my cpu, as I was aware the psu might be a limitation. Interestingly, though, evga had always reported power draw even below 100%. I figure a mere +100 MHz oc (even on stock throttle bios without ov) shouldn't overburden the psu.
My assertion still stands, though. If the AW17 cannot provide a proper environment for the 780m to unleash it's power, be it from a power or thermal perspective, it is of no use from a customer's perspective.

 

No worries Brother Ship!
It, sort of can, but the way things are going now...No way in heck it's going to do so.
They just do not have the settings even remotely close to being right. Your cpu will auto down clock and that throws off your gpu power and clocks.
The power set for stock is something like 85 tdp no where near enough to do anything right. Ergo throttling. And that tdp is like a sort of a guide line....it can get as high as 170 or as low as 40, but not really adjacent to watts....
But since i have a killowatt meter and can pretty much tell where it's sitting at in terms of watts and draw.

So until they decide to fix that bios, the AW17 and AW18 will have issues. But the AW18 can be over come because xtu can change the settings needed for the system to run. I think the 17 as well. Wont know till you get yours back.

 

But please explain why there's such a big difference between 680m and 780m. Basically, it is the same chip with one more shader cluster enabled. There can't be THAT much of a difference in power draw. As we are all aware, the trustly M17x R4 can happily support a heavily overclocked and overvolted 680m, thanks to dedicated people from the community like you. Mine is running at 1019/2350 MHz 24/7 fully stable. What astonished me and made me send back the AW17 right away was the fact that I could not get the 780m to even match that performance! I'm not talking about any major increase, I'm not talking 10K 3DM11 GPU, just performance en par with the older genertion. shouldn't both cards draw more or less the same amount of power for equal performance? And shouldn't the 780m just downclock when it hits the power limit, instead of crashing?

 

The simple answer...
The 780 can draw, from what we can tell...about 160+ Watts. Still looking for a solid way to prove this.

To show you about your clocks theory...Point me to you best benchmark run at these 1019/2350 clocks.. and I will beat that not even moving the memory nor the voltage nor go above 1 ghz and at stock cpu. Not bragging( And you could do the same), but allot of time it depends on the machine you are using. The 18X is better equip at this over the 17X. They would need someone to commission them to allow the 330W psu to be fully supported. A single gpu in the M18x can more than likely hit 11k single gpu. That will never happen in the 17X. Why? because 240W psu is no where near enough power.

As to 10k gpu in 3dmark11. since it looks like there are only about 3 people who can do this...That can't really be your mark...You have to go by what you can get. So take your best 680 score and then put your best 780 score up against it.
Also, this clock for clock stuff is also throwing you off. No way you need matching clocks to match your 680's. If that is the case, then there is something seriously wrong. The object is to do better with less. Not do the same with more. . Also, these are not harsh words...Just words coming from someone who benches allot of parts.

 

@johnksss:

As per your request, I just ran this test:

NVIDIA GeForce GTX 680M video card benchmark result - Intel Core i7-3720QM Processor,Alienware M17xR4

This is with my M17x R4 24/7 setup with the 680m @ 1019/2350 MHz (OV 1050 mV vBIOS) and the i7-3720qm @ 40/39/38/38.
I know a lot of people score even better than this, and I could probably go up a little higher still for benchmark purposes, but that result is fine with me.

My AW17 would barely exceed P8000 at somewhat stable oc settings with temps in the high 80ies already. As for the standard vbios, my non-optimus 3D Version throttled at 73 degrees, so I couldn't even reach P7000. Relative performance was the same for 3DM Firestrike and Heaven 4.0.

If I understood what you said correctly, a properly overclocked 680m more or less represents the maximum that can be achieved on the M17x/AW17 or any notebook with a 240W psu. Thus, there is really no point in moving from single 680m to single 780m in any of these platforms.

BTW, I'm fully aware that I don't need to achieve matching clocks to achieve performance parity between 680m and 780m. If I do my math correctly and leave memory speed out of the equation, 780m should reach parity at 1/(1536/1344) = 0,875 times the 680m's clock rate, which, in my case, would have been 892 MHz. Interestingly, it did *not* achieve performance parity at that clockspeed.

