[techpowerup] Radeon HD 7970 FOB Price Cut to $475

[BallaTheFeared]

Just use FXAA?

I've been using a ramped up version of the BF3/Skrim ones for DX11/9 titles.

Edit: I understand what you're saying, sometimes they get around to adding compatibility bits, sometimes they don't.

It's a pick your poison type deal, either wait or never get the ability to force different types of AA - or go with AMD and wait weeks, even months for performance acceptable drivers and CF support.

Nobody is perfect, that's for sure.

 

[blackened23]

Correct answer there, FXAA is the best bet unless you're playing an older game. All of the fancy AA is great but AA still has a huge performance hit regardless of what type of GPU you're using.....to be honest i'm glad FXAA is so common in new games, it looks pretty good for the relatively minor performance hit.

 

[LiuKangBakinPie]

Are you serious? From your numbers (3.79 - 2.7)/2.7=0,404

That's ~40% more TFLOPS. According to you, that 'roughly' translate into 13% difference?

No way thats 40 percent in real world.
Anyways thats not my point I wanted to make or ask you. What I really want to know is do you think is the money worth for gpu owners on gpus like 6950 or 6970 or 560ti or a bit lower or do you think they dont need that extra power at the moment as the cards is just released and the prices wont be as low as lets say a couple of months down the line? And if yes from what cards from Nvidia and Amd would a person ditch right now and do the upgrade?

 

[3DVagabond]

No way thats 40 percent in real world.
Anyways thats not my point I wanted to make or ask you. What I really want to know is do you think is the money worth for gpu owners on gpus like 6950 or 6970 or 560ti or a bit lower or do you think they dont need that extra power at the moment as the cards is just released and the prices wont be as low as lets say a couple of months down the line? And if yes from what cards from Nvidia and Amd would a person ditch right now and do the upgrade?

I would think that anyone contemplating this card would be a high end buyer. People who bought 5870, 5970, 6970, 480, 570, and *580 on release would be prime candidates for an upgrade and would see a sizable performance increase. Others, of course would be those who just need the latest and greatest.


*The 580 not so much because it's still a very high performance card once O/C'd. Although, if O/C numbers on the 7970 we are seeing now end up the norm, the 7970 could still be an appreciable upgrade.

 

[LiuKangBakinPie]

I would think that anyone contemplating this card would be a high end buyer. People who bought 5870, 5970, 6970, 480, 570, and *580 on release would be prime candidates for an upgrade and would see a sizable performance increase. Others, of course would be those who just need the latest and greatest.


*The 580 not so much because it's still a very high performance card once O/C'd. Although, if O/C numbers on the 7970 we are seeing now end up the norm, the 7970 could still be an appreciable upgrade.

Forgot about High end eeek as we are so use to see the mid range to low end get released first.
I notice some of the benches it did beat the older cards but not by much at times some by some margin. But one is a little wary as I see a new thing has creeped into hardware reviews. Certain hardware get tested with certain games to create a situation where the one piece of hardware might excel. Its a very clever tactic as most people dont notice it and the review looks good. Sad to see though thats why I will rather fish around or go take the gamble and see for myself.

 

[3DVagabond]

Forgot about High end eeek as we are so use to see the mid range to low end get released first.
I notice some of the benches it did beat the older cards but not by much at times some by some margin. But one is a little wary as I see a new thing has creeped into hardware reviews. Certain hardware get tested with certain games to create a situation where the one piece of hardware might excel. Its a very clever tactic as most people dont notice it and the review looks good. Sad to see though thats why I will rather fish around or go take the gamble and see for myself.

This isn't so new. That's why it's best to read multiple reviews. I've seen it done to make cards look bad as well. I'm sure both camps ask for their hardware to be reviewed using conditions that will make them look as good as possible and the competition to look worse. Why wouldn't they? Likely the main reason for "Reviewer's guides".

