7970 Lightning now available

[railven]

That's not the point, the 1300mhz overclock is not a 30% overclock since reference cards boost to 1110mhz. This means we shouldn't see scaling beyond 17% in the first place, which is exactly what the benchmarks are showing.

Wait, let me get this right...so I understand the arguments better...

We count the OC only after the initial OC from GPU Boost? Can i ask why? If the card is set stock at 1006mhz, to me that is stock. The Boost will vary depending on various conditions. So, if I have a GTX 680 and I'm a newb and stick it into a Dell case, just playing WoW I'm hitting +95c on the card - what's my GPU boost?

If I decide to get a nice HAF-X and now my card is in WoW 65c - what's my boost?

And if I overclock it, where do I start counting from the 95c scenario or the 65c scenario?

Just curious because this is starting to get murky.

 

[aaksheytalwar]

Bottomline
At stock 680 is slightly better in most games (not all), often >10% but definitely not enough for you to side grade from a 7970 especially if it is overclocked. Also, this difference tends to finish or become nearly 0 or 0 at 1440p or 1600p, but at 1080p is still a bit there, but since both are fast enough it doesn't matter.

Once both are overclocked, it is a tie. They trade blows and sometimes one wins sometimes the other.

On water cooling 7970 is probably better by an unnoticeable margin but that isn't a proven fact yet.

The 7970 is ideal at 460-480 bucks whereas 500 is the max they should consider. $501 for the 7970 makes no sense because the 680 doesn't lack anywhere and sells at $500.

If AMD and nVidia were equally stron, $500 would be okay. But since AMD is weaker in terms of sale, $450 is the sweet spot for them.

 

[BallaTheFeared]

We count the OC only after the initial OC from GPU Boost? Can i ask why? I

Because it's a boost and part of the stock performance, it is not an overclock.

 

[railven]

Because it's a boost and part of the stock performance, it is not an overclock.

So where do you count an OC on an Intel processor? Again, I'm just curious since this is getting murky.

And, if everything I read right is accurate, my Dell scenario doesn't give the GTX 680 any boost.

EDIT: Goddamn time warps! haha.

 

[notty22]

My I5 - 750 if run with turbo boost is never a 2.66 cpu, it always would run with at least 1 extra bin of boost or 2.8ghz
As I understand it, the gtx 680 will always run at 1056 or there abouts and not 1006.
You could probably force that, but it's not it's stock behavior.
It's stock behavior is like my Intel example I believe.

 

[BallaTheFeared]

The 680 runs at whatever speed it can within it's allotted TDP limit.

That means from game to to game you can't even count on the same stock clock speeds.

 

[railven]

The 680 runs at whatever speed it can within it's allotted TDP limit.

That means from game to to game you can't even count on the same stock clock speeds.

Of course you can because it is registered to operate safely at 1006. They found this by adjusting the temperature on the card while it ran the same game. As the temperature increased the the GPU Boosts dropped.

So to clear this up - where does one start measuring the OC? Because in the two scenarios I gave, the final clock read will be completely different.

Do we peg it to a moving target or to a "stock" reading set by the manufacturer?

 

[BallaTheFeared]

Of course you can because it is registered to operate safely at 1006. They found this by adjusting the temperature on the card while it ran the same game. As the temperature increased the the GPU Boosts dropped.

So to clear this up - where does one start measuring the OC? Because in the two scenarios I gave, the final clock read will be completely different.

Do we peg it to a moving target or to a "stock" reading set by the manufacturer?

If it's moving it's moving.

There are many factors that increase TDP, a hot GPU is one of them. With Kepler things that increase power draw reduce performance, games themselves have different power requirements thus changing from game to game, from area to area, it is what it is.

If you can't pinpoint something then you can't make up a number to use as a reference point.

 

[railven]

If it's moving it's moving.

There are many factors that increase TDP, a hot GPU is one of them. With Kepler things that increase power draw reduce performance, games themselves have different power requirements thus changing from game to game, from area to area, it is what it is.

If you can't pinpoint something then you can't make up a number to use as a reference point.

I didn't make up a number. That number is set by nVidia (the makers of the card.) So I guess your basically saying we can't take accurate readings for the OC since "if it's moving it's moving."

