The GTX 780, 770, 760 ti Thread *First review leaked $700+?*

[BallaTheFeared]

You also have a .001% card

 

[Face2Face]

I think there's something about DirectX-9 that's harder on the ROPs at least on the 79xx series.

That's what makes MechWarrior Online such a good load test, since its Cryengine 3 + DirectX 9.

Only thing you have to do to reproduce is to make sure you are not CPU or Shader bottlenecked.

This is also why I think AMD/ATi dropped the ball on not doing the refresh.

AMD/ATi was well aware that the achillies heel of the 7xxx series was the Memory controller and ROPs.
The 7790 demonstrates clearly what they could have done with a refresh.
My 7970 chip would probably clock to 1150-1170+ 100% stably instead of only about 1075 if it had the ROPs and Integrated Memory Controller like the 7790 instead of what it has now.

Their 7xxx series would also all be able to use 6000 MT/s GDDR5 100% stably without ASIC binning, allowing even more efficient actual GPU utilization per clock speed.

Way over my head on this one buddy..

You also have a .001% card

Maybe? I think the people that bought the MSI 7950 TF3's when they first were released with the 7970 PCB (High ASIC Cards 85% and higher) Could maybe do the same??? Zargon and Zaydq on the forums. They would need to go for a better cooler or on water.

 

[24601]

You also have a .001% card

The same issues pop up when I'm testing both my 7970s and both my 7950s.

The cards ALWAYS run in to ROP and Integrated Memory Controller issues before they run into Shader issues.

This is with literally every single card being different in just about every way possible.

I even replicated the tests on a completely different PSU to make sure it wasn't just some sort of weird load issue and came with the exact same results.

 

[3DVagabond]

I'm starting to wonder if it will really be starting at $500 (reference) up to $600 with more memory etc. Maybe they tested sentiment with the titan and notice people didn't like the price so this will feel like a value (and potentially lure customers that would normally not buy the high end). Won't be holding my breath on this though.

I didn't see any great outcry about the price from nVidia's customers here. People still complain more about the $550 release price of the 7970.

 

[3DVagabond]

So 780 vanilla should be $549 and 770 probably $449. Within the range that I expected and not $700 like some have been saying.

The 770 needs to be more than $100 less than the 780.

 

[toyota]

Your post is what we call in the UK "gobshite"

Let me explain to you first of all that the GE is just an overclocked 7970. There is nothing different which means they CANT be called different cards. ITS THE SAME GPU.

Also you can cherry pick one websites information at 2560x1600 to support your theory. When in fact the 7970 is just as fast as the 6990. Which is 2x6970 so it must be more than 48% faster than a single 6970. Also It was a general comment about performance and im sure in some games it could be 50% and others closer to 70%.

The POINT was that performance increased ALOT as the drivers matured because GCN was new.

The GK110 ISNT NEW. Its just kepler and the drivers are very mature which means you wont see huge increases in performance like with GCN.

So keep gobbing off but im right.

Oh Anyone can do this too http://www.techpowerup.com/reviews/AMD/HD_7970_GHz_Edition/14.html

78%

http://tpucdn.com/reviews/AMD/HD_7970_GHz_Edition/images/crysis2_1920_1200.gif

80%

http://tpucdn.com/reviews/AMD/HD_7970_GHz_Edition/images/bf3_2560_1600.gif

69.4%

http://tpucdn.com/reviews/AMD/HD_7970_GHz_Edition/images/skyrim_2560_1600.gif

you are not making sense. you said the 7970 was 80% faster than 6970 with driver improvements. well the 7970 GE is NOT the same card as its already overclocked. and he compared the overall average not the handful of games where there was huge improvement.

 

[Fx1]

you are not making sense. you said the 7970 was 80% faster than 6970 with driver improvements. well the 7970 GE is NOT the same card as its already overclocked. and he compared the overall average not the handful of games where there was huge improvement.

i just listed 3 examples of where the 7970 is 70-80% faster than the 6970. Its also equal or faster than the 6990.

As far as im concerned the 7970 Ghz is 1000-1050mhz clock speed and MSI reference is 1010mhz so they are equal. Especially since you can just run it at 1050mhz 24/7 anyway. They are the same GPU its just if you want to be lazy and pay £80 for a 40mhz factory OC then what ever.

The 7970 is an awesome upgrade of the 6970 what ever you say.

 

[tviceman]

There is still a possibility that 780 might cost $500. Fingers crossed.
http://videocardz.com/41552/exclusive-nvidia-geforce-gtx-780-picture



Sure, then let's start comparing $300 HD7950 @ 1225mhz vs. GTX680/780 on price/performance. The point is HD7970 was a huge increase in performance over GTX580 OC or HD6970 OC.