That said, I think the 7970 is appreciably faster than the older tech. Keep in mind that not all games will be well optimized this early on. Some just work better on different architectures, like nVidia's or even VLIW.

 

[LiuKangBakinPie]

This isn't so new. That's why it's best to read multiple reviews. I've seen it done to make cards look bad as well. I'm sure both camps ask for their hardware to be reviewed using conditions that will make them look as good as possible and the competition to look worse. Why wouldn't they? Likely the main reason for "Reviewer's guides".

That said, I think the 7970 is appreciably faster than the older tech. Keep in mind that not all games will be well optimized this early on. Some just work better on different architectures, like nVidia's or even VLIW.

Yeah its faster and look better only the memory config Amd chose Im little dissapointed in. I mean going the way Nvidia do it where Nvidia struggled with the ram connection and altough a smaller bus Amd were superior with the connection to the ram.

The thing is its now simple to choose but when they really going to influence the users is when the competition release their products and its the Brands going at each other and a carefully chosen situation might give superiority to 1 where they were in fact basically all on a equal footing. Sad to see it and from respectable ones as well.

But back to the card I notice that the ROP characteristics from RV 7 to the current are very close. Why changing a working thing. But the caches of the Z-stensil were doubled and sort of a hidden secret from AMD. ROP stayed the same wonder if they doubled the caches again

 

[Suspicious-Teach8788]

Why are they even doing this?

Now the NV fanbois can STFU once and for all?

 

[3DVagabond]

Now the NV fanbois can STFU once and for all?

Nah! They'll take all the credit.

 

[LiuKangBakinPie]

No one gets any discount for being a dedicated fanboy. We all pay the same price

 

[Arkadrel]

No way thats 40 percent in real world.

GPGPU stuff:

CiV -DirectCompute texture compression:
6970 = 210 --> 7970 = 313 (49% gain)

Luxmark - Luxball HDR - OpenCL rendering:
6970 = 8732 --> 7970 = 16155 (86% gains)

SmallLuxGPU 2.0d4 - raytraceing:
6970 = 11900 --> 7970 = 20700 (74% gain)

want to do Raytraceing?:

@LiuKangBakinPie

Theortical flops arnt everything.... look at the ACTUAL differences in performance,
between the 6970 and 7970, their pretty big (sometimes 50%-100% differnce (ei twice as fast)).

* these are all at stock settings.

 

[LiuKangBakinPie]

GPGPU stuff:

CiV -DirectCompute texture compression:
6970 = 210 --> 7970 = 313 (49% gain)

Luxmark - Luxball HDR - OpenCL rendering:
6970 = 8732 --> 7970 = 16155 (86% gains)

SmallLuxGPU 2.0d4 - raytraceing:
6970 = 11900 --> 7970 = 20700 (74% gain)

GPGPU who said anything bout GPGPU. Single precision your automatically thinking about GPGPU? Im talking about calculating GPU performance in things like graphics

Then take away the bandwith advantage the newer card has and youll end up with roughlly a gpu performance increase of what? Around 13 to 20 percent

 

[blackened23]

No one gets any discount for being a dedicated fanboy. We all pay the same price

20-50$ discounts for being an online apologist would be pretty cool I'll become a 7970 fanboy for a week!

 

[Arkadrel]

GPGPU who said anything bout GPGPU. Single precision your automatically thinking about GPGPU? Im talking about calculating GPU performance in things like graphics

Then take away the bandwith advantage the newer card has and youll end up with roughlly a gpu performance increase of what? Around 13 to 20 percent

Or around ~48% incrase in 3Dmark? (@ stock speeds)
(~48% faster than a 6970)
(~33% faster than a 580)





Overclocked:
A single 7970@1700mhz core in 3dmark vantage:

53,821 GPU score ^-^ in 3dmark vantage

 

[maddie]

GPGPU who said anything bout GPGPU. Single precision your automatically thinking about GPGPU? Im talking about calculating GPU performance in things like graphics

Then take away the bandwith advantage the newer card has and youll end up with roughlly a gpu performance increase of what? Around 13 to 20 percent

A word of advice. REAL WORLD DATA TRUMPS THEORETICAL PROOFS.