So no one can claim someone's else findings wrong since, well, there is no reference point. I sort of like where this is going haha.

 

[BallaTheFeared]

The boost, not the OC. It's not an OC because it's not overclocking past where it's meant to clock to in the first place. Hence it's called a boost.

The only constant is the TDP, and even that seems to be suspect at time.

Personally at this point I just haven't seen enough information that is consistent to really say anything on the matter. There is a decent difference between what people can get with +132% TDP "Overclocked" some can get +100, others +170, still others something else. The overclocking of Kepler is based solely on your samples ability to perform within a 225w TDP.

Yet that still has questions, because people using precision X with +132% still have the program report power usage percents at different levels, some set +132% and only see around 100% power usage with their offset, others do the same and see +125% or more power used..

 

[railven]

The boost, not the OC. It's not an OC because it's not overclocking past where it's meant to clock to in the first place. Hence it's called a boost.

I get that, but what I'm trying to determine is where people are starting to count the OC.

So far I've read you can't use the reference 1006mhz since the card boost to, seen the number thrown around 1110mhz, so all calculations on the OC have to originate on the 1110mhz.

As you just explained even that 1110mhz isn't ideal since, well, so many variables to consider. Even in the simple scenario I gave, the boost won't be 1110mhz.

It just seems either use the reference point set by the manufacturer or get ready for no one's findings to be legitimate since there is no set reference point. If John Poster wants to say "you can't use 1006, because it boosts to 1110" well then Jane Poster can easily say "but it can also boost to 1050 or even less."

Murky water is murky. With no real answer outside of "if it's moving it's moving" I'd say it's safer to use the 1006 point since it is reference on all cards - but I can see why people wouldn't want to use it since, it hurts their position. To each their own I guess.

 

[BallaTheFeared]

It would have to be case by case imo, since each card/game is different.

 

[railven]

It would have to be case by case imo, since each card/game is different.

Which makes things murky. /shrug.

You guys continue the OC wars. I'm pissed UPS won't deliver my card that according to their site is sitting in their warehouse down the street from me, and after a nice call I can't even pick it up. WTF UPS. W.T.F.

 

[VulgarDisplay]

I'm more interested in price wars than OC wars. If AMD says "Oh Yeah!?!?" and prices the 7970 at $450-475 I would fully expect nvidia to basically undercut the crap out of them because the gtx680 is so much cheaper to produce.

 

[Grooveriding]

Russian, the same thought occurred to me after my post in relation to stock benches having some boost applied to them. My cards arrived and I will see if they boost differently or the same later tonight. I unfortunately have more on my plate today than expected

Will do some benches tonight and see how they scale with overclocks.

 

[Quantos]

Probably counting the OC from the base + boost will give you the best figure. The base + boost isn't necessarily what you get, but it's what you're aiming for. There's really no way to know what you'll get depending on the application.

 

[railven]

I'm more interested in price wars than OC wars. If AMD says "Oh Yeah!?!?" and prices the 7970 at $450-475 I would fully expect nvidia to basically undercut the crap out of them because the gtx680 is so much cheaper to produce.

I'm with you. In the end all these arguments don't mean jack to me, the end user, outside of helping me pass time at work. I like to get behind the technology, and sometimes want to understand people's reasoning (you can't use X-model as a comparative, because I said so! - one of my all time favorites) for defending their position.

I truly wish AMD drops the price. I want to collect the difference and buy something. And I truly wish UPS wouldn't be jerks and delivered my GTX 680 today since I'm sitting at home with nothing to do (forever lonely...) haha.

Ah well. Maybe I'll grind another alt on le WoW.

 

[Elfear]

So if the Boost feature can continually move the clocks to stay under a certain TDP, does the same happen when manually overclocking? So if a user said "Hey my card will do 1300Mhz", it might be boosting to 1300Mhz in certain situations but in others it could be 1225 or 1250?

 

[Crap Daddy]

Whatever you do, you are not alowed to exceed the TDP on the 680. This is the limit of your overclocking. Under load an oveclocked GTX680 consumes a similar amount of Watts like a stock HD7970 while a 7970 at 1200MHz needs 100W more.

 

[RussianSensation]

Wait, let me get this right...so I understand the arguments better...

