How to convince myself to go red

[ICDP]

That is a silly way of looking at it. What if the gtx 670 you buy is a terrible overclocker, but the 680 that you might have bought instead was a great clocker. You then might have had spent more but got a card a good 30% faster. Obviously the opposite might happen and you could have saved a decent amount but it is all just a lottery.

Therefore it's only really worth comparing them stock for stock in my opinion(unless it's a factory overclock).

Therefore the 770 is a decent amount faster than a 7950. Even a 660ti or 760 can match a stock 7950.

I did state that I was referring to potential OC, so I already conceded it is a lottery. I'm sorry but it's not silly to compare overclocking potential of the graphics cards available. You are on the Overclockers GPU forum, do you think it odd that people compare OC vs OC here? You might compare stock vs stock and conclude that "shock horror" the £100 more expensive card is faster but most people here will look at overclocking as a massive factor in their purchasing decision.

My point was never that GTX 670 or HD 7950 are as fast as a GTX770 or HD 7970GE stock, I'm saying that with decent overclocks on all cards they will be so close you would need an FPS counter to tell the difference. So paying the extra £100 is nowhere near worth the extra 10% performance.

GTX670 and HD 7950 stock clocks are actually artificially lower than GTX680/770 or HD7970 GE. So GTX670 and HD 7950 will close the gap quite significantly when all the cards are set at similar clocks.

 

[thisiswar88]

I think the original update has clocks of 800/925, but most (if not all) released cards will be a bit higher at stock.

yeah most of them are factory oc'd to 925mhz. Anyway like i have said in previous posts a 7950@1100mhz is like 5 fps behind the 770. LinusTechTips done a review on the 760 @1080p pretty interesting. It looks like AMD are going to beat nvidia with there new series. Nvidia can't even get there drivers right for the 700 series there's a tone of people complaining about crashes/artefact's. I don't blame people getting annoyed at nvidia you would think the would be more organized for there new cards. All nvidia is good at is smooth frames that's about it really. There drivers are soooo over hyped.

 

[gpuerrilla]

I did state that I was referring to potential OC, so I already conceded it is a lottery. I'm sorry but it's not silly to compare overclocking potential of the graphics cards available. You are on the Overclockers GPU forum, do you think it odd that people compare OC vs OC here? You might compare stock vs stock and conclude that "shock horror" the £100 more expensive card is faster but most people here will look at overclocking as a massive factor in their purchasing decision.

My point was never that GTX 670 or HD 7950 are as fast as a GTX770 or HD 7970GE stock, I'm saying that with decent overclocks on all cards they will be so close you would need an FPS counter to tell the difference. So paying the extra £100 is nowhere near worth the extra 10% performance.

GTX670 and HD 7950 stock clocks are actually artificially lower than GTX680/770 or HD7970 GE. So GTX670 and HD 7950 will close the gap quite significantly when all the cards are set at similar clocks.

Well said. I think it is safe to say some are brand snobs no matter what way you look at it.

Value is the best performance for price. How people skew this is beyond me.

 

[spoffle]

That is a silly way of looking at it. What if the gtx 670 you buy is a terrible overclocker, but the 680 that you might have bought instead was a great clocker. You then might have had spent more but got a card a good 30% faster. Obviously the opposite might happen and you could have saved a decent amount but it is all just a lottery.

Therefore it's only really worth comparing them stock for stock in my opinion(unless it's a factory overclock).

Therefore the 770 is a decent amount faster than a 7950. Even a 660ti or 760 can match a stock 7950.

The issue with your observations here is that it's much more likely that you will get a 7950 that will overclock very well than a 770/670 as you can adjust the voltages on a 7950.

Therefore, overclocking is less of a factor for nVidia cards and more of a factor for AMD as voltage tweaking just isn't an option.

So getting a decent clocker is a much bigger gamble with nVidia, whereas it seems that it's much more likely than not that you'll get a 7950 that can do 1100Mhz on the core, which is in excess of 770 performance.

 

[Kaapstad]

The issue with your observations here is that it's much more likely that you will get a 7950 that will overclock very well than a 770/670 as you can adjust the voltages on a 7950.

Therefore, overclocking is less of a factor for nVidia cards and more of a factor for AMD as voltage tweaking just isn't an option.

So getting a decent clocker is a much bigger gamble with nVidia, whereas it seems that it's much more likely than not that you'll get a 7950 that can do 1100Mhz on the core, which is in excess of 770 performance.

There are three GTX 770s on the heaven bench thread that are above my Asus Matrix Platinum HD 7970 (which allows overvolting up to 1.4v). My HD 7970 in turn sits above all the HD 7950s, I would say the GTX 770s are quite capable of seeing off all the HD 7950s and most of the HD 7970s.

