TechPowerup - Nvidia Kepler GK104 PCB Drawings and power connector pics

[BallaTheFeared]

So, about Kepler...

 

[AnandThenMan]

For me it greatly depends on the game. Some games 40fps is perfectly fine I don't notice any motion problems that detract from gameplay, it feels perfectly smooth to me. Others, I don't like the experience at all. That's what [H] tries to find, the compromise between a good gameplay experience and visuals. I believe this is easily the best way to bench hardware, getting 150fps is useless, and scripted benches are almost as bad they may have no correlation to real world.

It all comes down to how much your experience matches up with what [H] finds. Some insist that you must have 120fps for smooth gameplay.

 

[Silverforce11]

40 fps in batman is very playable. 43 fps in bf3 is borderline, you could disable the 2x MSAA and get a big perf boost, 50 fps would be playable. would it alter the outcome?

srly, someone buying a $550 graphics card would care about Far Cry 2, Hawx why? It makes no difference 100 fps+. Its the quality of games being tested thats far more important than quantity, especially considering older games already run too fast for the user to tell the difference. Adding those games in would skew the final % at no benefit to the user.

 

[omeds]

125 fps cap for online play and 120 or 60 fps vsync for single player games, if its not running at those I dont find it smooth.

That said I can see what H try to do, but find what they consider playable next to useless. Personally I would like to see them adjust apples to apples settings until one of the 2 cards is running at ~60fps average. If they want to test Vram limits, run the hardware in SLI, Tri etc and crank up the SSAA or other filtering.

 

[Grooveriding]

40 fps in batman is very playable. 43 fps in bf3 is borderline, you could disable the 2x MSAA and get a big perf boost, 50 fps would be playable. would it alter the outcome?

srly, someone buying a $550 graphics card would care about Far Cry 2, Hawx why? It makes no difference 100 fps+. Its the quality of games being tested thats far more important than quantity, especially considering older games already run too fast for the user to tell the difference. Adding those games in would skew the final % at no benefit to the user.

:thumbsup:

This is why overall conclusions on modern cards that are inclusive of archaic games running at 100+FPS are worthless.

Per FPS, SP games are just fine at 40fps, heck Crysis was fine at 30fps. MP games at least 50+.

 

[tviceman]

40 fps in batman is very playable. 43 fps in bf3 is borderline, you could disable the 2x MSAA and get a big perf boost, 50 fps would be playable. would it alter the outcome?

srly, someone buying a $550 graphics card would care about Far Cry 2, Hawx why? It makes no difference 100 fps+. Its the quality of games being tested thats far more important than quantity, especially considering older games already run too fast for the user to tell the difference. Adding those games in would skew the final % at no benefit to the user.

Note to Silverforce: there are more than just 4 demanding games available. Hi there every Crysis game, Metro 2033, Witcher 2, Total War: Shogun 2, Starcraft II, Alien Vs. Predator..............
Second note to silverforce: the entire time between 190 seconds and 290 seconds was consistently dipping below 25fps. 25-30fps for that amount of time in a benchmark is easily disguised when you say "40 fps is playable." Yeah, when it's consistently 40fps. That benchmark certainly was not.

 

[Silverforce11]

Crysis nobody cares, online its been dead for years. Crysis 2 was an aborted game, DoA. Metro and AvP are both dead.

Out of the games you mentioned, only SC2 is thriving, and its a highly CPU limited game.

If you cared about those older games as I've said, look at other review places, they have them in plenty. In fact, most review sites just do max settings 4x AA at multiple different res. [H] just focuses on 2560 and 3 monitor gaming, if you have a problem with it, don't go there, but to call it a poor review site is ignorant and spiteful.

ps. [H] is not biased to one company over another as some of you may feel, he bashes AMD regularly and prolly more than NV.

And oh, if Kyle ran that BF3 bench with no MSAA to make the lead for the 7950 even bigger, you ppl would claim he's biased. In fact, he picked 2x MSAA and not 4x MSAA as that would cripple the gtx580 since it lacks vram for that setting, it would have been a slideshow.

 

[MrK6]

That would be pretty terrible actually. Very small selection benchmarks, only 1 apples-to-apples comparison per game at settings which are usually too demanding to be playing at in the first place.

But they're actual games, which gives more useful data for generalization than a canned fly-by.