We count the OC only after the initial OC from GPU Boost? Can i ask why? If the card is set stock at 1006mhz, to me that is stock.

That's how we have always counted any product with dynamic OC --> see CPUs.
Please take 10 min to read the PureOverclock 1 page, it explains it much better than I do.

GTX680 has less overclocking room because dynamic OC already overclocks part of the way there. That has nothing to do with GTX680 having poor overclocking scaling. What it rather means is 680's have lower overclocking room to gain performance from dynamic base clocks.

GTX680 has 1006 factory clocks.
Its operational clocks are 1110mhz.
Manual overclock is 1300mhz.

How much is that overclock in the real world? 1300mhz / 1110mhz = 17%

Therefore, you should see at most a 17% performance increase from stock benchmarks, which is exactly what's being shown.

It's the same as we've always counted it with CPUs. This applies to any product with dynamic OC functionality built in. For example, while Core i7 860's stockspeed is 2.8ghz, it turbos at a minimum to 2.93ghz on all 4 cores. If I were to run all my 4 threaded benchmarks with i7 860 and get my Base scores and assign them 100% reference, then compared to those scores a manually overclocked 860 @ 3.2ghz is not 18% faster! It would correct to say that my CPU was overclocked by 14% from "stock clock of 2.8ghz" but it would be INCORRECT to compare benchmarks and look for a 14% growth since the stock 860 never operates at 2.8ghz. It operates at a minimum of 2.93ghz. The largest increase I would see is 9%. Alternatively, imagine if you ran a single threaded bench such as SuperPi. A stock i7 860 would beat a manually overclocked 3.2ghz i7 860. Using your logic, you'd get negative CPU scaling. That's because it's incorrect to use 2.8ghz as the base clock for a Turbo enabled CPU.

For mobile CPUs, the CPU boost is even more. You can have a mobile CPU boost 1ghz from 2ghz. Are you going to say that if you overclocked your CPU to 3.2ghz, that's a 1.2ghz overclock from stock? NO. That's only a 200mhz overclock from factory operational clocks.

So if you are going to look at frequency scaling, the base clock = highest frequency scaling out of the box vs. overclocked highest frequency.

We wouldn't say that a 4.0ghz 2600K is 18% faster clocked than a reference 2600K with Turbo. We would say a 4.0ghz 2600k has 18% higher frequency than 2600K operating at 3.4ghz. This is because 2600K can turboboost to 3.8ghz. So obviously you'll never see 18% scaling on 2600K @ 4.0ghz over a 2600K with Boost up to 3.8ghz.

The benchmarks are showing this exactly. The base = 100% is likely at minimum 1110mhz, not 1006. So it's 100% incorrect to compare reference benchmarks and then say it's a 30% overclock from 1006 to reference #s because the GPU never operated at 1006mhz. You would say GTX680 has less overclocking headroom since Dynamic OCing has eaten into that headroom with a 7-10% factory dynamic OC.

There is no apparent GPU scaling problem with Kepler, it's simply incorrect use of mathematics using an incorrect 1006mhz as the base for the benchmarks. What Kepler has is a lower overclocking headroom in % terms vs. 7970.

 

[RussianSensation]

Russian, the same thought occurred to me after my post in relation to stock benches having some boost applied to them. My cards arrived and I will see if they boost differently or the same later tonight. I unfortunately have more on my plate today than expected

Will do some benches tonight and see how they scale with overclocks.

NV made it really difficult to compare benchmarks now. As you will likely see shortly, you may have a situation where your 680s may not boost to the same clocks. This goes back to all the Kepler reviews - temperatures, ASIC quality/variability will determine maximum boosting. It's probably going to feel a little weird to see your 680s operating at different dynamic frequencies at load since GPU speed is no longer a constant, and neither are voltages. Crazy NV took a page out of Intel's book but instead of it being predictable such as boosting on a per core basis (1-4 threads), this is some voodoo for sure since it will vary on a per game basis and on a per individual system basis! ^_^

 

[RussianSensation]

So if the Boost feature can continually move the clocks to stay under a certain TDP, does the same happen when manually overclocking? So if a user said "Hey my card will do 1300Mhz", it might be boosting to 1300Mhz in certain situations but in others it could be 1225 or 1250?