 

[Arkadrel]

A word of advice. REAL WORLD DATA TRUMPS THEORETICAL PROOFS.

Their not proofs, their just "calculations" useing a scewed, score vs ratio.
from a synthetical benchmark vs "read off box, therotical single persion flops".

But I agree with your sentiments, made up theory of how to calculate a gpus performance = non-sense.

why? because you can actually *test* the card, and see its *actual* performance,
instead of some bad made up math theory to "try" to calculate it.

 

[JAG87]

Did you really just use this argument? It boggles my mind that you bring this up as an argument. AMD has better AA than nvidia. Transparency SS is not supersampled AA and it is not in the nvidia driver (AMD does include SSAA.)
True SSAA has a *huge* performance hit and nvidia doesn't include it. TrSS is a cheap form of AA just like FXAA.

Supersampling renders the scene at a considerably higher resolution and then down-samples pixel samples to the required resolution. This costs a lot of performance and isn't practical in most scenarios. However one benefit of supersampling is that it does collect (and sample) every single pixel in a scene, allowing for optimum image quality. SSAA can't be used in 95% of games because it comes at a considerable performane cost. AMD offers it and nvidia doesn't, unless you want to muck around in nvidia inspector. Even then, it only works about 20-30% of the time and with ridiculously low framerates.

SS Transparency is essentially a complement to conventional sampling patterns. It works by anti-aliasing samples taken from within polygons where transparent textures are used to create effects that would cost too much if replicated with raw geometry. These transparent textures include objects like trees, grass and chain-link fences. The nice thing about SS transparency is that there is little performance cost. But the downside is that SS transparency is a joke and looks like garbage compared to SSAA. For comparison sake, even regular regular AA looks better than transparency anti aliasing. But the draw of SS transparency is the low performance hit for it. Which makes complete sense since it doesn't really improve image quality at all.

review / TLDR

Nvidia doesn't support SSAA in the driver. (TrSS is substantially worse) AMD does include SSAA. You talk down of FXAA but Transparency SS is a cheap form of AA just like FXAA is -- thus your assertion that nvidia has better AA within their drivers than AMD does is incorrect.

TrSSAA is cheap AA and substantially worse? What? You really have no clue what you're talking about.

No where did I say that TrSSAA = FSSSAA and no where did I say that you should use FSSSAA. MSAA+TrSSAA has no where near the performance hit of FSSSAA, and it is the single most obvious IQ improvement you can make after basic MSAA.

Try playing any Source engine game (CS, DOD, TF2, L4D, Portal) with and without TrSSAA and stare at the massive difference. You can test this yourself because Source is DX9 and you can enable TrSSAA on DX9 titles with your AMD card. Too bad you can't for DX10+ games...

4xMSAA

4xMSAA+4xTrSSAA

Even the review on AT mentions how this is a shortfall of AMD cards. AMD's response is, AA focus is shifting to shader-based AA like FXAA and therefore we excuse ourselves for not work on implementing TrSSAA. Well guess what, FXAA looks mostly like shit, and it does nowhere near good of a job on transparent textures. It helps with shader aliasing, but it's pretty awful on geometry and texture aliasing. Ideally you need a combination of the three, MSAA, TrSSAA and FXAA.

While we’re on the subject of image quality, had you asked me two weeks ago what I was expecting with Southern Islands I would have put good money on new anti-aliasing modes. AMD and NVIDIA have traditionally kept parity with AA modes, with both implementing DX9 SSAA with the previous generation of GPUs, and AMD catching up to NVIDIA by implementing Enhanced Quality AA (their version of NVIDIA’s CSAA) with Cayman. With between Fermi and Cayman the only stark differences are that AMD offers their global faux-AA MLAA filter, while NVIDIA has support for true transparency and super sample anti-aliasing on DX10+ games.

