GTX680 to arrive at the end of February!

[dave_beast]

So the 7950 1.5GB won't be upto scratch for 1920x1080 ultra?

I'm getting 110fps on ultra with all settings maxed, so the 1.5gb 7950 might lose a few fps because of less vram available, but would still easily reach 70-80fps on ultra.

Between the first 79** driver and the 2nd driver I gained 8fps

 

[xsistor]

This is better than a few hours on damnyouautocorrect...

Seriously, you use another completely incorrect post by Xsistor to try and make your point that I'm wrong.

Here's a hint, if you can get 800Mhz at 0.9v, 900Mhz at 0.95v, 1000Mhz at 1.05v, 1100Mhz at 1.2v, 1200Mhz at 1.5v, etc, etc....... you'll find that is most definately exponential. His entire diatribe is both hilariously stupid and hilariously wrong as is everything he ever responds to me. I called him stupid and got banned because him and his dupe account both laughed at what can really nicely be put as, complete inability to read. I stated that I specifically DID NOT have to post a mathematical proof to be able to then state that 9.8 the value for gravity. TO which I think there were maybe 4 posts of Krugga and Xsistor incorrectly assuming I said that WAS a mathematical proof.

You shouldn't have gone there because Duff-Man's apologetic for your statements in that previous thread absolved you of looking completely and utterly ignorant. You only looked somewhat ignorant. Or you looked like a layperson without specialist knowledge. Now, going back to that thread and defending your ridiculous argument for an exponential voltage-frequency relationship has shown you for what you really are. As both Lisa Simpson and conventional wisdom say, "It is better to remain silent and be thought a fool than open your mouth and remove all doubt." Advice you should strongly consider.

If you think, "800Mhz at 0.9v, 900Mhz at 0.95v, 1000Mhz at 1.05v, 1100Mhz at 1.2v, 1200Mhz at 1.5v, etc, etc....... " suggests an exponential relationship, you need to go back to school and pay attention this time. While any mathematician could tell you that you can form fit any function through a series of points (within reason, given that properties such as monotonicity is preserved) -- for e.g. a polynomial curve "looks" like an exponential curve if you only look at a few values -- there is NO exponential function in the relationship between voltage and frequency.

it's not worth arguing this point further because anyone who knows a thing or two would know how wrong you are. Anyway, I doubt Duff-Man had to read my correction of your words to know that this isn't an exponential relationship.
As every engineer worth his salt knows, BJT (bipolar-junction transistor) and CMOS logic, by design, tend to use circa 5v, and it is microelectronic CMOS voltages that are reduced to circa 1v by design... Most of these smaller VLSI circuits run at a much higher frequency than their typical LSI CMOS counterparts. So obviously there is no such thing as an exponential relationship between frequency and voltage.

The actual relationship between voltage and frequency in a circuit is an engineering design optimization issue and is much more complicated. it depends on resistive, capacitive and inductive impedances, skin effect and other transmission line effects, slew rate and clock skew, white noise, thermal noise, power, etc.

 

[drunkenmaster]

Asides from right or wrong, you just tried to prove that an INDIVIDUAL CHIPS voltage/frequency relationship can't be exponential because you can have low frequency high voltage chips, and high frequency low voltage chips........

This is ultimately your problem, using a multitude of different arguments, unconnected to anything I say, or the matter at hand, to prove a different point.

Please quote where I said ALL chips follow the same voltage/frequency relationship, are all designed the same and all therefore need the same voltage to run at the same speed.

Because had I said that, your argument would possibly make sense, as I never said anything even remotely slightly loosely mistakeable for that, you've made another irrelevant argument.

In reality, in "theory" I should have said something along the lines of(its 1am and have work to do and don't care that much) voltage and frequency in real use on a single chip(explicitly even though a 3 year old child could work out I was talking about an individual gpu) follow an exponential relationship more to do with the heat output increase from upping voltage, the general instability at higher temperatures, the actually increased efficiency in cooling(more heat..... same air temp, better cooling) but still effective increase in temp meaning that for MANY reason all together almost every chip needs a greater and greater voltage bump to get the same bump in clock speed.

