[Semiaccurate] GK104/Kepler/GTX680 Next Week?

[boxleitnerb]

Seems to me he's just trolling. 10fps...yeahsureok. As if that would be any accomplishment that Mark Rein would tout.

 

[blackened23]

Not sure what the point of that graph is...

99th percentile frame time is a general indication of possible microstutter. Its not conclusive, but there is no MS on 7970 xfire that I see. I have however seen microstutter on 5000 series and GTX 400 series.

 

[notty22]

Not sure what the point of that graph is...

It was to avoid backing up his 10fps comment.

 

[blackened23]

Seems to me he's just trolling. 10fps...yeahsureok. As if that would be any accomplishment that Mark Rein would tout.

You don't have to believe it, like I said videos will be out soon enough for you to witness for yourself. Doesn't bother me much either way Or do you think a single kepler would really trump tri sli 580s? :thumbsup:

 

[MrK6]

We get it, you don't like nVidia cards, move on!

No, I like NVIDIA cards. I already said I'd get a Kepler model if it outpaces my 7970 by enough (30%+). The thing is, certain fanboys can't stand AMD, much less that AMD is the current performance king. I like to point that out, especially when they fail abysmally. :whiste:

Welp, we've hit ground floor.

The only hope some people have is to try to drag others into the mud with them.

 

[boxleitnerb]

You don't have to believe it, like I said videos will be out soon enough for you to witness for yourself. Doesn't bother me much either way Or do you think a single kepler would really trump tri sli 580s? :thumbsup:

If you had stopped to think before you post you would have noticed that:

1. There is no info on fps. Could have been 45fps back then and 35 now
2. GK104 has roughly double the shader power as a 580
3. The demo most likely ran on 2-way SLI with 1 580 dedicated to GPU-PhysX. That means that the performance increase over a single 580 may lie between 2 and 3 and is not necessarily close to 3
4. They use FXAA instead of MSAA
5. They had a year to optimize the code. We don't know about the state of optimization back then
6. There could have been a VRAM limit on the 580s

But instead you claim something with no proof whatsoever, because you're biased and/or because you are not able to analyze the info at hand (or lack there of).

 

[blackened23]

If you had stopped to think before you post you would have noticed that:

1. There is no info on fps. Could have been 45fps back then and 35 now
2. GK104 has roughly double the shader power as a 580
3. The demo most likely ran on 2-way SLI with 1 580 dedicated to GPU-PhysX. That means that the performance increase over a single 580 may lie between 2 and 3 and is not necessarily close to 3
4. They use FXAA instead of MSAA
5. They had a year to optimize the code. We don't know about the state of optimization back then
6. There could have been a VRAM limit on the 580s

But instead you claim something with no proof whatsoever, because you're biased and/or because you are not able to analyze the info at hand (or lack there of).

You're correct on a lot of these points, I can't argue that. I guess I spoke out of turn because some suggested that 1 Kepler outperformed 3 580s in the demo. I don't believe that was the case.

 

[SolMiester]

No, I like NVIDIA cards. I already said I'd get a Kepler model if it outpaces my 7970 by enough (30%+). The thing is, certain fanboys can't stand AMD, much less that AMD is the current performance king. I like to point that out, especially when they fail abysmally. :whiste:

The only hope some people have is to try to drag others into the mud with them.

You dont have to be a fanboi to be obnoxious....MrK6

 

[badb0y]

Yea if stock for stock Kepler is 20-30% faster than HD 7970 I will jump ship but if we are looking at a 10-20% advantage....I don't think I will get nVidia this round.

I'll probably grab a 3D Vision ready monitor while I wait for GK110(?).

 

[Vesku]

4. They use FXAA instead of MSAA
5. They had a year to optimize the code. We don't know about the state of optimization back then
6. There could have been a VRAM limit on the 580s
- boxleitnerb

NVIDIA's PR release on the tech demo states these points pretty clearly. Going from MSAA to FXAA was not only less costly in processing and implementation complexity but apparently it is also very memory light in comparison with last years demo.

3. The demo most likely ran on 2-way SLI with 1 580 dedicated to GPU-PhysX. That means that the performance increase over a single 580 may lie between 2 and 3 and is not necessarily close to 3

Apparently the APEX particle system is underpinned with Physx.

In my search for Physx info I read an in depth on the 2011 demo. It mentions how they used 3-way SLI to save time on the optimization side. I think it likely they've optimized more since, especially since NVIDIA has been the primary developer on some of the features showcased.

 

[OILFIELDTRASH]

Yea if stock for stock Kepler is 20-30% faster than HD 7970 I will jump ship but if we are looking at a 10-20% advantage....I don't think I will get nVidia this round.

I'll probably grab a 3D Vision ready monitor while I wait for GK110(?).

Why do you do this to yourself? Just be patient next time and buy the flagship nvidia card. I remember you struggling last go round about wanting a 580 while you had a slower 6950. Now it's looking like the same situation will play out.

 

[Grooveriding]

Why do you do this to yourself? Just be patient next time and buy the flagship nvidia card. I remember you struggling last go round about wanting a 580 while you had a slower 6950. Now it's looking like the same situation will play out.

Sometimes you just don't want to wait. I would have had two 7970s by now if I hadn't spent $1000+ on water cooling and case mods last month. Now I have to lay low and wait or my wife will raise some hell.