Thus I had expected AMD to close the gap from their end with Southern Islands by implementing DX10+ versions of Adaptive AA and SSAA, but this has not come to pass. AMD has not implemented any new AA modes compared to Cayman, and as a result AAA and SSAA continue to only available in DX9 titles. And admittedly alpha-to-coverage support does diminish the need for these modes somewhat, but one only needs to fire up our favorite testing game, Crysis, to see the advantages these modes can bring even to DX10+ games. What’s more surprising is that it was AMD that brought AA IQ back to the forefront in the first place by officially adding SSAA, so to see them not continue that trend is surprising.

As a result for the time being there will continue to be an interesting division in image quality between AMD and NVIDIA. AMD still maintains an advantage with anisotropic filtering thanks to their angle-independent algorithm, but NVIDIA will have better anti-aliasing options in DX10+ games (ed: and Minecraft). It’s an unusual status quo that apparently will be maintained for quite some time to come.

Update: AMD has sent us a response in regard to our question about DX10+ SSAA

Basically the fact that most new game engines are moving to deferred rendering schemes (which are not directly compatible with hardware MSAA) has meant that a lot of attention is now being focused on shader-based AA techniques, like MLAA, FXAA, and many others. These techniques still tend to lag MSAA in terms of quality, but they can run very fast on modern hardware, and are improving continuously through rapid iteration. We are continuing work in this area ourselves, and we should have some exciting developments to talk about in the near future. But for now I would just say that there is a lot more we can still do to improve AA quality and performance using the hardware we already have.

Regarding AAA & SSAA, forcing these modes on in a general way for DX10+ games is problematic from a compatibility standpoint due to new API features that were not present in DX9. The preferred solution would be to have games implement these features natively, and we are currently investigating some new ways to encourage this going forward.

http://www.anandtech.com/show/5261/amd-radeon-hd-7970-review/12



The fact that AMD has an option for FSSSAA for DX9 means absolutely nothing when they don't even provide TrSSAA in DX10+. NV might not have FSSSAA at all, but like you said it's pretty pointless seeing how it kills performance in most scenarios. But if you really wanted it, like you mentioned there is a way to have FSSSAA in all DX versions, not just DX9, something AMD can't do with or without tweaking programs.


Why would I pay top dollar for worse IQ?

 

[SirPauly]

Yes Nvidia has big deal of the pie for GPGPU market, it makes alot of money there.

AMD makes a card that does GPGPU (much better than their 69xx cards, and better than the 580)
, and Intel has Knights Corner, whats Nvidia to do?

same thing... make a card much much more GPGPU focused than before.

Thats one of the reasons, I dont expect the 680 to be masivily faster than the 7970. I think this card will have alot of focus dedicated to GPGPU with good performance/watt at GPGPU, not nessarly for gameing.

It's not just hardware but the software as well with the professional and compute products.

 

[SirPauly]

Another thing about SSAA. What it does is it takes the image internally and processes it at a substantially higher resolution. Say you have an image that is 1920x1080 on screen, SSAA will be rendered internally at 8 times that if you're using 8x SSAA. So you're basically using an effective resolution in excess of 15000x8000. And then it scales it back down and uses excess pixels in the calculation, smoothing over the colors and aliasing.

Obviously when the card is rendering internally at 2x-8x higher resolution than what appears on the screen, the performance hit is substantial. This is why nvidia does not include it in the driver. With Transparency SS, it does not do ANY of this. It only takes bits and pieces of the image and covers polygons with transparent textures. Thus the performance hit substantially lower than SSAA. But it also looks like crap compared to SSAA.