I could have said all that, but THIS IS A FORUM, its not work, I'm not providing proofs, its as easy to say what I said and find that in reality that is how you will indeed find chips work in the real world for the most part.

The relationship between clock speed and voltage, even with LN2 is in NO way linear, even then the more heat your gpu is using at a higher voltage, even LN2 has its limits, a chip cooled via LN2 will still run hotter at 1.4v and 1200Mhz than at 1.3V and 1200Mhz, it will still need ever more voltage for the same clock speed improvements.

You're basically the ONLY person on here who can't let something that is easily simplified for talking on a forum, but this is your problem, you won't let the simple crap slide, you then go off on proving a COMPLETELY DIFFERENT POINT, and spending a half a thread laughing at YOUR OWN READING MISTAKE while trying to call other people idiots.

Every single discussion on cooling I've seen for CPU's and GPU's for a decade go along the same line... it scales fairly well TO A POINT, then it needs massively increasing voltage for such a small clock speed bump its just not worth it, under air, water, ln2 cooling, it's funny how EVERY chip I've seen talked about for a decade has displayed this trait....... odd.

 

[Duff-Man]

Okay, this is the last time I try to dissect your nonsense. It simply isn't worth my time.

Here's a hint, if you can get 800Mhz at 0.9v, 900Mhz at 0.95v, 1000Mhz at 1.05v, 1100Mhz at 1.2v, 1200Mhz at 1.5v, etc, etc....... you'll find that is most definately exponential.

First of all, that is not an exponential function. An exponential function is one where the function grows as a power series, i.e:

f(x) = k*a^x + const

But lets pretend for a second that it is. What you're saying is that IF these were the voltages required for a stable overclock, then they would grow exponentially. Well, what evidence do you have to suggest that your numbers are reflected in reality?! There certainly is no reason, in terms of electrical physics, that you would see an exponential growth of voltage with clockspeed.

His entire diatribe is both hilariously stupid and hilariously wrong as is everything he ever responds to me. I called him stupid and got banned because him and his dupe account both laughed at what can really nicely be put as, complete inability to read. I stated that I specifically DID NOT have to post a mathematical proof to be able to then state that 9.8 the value for gravity. TO which I think there were maybe 4 posts of Krugga and Xsistor incorrectly assuming I said that WAS a mathematical proof.

Every single time Xsistor has said I was wrong, he's been incapable of reading what I said. Which makes you basing your arguments on what he said as.. misguided.

I have not based anything I've said to you on anything that xsistor, or anyone else, has said. And what has any of this above got to do with anything?

As for your diatribe of more nonsense, no, shaders are, both complex and very simple, you can without question speed up ANY calculation by adding more complex units that do several calculations in one, magnitudes faster, the issue is always, how much faster, how often is it used, overall what is the better option, more smaller wider use compute parts or less, bigger, specific and not always useful parts. Saying you can't proves you know absolutely nothing.

Just because better algorithms would speed up fluid dynamic calculations on gpu's, DOES NOT EQUATE to better designed, more complex and more specific hardware couldn't do it faster.

This is the case for almost anything you do on a computer. It's really as simple as this, you can add 1 + 1 + 1, or have a more complex shader that can do that three times. I'd be not shocked but downright amazed if there wasn't a single calculation that couldn't be speed up.

But that is where AMD/Intel with cpu's and gpu's are at, multifunction, simple enough that almost any software can be ported to work on it even if many calculations take many clocks and many separate calculations.

Isn't this the entire principle of GPU compute? That you take away fixed function units in order to allow a wider range of use?

Of course you can design processors to perform simple, predictable tasks more efficiently. But the entire point of GPGPU applications is that they're inherently unpredictable in terms of their specific needs. That's WHY you design a flexible compute architecture in the first place!

