[Mintmaster]

Originally Posted by Bill

That review shows stock X1800Xt going toe to toe with 512 GTX at highest settings. 512 GTX being 799.

Bodes well for ATI's future. Considering ATI's pipeline deficit and near eqaul clocks.

I think R580 vs. 32-pipe G71 won't be the blowout that everyone expects it to be. Remember, though, that this review is showing newer games, and ATI does better on these games (for the most part anyway).

Still, if the Inquirer is right, things won't be easy for ATI. I thought 750MHz would be a 24-pipe chip, and 32-pipe, if true, would be clocked lower, at say 600MHz. ATI would need dynamic branching to appear in at least one game, or else they'll keep losing marketshare. Of course, we all remember what the Inquirer was saying about R520.

Nonetheless, if G80 is indeed coming out this summer, even if late summer, I think a 24-pipe card would be smarter. They could kill ATI's margins, keep power consumption under control, and use the same chip for a long time.

[ondaedg]

Originally Posted by Skrying

Laugh, you're point is semi right, but you thinking that $300 cards bring in the bucks is a mistake. Yes they bring in more than the $450+ ones, but the real monkey makers are the $80~$200 range. The X1300/X1600 and 7200/7600 (when they come) are what makes the big money, and grabs the OEM deals.

You also under estimate the mind share value of high end cards. People will think bad of your range of products if you can not produce enough of your high end to satisfy the want, to a point.

I disagree with you. OEM means smaller margins on low cost high volume products. Wasn't it Jen who stated that the 6800 GT was bringing in great sales and amazing margins and did he not state that they were a major contributor to their positive earnings approximately 3 quarters ago?

In the past, OEM wins would typically indicate the strength of the financial earnings of the IHV. Whoever had the most press releases announcing wins would typically have the best quarter. I don't believe that is the case right now. It still plays a significant role in the overall earnings of the IHV, but it no longer dictates who is profiting the most. The sales of the 6800 GT and the 6600 GT to the retail market is evidence of that.

edit: I just realized you included the 200.00 segment. I should correct myself and refer to the 200-300.00 market.

[ToxicTaZ]

If the spring refresh 90nm G71 @750MHz with 24/32 pipes on it what would be the cooling for that GPU? would it be the same as the GTX 512 and use a dual-slot aluminium or copper cooling with at least four heatpipes tech.


Air or Water cooled?

Help give some info on what kind of cooling they would need?

[AlphaWolf]

Originally Posted by ToxicTaZ

If the spring refresh 90nm G71 @750MHz with 24/32 pipes on it what would be the cooling for that GPU? would it be the same as the GTX 512 and use a dual-slot aluminium or copper cooling with at least four heatpipes tech.


Air or Water cooled?

Help give some info on what kind of cooling they would need?

You aren't going to see water cooling for mainstream PC parts anytime soon, its only going to be offered as a speical purpose by AIBs. I'd expect a dual slot air cooler, probably something lower cost than the 7800gtx512s cooler.

[ANova]

I expect the G71 and R580 to be clocked around 650-700 each.

[trinibwoy]

Originally Posted by ANova

I expect the G71 and R580 to be clocked around 650-700 each.

Hmmm when was the last time these two were on the same process? ATi did great with 150nm and Nvidia squeezed the life out of 110nm. It's gonna be interesting to see who gets more out of the process although ATi has one hell of a headstart. They've already got R520, RV530 and Xenos under their belt. What's Nvidia got - a motherboard chipset ?

[Geo]

Originally Posted by ANova

I expect the G71 and R580 to be clocked around 650-700 each.

Dude, did you play poker with Orton and Jen-Hsun one night recently? You're just laying down the law of late. You should sig-up so we can check it later.

[mrcorbo]

Originally Posted by Mintmaster

You gave NO evidence whatsoever for your statement. What did you say? "HDR+FSAA isn't worth the time it took to mention it"?

Actually what I said was this and I'm going to quote since you seem to be unwilling to go back and look yourself so you might actually know what you are talking about.

Originally Posted by mrcorbo

It doesn't matter the % hit you take if the end result is that you don't get playable framerates at the resolutions you expect to be able to play at. You have to figure that anyone shelling out the cash for one of these cards is likely to have a display capable of at least 1600X1200 and a hard-core gamer would be loathe to sacrifice resolution or playable framerates for eye-candy.