Your argument is busted

You really should give NVidia cards credit where they deserve it, despite your dislike for NVidia.

Remember the Heaven 4 bench is more demanding on GPU graphics performance than any of the other benches.

 

[ICDP]

There are three GTX 770s on the heaven bench thread that are above my Asus Matrix Platinum HD 7970 (which allows overvolting up to 1.4v). My HD 7970 in turn sits above all the HD 7950s, I would say the GTX 770s are quite capable of seeing off all the HD 7950s and most of the HD 7970s.

Your argument is busted

You really should give NVidia cards credit where they deserve it, despite your dislike for NVidia.

Remember the Heaven 4 bench is more demanding on GPU graphics performance than any of the other benches.

I agree with your point that a GTX770 will also overclock, but the problem is there is less overhead compared to a lower clocked GTX670 or GTX680. GTX680, GTX670 and GTX770 all use the same core GPU and as such have almost identical max OC potential. At best you might get a ~10% performance difference when all three are overclocked to max potential. For the record (I know you were replying to Spoffle), my position is that while the GTX770 or HD 7970 will remain faster than a GTX670 or HD 7950, you would need to run an FPS counter to tell the difference.

A GTX770 is ~17%-20% faster than a stock GTX670 depending on the game tested.

Stock GTX670 is 915MHz core 6GHz VRAM
Stock GTX770 is 1046MHz core 7GHz VRAM

So straight away most of the performance advantage is down to higher core and VRAM clock speeds. If we consider good OC potential, both will reach a similar core clock of ~1270+ and 7GHz-8GHz VRAM. Right away that ~20% performance advantage is eroded and might only be ~10%.

I would also hold off declaring a winner based on some almost pointless synthetic benchmarks knowing they are in no way representative of most actual game engines. Why do you think any serious review sites avoid them like the plague? While they have their uses they aren't the force they used to be when deciding the best GPU for your needs. They aren't even good for finding stable overclocks anymore. So many times I could run Heaven 4.0 overnight as a stress test, only for my "stable" overclock to lock up/artifact after 20 minutes playing Crysis 3.

To backup Spoffles remarks, my own tests show that a GPU will have more OC potential if it has unlocked voltage controls. In general that means unlicked voltage Tahiti based cards (or early BIOS MSI Lightning GTX680) will give higher and more stable overclocks.

 

[ICDP]

Another very important point to remember is that the new Nvidia boost (and to a lesser degree AMD boost) renders sythetic benchmarks almost pointless when it comes to measuring actual real world gaming performance. 6x0, 7x0 series GPUs will dynamically reduce clocks to maintain a set temperature. A five minute benchmark will be over long before this throttle point is reached. It is very, very common for someone doing synthetic benchmarks to run much higher clocks for the short duration the benchmark is running. Their objective is to score highly regardless of if the overclock is stable 24/7.

I would frequently see my 1370 core clock on my GTX680 Lightning drop to 1280 after prolonged periods of Far Cry 3 or Hitman Absolution. Running a short Heaven benchmark showed no such throttling and as such was not representative of the GPUs actual gaming performance.

 

[Kaapstad]

I agree with your point that a GTX770 will also overclock, but the problem is there is less overhead compared to a lower clocked GTX670 or GTX680. GTX680, GTX670 and GTX770 all use the same core GPU and as such have almost identical max OC potential. At best you might get a ~10% performance difference when all three are overclocked to max potential. For the record (I know you were replying to Spoffle), my position is that while the GTX770 or HD 7970 will remain faster than a GTX670 or HD 7950, you would need to run an FPS counter to tell the difference.

A GTX770 is ~17%-20% faster than a stock GTX670 depending on the game tested.

Stock GTX670 is 915MHz core 6GHz VRAM
Stock GTX770 is 1046MHz core 7GHz VRAM

So straight away most of the performance advantage is down to higher core and VRAM clock speeds. If we consider good OC potential, both will reach a similar core clock of ~1270+ and 7GHz-8GHz VRAM. Right away that ~20% performance advantage is eroded and might only be ~10%.

I would also hold off declaring a winner based on some almost pointless synthetic benchmarks knowing they are in no way representative of most actual game engines. Why do you think any serious review sites avoid them like the plague? While they have their uses they aren't the force they used to be when deciding the best GPU for your needs. They aren't even good for finding stable overclocks anymore. So many times I could run Heaven 4.0 overnight as a stress test, only for my "stable" overclock to lock up/artifact after 20 minutes playing Crysis 3.

To backup Spoffles remarks, my own tests show that a GPU will have more OC potential if it has unlocked voltage controls. In general that means unlicked voltage Tahiti based cards (or early BIOS MSI Lightning GTX680) will give higher and more stable overclocks.

I agree with nearly everything you have said.