 

[wahdangun]

AMD lost market share from about 50/50 to 36/63. So I am not sure how the small die strategy allowed AMD to gain market share.

HD5970 was a great card. HD6990 was a very loud card (2nd loudest card of all time IIRC). But not sure how those cards matter in the single-GPU context.

I

hey before they launch HD 4870 AMD market share was very low if i remember it was around 10% and now it was almost 50% (I'm counting only discrete market share) and its actually surpassing nvdia in notebook market, and all that because they use small die strategy and now after AMD establish market share and maybe have more brand recognition they shift their strategy to become more profitable.

you talked in the context of average joe so dual GPU or not its really doesn't matter much, what they see is the best performance card. and btw HD 6990 is better engineered than GTX 590, it can even be overclocked while GTX 590 struggle to keep its reference speed.

 

[wahdangun]

I think you missed the main point tviceman was making. Kyle is trying to test "playable" performance but what he really shows is what's "Playable for him" since his standards are so low that often find cards running at 35-42 fps in FPS (BF) or racing games (Dirt 3) as playable. That's not playable in a competitive FPS or racing genre to most people.


.....................................

its still doesn't matter at all, its opinion and after all [H] still showed the FPS so you can judge for your self if you think 40 FPS is not playable.

on topic :

so is it confirmed that nvdia will drop hot clock for kepler ?

 

[GaiaHunter]

hey before they launch HD 4870 AMD market share was very low if i remember it was around 10% and now it was almost 50% (I'm counting only discrete market share) and its actually surpassing nvdia in notebook market, and all that because they use small die strategy and now after AMD establish market share and maybe have more brand recognition they shift their strategy to become more profitable.

you talked in the context of average joe so dual GPU or not its really doesn't matter much, what they see is the best performance card. and btw HD 6990 is better engineered than GTX 590, it can even be overclocked while GTX 590 struggle to keep its reference speed.

It is hard to compare market share apples to apples these days, with APUs that have IGPs as fast as lower end GPUs.

On the other hand those APUs are a package so it is hard to be precise how many of those sales translate to a old GPU buy.

What we need is a price bracket breakdown for a more valid picture.

 

[tviceman]

Crysis nobody cares, online its been dead for years. Crysis 2 was an aborted game, DoA. Metro and AvP are both dead.

Out of the games you mentioned, only SC2 is thriving, and its a highly CPU limited game.

If you cared about those older games as I've said, look at other review places, they have them in plenty. In fact, most review sites just do max settings 4x AA at multiple different res. [H] just focuses on 2560 and 3 monitor gaming, if you have a problem with it, don't go there, but to call it a poor review site is ignorant and spiteful.

ps. [H] is not biased to one company over another as some of you may feel, he bashes AMD regularly and prolly more than NV.

And oh, if Kyle ran that BF3 bench with no MSAA to make the lead for the 7950 even bigger, you ppl would claim he's biased. In fact, he picked 2x MSAA and not 4x MSAA as that would cripple the gtx580 since it lacks vram for that setting, it would have been a slideshow.

I have never said in this conversation that [h] was biased. I'm not even sure if I've said that at all. I have said on many occasions their results do not vibe with most other reputable hardware sites and all I've said in this thread is that their reviews suck. Nothing about MSAA, nothing about being unfair, nothing about crippling performance by running out of vram. Soooooo, thanks there. I hope I cleared up something that didn't actually exist to begin with.

Your reasons to to run more game benchmarks are stupidly ridiculous. Less information is better, you suppose? Help me to understand the reasons.

1. If a particular game doesn't measure up to your barometer of what is good then it shouldn't be benchmarked, correct? Crysis 2, check.

2. You also think any game that is older than 4-5 months will be dead by then and no one else will ever buy or play it so it doesn't need to be benchmarked anymore, despite it being the most demanding game benchmark available. Do I have that right? Metro2033, check.

3. And then, as you have said like a broken record for the past year and a half, toss out every TWIMTBP game because it will OBVIOUSLY be crippled on AMD hardware. No need for you to confirm that one, you have said it so often that I almost believe it to be true. Also, along the same lines, you will probably want to cross off every game that has been rumored by your cousin's online friend to have been funded by Nvidia.