You then responded with this post: http://www.beyond3d.com/forum/showpo...&postcount=107

Which besides the fact that you decided to toss in the insults (which only addressed your ability to act like a tool), completely failed to address the original statement.

Fact is, at this point if my card couldn't do HDR+MSAA then I just wouldn't turn on the HDR. It's nice if you've got it, but hardly necessary. That was the eye-candy I was referring to. 1600X1200 w/ 4XAA and AF is pretty much a given as long as you can still get playable framerates.

You can disagree with me about how inportant this feature is, I don't really care. It's pretty damn ridiculous that this much came out of what was really an off-hand comment saying that implementing MSAA+HDR on G71 probably isn't that important relative to upping the quality on the AF and that there were performance issues (an opinion I formed from looking at the exact benchmarks you referenced).

In this context I don't get what all the XT and GTX performance comparisons were supposed to prove. They seemed completely OT to me.

[Mintmaster]

OK mrcorbo, I owe you an apology. I confused your original post with this one, which had a similar conclusion (AF is the only real advantage of the X1800XT, HDR+AA is not useful).

Then when you said percentage doesn't matter, I figured you implied all ATI HDR scores were too low. Why else would a low % hit for AA be meaningless?

Anyway, I'm sorry and let's ignore the past and start again. Are you saying it doesn't matter how good a game looks, you'd disable graphics options until you get playable framerates at 1600x1200? Even the GTX only gets 43fps at 1600x1200 in FarCry HDR, and 41fps in SS2. So is FP blending useless on G70/NV40?

If you're only talking about AA, then are you saying AA is useless unless 1600x1200 is playable?

From your post, these are the only conclusions I can make. I don't see how you can generalize hardcore gamers to have such a one-dimensional outlook on image quality.

[mrcorbo]

Originally Posted by Mintmaster

OK mrcorbo, I owe you an apology. I confused your original post with this one, which had a similar conclusion (AF is the only real advantage of the X1800XT, HDR+AA is not useful).

Then when you said percentage doesn't matter, I figured you implied all ATI HDR scores were too low. Why else would a low % hit for AA be meaningless?

Anyway, I'm sorry and let's ignore the past and start again. Are you saying it doesn't matter how good a game looks, you'd disable graphics options until you get playable framerates at 1600x1200? Even the GTX only gets 43fps at 1600x1200 in FarCry HDR, and 41fps in SS2. So is FP blending useless on G70/NV40?

If you're only talking about AA, then are you saying AA is useless unless 1600x1200 is playable?

From your post, these are the only conclusions I can make. I don't see how you can generalize hardcore gamers to have such a one-dimensional outlook on image quality.

Thanks for the apology, I figured that must have been carry-over from another thread. Anyway, I agree, let's move on.

Let me ask you this; Given a theoretical R580 and G71 where G71 had equal or better AA/AF quality and a noticable performance lead at all resolutions, but couldn't do MSAA+HDR, which card would you buy? And will there be a lot of titles supporting this version of HDR rendering or will more developers go Valve's route and make it compatable with both?

As for resolution vs. AA settings, it depends on the game. If the game looks like crap without 4XAA then I probably would sacrifice some resolution. Personally I doubt that this will ever be an issue on this coming generation of cards. I think we're likely going to get 16X12 4XAA/16XFF from even F.E.A.R. on R580 and probably G71, too.

Oh, yeah. And another bit of irony. My display limits me to 1360x768, so the X1800XL actually looks very attractive to me in my situation, because of the IQ. Just waiting to see what they do with AVIVO (PureVideo is a very important feature for me) and see what happens with the Spring refreshes (7600 and maybe, just maybe an 8-1-3-1 part from ATI).

[AlexV]

Valve`s HDR is a bit hackish, IMHO, at least from what is known about it.FP16 blending based HDR OTOH is here to stay so yes, this will probably be the preffered method of doing HDR rendering for some time to come.

[Mintmaster]

Regarding your hypothetical scenario, I'd need to know what the speed advantage was for G71. However, I'm pretty sure I'd pick R580 because I want to keep the card for a few years. I want low resolution to look good, and AA is the only way to do that. Compare 800x600 without AA and contrast that with 640x480 with 6xAA. The latter looks enormously better, because the pixels are relatively huge. I also have ideas about using AA to improve shadow maps (which need high precision), so FP16/I16+AA is important.