So bottom line that entire thing you wrote is ridiculous. AMD has better AA than nvidia, end of.

mini frustration / tangent

The other frustration with nvidia CP is that for roughly 60% of games, the override setting does nothing. I'll throw an example out: Dead space 1, dead space 2, dead island, among others. Its completely frustrating that trying to manually enable AA in an older game results in nothing. I'm not a huge fan of AMD these days but CCC always obeys 99% of what you override in their control panel applet. Nvidia control panel generally ignores what you put in, especially for newer games. Its EXTREMELY annoying. For example, with CCC / AMD you can enable SSAA in dead space 2 which plays at great frame rates on Xfire 6970's. With nvidia CP, you cannot under any cirumstance enable SSAA. You can enable regular AA with nvidia inspector, and anything higher than 8x gets you a choppy framerate. And weird shadows sometimes.

Say what you will about AMD but at least their AA override works the great majority of the time. With NV control panel its loads of fun trying to get override AA in dead space 2 and finding it does nothing. Yet with AMD CCC you can simply select SSAA and thats it, it will obey what you put in the control panel. Nvidia doesn't have SSAA, and even if it it did, override doesn't work for dead space 2. It also works via override settings most of the time in CCC, the same cannot be said of nvidia. Look for yourself in nvidia inspector -- nvidia has game profiles for all game exe's in the registry and you can view them in nvidia inspector. Most games (including dead space, dead island like I mentioned above) have the setting "treat override as use applicaton preference" flagged. Thus for the great majority of games override does nothing. Thats fun stuff isn't it? Thats my biggest annoyance with my 580s. Here let me show you what i'm talking about on my system:

If it bothers you so much one may change the behavior flag with a third party tool.

 

[SirPauly]

Did you really just use this argument? It boggles my mind that you bring this up as an argument. AMD has better AA than nvidia. Transparency SS is not supersampled AA and it is not in the nvidia driver (AMD does include SSAA.)
True SSAA has a *huge* performance hit and nvidia doesn't include it. TrSS is a cheap form of AA just like FXAA.

Supersampling renders the scene at a considerably higher resolution and then down-samples pixel samples to the required resolution. This costs a lot of performance and isn't practical in most scenarios. However one benefit of supersampling is that it does collect (and sample) every single pixel in a scene, allowing for optimum image quality. SSAA can't be used in 95% of games because it comes at a considerable performane cost. AMD offers it and nvidia doesn't, unless you want to muck around in nvidia inspector. Even then, it only works about 20-30% of the time and with ridiculously low framerates.

SS Transparency is essentially a complement to conventional sampling patterns. It works by anti-aliasing samples taken from within polygons where transparent textures are used to create effects that would cost too much if replicated with raw geometry. These transparent textures include objects like trees, grass and chain-link fences. The nice thing about SS transparency is that there is little performance cost. But the downside is that SS transparency is a joke and looks like garbage compared to SSAA. For comparison sake, even regular regular AA looks better than transparency anti aliasing. But the draw of SS transparency is the low performance hit for it. Which makes complete sense since it doesn't really improve image quality at all.

review / TLDR

Nvidia doesn't support SSAA in the driver. (TrSS is substantially worse) AMD does include SSAA. You talk down of FXAA but Transparency SS is a cheap form of AA just like FXAA is -- thus your assertion that nvidia has better AA within their drivers than AMD does is incorrect.

One can go to nVidia's site to download a tool for SSAA:

http://nvidia.custhelp.com/app/answ...sion/L3RpbWUvMTMyNTg2MTAxNS9zaWQvd0RPQzl0Tms=

I differ because of the ability to add transparency and super-sampled to DirectX 10 and 11 titles and there is flexibility with these modes and not locked to multi-sampling, like the past, so one can add, x2, x4 x8 TAA or FSAA to x8 MSAA.

And there are compatibility flags for many titles:

http://www.rage3d.com/board/showthread.php?t=33972293

 

[blackened23]

If it bothers you so much one may change the behavior flag with a third party tool.