This really isn't rocket science...

You know, as far as dataloads go, 3D rendering is incredibly predictable and uniform. Geometry processing comes from a triangular mesh loaded in fixed format. Pixels are passed individually, and have relatively simple queues of pre-defined multiply-add operations applied. And still, with all this uniformity, we see fairly large variations in performance between architectures (i.e. in games).

The range of dataset conditions encountered in GPGPU compute, even within a fairly tight field such a fluid dynamics, is just immense by comparison. You couldn't possibly design fixed function hardware to cope efficiently with all the needs. Sure - you could design a GPU to perform very efficiently on a single specific code running a single specific model - BUT, as soon as you try to run anything else your needs would change so dramatically that the highly-tuned, fixed function hardware would be all-but useless.

As an example, using the exact same commercial CFD code: What I would need from a GPU when running a 1-million cell mesh for a laminar flow simulation is completely different to what I need when running a 20-million cell mesh with turbulence. The balance between raw processing power and inter-cell communication is completely different. It simply isn't possible to write fixed-function hardware to cope with the entire range of requirements. Hence the need for GPGPU, and the reason AMD and Nvidia have been investing billions (of dollars and transistors) in GPGPU compute.

 

[xsistor]

Show me an equation showing exponential voltage-frequency relationships. You can't because there isn't one. The examples I show are counter-examples to show how ridiculous your statements are (but I also showed you the CMOS power equation which admits only a quadratic f-v relationship). As what you made is a mathematical statement, I showed you counter-examples of why this statement can't hold. It's used in a perfectly valid formal mathematical proof technique called proof by contradiction. There are plenty of examples that contradict what you said, and I've shown you many that can easily be understood even by those without a grasp of engineering (only requiring some pre-calculus/calculus background-- even a conceptual grasp being sufficient). You obviously have no clue about exponential functions or you wouldn't be talking this garbage. What a waste of time. Talking to you is like talking to a goat.

 

[drunkenmaster]

Isn't this the entire principle of GPU compute? That you take away fixed function units in order to allow a wider range of use?

No, and that's the whole point, GPGPU's ALREADY have MANY hardware accelerated parts in them, PART of them(the shaders take up a smidge over half the transistors iirc) are multifunction, a lot AREN'T and are for very specific functions. A GPGPU is a MIX of these to get the best overall performance. If you could speed up half of your fluid calculations by 70% with 1/4 die increase with fixed function units, which wouldn't increase the other at all, or you could add more multifunction shaders that would speed up everything by say, 20%. Well this is the thing, there will be many people who find and overall increase in performance with method one, and many people that wouldn't see any increase at all unless with method two.

Its all about, who wants the cards, how much they will pay, how much a feature takes up size wise, what it speeds up and how much. CPU's have fixed function hardware and constantly add new bits, in terms of quicksync you've got a crapload of fixed function stuff that takes up a relatively large amount of die space that could have gone to other things. But for Intel obviously most people both require a gpu and also benefit from video hardware encode/decode functions that it was more than worth it.

Again the initial argument was more about the article being completely wrong, when in reality its not particularly likely but could EASILY speed up certain tasks, the ARTICLE mentioned fluid dynamics therefore I commented on them, its much more likely to be Physx and physx based fluid dynamics, I was merely providing an example of both saying I think its unlikely but a possible reason Nvidia would invest that much die space.... the people who WOULD pay for that speed up.

As an example, using the exact same commercial CFD code: What I would need from a GPU when running a 1-million cell mesh for a laminar flow simulation is completely different to what I need when running a 20-million cell mesh with turbulence. The balance between raw processing power and inter-cell communication is completely different. It simply isn't possible to write fixed-function hardware to cope with the entire range of requirements. Hence the need for GPGPU, and the reason AMD and Nvidia have been investing billions (of dollars and transistors) in GPGPU compute.