Valve's HDR in source is essentially a hack. It doesn't even use FP blending, so if you think developers should adopt this method, ATI should have just doubled R480 this gen and paired that with the new memory controller. This hack won't be able to offer all the benefits of HDR, so I doubt it'll be the way of the future.

About your argument that some games don't need AA: If a game doesn't look bad without AA, then the sharp transition of an edge is not really visible. Clearly that means you can't see details as well, so what does higher resolution get you anyway?

Oh, and you didn't answer my questions yet. Do you really think gamers will reduce graphics setting in game just to have 1600x1200? There aren't many games where GTX will get playable HDR at this resolution. There's a lot more to image quality than resolution. I'd rather see photorealistic results at 640x480 than FEAR at 2048x1536.

[Mintmaster]

While on the topic of image quality, I think I should say a bit about AF in general.

I will not play without AF under any circumstances. It sharpens textures far better than resolution does. Even for areas on the screen requiring only 4xAF, 640x480 will be sharper than 2048x1536. (The only exception is high frequency specular normal mapping, but that's another matter.) To me it is absolutely incomprehensible why you would play at a higher resolution without AF than at a lower resolution with it. The whole point of upping the resolution is to increase detail. I simply will not accept that anyone will choose the highest playable resolution without AF instead of the highest playable resolution with AF, if they actually honestly compared the image.

So all the benchmarks we see on the internet without AA/AF are pointless. Once you throw AF in there, AA only has a small additional hit. Even for G70, the real AA performance hit (i.e. from 0xAA/16xAF to 4xAA/16xAF) is probably not much more than 20%, whereas increasing the resolution one notch costs 36% (both numbers will be lower if partially CPU or geometry limited).

Basically my point is that the whole AA vs. resolution argument, as portrayed by 99% of the benchmarks, isn't even an apples to apples comparison. It's actually a resolution vs. AA and AF argument, and the former just can't win in an honest comparison.

[Skrying]

Funny that Valve's HDR is a "hack" yet its by FAR (and by FAR I mean REALLY REALLY FAR) better looking than any other HDR I've seen in other games. Doesnt show near the freaking bugs, nor near the side effects, yet adds tons of subtile lightning enhancments. Playing de_nuke with HDR on is really a treat, and its very playable with decently good settings (1280x1024, 2xAA, 8xAF, rest settings maxed).

Valve amazes me. By the way some talk about their engine you'd think its crap, but in the hands of a talented map designer and texture artist it really wows the crap out of me.

[Moloch]

Originally Posted by Mintmaster

Regarding your hypothetical scenario, I'd need to know what the speed advantage was for G71. However, I'm pretty sure I'd pick R580 because I want to keep the card for a few years. I want low resolution to look good, and AA is the only way to do that. Compare 800x600 without AA and contrast that with 640x480 with 6xAA. The latter looks enormously better, because the pixels are relatively huge. I also have ideas about using AA to improve shadow maps (which need high precision), so FP16/I16+AA is important.

Valve's HDR in source is essentially a hack. It doesn't even use FP blending, so if you think developers should adopt this method, ATI should have just doubled R480 this gen and paired that with the new memory controller. This hack won't be able to offer all the benefits of HDR, so I doubt it'll be the way of the future.

About your argument that some games don't need AA: If a game doesn't look bad without AA, then the sharp transition of an edge is not really visible. Clearly that means you can't see details as well, so what does higher resolution get you anyway?

Oh, and you didn't answer my questions yet. Do you really think gamers will reduce graphics setting in game just to have 1600x1200? There aren't many games where GTX will get playable HDR at this resolution. There's a lot more to image quality than resolution. I'd rather see photorealistic results at 640x480 than FEAR at 2048x1536.

640x480 looks completely horrible on a 19" display.
I would prefer the former... well atleast 1024x768 for a 19"

[mrcorbo]

Well with FPS games resolution can impact playability. And in this type of game I'm not there to admire the scenery . For that matter, even in a game like AOEIII higher resolution is desirable, though you can certainly get away width lower framerates. So, truth be told, unless there is a really noticable flaw in the visuals that ruins the immersion effect of the game, I probably would take higher resolution. Now if we're talking only one "step" like 16X12 to 12X10 I could go either way, again it would depend on the game, but probably not more than that. Hell, if I'm spending more time analyzing IQ than fragging maybe I'm playing the wrong game.