 

This is not even my best score NVIDIA GeForce GTX 780M video card benchmark result - Intel Core i7-4700MQ,Micro-Star International Co., Ltd. MS-16F4 which I got without setting my CPU at 34x, and not the highest OC. That test was done at 980/3000 i think.

With full on fan speed, I can play games at 980/2900 non stop with 72-80 temps in average on the gpu. I need to change my vbios to OC highr, I just need my USB haha.

This is merely for comparison, not trying to argue which one is better and what not.

 

Yeah, It doesn't work like that. You would need nvidias equations and not your own. . So I gave up on trying to work the numbers like that.
This is an experience game. Numbers on paper are just that. Since the minute you change something you change the outcome...What we are doing now...they said it couldn't be done. If we stuck with that...back to square one and a 5870 would still be king.

 

Over at the AW17's owner lounge, one guy just posted a 3DM11 result of P5600 at default settings. This just to show how ridiculously Dell has gimped the 780m out of the box.

 

Yep. This was experienced as well.

Side note: Did a quick test and at 1100/2375 results in a max wattage of 245. The 780m is well into the 340+ range.
So it's safe to say...the 780m draws far more watts.

 

That's ... a lot. I take it you are referring to the total system power draw though, not just the mxm board?

 

Total system
680M
Idle watts 60
over clocked watts 245


780M
idle watts 70
over clocked watts 340

 

Do you have any idea why the 780m draws so much more power? It's easy to see that those demands will quickly overwhelm a 240W psu though. A 780m sli setup should then just barely make to with a 330W psu non-oced, right?

 

Our psu shuts off at 438W. So there is still room to run, just not as much.

And as far as i can tell...The 780M can really compete with desktop cards.

 

but how can it be then that the gtx780m works fine in a p150sm with "only" 180 watt psu???

 

^ I hear it uses the battery in conjunction, crazy lazy imo.

 

so by using the battery it makes like 240 watts? how does this work, especially on long gaming sessions? Wouldn't be the battery empty then at one point? Please explain, you made me curious how it works.

 

Please correct me if I'm wrong, but Sager models (P17xSM, P15xSM) do NOT have the supplemental battery support -- that's only MSI models (part of what they call the "NOS" feature).

In regards to how much extra wattage can be pulled from the battery when the load tops the PSU limit, I'm not sure that's been exactly determined (I'd reckon no more than 40W), but that might be better answered by a GT70 owner (or maybe DJ "Why are you making me wear this hat" Blu-Ray can help YOLO)

 

You're probably going to need to define what you mean by "it works fine"

 

cpu on full turbo boost while the gpu is also on full max speed (standard, no manual overclock)

 

You have any links to any numbers on this?

 

numbers of/for what?

 

Are those wattage readings for a single 780m? Because you have 780m SLI and the majority of us will only be able to use (afford) a single 780m.
Thx

 

So, I'm confused, whether it be what I read or maybe how I interpreted it. Anyway, the 780M (when overclocked) exceeds 240W? The Alienware 17 comes with a 240W PSU, so, I won't be able to overclock it as much as the 680M?

Hope someone can clarify this for me.

 

Probably wall readings so you can take the losses from the brick out to get the system power.

 

Nevermind. I checked already.

Before I had sli I had single. These are single card readings.

It has been gone over and we are hoping Dell will address it.

This is where things start to get funny to me.

Care to "rationalize" an answer? I mean I already know the answer because well i tested it, but it's always good to hear others opinions on the matter.

 

@johnksss

So now there are two "unofficial" confirmed restrictions on Alienwares with 780m?
1) Temperature throttle (Can be overcomed with EVGA Precision or modded vBIOS)
2) PSU limit (How can this be overcomed? Purchase an AW18 PSU?! But I guess this will only "Technicially" solve the problem on AW17. Imaging what PSU is required for 780m SLI in AW18)

If those are overcomed can 780m be stretched (OCed) like 680m could? I.e 30%?

P.S. I think people like you who can find out problems on new releases in such a short time period are really awesome +1

 

I've run some further tests on my P375SM. It was really bugging me as to why I couldn't budge past 100+/300+ without freezing in SLI so wanted to eliminate the variable that I may have a somewhat handicapped GPU. I started by switching SLI off and retesting my master card: 135+/450+ no hiccups. I then replaced my master with my slave card, and tested that single card: again, 135+/450+ no problem. Something is definitely limiting my GPUs while in SLI, whether that be PSU limitations or something else I'm not sure. I don't think, however, that it's not a thermal limit as the hottest they hit is 81c.