Can you find more than one sentence that you can respond to? I guess it was the only one you can easily refute?

His post was two sentences long, and neither sentence had anything to do with the other.

????

 

[3DVagabond]

The 7970 was $550 at launch. From a comparison perspective, lets compare it to gtx580. According to techpowerup's original 7970 review @1600p (easy reference graphs), 7970 was 19% faster than gtx580 at launch.

If gtx780 is 15% faster than the current hd7970GE according to TPU's Titan review, that puts it at 62% faster than a gtx580.

GTX580's were going for right around $480 or so at 7970's launch. So, the 7970 at launch offered very little (if any) perf/$ when it first launched over the gtx580. If GTX780 launches at $599, that would be a substantial perf/$ improvement over what the 7970 launched at. If it launches at $649, it would be a modest improvement in perf/$. If it launches at $699, it will be a negligible improvement. Compared to 7970GE, which launched at $499, a $599 MSRP would offer a worse perf/$ over the 7970GE's launch curve.

This is all based on if gtx780 is 15% faster than 7970GE. I personally think gtx780 will end up 20% faster across the board, which would make a $599 launch identical to the perf/$ curve that the hd7970GE launched at.

Why are we comparing the 7970 to the EOL 580 price, but using the 7970 launch price for the nVidia comparisons? Going just from memory the 7970 launched at exactly the same price as what the 580 3gig was selling for at the time.

 

[blackened23]

GTX580's were going for right around $480 or so at 7970's launch. So, the 7970 at launch offered very little (if any) perf/$ when it first launched over the gtx580. If GTX780 launches at $599, that would be a substantial perf/$ improvement over what the 7970 launched at. If it launches at $649, it would be a modest improvement in perf/$. If it launches at $699, it will be a negligible improvement. Compared to 7970GE, which launched at $499, a $599 MSRP would offer a worse perf/$ over the 7970GE's launch curve.

I'm not going to get caught up in the performance argument (in the GTX 700 thread? Really? and I know that's not your fault, you didn't start it.). Anyway, just a minor correction: the 7970 3GB was actually a bit cheaper than the 3GB variants of the GTX 580. The MSRP of 3GB 580s were around 580$ at the time, and perhaps AMD was counting on price comparisons to be gauged in that manner? I don't know. Personally I think they should have made both 1.5/3GB versions at launch to at least give the appearance of a clear value winner, but alas AMD didn't. Just think if AMD had released a 1.5GB 7970 for 430$ - with proper launch drivers (which they also didn't do..) Instead they released just a 3GB card which was widely considered too expensive at launch, even though the initial price comparisons may not have been completely apples to apples. Everyone based their comparison on a 1.5GB GTX 580 to the 3GB 7970. When it comes to VRAM, I think consumer choice is the best option. Clearly 3-6GB is great for surround, but maybe for a single screen 1080p user they don't need that much!

Oh well. Hopefully AMD learned a lesson from that, but with AMD having AMD's poor management you never know.

 

[tviceman]

Why are we comparing the 7970 to the EOL 580 price, but using the 7970 launch price for the nVidia comparisons? Going just from memory the 7970 launched at exactly the same price as what the 580 3gig was selling for at the time.

I'm not going to get caught up in the performance argument (in the GTX 700 thread? Really? and I know that's not your fault, you didn't start it.). Anyway, just a minor correction: the 7970 3GB was actually a bit cheaper than the 3GB variants of the GTX 580. The MSRP of 3GB 580s were around 580$ at the time, and perhaps AMD was counting on price comparisons to be gauged in that manner? I don't know.

I was using gtx580 1.5gb prices, and I think $480 was probably on the high side of a gtx580 at 7970's launch. I didn't use 3gb prices because 3gb 580's were not the norm and came with an abnormal price premium. Just like 7970 6gb cards are not the norm, come with a large price premium, and no one uses that model to compare with Titan.

Anyways, on topic. I don't think $499 is going to happen. I think it is going to be $599. $549 is probably the maximum price Nvidia can ask to stay with the perf/$ curve of the 7970GE, but Nvidia's single fastest card has always come with a little premium over it's competitor.

 

[blackened23]

Well, fortunately consumers are getting choices with VRAM. I was very pleased that the 780 will not be a 5-6GB VRAM card as a baseline, as the price would be considerably higher.

When it comes to VRAM, consumer choice is where it's at. 2/4GB 770s and 3/6GB 780s sounds good to me.

 

[3DVagabond]

7950s do not clock to 1225 mhz 100% stably.

Me and all my bitcoin/litecoin compatriots that game as well know this all too well.

You get 1050-1100 100% stably, any higher and you have crash after crash.