[Ailuros]

Mintmaster,

Screen resolution is directly connected to the size of the monitor. Ideally I'll have the highest possible resolution combined with AA/AF. If I'd have to reduce something it would be Multisampling first, but I wouldn't go below 1152 on this 21" CRT. 640*480 or 800*600 lookslikeass(tm) on this one and only Supersampling with an insane sample density could possibly make any of those look slightly better, but it won't change the fact that in 640 I could if I try hard enough to start counting pixels (yeah I know it's an exaggeration, but there's truth to it).

Even for G70, the real AA performance hit (i.e. from 0xAA/16xAF to 4xAA/16xAF) is probably not much more than 20%, whereas increasing the resolution one notch costs 36% (both numbers will be lower if partially CPU or geometry limited).

In games like Fear? I lose from 2x to 4xAA ~30%; can you imagine how much I lose from 1x to 4xAA?

Compare 800x600 without AA and contrast that with 640x480 with 6xAA.

See above. On mine here I get ~50dpi with the first and around ~40dpi with the latter. Apart from polygon edges and intersections 6x Multisampling is going to do what exactly to polygon interior data and since when does antialiasing replace any resolution? When you're polygons are huge they'll stay huge in a very low resolution; Multisampling in the best case will try to cure some "edge aliasing" but even 6x isn't enough there. Try rather something beyond 32x samples and then we can talk.

I'd rather see photorealistic results at 640x480 than FEAR at 2048x1536.

That's a huge ballpark you're setting there. In connection to the above I get on the horizontal axis over 130dpi in 2048*1536, and that's more than 3x the value of 640*480.

R580 type of hardware will most likely be playable in 1600 with 4xAA/16xAF so you'd theoretically need neither/nor of rather extreme paradigms (unless of course we're talking about a 10" monitor).

If you want something close to 3-dimensional "photorealism" re-watch one of Pixar's last movies. Rendered most likely in HDTV resolutions with 64x stochastic SSAA and yes we're many lightyears from that apart.

[Mintmaster]

radeonic, that's nonsense. Are you saying you can't watch a DVD on a PC monitor? Are you saying instead of watching Finding Nemo at 720x480, you'd rather see it rendered on a graphics card at 1600x1200 with rendering options set to make it 24fps? If you're talking about electron gun scanlines at 640x480, then that's solved easily by upscaling the final image. (These questions apply to you as well Ailuros.)

As another example, take any tack sharp high res photo (DSLR 8MP+) and downscale it to 640x480 with bilinear filtering in Photshop. Then take the same original photo and downscale to 800x600 with nearest neighbour filtering. Run your monitor at a high res, and view the images at full screen size. Which looks better? The difference would be even more dramatic in motion.

True, this example overstates the difference because texture filtering gets rid of aliasing (well it's supposed to!) in the interiors of surfaces, but you get my point.

[Ailuros]

Originally Posted by Mintmaster

radeonic, that's nonsense. Are you saying you can't watch a DVD on a PC monitor? Are you saying instead of watching Finding Nemo at 720x480, you'd rather see it rendered on a graphics card at 1600x1200 with rendering options set to make it 24fps? If you're talking about electron gun scanlines at 640x480, then that's solved easily by upscaling the final image. (These questions apply to you as well Ailuros.)

Yes they do apply to me also, but a movie is a fundamentally different thing than a interactive 3D game to start with.

Second while the video source resolution is 720*586, playback software will use my desktop's resolution and not anything below.

I can use doublescan through ForceWare drivers in 640 and add maybe 16xAA, it'll still looklikeass. Again I said on a 21" CRT; is it so hard for some of you folks to understand the monitor size to resolution relation here?

Third movies use real motion blur and aren't interactive.

True, this example overstates the difference because texture filtering gets rid of aliasing (well it's supposed to!) in the interiors of surfaces, but you get my point.

This would only be a valid point if I'd run anything w/o texture filtering. 16x full trilinear with no texturing stage optimizations in 1600*1200 with 4xAA for me if possible. If not then the next best candidate is either 1600 + 2xAA or 1280 + 4xAA.