 

This is the main reason why we seek out early adopters who are not afraid to run the machine. This is the only way to find out if it is really ready for prime time or just a shut up and take this product and be happy with it. Which we most certainly will not be! If it does not run as expected. Your problem is the power supplied to your cards is not enough. Sort of like trying to run you stick shift car in 2nd gear when it has 5 gears.

This is for both post
Example:
100 percent tdp=110W from card. This might be near, below or above stock. Depends on which vbios is on it at the time of purchase.
When you ramp up the cards the tdp rises. If you hit it's 100% mark (110W) the card will dial back or slow down or throttle to keep you under it's 110W max. And some cards can hit this on stock clocks. (depending on vbios)
Until the card is told that it can go above this value, you will never be able to get higher clocks. And no amount of over volting is going to help that.

And since none of the factory vbios files are set like this (Yet), it will never happen.

Also, You can run on a stock 240W psu which maxes out around 300 or so watts. You just can't go crazy with it. Not even using just a single 780M. This card when trying to max it out will max out your psu first, but it will get you the desired performance you seek though before it will give out. That's why svl7 unlocked the sliders. To put you in control! Now go make those better numbers happen.

 

Thanks, Johnkss. That makes sense that the card throttles when TDP is reached, but why the disparity in headroom I (and others seem to be) seeing between a single card and SLI?

 

That's easy. Single uses less power to run vs the power need to run two.
And of course the machine it's in plus vbios and power ratios in play.

 

Right, I guess I don't understand why the GPUs would throttle at a lower speed e.g. 100+/300+ in SLI, whereas I can push a good deal further e.g. 135+/450+ (and likely beyond) in single mode if the limiting factor is the TDP on the vbios of each card. In SLI, wouldn't they just top out at the weakest cards limit? Is your proposition that the cards are pushing my PSU past its limit and somehow they "know" to throttle back? And if the GPUs aren't throttling, and I'm hitting the PSU's ceiling, wouldn't the PSU trip and the system power down (to be clear, this is not what I'm seeing by the way: my system freezes but remains powered when I go above 100+/300+ in SLI)? Further, if I am in fact hitting the PSU's limit at this mediocre OC, what good will yours and svl7's vbios do me?

Sorry for the lack of comprehension. It's just in nearly nine years of working with SLI configs I've never seen dual GPUs not be able to run at least as fast as the lowest-denomintor GPU (provided they had proper juice to run). I have seen SLI trip power supplies, and oddities in the early days where one GPU would run slower than another, but that doesn't seem to be the case here.

 

Going to break this down...

You would need to forget old school methods and start changing with the times...
It's not just one limiting factor, its a few all rolled up together.
psu
gpu clocks not matching voltage
to much voltage
not enough voltage
board limitations
heat limitations
power limitations
What is the actual cutoff of your psu.


At the end of the day, Im only giving you possible answers to your questions. What you do with that information is up to you. It's not a what you should do or up for argument....Only what you can take and build on...

 

I miss the "old days" when the only variable was your silicon and the voltage it could handle

Will give that vbios a shot.

If my limiting factor is, in fact, the PSU, I'm wondering now if the mobo can handle a daisy chain (like on my x7200). Just thinking ahead... If I'm stuck at these speeds, I'm starting to have buyers remorse for giving up my m18x r2, or I should have just stuck with a single GPU (and upgraded the PSU) and clocked the hell out of it.

 

Depends on the end result. Lots of stuff to try before calling it quits.

 

@Johnkasss

Someone in the AW17 thread said the latest Nvidia driver 320.49 is working now. Can you please test it out?

Thx

 

Im on that driver now. What are we suppose to be testing?

 

Sry got it wrong. I meant to say the latest beta 326.19 NVIDIA DRIVERS 326.19BETA
It says it supports 780m

Testing stability in OC of course. It just doesn't make sense 780m can't OC as well as 680m. Each new gen flagship should be able to do +30% on top of the stock performance.

 

Ok, got it. Will try them out.

 

Oh and please prioritize single 780m testing.