NVIDIA claims top DX9 GPU marketshare spot.

[digitalwanderer]

Maybe so, but vertex shaders can be emulated on a fairly fast CPU and Pixel Shader Speed of a FX 5200 Is too Slow to be considered a 'feature' as enabling Pixel Shader effects in a game would bring the game to unplayable levels, and that is all that matters.
The Pixel Shader speed is more imortant than Vertex Shader support IMO, there is still lots of game engines that use very little of Vertex Shaders, in fact UT 2003 engine doesn't use them at all.

I'm glad you have your opinion that they're too slow to be considered a feature.

That wasn't the question asked.

The question was "are PS2.0 instructions done in hardware on the 5200?"

The answer is clearly yes.

But, to address your "too slow" opinion, consider that they are, even by your graph, half the speed of the 5600, and 1/4 the speed of the 9500. This would lead one to believe that you could run at 1/4 the resolution of the 9500 and still get the same framerates. That would be, for example, the difference of no AA at 1024x768 on the 5200 compared with 4x AA at 1024x768 on the 9500. Or no AA at 800x600 at a higher frame rate. (Assuming framerates are solely a ratio that mirrors the ratios of PS2.0 scores in 3dmk03). In that light, the it seems it is, indeed, a feature that has usefulness. Unless you're proposing that the 9500 is useless also, just 1/4 as useless.

And that doesn't even address that PS2.0, for the near future, will only influence a fraction of the framerate simply because no card can run them at truly acceptable speeds. Even more so that the bulk of the cards sold that can do DX9 PS20 shaders are in the "too slow" category.

I beg to differ. If it can't run DX9 functions well enough to be useful it ain't truly a DX9 part, it's just PR window dressing.

Sorry, but that's my opinion and I'm sticking to it. The 5200 is a sadly under-powered excuse for even an entry-level gaming card.

 

[RussSchultz]

So, 1024x768 isn't useful? 800x600? Where does 'useful' begin?

 

[digitalwanderer]

So, 1024x768 isn't useful? 800x600? Where does 'useful' begin?

But what kind of performance are we talking about at those levels right now, not to mention that titles are only going to get more and more demanding.

My V5 5500 can game at 1024x768, but it can't do it very well.

Sorry, I really shouldn't be making an opinion since I haven't really had a chance to try one yet...but from the reviews I've seen I really don't think it's got enough power to be considered a good card.

 

[RussSchultz]

If the 9500 can run at 4xAA in 1024x768, then the 5200 can run at the same framerate without AA.

If the 9500 can run at 2xAA in 1024x768, then the 5200 can run at a slightly lower framerate at 800x600.

(Again, assuming that the game framerate directly by the ratio of the PS2.0 scores in 3dmk03 and the ratio of the resolution resolution)

Is that still useless for an entry level card?

You or I might not buy one, but for the average person they're "OK", and they do their job.

 

[andypski]

If the 9500 can run at 4xAA in 1024x768, then the 5200 can run at the same framerate without AA.

If the 9500 can run at 2xAA in 1024x768, then the 5200 can run at a slightly lower framerate at 800x600.

I understand where you are coming from, but I really wouldn't make a sweeping generalisation like that. Depending on how limited a particular application is by factors such as the pixel shader and vertex shader then 4xAA can potentially come for very little frame rate hit (multisampling, remember).

[edit]
Quick example - a 9600 Pro might do about 29fps in the 3Dmark03 PS2.0 test without AA and 23fps with 4xAA (about 79% of the noAA performance). A 9500 would be a bit slower than this, but not much.

I don't think that a 5200 would get close to 20 fps in that test even without AA (well, not if it's rendering stuff correctly ) It would probably somewhere closer to 10 fps at best.

The same thing could easily be the case in other applications that make extensive use of longish pixel shaders.
[/edit]

 

[digitalwanderer]

You or I might not buy one, but for the average person they're "OK", and they do their job.

The question I ask myself is, "Would I recomend it to someone as an entry level card?"; and to that I answer "NO!".

Can we agree to disagree on this one since it is just an opinion call? I'm not saying I am absolutely right here, I'm just giving me most honest opinion that I do not think it has sufficient power to even be considered an entry level card for gaming.

 

[RussSchultz]

If the 9500 can run at 4xAA in 1024x768, then the 5200 can run at the same framerate without AA.