You're not using only vast generalizations you're also overlooking screen size which has to be factored in. 640 looks gorgious on a large screen handheld device these days, but that's about it.

***edit: and since this debate actually was about float HDR, I will most likely be able to use it with MSAA on true next generation hardware and I expect it to be out either before or at the same time as games like UT2k7. I have severe doubts that I'll be forced to go lower than 1280 with AA, HDR and AF enabled in that one; so any 640 exaggerations are more than out of place here.

[ANova]

Originally Posted by Mintmaster

Valve's HDR in source is essentially a hack. It doesn't even use FP blending, so if you think developers should adopt this method, ATI should have just doubled R480 this gen and paired that with the new memory controller. This hack won't be able to offer all the benefits of HDR, so I doubt it'll be the way of the future.

Who's to say one method is the "right" method above any other if the end result is the same or the differences are unnoticable to the naked eye? Valve's method seems to be quite effective considering it works on any card that supports SM2 functionality, rather then only those that support FP16 blending.

[Mintmaster]

Ailuros, regarding lookslikeass(tm), are you talking about how a computer CRT looks (due to scanlines) or are you complaining about the lack of spatial information? The latter is what I'm arguing about. Simple upscaling solves the former.

Originally Posted by Ailuros

In games like Fear? I lose from 2x to 4xAA ~30%; can you imagine how much I lose from 1x to 4xAA?

You sure that's with AF enabled? Anyway, of all current games in XBit's recent roundup, FEAR is the only one with an AA hit anywhere near that high, and their "Pure Speed" scores don't have AF either.

See above. On mine here I get ~50dpi with the first and around ~40dpi with the latter. Apart from polygon edges and intersections 6x Multisampling is going to do what exactly to polygon interior data and since when does antialiasing replace any resolution? When you're polygons are huge they'll stay huge in a very low resolution; Multisampling in the best case will try to cure some "edge aliasing" but even 6x isn't enough there. Try rather something beyond 32x samples and then we can talk.

Regarding polygon interiors, that's exactly why scores without AF are pointless. For edge quality, if you place such low value on MSAA, then how can you find so much merit in 32xAA over 6xAA? Going from no AA to 6xAA is an order of magnitude larger leap in quality than this. There's only a small subset of line angles and contrast combinations that will be visibly better with 32xAA. In fact, aliased 1080p downsampled to 320x180 will have edge quality similar to 320x180 with 6xAA.

Resolution and antialiasing are very related. Both aid in figuring out where an outline is positioned, and outlines are a major component of visual perception. Supersampling (equivalent to MSAA under most scenarios for realtime graphics) displays the information of higher resolution but the averaging smears it out a bit. I do not place a 1:1 relationship between the values of more pixels vs. more samples, but certainly it's not close to 3:1.

[Mintmaster]

Originally Posted by ANova

Who's to say one method is the "right" method above any other if the end result is the same or the differences are unnoticable to the naked eye? Valve's method seems to be quite effective considering it works on any card that supports SM2 functionality, rather then only those that support FP16 blending.

Well, the end result isn't the same. You won't get any HDR effects through a window, and you can't get antialiasing on HDR to LDR transitions (Valve masks this deficiency with bloom).

It's funny that with you I'm debating whether 8-bit is enough for the framebuffer, but earlier in this thread bloodbob and Chalnoth were arguing that 16-bit (non-FP) is not enough even for textures.

Here's a quote from an interview:

HDR version 3 was unveiled at E3 this year. According to Valve, it is the HDR implementation of the future - just not of the present. It included refraction, but could be difficult for some hardware, as floating-point alpha blending was used; this was not supported on many devices and was quite a RAM hog.

Who's to say what's right? I would say valve is pretty qualified to judge their own method.

[nAo]

What is Valve's method?! How it works? I'm really curious..

[Jawed]

This is the best description I know of:

http://arstechnica.com/articles/culture/lostcoast.ars/3

It appears that it isn't uniform across ATI and NVidia hardware.

Jawed

[nAo]

Originally Posted by Jawed

This is the best description I know of:

http://arstechnica.com/articles/culture/lostcoast.ars/3

It appears that it isn't uniform across ATI and NVidia hardware.

Jawed

unfurtunately it doesn't explain much about their final method..even though it sounds a bit hacky (disclaimer: hacky doesn't mean inferior, ok? )