Perfect! :thumbsup:

That means Kepler's clocks and voltages will automatically adjust on the fly, depending on the load required (or not). The card's OC and voltages will also downclock if the horsepower isn't needed, while our current cards are stuck at whatever speeds and voltages we specified in Afterburner, etc.

"The fact is that "scientific benchmarks" no longer hold any relevance to gameplay performance. The old way of thinking where GPU frequency has to remain the same is gone, there is no way to ensure or lock the GPU to the exact same frequency for every test or game. The fact is that the GPU clock speed is now dynamic, and will be changing on you in every game, and every game will be different, with a different clock speed.

A canned or synthetic benchmark is now even further removed from being relevant to gameplay performance; a benchmark won't represent the real-world clock speeds you will get in games. Since you cannot disable GPU Boost and lock the frequencies, the old way of comparing performance based purely on framerate no longer applies." - HardOCP

Yup, manual overclocking is still dynamic right now. From what I understand, you cannot fix the speed at 1300mhz. Unless new utilities figure out a way to fix the speed.

Notice how the card @ Xbitlabs pulls 1277mhz but sometimes it went to 1304mhz?

Even though the card sometimes boosted to 1304mhz, would it be fair to say it operates at 1304mhz? Imo, not really, unless it can do that in every game most of the time. The GPU frequency changes throughout the benchmark even. Making comparisons on a clock per clock basis too ambiguous.

That's why if we just compare reference 680 vs. an OCed 680, and try to derive exact frequency scaling in benchmarks, it's very difficult now. It's almost like you'd need to find the average operating "stock" frequency in a benchmark and plot that against an average "overclocked" boosted frequency in the same benchmark on the same test bed. That in itself will vary among users since GTX680's maximum boost can only work at or below 70*C. So if someone has a 1150mhz 680 and it works at 85*C, it'll be slower than an 1150mhz 680 at 70*C, but not enough to make it a deal breaker. But the key takeaway is that two 680s at any frequency may or may not perform exactly the same. Since we are talking about 1100mhz GPUs, small 20-30mhz fluctuations aren't going to be material though.

 

[BallaTheFeared]

The old way of thinking where GPU frequency has to remain the same is gone, there is no way to ensure or lock the GPU to the exact same frequency for every test or game. The fact is that the GPU clock speed is now dynamic, and will be changing on you in every game, and every game will be different, with a different clock speed.

Perhaps blue letters conveyed the point better than my gibberish

Bring on the 8+8 pin monstrosities, then GK100 please.

 

[Grooveriding]

NV made it really difficult to compare benchmarks now. As you will likely see shortly, you may have a situation where your 680s may not boost to the same clocks. This goes back to all the Kepler reviews - temperatures, ASIC quality/variability will determine maximum boosting. It's probably going to feel a little weird to see your 680s operating at different dynamic frequencies at load since GPU speed is no longer a constant, and neither are voltages. Crazy NV took a page out of Intel's book but instead of it being predictable such as boosting on a per core basis (1-4 threads), this is some voodoo for sure since it will vary on a per game basis and on a per individual system basis! ^_^

Well both my cards without any offset added boost to 1080, but the caveat, is that the boosting is highly dependent on gpu usage and load. It is a major pain to create the right conditions to see how the boost reacts. It does eat into OC headroom for sure. One card I can do 200 offset the other can do 135. So for now I've settled on 125 on both of them.

I'll probably have to play with furmark to see how gpu boost works and reacts. I've seen some people post that just at stock with no offset being used but with the power limit pushed to 132, their cards have gone up as high as 1200+

It's pretty difficult to bench a game consistently because the clocks jump around a lot. In Skyrim I run vysnc on and my second card sat around 800mhz most of the time, while the first one was 1200+.

It's a nice setup, in some games it's clearly faster than 580 tri-sli speeds, particularly BF3. The only game I've found slower numbers on with this setup is crysis/warhead benches about 10% slower than my other setup.

Man they are quiet under load though and quick. I have some blocks coming in the mail for them. At this point they have my monitor beat down for any game out there. I'll upgrade again once I get this performance out of a single card or we actually get some games that push the envelope again. :thumbsup:

 

[Avalon]

It's kind of different with CPU OC's, because generally most people disable turbo boost when they OC, at least that I'm aware of.