Same thing for 7970s, you get 1050-1100 100% stably, any higher and you have crash after crash.

Not everyone bought a MSI 7970 Lightning. Stock VRMs on the 7970 was never very good and 80% of the current cards are sporting inferior to stock 7970 VRMs.

These cards most likely can do higher at lower volts if they were equipped with MSI 7970 Lightning VRMs and MSI 7970 Twin Frozr IV, but they are not.

All you people and reviewers out there who think your cards are stable at higher clocks than this... You are all wrong because Communism says so. No other valid reason needed. He says so, so it will be from now on. Any O/C comparisons we make must be at 1100MHz max.



Nice try, but there are too many people and reviewers who know what these cards are capable of. It's been very well documented. You and your friends must have just got crap O/C'ing cards, or aren't as capable as others at O/C'ing.

 

[tviceman]

Well, fortunately consumers are getting choices with VRAM. I was very pleased that the 780 will not be a 5-6GB VRAM card as a baseline, as the price would be considerably higher.

When it comes to VRAM, consumer choice is where it's at. 2/4GB 770s and 3/6GB 780s sounds good to me.

Do you think we could lobby MSI to use larger fans with their twin frozr heatsink? 90mm fans make so much noise when spinning up. I swear on my life that the twin frozr series back in 2010 (with the gtx400 series cards) had larger fans, because I had a gtx465 that was dead quite up to 50% fan speed. My PE gtx670 is audible at 45% and above.

I think there is room for an easy redesign of their shroud to accommodate 100mm fans. Lets do it.

 

[3DVagabond]

I was using gtx580 1.5gb prices, and I think $480 was probably on the high side of a gtx580 at 7970's launch. I didn't use 3gb prices because 3gb 580's were not the norm and came with an abnormal price premium. Just like 7970 6gb cards are not the norm, come with a large price premium, and no one uses that model to compare with Titan.

Anyways, on topic. I don't think $499 is going to happen. I think it is going to be $599. $549 is probably the maximum price Nvidia can ask to stay with the perf/$ curve of the 7970GE, but Nvidia's single fastest card has always come with a little premium over it's competitor.

You never answered the question of why you are comparing 580 EOL pricing, which I don't believe you are accurate about anyway, for the 7970, but for the 700 series you are going back to 7970 release pricing for the comparison. Why don't you use the current 7970 pricing?

 

[24601]

All you people and reviewers out there who think your cards are stable at higher clocks than this... You are all wrong because Communism says so. No other valid reason needed. He says so, so it will be from now on. Any O/C comparisons we make must be at 1100MHz max.



Nice try, but there are too many people and reviewers who know what these cards are capable of. It's been very well documented. You and your friends must have just got crap O/C'ing cards, or aren't as capable as others at O/C'ing.

There's a reason you see the people bragging about OCing to super high running at hilariously low VRAM speed or using shader bound tests to justify their clocks.

7970 and 7950 ROPs get unstable at lower clocks than the Shaders can go.

If you load up the ROPs you start getting artifacts very quickly, these artifacts also eventually lead to crashes if the program is not fault tolerant enough.

DirectX-9 seems to just be less data corruption fault tolerant than DirectX-11 when running similar code-paths.

If you think this is not the case then you must think AMD's original refresh plan would be a rebadging, since their changes to GCN with pretty much nil other than the improved ROP and Integrated Memory Controller.

 

[tviceman]

You never answered the question of why you are comparing 580 EOL pricing, which I don't believe you are accurate about anyway, for the 7970, but for the 700 series you are going back to 7970 release pricing for the comparison. Why don't you use the current 7970 pricing?

I did answer your question. I was using 1.5gb gtx580 pricing because 1.5gb gtx580's were the common card that were used in most reviews for comparison to the 7970GE. And you're kind of right, I'm probably not accurate about the price of the 1.5gb gtx580 when the 7970's launched. I think it was probably cheaper than the $480 figure I used, I was just trying to be "fair". And I did answer the other part to your question. You apparently just decided to quote my entire post without actually reading it all.

Anyways, on topic. I don't think $499 is going to happen. I think it is going to be $599. $549 is probably the maximum price Nvidia can ask to stay with the perf/$ curve of the 7970GE, but Nvidia's single fastest card has always come with a little premium over it's competitor.

In other words, at 7970GE's current prices, GTX780 would have to probably come in at $549 to match 7970GE's current perf/$.


Are you satisfied?

 

[f1sherman]

There's a reason you see the people bragging about OCing to super high running at hilariously low VRAM speed or using shader bound tests to justify their clocks.

yeah, and when they report an issue/bug along with their rock stable overclock, you have to beg them to run everything stock

 

[3DVagabond]

There's a reason you see the people bragging about OCing to super high running at hilariously low VRAM speed or using shader bound tests to justify their clocks.

