The Official Graphics Card and PC gaming Thread

Oh, thanks. The light went off, and I guess I understand now. So taking a look at your top-of-page chart again:

As I think I now understand, we're looking at the relative performance of those cards, on those games (more or less based on a 4870 at 100.)

So for example, if a person's main interest is Battlefield 3, he's not going to see a huge jump up from a 6970, to a 7950 to a 7970, the jump will be from 210 to 240 to 285. So depending on price, a 6970 might be the correct investment, if a number of 200 performance satisfies what that person needs for his type of gaming depending on resolution and acceptable frame rates. For crossfire, apply 90% scaling.


Sam, if that is basically correct, then, if you have the figures handy, could you quote those numbers for two additional games, for the new Metro, and for any Crysis? And at the same time that you state the relative numbers, could you quote the actual performance number required for max quality (but maybe just 4xAA) at - let's say average 40 fps, or 60 if you prefer, (or maybe you'd prefer to quote minimums) - for 24" and for 30" gaming?

I realize that this would likely be a rehash of prior posts, and maybe I'm lucky and the information is already saved in PaperPort, but I have the feeling I might have to dig through the back posts to find it. If you have the numbers handy, this time it will definitely go into the appropriate game folder.

I'm trying to zero in, because for the first time in the last 3 or 4 years, I actually have $1000-1200 that I could invest in a couple of 7970s, plus maybe a small SSD to work better with Windows 7 (I already have maxed the P5E with 8 gigs of memory - only $100 after rebate) even if this new sale I'm working on doesn't pan out.

But rather than jump into that now, if you think there could be a significant price drop by summer, I have plenty of gaming to keep me busy until then with the 8800GTX and MW2 and MW3, even in solo mode on the special ops, plus Left 4 Dead 1 and 2. (I'm trying to balance out the gaming with business, thinking in terms of one week a month - one day a week didn't really work, lol.)

Then maybe in the summer, I could join you guys and Jeff on the BF3 battlefield.

So what do you think - do you think we might see a $300 7970 by summer?

Rich

 

Crysis (Very High): 260/490 (NoAA), 280/540 (2xAA), 355/670 (4xAA)
Crysis 2: 290/375 (DX9), 345/485 (DX11)
Metro 2033: 145/165 (Low10 AAA), 160/170 (Medium10 AAA), 215/255 (High10 AAA), 310/570 (VH10 AAA), 470/830 (VH11 AAA)

10/11 refers to the DirectX version used. MSAA (as per 2x, 4x in crysis) has not been tested, due to its hardware demands.


I have not yet thoroughly benched the new Metro title (because it is not released until Mid-2013!)

These results apply for 1920x1080. For 2560x1600 results multiply by approximately 1.9.

Note - the requirement for an average of 60fps is displayed first, then the requirement for a minimum of 60fps. e.g. If a title needs a score of 100 to average 60fps, and 200 for a minimum of 60fps, it will be displayed as 100/200.


The lowest you will see an HD7970 by summer will be $500. Even that would surprise me.

 

Ok, magnificent. Thanks Sam.

These 10 sets of numbers, each set consisting of A. an average 60 fps, followed by B a minimum 60 fps, are all frame rate numbers, at 24" gaming, and I can multiply by 1.9 to get to 30" gaming.

So I have half the puzzle - I have the frame rate numbers and the required graphic power to achieve those numbers.

Now, the only thing left that I need, I believe, is the other half of the puzzle - the relative performance of each card, on those games, like your chart #7876 at the top of this page. That would allow me to correlate the frame rate data of this last post, with the available cards, 6970,7950, or 7970, and see whether I can run these games at decent frame rates, depending on which card and depending on crossfire, assuming 90% scaling.

In other words I need the same info for each card that you provided for the five games in your earlier post #7876 at the top of this page:

So if you could please give me a crysis (very high) set of three relative performance numbers just like above, and a crysis 2 (enthusiast) set of 3 numbers, and a Metro 2033 set of three numbers, then I will produce the chart for 30" gaming.

Rich

 

The whole point of providing a relative index instead of the actual frame rate numbers, is so you can get a general idea of how cards perform without needing a game-specific relationship.
You can apply those numbers to any card based on how it normally performs and get a reasonable idea of how it will run.

If you know the requirement index of the game, and the performance index of the card, that's all you need to work it all out.

For example, if we have Metro 2033, at very high detail in DirectX11 with AAA applied, and we want a minimum of 40fps, that's a requirement of (40/60)*830 = 553.
We know that two HD7970s have a performance index of between 540 and 710, so they will be able to cope. Pretty well I would expect, as the very heavy loading of Metro 2033 should push HD7 series performance towards the upper end of the band.