This being the point, I didn't suggest anywhere they would somehow include a fix function for everything, but we're talking about balance, what if a very small portion of the die could help drastically with ONE PART that speed it up to the tune that all together it was worth it.

I doubt one particular segment of the industry would make it worthwhile using the specific functions required, the question is what functions can they introduce to speed up enough things where by the die size cost is worth it, this is a question for MANY features EVERY generation.

The likelyhood is that physx acceleration ON DIE could drastically improve PHYSX performance, the very simple question is, can they use that well enough that its worth the die size cost, is the die size cost even that big, and potentially IF the die size cost was SO big it would actually harm gaming performance enough that it wasn't worth it, would potentially added fixed fuction units help them sell more GPGPU's.

Again read the article quoted and Rroff's post, you and xsistor react to every post like I'm automatically making some claim, I didn't, I WONDERED something about WHAT COULD MAKE IT PROFITABLE and worthwhile, nothing more or less, and also to point out that while I fully got that its not particularly likely I thought Rroff was slightly misunderstanding what was being said in the article..... which does come across as (make physx work on software on the GPU) as Rroff took it, but I think was really saying, stick physx hardware on die to accelerate it greatly.

 

[drunkenmaster]

Show me an equation showing exponential voltage-frequency relationships. You can't because there isn't one. The examples I show are counter-examples to show how ridiculous your statements are (but I also showed you the CMOS power equation which admits only a quadratic f-v relationship). As what you made is a mathematical statement, I showed you counter-examples of why this statement can't hold. It's used in a perfectly valid formal mathematical proof technique called proof by contradiction. There are plenty of examples that contradict what you said, and I've shown you many that can easily be understood even by those without a grasp of engineering (only requiring some pre-calculus/calculus background-- even a conceptual grasp being sufficient). You obviously have no clue about exponential functions or you wouldn't be talking this garbage. What a waste of time. Talking to you is like talking to a goat.

Your example was a counter argument to an argument I both never made, and one you see in your own head that I was making.

Again post where I claimed there was an equation to prove this, in reality using a GPU this is what you find for MANY reasons all added together.

Mr wants to be scientist, your counter proof was not proof, it is an argument to counter prove something UNRELATED that I DIDN'T CLAIM.

EDIT:- I'm still awaiting a response to the post where you both, claimed Charlie was a joke with ANY proof he is somehow always wrong, an explanation of what Charlie was refering to with his Ghandi comparison that makes my implication absolutely outlandish and inaccurate, and a response to how exactly I was blowing my own trumpet by refering to someone else's rumour......

It's funny, when you talk crap and get called on it..... you again fight another argument and prove some random claim as false...... when no one made the argument or the claim and then claim yourself to be victor.

Got any response to the whole, when you managed to read "I don't need to post a mathematical proof to STATE that gravity is 9.8m/s2" and you somehow, misread this(exactly as ONE other person did at the same time...... no not hinting at you having a dupe account to agree with yourself at all) as me stating this was a mathematical proof....... I still can't understand how ANYONE could misread that.

 

[Jakus]

The actual relationship between voltage and frequency in a circuit is an engineering design optimization issue and is much more complicated. it depends on resistive, capacitive and inductive impedances, skin effect and other transmission line effects, slew rate and clock skew, white noise, thermal noise, power, etc.

Think you might have meant "Inductive reactance" , but you are correct, upping the voltage does not make the chip run faster, it just helps to overcome the induced issues of trying to make it run faster, and for sure it's too complex to be able to have any exponential or even linear correlation to applied voltage.

 

[xsistor]

Your example was a counter argument to an argument I both never made, and one you see in your own head that I was making.

Again post where I claimed there was an equation to prove this, in reality using a GPU this is what you find for MANY reasons all added together.

Mr wants to be scientist, your counter proof was not proof, it is an argument to counter prove something UNRELATED that I DIDN'T CLAIM.