4. FINALLY, CPU-bound games must be worthless too, even though Anandtech found a wide range of performance among all GPU's with the same CPU being used in Starcraft II. http://www.anandtech.com/show/5476/amd-radeon-7950-review/13

So basically, with your rationale, you are saying is that there are only 3 games worth benchmarking. Can I add to the list too, then? Lets not benchmark any new DX9 games, because even though the game might be amazing and hardware demanding, DX9 is broken old tech not worth including in benchmarks. Also, any game with letters C, R, and can I buy a vowel with the letter A - scratch 'em off the list of potential benchmarks. Those letters are so last year no one uses them anymore.

Good stuff!

 

[Quantos]

:thumbsup:

This is why overall conclusions on modern cards that are inclusive of archaic games running at 100+FPS are worthless.

Per FPS, SP games are just fine at 40fps, heck Crysis was fine at 30fps. MP games at least 50+.

Or modern games almost completely void of GPU dependence. Techpowerup's inclusion of SC2 makes me facepalm every time. It just dilutes the results so much! :\


Edit: with MSAA disabled

 

[railven]

So basically, with your rationale, you are saying is that there are only 3 games worth benchmarking. Can I add to the list too, then? Lets not benchmark any new DX9 games, because even though the game might be amazing and hardware demanding, DX9 is broken old tech not worth including in benchmarks. Also, any game with letters C, R, and can I buy a vowel with the letter A - scratch 'em off the list of potential benchmarks. Those letters are so last year no one uses them anymore.

Good stuff!

Haha, nice! Kudos guy!

On the subject of benchmarks, there are plenty to choose from. Find a site that reinforces your preference and you'll never be wrong

I personally love AT and their articles drip with information most other sites don't even touch on.

 

[tviceman]

Haha, nice! Kudos guy!

On the subject of benchmarks, there are plenty to choose from. Find a site that reinforces your preference and you'll never be wrong

I personally love AT and their articles drip with information most other sites don't even touch on.

I think Anandtech is great as well - the most informative and nonbiased site (in my opinion). I also like techpowerup's reviews, if for no other reason other than how well they aggregate all the information. I wish they would stop running benchmarks at anything lower than 1680x1050 though.

 

[Homeles]

I think Anandtech is great as well - the most informative and nonbiased site (in my opinion). I also like techpowerup's reviews, if for no other reason other than how well they aggregate all the information. I wish they would stop running benchmarks at anything lower than 1680x1050 though.

Well, for their mid range and high end cards at least.

 

[f1sherman]

Hardocp - benchmarking and pitting Nvidia's Transparency anti-aliasing vs AMD Adaptive

in F1 2010, Dirt 3, BFBC2, Civilization V

Clueless doesn't cut it, you need to be at least semi-blind too,
not to notice that Adaptive AA does not work in DX10/11.

Their subjective method has its values, but you really need to have cold-headed reviewer for that,
and Hardocp is more like a bunch of friends having fun with sponsored hardware.

 

[blackened23]

Hardocp - benchmarking and pitting Nvidia's Transparency anti-aliasing vs AMD Adaptive

in F1 2010, Dirt 3, BFBC2, Civilization V

Clueless doesn't cut it, you need to be at least semi-blind too,
not to notice that Adaptive AA does not work in DX10/11.

CCC version 12 notes:

Super Sampling Anti-Aliasing is now available for DirectX® 10 and DirectX 11 applications

Supported on the AMD Radeon HD 7900 and 7700 Series
Users can now enable Super Sample Anti-Aliasing and Adaptive Anti-Aliasing through the AMD Catalyst Control Center for DirectX 10 and DirectX 11 applications

Its also pretty convenient that you can't enable supersampled AA with nvidia unless you use an external application such as nvidia inspector, for users that don't know what external application to get you can't use SSAA. Nvidia also does not have "fast AA" injection within the driver, AMD has MLAA (similar to FXAA) while nvidia does not have a fast AA injector. You must , again, use an external application to inject FXAA in non supported games. On top of this, AMD AA override is on average more successful than nvidia override, if you want a citation TH did a comparison on this a while back. AMD supports FXAA as well, but offers MLAA injection on top of that within the driver.

I thought nvidia was all about user experience? Why no SSAA enabling within the driver? What say you nvidia apologists? I need to hear the excuses, I thought nvidia always had more features than AMD? Whats going on? I want to hear the excuses as to why AMD has more AA features in their driver than nvidia does.