If the 9500 can run at 2xAA in 1024x768, then the 5200 can run at a slightly lower framerate at 800x600.

I understand where you are coming from, but I really wouldn't make a sweeping generalisation like that. Depending on how limited a particular application is by factors such as the pixel shader and vertex shader then 4xAA can potentially come for very little frame rate hit (multisampling, remember).

Of course, that's why my disclaimer was there.

Though, presuming the 9500 would get its "free" 4xMSAA, then the 5200 would too. Or not?

 

[andypski]

Though, presuming the 9500 would get its "free" 4xMSAA, then the 5200 would too. Or not?

It might do, but the problem is that it wouldn't necessarily get close to the same performance as the 9500 even without AA, so the fact that it might get AA close to free wouldn't necessarily be of any use.

 

[digitalwanderer]

If the 9500 can run at 4xAA in 1024x768, then the 5200 can run at the same framerate without AA.

If the 9500 can run at 2xAA in 1024x768, then the 5200 can run at a slightly lower framerate at 800x600.

I understand where you are coming from, but I really wouldn't make a sweeping generalisation like that. Depending on how limited a particular application is by factors such as the pixel shader and vertex shader then 4xAA can potentially come for very little frame rate hit (multisampling, remember).

Of course, that's why my disclaimer was there.

Though, presuming the 9500 would get its "free" 4xMSAA, then the 5200 would too. Or not?

Not.

 

[RussSchultz]

Though, presuming the 9500 would get its "free" 4xMSAA, then the 5200 would too. Or not?

It might do, but the problem is that it wouldn't necessarily get close to the same performance as the 9500 even without AA, so the fact that it might get AA close to free wouldn't necessarily be of any use.

Sure, but then we're back to the 1/4 ratio, no?

(Even though, I think historically games have shown us, that its never a direct ratio of the fillrate, but that the ratio in performance delta seems to be much less than the ratio of fillrate would suggest. In otherwords, a 2x improvement in fillrate doesn't translate to a 2x improment in framerate, it translates to much less than a 2x improvement in framerate.)

 

[demalion]

This is not intended as a flame or to be disrespectful, but what in the heck are the two of you even arguing over anymore?

Could you just say it in simple words, and only a few of them?

Russ requires that I believe things that I propose don't make sense when considering: logic, English, and the discussion I've proposed to him multiple times connected with these things, and he refuses to provide support for why he is right in this requirement (he simply repeats that he is right). He has flatly refused every alternative to continuing this in this thread that I've offered.

What purpose a discussion serves, because Russ is simply demanding that he isn't accountable to provide support for his viewpoint except his belief in it, or even to simply show why my own assertions counter to it are wrong? The only one I ssee is to allow Russ to say whatever he wants without possibility of him being wrong until such time as it suits him to believe it (in this particular case, that seems to be "never"). I.e., I can't prove him wrong with a discussion of the objective criteria above as long he ignores them.

That's where we are.

 

[demalion]

So, 1024x768 isn't useful? 800x600? Where does 'useful' begin?

Hey, you know someone tried to hold a discussion with you to answer this. Why are you are ignoring it? Math too inconvenient?

 

[digitalwanderer]

What purpose a discussion serves, because Russ is simply demanding that he isn't accountable to provide support for his viewpoint except his belief in it, or even to simply show why my own assertions counter to it are wrong?

So y'all mean you're just arguing over if Russ' opinion is a fact or in fact his opinion y'mean...right? :|

 

[indio]

My question is this. Is the 5200 really doing DX9 level PS 2.0 (that means at minimum precision) ? Is the 5200 doing a minimum of FP24?
This smells like the next-gen GF4mx.

 

[digitalwanderer]

This smells like the next-gen GF4mx.

That's my big problem with this card. I recomend a GF4ti over it any day! (How come I have a feeling not a whole lot of people are going to jump up and disagree with me on that one... )

 

[demalion]

This is supposed to be what I'm proposing, right?

The concept of saying "No" to someone saying "Properly can include performance perfectly validly, and you are wrong to provide an argument based on the precept that it cannot".

So, you're suggesting that simply by offering my interpretation and my reasoning, I'm somehow stating that your interpretation cannot be, with certainty.