Simply changing the setting doesn't work. If you're under the impression that you can simply change the flag to make it work, you are mistaken. It still does not work. Trust me, i've tried it with all sorts of games, I have 200+ games in my steam library. AMD is just more reliable than nvidia when it comes to overriding AA settings. Try it with Dead Space 2, the only way you can actually get it to work is by searching the net for the AA compatibility bit, and then you can enable up to 8xAA. Anything higher than that results in a slideshow. In contrast, you can enable SSAA with AMD CCC and it works with high frame rates. With crossfire 6970s its a constant fluid 60 fps, and it looks substantially better than what nvidia produces. (trying to enable SSAA with nvidia's SSAA tool results in a slideshow, btw)

The fact remains that SSAA via nvidia inspector is not reliable, and it is not a proper means for an end user to enable SSAA. AMD's CCC provides SSAA which is generally reliable and works -- you can view the write up on tomshardware regarding override AA. AMD is more reliable than nvidia in this respect, and AMD offers SSAA in the driver while nvidia does not.

 

[SirPauly]

I don't understand, I offered a link with a lot of compatibility flags for titles. You would have a leg to stand on if AMD offered transparency and super-sampled for DirectX 10+.

 

[blackened23]

I don't understand, I offered a link with a lot of compatibility flags for titles. You would have a leg to stand on if AMD offered transparency and super-sampled for DirectX 10+.

Lets be clear here. You will never use SSAA in a DX11 game unless you have something like 2-3 new GPU's - the performance hit is ridiculously high due to how SSAA works. Even if nvidia supports SSAA, it is NOT in the driver for users to enable. It is also not realistic to use because SSAA will easily lower your performance by 60% or more. It is far more demanding than any other type of AA.

I think you misunderstand what i'm saying. This is not a proper means for an end user to enable SSAA. First of all, it is highly unreliable. I've tested this on various games with both sets of hardware (6970 xfire and GTX 580 SLI) and in the end AMD is far more reliable. Aside from that, this is not how an end user should have to enable SSAA - with nvidia the only way to enable it is with a combination of a huge list of AA compatibility bits (for applicable titles) and entering it in nvidia inspector. This is probably less than 1% of enthusiasts, nobody should have to go through this hassle. And even when you do, its still unreliable.

Secondly, SSAA should work for DX10+ with CCC version 12 that was just released. This I will double check on. Nvidia has their AA, and AMD has their own (EQAA). Quality wise, they are roughly equal but AMD is far more reliable in terms of successfully overriding AA.

 

[Grooveriding]

Lets be clear here. You will never use SSAA in a DX11 game unless you have alien hardware from 2015

I fixed that for you. Besides that, you can't even force AA in DX11 games using deferred lighting/shading, period.

 

[JAG87]

I think you misunderstand what i'm saying. This is not a proper means for an end user to enable SSAA. First of all, it is highly unreliable. I've tested this on various games with both sets of hardware (6970 xfire and GTX 580 SLI) and in the end AMD is far more reliable. Aside from that, this is not how an end user should have to enable SSAA - with nvidia the only way to enable it is with a combination of a huge list of AA compatibility bits (for applicable titles) and entering it in nvidia inspector. This is probably less than 1% of enthusiasts, nobody should have to go through this hassle. And even when you do, its still unreliable.

Secondly, SSAA does work for DX10+ with CCC version 12 that was just released. Nvidia has their AA, and AMD has their own (EQAA).

Please stop viewing the world through nvidia glasses. I know that regardless of what is said, you will view AMD as a big bad evil company. And you must defend nvidias honor. Whatever dude.

Who the hell uses full scene SSAA? Even YOU said that it's way too much of a performance hit, so why do you keep bringing it up? Even if AMD's FSSSAA worked in DX10+ (which it doesn't, but you keep arguing), you wouldn't use it because no demanding DX10+ title will run at acceptable fps with FSSSAA.

Instead you down play the importance of TrSSAA, which gives like 90% of the visual improvement of FSSSAA without the penalty. The only thing it doesn't fix is shader aliasing, and that's what FXAA/MLAA is for.

Does AMD have TrSSAA in DX10+? No.