Works out in my favour though, as GK110 should be very nice x2 whereas 7970x2 or 680x2 if rumours are accurate need to be overclocked to give the sort of gains I am looking for in two cards.

Later this year when we start getting closer and have some real info on GK110 and its release, the 7970s should still fetch $400 or so used. Unless one were to believe the rumours of a $300 to $400 GTX 680

 

[badb0y]

Why do you do this to yourself? Just be patient next time and buy the flagship nvidia card. I remember you struggling last go round about wanting a 580 while you had a slower 6950. Now it's looking like the same situation will play out.

Last time around I was much more limited with funds so I couldn't grab the GTX 580 but this time around I have much more money. If GTX 680 is atleast 20-30% faster than HD 7970 I won't have any problem selling off my HD 7970 and picking up a GTX 680. I guess I just like to have the latest and greatest .

 

[MrK6]

You dont have to be a fanboi to be obnoxious....MrK6

No, however it is obnoxious to disregard the rules of the forum.

 

[Stuka87]

If you had stopped to think before you post you would have noticed that:

1. There is no info on fps. Could have been 45fps back then and 35 now
2. GK104 has roughly double the shader power as a 580
3. The demo most likely ran on 2-way SLI with 1 580 dedicated to GPU-PhysX. That means that the performance increase over a single 580 may lie between 2 and 3 and is not necessarily close to 3
4. They use FXAA instead of MSAA
5. They had a year to optimize the code. We don't know about the state of optimization back then
6. There could have been a VRAM limit on the 580s

But instead you claim something with no proof whatsoever, because you're biased and/or because you are not able to analyze the info at hand (or lack there of).

Where did you find out that Kepler has twice the shader power? The number of shaders means nothing at this point as the shaders have been redesigned since fermi, so currently there is no baseline to judge the new shaders on until cards get into testers hands.

 

[frostedflakes]

Raw compute power is rumored to be a ton higher, 4.3Tflops compared to 1.6Tflops for GTX 580 and 3.8Tflops for 7970. Memory bandwidth, fillrate, etc. is also rumored to be a lot lower than 7970 and more comparable to the 7950, though, so that will bottleneck things despite having more theoretical shader power. Keeping this in mind the rumors of the 680 falling somewhere between 7950 and 7970 in gaming performance (excluding TWIMTBP titles and other programs that are heavily optimized for nVidia I'm sure) seems pretty reasonable to me. But again these are all rumors, so take them with a grain of salt.

http://en.expreview.com/2012/03/08/nvidia-kepler-gk104-specifications-surfaced/21693.html

 

[SolMiester]

No, however it is obnoxious to disregard the rules of the forum.

Yes, you walk the thin blue line well......

 

[BallaTheFeared]

Wrong thread!

 

[Will Robinson]

Lots of people on here thought Nvidia couldn't catch or beat AMD GPU's on performance per mm^2. Lots of people on here also thought GK104 wouldn't be able to keep up with or beat Tahiti. Lots of people on here think Nvidia isn't going to upset the current pricing tier's set by AMD.

I hope lots of people are wrong 3 out of 3 times.

One person....bet everyone here that NVDA's high end card would be released within 2 months of HD7970...ie Mar 9th.
So here we are...and no Kepler.:whiste:

FAIL

 

[boxleitnerb]

Where did you find out that Kepler has twice the shader power? The number of shaders means nothing at this point as the shaders have been redesigned since fermi, so currently there is no baseline to judge the new shaders on until cards get into testers hands.

1536*2*x MHz. Just talking FLOPs - I didn't say how effective that translates into performance.

 

[skipsneeky2]

Last time around I was much more limited with funds so I couldn't grab the GTX 580 but this time around I have much more money. If GTX 680 is atleast 20-30% faster than HD 7970 I won't have any problem selling off my HD 7970 and picking up a GTX 680. I guess I just like to have the latest and greatest .

I would think most people will be stoked to own a single 7970,me i love mine and for once i am actually pretty happy knowing i need nothing else for a good while prob well into the next generation of lousy coded console ports for the ps4 and xbox 720.

BF3 1080p is my most demanding title all max ultra settings enabled including vsync and motion blur and 2x msaa and its just perfect,performance,visuals just freaking perfect.

Is there any game out right now that would bring a 7970 down to its knees at 1080p before even enabling msaa or aa besides Crysis?

 

[Tempered81]

crysis and metro 2033

 

[X98]

One person....bet everyone here that NVDA's high end card would be released within 2 months of HD7970...ie Mar 9th.
So here we are...and no Kepler.:whiste:

FAIL

mar 9th or not, tahiti will still get dominated in the rear XD

 

[Keysplayr]

One person....bet everyone here that NVDA's high end card would be released within 2 months of HD7970...ie Mar 9th.
So here we are...and no Kepler.:whiste:

FAIL

One should never gamble. March is still young yet. If it goes into April, then I'd say he really lost the bet. Have patience Will.

 

[Madcatatlas]

Here is what happened: this is nvidias 680 which in turn is half of nvidias 780 which nvidia hope to release later this year, around december.

Im gambling on this being 5% faster/slower than the 7970 with both being stock. The 7970 will walk all over this when overclocked.