I do definitely agree about needing a FPS counter to tell the difference between cards and think people get to obsessed with 1 or 2 extra FPS. I recently ran the Tomb Raider bench for another thread on my Titans and it looked exactly the same on one Titan as it did on all 4 even though the difference in fps was 76fps on one compared to 268fps on all four. I would have still needed a fps counter to tell the difference.

I think someone would have to have a really golden HD 7950 if they were going to catch a GTX 770. Having said that though it would be hard to tell which one was being used without a fps counter and then you would need to run several games to be sure.

I think Spoffle needs to step back from his dislike of NVidia and give more credit to their products when they perform well.

 

[Rroff]

even though the difference in fps was 76fps on one compared to 268fps on all four. I would have still needed a fps counter to tell the difference.

I think someone would have to have a really golden HD 7950 if they were going to catch a GTX 770. Having said that though it would be hard to tell which one was being used without a fps counter and then you would need to run several games to be sure.

Once you get upto around 80fps a lot of people's ability to gain anything from higher framerates drops off a lot, once you hit 120fps you've weeded out most casual players. However what your buying high end GPUs (and I hope a good part of the reason you bought multiple Titans) is to get the best possbile minimum framerates - there can be a big difference in the experience between 2 cards that might be averaging say 68 and 72fps but one hits minimums of say 31 and the other 38.

I wish more GPU reviewers would present atleast one lot of benchmarks, even if they don't do it for every test, with numbers for how long each card was below 30fps and how long below 60fps as those are the 2 of the more useful statistics to know when comparing to otherwide similiar cards.

 

[Kaapstad]

Another very important point to remember is that the new Nvidia boost (and to a lesser degree the AMD boost) renders sythetic benchmarks almost pointless when it comes to measuring actual real world gaming performance. 6x0, 7x0 series GPUs will dynamically reduce clocks to maintain a set temperature. A five minute benchmark will be over long before this throttle point is reached. It is very, very common for someone doing synthetic benchmarks to run much higher clocks for the short duration the benchmark is running. Their objective is to score highly regardless of if the overclock is stable 24/7.

I would frequently see my 1370 core clock on my GTX680 Lightning drop to 1280 after prolonged periods of Far Cry 3 or Hitman Absolution. Running a short Heaven benchmark showed no such thorttling and as such was not representative of the GPUs actual gaming performance.

I can not comment on GTX 670/80s as I have not used them but the GTX 690 is a little different. The GTX 690s run hot by default as the temps go over 80c and even 90c they will lose 13mhz each time off of their max boost available but they don't throttle like the other 6 and 7 series cards, they hold a constant clockspeed.

I did not really understand what people were going on about with this throttling business until I got my Titans. As soon as they hit 80c the clocks nose dived. This is easy (if a pain in the behind) to get over by setting a modded fan profile or raising the temp limit or both on the Titan and 7 series. Having said that when I watercooled my Titans one of the best benefits was no longer having to mess with fan profiles.

I think the GTX 770 should be able to hold a constant high overclock by using a modded fan profile/raising the temp limit but until I get my hand on one and try it I don't know for sure. Perhaps someone who owns one could answer this.

 

[Kaapstad]

Once you get upto around 80fps a lot of people's ability to gain anything from higher framerates drops off a lot, once you hit 120fps you've weeded out most casual players. However what your buying high end GPUs (and I hope a good part of the reason you bought multiple Titans) is to get the best possbile minimum framerates - there can be a big difference in the experience between 2 cards that might be averaging say 68 and 72fps but one hits minimums of say 31 and the other 38.

I wish more GPU reviewers would present atleast one lot of benchmarks, even if they don't do it for every test, with numbers for how long each card was below 30fps and how long below 60fps as those are the 2 of the more useful statistics to know when comparing to otherwide similiar cards.

Yes high minimums was a big part of it.

When it comes to min avg and max fps for me I find it more annoying if there is a wide variation. If I run Crysis 3 @1600p maxed on a single Titan I get about 22fps but for me it is quite playable as the min and max are also pretty close to 22fps as well and everything is constant. What I find really annoying is if something is running at well over 100fps and I get a dip for a split second even if this only happens every few minutes.

BTW I am rubbish at Crysis 3 if I was a good player I might be more concerned about higher fps.

 

[yannis76]

I have both 7970 and 760. Apart from physx in very few games, i would say there is no difference between the 2 brands. Of course 7970 is faster than 760 but regarding the drivers, i see no big difference. I would choose the one with the best, most silent coolig system at the best price.

 

[ICDP]

I think the GTX 770 should be able to hold a constant high overclock by using a modded fan profile/raising the temp limit but until I get my hand on one and try it I don't know for sure. Perhaps someone who owns one could answer this.