 

[blackened23]

I think Anandtech is great as well - the most informative and nonbiased site (in my opinion). I also like techpowerup's reviews, if for no other reason other than how well they aggregate all the information. I wish they would stop running benchmarks at anything lower than 1680x1050 though.

I like AT but I have to disagree about TPU. Their 7970 review used explicitly old drivers and who the hell benchmarks at 1024x768? Their entire benchmark suite is a joke because they're benchmarking 1996 resolutions. Its ridiculous.

 

[f1sherman]

So now that 12 something Cata supports Adaptive
(but only if MSAA can be set from within game )

that makes [H] previously bogus benchmarks, when Adaptive did not work,
all of a sudden valid.......

 

[MrK6]

I like AT but I have to disagree about TPU. Their 7970 review used explicitly old drivers and who the hell benchmarks at 1024x768? Their entire benchmark suite is a joke because they're benchmarking 1996 resolutions. Its ridiculous.

I agree. Their benchmark suite is a mess and their methods are ancient. To top it off, their process integrity is questionable at best, especially the driver situation. I've stopped looking at TPU articles really, it's mostly [H] and AT I stick with.

 

[blackened23]

So now that 12 something Cata supports Adaptive
(but only if MSAA can be set from within game )

that makes [H] previously bogus benchmarks, when Adaptive did not work,
all of a sudden valid.......

I'm not sure what benchmark you're comparing (citation needed please), but understand that AdAA has worked in dx11 for a long time. CCC 12 has been out for a very long time.

Don't dodge the question. Why does AMD have more features in their driver than nvidia does? You guys always preach about how great the features of nvidia are, yet AMD has more AA features built into CCC -and- override is more successful. Come on, give me the excuses. I want to hear them. The only thing nvidia can boast about is having AO in the driver, yet it doesn't have fast AA or SSAA override. Pretty silly.

 

[f1sherman]

I'm not sure what benchmark you're comparing (citation needed please)

Just browse Hardocp for a collection of failed benchmarks using Adaptive AA in DX11.

http://henterprise.hardocp.com/article/2011/03/07/amd_radeon_hd_6990_antilles_video_card_review/4
http://www.hardocp.com/article/2010/10/21/amd_radeon_hd_6870_6850_video_card_review/4
http://www.hardocp.com/article/2010/11/09/nvidia_geforce_gtx_580_video_card_review/3
http://www.hardocp.com/article/2011/06/13/dirt_3_gameplay_performance_review/8

AMD only recently brought full screen and alpha AA in DX11,

but understand that AdAA has worked in dx11 for a long time. CCC 12 has been out for a very long time.

So what's this about:

Users can now enable Super Sample Anti-Aliasing and Adaptive Anti-Aliasing through the AMD Catalyst Control Center for DirectX 10 and DirectX 11 applications


in 12.2 Catalyst Release Notes :whiste:

Don't dodge the question. Why does AMD have more features in their driver than nvidia does? You guys always preach about how great the features of nvidia are, yet AMD has more AA features built into CCC -and- override is more successful. Come on, give me the excuses. I want to hear them.

Please 1st try to understand the basics, like Adaptive, Super Sample, Full Screen anti-aliasing has never/ever worked on AMD gpus in DX10/11 until 12.2 Catalyst.

And even now sounds more like a BETA feature, but better then nothing I guess.

As to more features... From the top of my head:
AMD still has no way to downsample, force Ambient Occlusion, fiddle with compatibility bits if AA does not work, frame or temperature limit FPS or upgrade profiles automatically.

Can they hot-install driver without the need to reboot like Nvidia?

 

[maddie]

I think Anandtech is great as well - the most informative and nonbiased site (in my opinion). I also like techpowerup's reviews, if for no other reason other than how well they aggregate all the information. I wish they would stop running benchmarks at anything lower than 1680x1050 though.

Just ignore the resolutions you don't like.

Why do people complain about this? How does this impact the higher resolution results?

We should all be aware that there are MANY gamers in the developing world with low resolution systems. The net is global.

 

[Homeles]

Just ignore the resolutions you don't like.

Why do people complain about this? How does this impact the higher resolution results?

We should all be aware that there are MANY gamers in the developing world with low resolution systems. The net is global.

Yeah, but what moron would buy a 7970 for 1024x768? They need to drop irrelevant resolutions from high end reviews, as they're factored into their fabled "relative performance - all resolutions" chart.