7970 and 7950 ROPs get unstable at lower clocks than the Shaders can go.

If you load up the ROPs you start getting artifacts very quickly, these artifacts also eventually lead to crashes if the program is not fault tolerant enough.

DirectX-9 seems to just be less data corruption fault tolerant than DirectX-11 when running similar code-paths.

If you think this is not the case then you must think AMD's original refresh plan would be a rebadging, since their changes to GCN with pretty much nil other than the improved ROP and Integrated Memory Controller.

So, if you crank the piss out of the ram the card crashes above 1050 to 1100? Am I understanding you correctly?

Here's something to try, you push the core as far as it will go stable, then you push the RAM as far as it will go without crashing or artifacts. Not the other way around.



Completely separate advice, For BC mining underclock your RAM.

 

[BallaTheFeared]

I found my 7950 "frustrating" to overclock, and often referred to it as "finicky". I could be stable at a set clock and voltage in several games and then require a huge bump to get it stable in another.

It wasn't anything like overclocking my 470s, I don't know why that was. But this is the first time I've seen anyone attempt to explain it, pity it's really far off topic. I'd really like to see that topic explored more, but in another thread.

 

[Smartazz]

I found my 7950 "frustrating" to overclock, and often referred to it as "finicky". I could be stable at a set clock and voltage in several games and then require a huge bump to get it stable in another.

It wasn't anything like overclocking my 470s, I don't know why that was. But this is the first time I've seen anyone attempt to explain it, pity it's really far off topic. I'd really like to see that topic explored more, but in another thread.

I'm a little more liberal with my volts. I use 1.18 volts for 1100MHz, which is a bit above what I found to be the minimum to be stable.

 

[24601]

So, if you crank the piss out of the ram the card crashes above 1050 to 1100? Am I understanding you correctly?

Here's something to try, you push the core as far as it will go stable, then you push the RAM as far as it will go without crashing or artifacts. Not the other way around.



Completely separate advice, For BC mining underclock your RAM.

That advice is directly supporting my point that I've had to make several times just for you.

Thanks.

Learn how GPUs work.

 

[24601]

I found my 7950 "frustrating" to overclock, and often referred to it as "finicky". I could be stable at a set clock and voltage in several games and then require a huge bump to get it stable in another.

It wasn't anything like overclocking my 470s, I don't know why that was. But this is the first time I've seen anyone attempt to explain it, pity it's really far off topic. I'd really like to see that topic explored more, but in another thread.

Yes, if you target synthetic tests that don't stress the ROPs you can get the core clock crazy high with way less voltage. This is directly reflected when you look at the bottlenecks of the various games that crash/artifact vs the ones that don't.

The suggestion to lower RAM speeds when its not the RAM itself that's corrupting is very telling as to the ROPs being the problem.

I actually recompiled the memtestCL with the 79xx bug fixed and found I can get the memory completely error free up to 1800-1850, no matter the load.
The problems come in when the ROPs try to accommodate that load when you push the load closer to 100% actualized load.

This is a big problem since the whole point of getting a 7970 or 7950 instead of going NVidia is to utilize that extra memory bandwidth.

 

[Grooveriding]

LOL another thread going to disaster-zone. This is just describing any overclocking. Stable in some cases, but not in others ? Your overclock is not stable, dial it down until it's stable in all cases.

 

[BallaTheFeared]

Yes, if you target synthetic tests that don't stress the ROPs you can get the core clock crazy high with way less voltage. This is directly reflected when you look at the bottlenecks of the various games that crash/artifact vs the ones that don't.

The suggestion to lower RAM speeds when its not the RAM itself that's corrupting is very telling as to the ROPs being the problem.

I actually recompiled the memtestCL with the 79xx bug fixed and found I can get the memory completely error free up to 1800-1850, no matter the load.
The problems come in when the ROPs try to accommodate that load when you push the load closer to 100% actualized load.

This is a big problem since the whole point of getting a 7970 or 7950 instead of going NVidia is to utilize that extra memory bandwidth.

This is what I said back in March with my first 7950.

Finding a stable overclock isn't bad, it's finding it in the next game, and the next game, and the game after that.

It's repetitive and tiring, also a bit disheartening imo. BF3 can do 1200/1750 no problem, run the entire SP campaign without a hitch, then Crysis 2 needs a big bump in vcore, while Tomb Raider will take Crysis 2's settings and hit 1250 no problem on the core, but need to drop vram to 1675 or artifact.

It's probably just my OCD min/maxing nature that doesn't allow me to just do something lower for clocks with more voltage than is necessary for every title.