EDIT:- I'm still awaiting a response to the post where you both, claimed Charlie was a joke with ANY proof he is somehow always wrong, an explanation of what Charlie was refering to with his Ghandi comparison that makes my implication absolutely outlandish and inaccurate, and a response to how exactly I was blowing my own trumpet by refering to someone else's rumour......

It's funny, when you talk crap and get called on it..... you again fight another argument and prove some random claim as false...... when no one made the argument or the claim and then claim yourself to be victor.

Got any response to the whole, when you managed to read "I don't need to post a mathematical proof to STATE that gravity is 9.8m/s2" and you somehow, misread this(exactly as ONE other person did at the same time...... no not hinting at you having a dupe account to agree with yourself at all) as me stating this was a mathematical proof....... I still can't understand how ANYONE could misread that.

I'm not interested in beating a dead horse. My arguments have been made. Your rebuttals fail to refute them. Anyone with any knowledge on the topic will see that. Anyone (and by anyone I mean you) who won't won't see it anyway, even if I were to write 50 pages -- which I'm not interested in doing. You'll always try to weasel your way out of stuff.

Think you might have meant "Inductive reactance" , but you are correct, upping the voltage does not make the chip run faster, it just helps to overcome the induced issues of trying to make it run faster, and for sure it's too complex to be able to have any exponential or even linear correlation to applied voltage.

Yes it's a reactance. But reactances are also impedances. I might've said reactance if that comma separated list didn't start with "resistive" which is not a reactance. Then again, it's context. In small signal analysis it's common to see specific references to capacative and inductive reactances, but in transmission line/stripline/microstrip/etc analysis it is typical to see all referred to purely as impedances. And ultimately it doesn't matter as the mathematics handles the intricacies (i.e. the impedance is either pure real, complex (real and imaginary), or pure imaginary rather than 'resistive' or 'reactive').

 

[straxusii]

I came in here to look at some Kepler discussion, instead I'm in physics 101!

 

[mountaincow]

I came in here to look at some Kepler discussion, instead I'm in physics 101!

So did I! This is far more entertaining though.

 

[chief barker]

So did I! This is far more entertaining though.

No it's not....i'm trying to read it after getting up from a night shift.

 

[J.D]

No it's not....i'm trying to read it after getting up from a night shift.

Then go have a coffee/tea/drink or something to eat and wake up before reading?.

 

[crinkleshoes]

Forum wars... any chance there's some real information about the GTX680/780 lurking somewhere in there that I've skimmed over?

 

[d_brennen]

Forum wars... any chance there's some real information about the GTX680/780 lurking somewhere in there that I've skimmed over?

Don't hold your breath

 

[CAT-THE-FIFTH]

Forum wars... any chance there's some real information about the GTX680/780 lurking somewhere in there that I've skimmed over?

Summary of rumours in thread so far:
1.)The first Kepler models will be released in either late February,or late March/early April or June 2011
2.)It will be faster than an HD7970,as fast as an HD7970,slightly slower than an HD7970 or slightly faster than an HD7950
3.)It will be priced around the same level as a lower end GTX570 or higher end GTX560TI or priced higher
4.)It will either have additional hardware for PhysX or not
5.)It will have a higher TDP than an HD7950 or a lower one
6.)Nvidia is not worried about GCN or is worried about GCN
7.)The GPU being released first is meant to be smaller than the Tahiti GPU used in the HD7900 series

 

[Emlyn_Dewar]

Solid concrete information on Kepler gained from this thread so far... None.
Actually that's not true, we've gathered that it has something to do with Nvidia...

 

[crinkleshoes]

Haha

 

[MichaelM]

Solid concrete information on Kepler gained from this thread so far... None.
Actually that's not true, we've gathered that it has something to do with Nvidia...

!!!
Careful, they might think you know someone who broke the NDA.

 

[d_brennen]

NDAs are easy to keep if you have nothing in your hands to blab about