I'm not so sure it would be that different to how a GTX680 already works. As you already mentioned, when we overclock a GK104 based card we are telling it to attempt to attain a desired boost, you are not setting a static clock speed in the classic sense. You can raise the temp limit and power limit to help offset any throttling but it will still throttle in graphicaly intensive games.

Once the temperature limit is reached your GPU throttles to the next lowest max boost. The longer you play the hotter the GPU gets and the lower the max overclock. Of course it will stabilise at some point but depending on the game it could be quite a bit lower than the "stable" overclock you assume you are getting.

I would frequently notice my max overclock speed on a GTX6x0 series card dropping as much as 50-90MHz during a prolonged gaming session in a very demanding game such as Crysis 2 or FC3.

 

[oldestgregg]

If you have to talk yourself into something that isnt super important or essential-dont do it. Just buy green if that is what you are drawn towards.

 

[Kaapstad]

When you overclock a GK104 based card you are telling it to attempt to attain a desired boost, you are not setting a static clock speed in the classic sense. You can raise the temp limit and power limit to help offset any throttling but it will still throttle in graphicaly intensive games.

Once the temperature limit is reached your GPU throttles to the next lowest max boost. The longer you play the hotter the GPU gets and the lower the max overclock. Of course it will stabilise at some point but depending on the game it could be quite a bit lower than the "stable" overclock you assume you are getting.

I would frequently notice my max overclock speed on a GTX6x0 series card dropping as much as 50-90MHz during a prolonged gaming session in a very demanding game such as Crysis 2 or FC3.

What you are describing is how the GTX 670 and GTX 680 works, what I found with the Titan is providing you keep it below 80c or raise the temp limit it will maintain a constant overclock and I think the GTX 770 is the same. Having said that we need someone to test it.

 

[krooton]

I've only ever had 2 Nvidia cards.

A Riva TNT2 (first standalone gfx card ever) and an 8800GT.

The rest have all been ATI/AMD.

Out of loyal fanboyism? Nope, purely because at every point I was buying, I needed the best bang for buck factoring in overclocking.

Every time it just happened to be ATI, apart from the stonkingly good deal on the 8800GT, which I was very happy with.

As it stands, unless there is a better value option at the time, my next step is a second 7950 when the prices drop.

 

[ICDP]

What you are describing is how the GTX 670 and GTX 680 works, what I found with the Titan is providing you keep it below 80c or raise the temp limit it will maintain a constant overclock and I think the GTX 770 is the same. Having said that we need someone to test it.

My apologies, I am not making myself clear I think. The GK104 works in a similar way you are describing for the Titan. The user sets a desired boost clock along with fan profiles and power settings. The GPU then attempts to maintain the desired boost and if it reaches a pre-defined temperature it will automatically drop the core clock speed to avoid the GPU going over that predefined temperature.

My point is that in synthetic benchmarks it is possible to set much higher boost clocks (or overclock on a Tahiti card) that could not be maintained in a prolonged gaming session. I was using it to show that synthetic benchmarks do not show how a card will necessarily perform in actual games. If someone is making their purchasing decision based on a synthetic benchmark table on any forum, they are doing it wrong

Edit: I will add that I think Nvidia boost is fine for keeping a GPU more stable. Previously if you set an aggressive overclock your GPU would just crash and lock up after it overheated. It isn't foolproof as the GPU can still be pushed too hard but it does provide some failsafe.

 

[LtMatt]

One of the review sites picked up on this. For the life of me i can't remember who it was but it was noted on 770/760 that after a while the clock speed kept getting lower and lower. Obviously good for showing off in benchmarks but not ideal for long term sustainable performance.

 

[arc@css]

I don't know all that much about boost technology, though from what I've read I think I would prefer the static overclocking we used to see, not all that long ago.

 

[Kaapstad]

My apologies, I am not making myself clear I think. The GK104 works in a similar way you are describing for the Titan. The user sets a desired boost clock along with fan profiles and power settings. The GPU then attempts to maintain the desired boost and if it reaches a pre-defined temperature it will automatically drop the core clock speed to avoid the GPU going over that predefined temperature.

My point is that in synthetic benchmarks it is possible to set much higher boost clocks (or overclock on a Tahiti card) that could not be maintained in a prolonged gaming session. I was using it to show that synthetic benchmarks do not show how a card will necessarily perform in actual games. If someone is making their purchasing decision based on a synthetic benchmark table on any forum, they are doing it wrong

I see where you are coming from, Titans are pretty good at maintaining their temps over a prolonged period, probably because they are not clocked that high in the first place. But GTX 770s are clocked very high out of the box and thinking about it I can see the temps creeping up over a prolonged period.

It would be interest if someone like whyscotty who owns a GTX 770 could test this.