No, I'm proposing what you just quoted me as saying in contrast to just saying "No" in isolation or to something completely different.. Adding the "simply" in there changes it from what I said in this quote, to the "saying 'No' to something else entirely" which I already discussed. I.e., what I say in the quote is specifically not what you just said.

"Should I buy this car?" "No"
"Is buying this car a good idea?" "No"

There is no "simply saying 'No'" without context, Russ, even if it is inconvenient to your not wanting to admit an error. English does not allow you to propose that the two statements of "No" above communicate the same thing. Even if you protest you were "simply saying 'No", so can't be accused of saying something about the car.
Maybe if you'd discussed my first reply to your initial post instead of simply skipping over it, this wouldn't be news to you?

Like I said: you seem unable to grasp the difference between a statement of differing interpretation and a statement of denial of differing interpretation.

No, Russ, you are simply unwilling to grasp basic English.

Russ, I already addressed exactly why this statement is wrong, untrue, unrelated to actuality. What do you accomplish by ignoring each and every time I do so to simply say it again? Is it supposed to be hard to see that you just skipped over my discussion in its entirety, again? To whom?

 

[demalion]

What purpose a discussion serves, because Russ is simply demanding that he isn't accountable to provide support for his viewpoint except his belief in it, or even to simply show why my own assertions counter to it are wrong?

So y'all mean you're just arguing over if Russ' opinion is a fact or in fact his opinion y'mean...right? :|

I prefer to use "fact" for something more concrete. I think it would be closer to say we're arguing over whether Russ was wrong when he said something. Proposing "fact" implies there is no possibility of Russ being right, when what I'm actually proposing is that Russ refuses to believe anything else but also refuses to prove it except by saying he is right over and over.
Of course, this, as well as English, logic, etc., leads me to believe that he is, in "fact", wrong, but all I'm insisting on is an actual discussion where the possibility of him being right and wrong can both be represented. He, on the other hand, seems to want to skip to the "done deal" of him being right, without having an accountability to the intervening discussion.
At the moment, the argument for him being wrong seems to be the only one with support provided that correleates to something outside of insisting "I'm right because I said so".

 

[Slides]

No more caffeine for demalion.

 

[Dave Baumann]

Though, presuming the 9500 would get its "free" 4xMSAA, then the 5200 would too. Or not?

Russ, its not actually a given that a chip that can support, say, 4X MSAA natively actually support 4 samples per cycle.

 

[andypski]

Sure, but then we're back to the 1/4 ratio, no?

No. I think your original example is flawed because these cards use multisampling AA - with multisampled AA increasing the AA ratio is not remotely equivalent in terms of performance hit to increasing the number of screen pixels rendered.

I'll go back to your earlier post to comment here -

But, to address your "too slow" opinion, consider that they are, even by your graph, half the speed of the 5600, and 1/4 the speed of the 9500. This would lead one to believe that you could run at 1/4 the resolution of the 9500 and still get the same framerates.

This part of your statement is definitely reasonable.

That would be, for example, the difference of no AA at 1024x768 on the 5200 compared with 4x AA at 1024x768 on the 9500. Or no AA at 800x600 at a higher frame rate. (Assuming framerates are solely a ratio that mirrors the ratios of PS2.0 scores in 3dmk03).

This part is the part that I think is wrong.

Using a 1/4 pixel shader performance ratio means that to get equivalent performance in a pixel shader limited application you need to run at a physical resolution 1/4 the size, not the same resolution without AA.

eg. Taking a hypothetical completely pixel shader limited application where a 9500 is getting 30fps at 1280x960 resolution you might expect that a 5200 with 1/4 the shader performance would get the same frame rate at 640x480 (1/4 the number of pixels)

If the 9500 turned on 4xAA it might then take a 20->30% performance hit, so say it would drop to 20 fps.

If you increase the resolution on the 5200 to 1280x960 (without AA) it would be rendering at around 7.5fps - this is only 1/3rd the performance of a 9500 running at the same resolution with 4xAA enabled.

And that doesn't even address that PS2.0, for the near future, will only influence a fraction of the framerate simply because no card can run them at truly acceptable speeds

The first part of this is a perfectly reasonable position - we can expect a gradual uptake of PS2.0 just as with any other new technology, however I take issue with the second part of the statement. Certainly I believe that our current cards are perfectly capable of running complex PS2.0 shaders at good frame rates - they were designed with this performance level in mind from the start.