If we have Crysis at 'Very High' with 4x AA applied, and we want a minimum of 40fps, that's a requirement of (40/60)*670 = 447. Two HD7950s (450-600) should cope with this, and in fact, two HD6970s (at 410), should get reasonably close. 60fps, however, is far more tricky, and remember, this is at 1920x1080 resolution.
To get to 2560x1600, we need to multiply these numbers by about 1.9 in many cases.

Let's say we'll settle for a minimum of 30fps at 2560x1600, and run the tests again. This means we can take half of the requirements.
In Crysis at 'Very High' 4xAA, that means a requirement of 640. Two HD7970s might manage that, just, but nothing less will do. If we turn the AA off, that's only 470. You might get away with two HD7950s if you could live without AA, as well as 30fps. But then of course, 'reckoning' may well be pretty miserable if you only aimed for 30fps for the rest of the game.
In Metro in DX11 with AAA, that's a requirement of 790. Beyond the scope of even two HD7970s. How about DX10 mode? 540. That will do with two HD7970s, but 7950s will struggle I suspect, as by removing DX11 you are eliminating a lot of what makes the HD7 series particularly capable and perform at the high end of the performance band. Metro 2033 at 2560x1600 maxed out with only two cards is way off. We didn't even consider MSAA here.
If we settled for a further reduction in detail for Metro, and went to 'High' in DX10 mode, the requirement plunges to 240, even at 2560x1600, enough for a single HD7950 to cope, or for two to cope at almost 60fps minimum.

 

Could you possibly be any more vague? Or are you about to edit your signature, to include spam...

 

Oooh, I like this site!
http://www.gpureview.com/show_cards.php?card1=577&card2=637
I'm actually considering an Nvidia card now. Cuda can't be ignored. I guess I shouldn't jump the gun though. I just acquired the applications for encoding with my GPU. Haven't actually ran it the first time yet. I do expect at least a 25% gain though, based on other reviews. Should be an interesting experiment. I can't run it for a few hours though I have other tasks to perform first...

 

You do realise you can encode on Radeons with OpenCL right?

 

Are you suggesting that the Radeon GPU's can utilize the same exact drivers and software?

 

It's not exactly the same technology, but it works the same way. There are plenty of encoding applications out there that support OpenCL now. Unlike the nvidia system, it's not proprietary, so it'll likely get supported by more developers eventually, it depends how many software devs nvidia manage to pay off into exclusively using CUDA, in much the same way they do with PhysX and driver performance.

 

I might be mistaken, but I don't believe the configuration that I HAVE to use, will work with radeons :S

 

What software do you have to use?

 

Bd rebuilder, DGDecNV(which references an Nvidia DLL file). And of course DGDecNV may be coded for solely Nvidia. I'm not seeing any ati support in his forum.

 

Well yeah DGDecNV is unsurprisingly an nvidia-only application, but why do you need this specific program? There are dozens of bluray encoders out there...

 

I love bd rebuilder DvdCrap(fab) is garbage! And at the moment, I can see no reason to try other apps. Although Multiavchd is on my list of "To Do's" LOL! I use Ripbot for specific needs. E.g. encoding subs into video(MKV) permanently. It has been a while since I searched applications. They're probably popping up like crazy by now

 

One of the reasons I ask is that it's well known that nvidia substantially cut the video quality of rendering when used via CUDA so they take the benchmark advantage (where have we heard that before?) - Best quality is on the CPU still, but AMD are right behind. Nvidia encodes are like choosing the 'quick encode' method - fast, but very dirty.

 

Ugh. Really? I guess it's something I'll have to compare then. It is in the works. I encoded a BD-25 last night, Via CPU. I'm running the same exact job through the GPU today. I'm gonna weigh the advantages when I get home, and the job is done. Compare some still images. As well as watch the output.

 

Generic CPU encoding:


Intel Sandy Bridge Quicksync:


AMD Radeon HD6870:


nvidia Geforce GTX460:


Extreme case maybe, but I wouldn't want to go anywhere near CUDA encoding at the risk of it ever putting out frames like this.

 

Extreme indeed. Surely those are old... In any case, I do realize that the software involved is nowhere near mainstream yet. It's something of a beta test. Though donations are required.
I use to think a processor is a processor. Even in the case of a GPU. But your screenies make me question certain factors. Coding is certainly the most important of factors. It's not the GPU's fault it's doing a p!$$ poor job

 

There are several cases of this being exhibited.
This test was from January 2011. Not that old really...