ATI - Full Tri Performance Hit

[Reverend]

But at least they prove in pain killer trylinear is better than brilinear .

Not that I care too much about all these new filtering-mucking-about (other than for the ultra-important "apples-to-apples" comparisons) but can someone tell me just wtf is the difference between "trylinear" and "brilinear"? They're not the same?

BTW, what's next? Instead of true anisotropic, we get ansiotropic?

 

[MuD]

That's true about the memory controller. Just look at the Catalysts 4.6 released yesterday. They state in the release notes that AA has improved for X800 cards thanks to memory controller tweaking.

 

[MuD]

But at least they prove in pain killer trylinear is better than brilinear .

Not that I care too much about all these new filtering-mucking-about (other than for the ultra-important "apples-to-apples" comparisons) but can someone tell me just wtf is the difference between "trylinear" and "brilinear"? They're not the same?

BTW, what's next? Instead of true anisotropic, we get ansiotropic?

Trylinear means that the driver tries to determine if a particular mipmap needs "optimizations" or not (which ATI does). If the transitions are too harsh, it'll revert to legacy or "true" trilinear filtering (no optimizations). The optimizations can range from "closer to full trilinear" to "near bilinear" in quality.

Brilinear means that the driver always uses an exact mixture of bilinear and trilinear (exact percentage unknown, perhaps 50%-50%, who knows) for the mipmaps (which nVidia does). There's never a change in the amount of optimizations done (the range "closer to full trilinear" or "near bilinear" never changes).

So it depends from the manufacturer which algorithms they use/prefer. Up to you to decide which one gives the best IQ. Hope this helpes.

 

[Reverend]

Gosh, your explanations sounds like those two words are official 3D terminologies...

 

[MuD]

Gosh, your explanations sounds like those two words are official 3D terminologies...

Nope, just trying to help. Anyone can surf to EB and look it up themselves, if they don't believe me.

 

[Bjorn]

If you compare the NV40 to the R300 it's nothing more then speed enhancments as well along with SM3 support.

And the R300 is nothing more then a R200 with MSAA and SM2.0 support ?

It is also unfair to ATI to overlook the fact that both the R420 and R300 are smaller and consume less power then nvidia's offerings. In fact the X800 XT, while being superior to the R300 speedwise as well as offering new features such as 3dc, TAA and SM2.0b, draws less power then the 9800 XT. The same can hardly be said for nvidia.

TAA isn't exlusive to the X800. And most reviews haven't overlooked the "less power consuming" part. And the fact that the X800 Pro is actually available now compared the the 6800, which "supposedly" is going to ship to the stores in a couple of days.

 

[Demirug]

Trylinear means that the driver tries to determine if a particular mipmap needs "optimizations" or not (which ATI does). If the transitions are too harsh, it'll revert to legacy or "true" trilinear filtering (no optimizations). The optimizations can range from "closer to full trilinear" to "near bilinear" in quality.

Sorry, but this is not right.

There are only two setting. "optimizations" ON and "optimizations" OFF.

The procedure that the make the choice is very simple. If you know how it works you can even show the optimizations with colored Mipmaps. And in this case you have "harsh transitions" that do not get full trilinear.

On the other hand the procedure have such a high threshold that for a normal game texture you will never get the optimazation off.

They only thing that the procedure can do is detect a "strange" use of mipmaps and as I said before even this works not very well.

 

[CarstenS]

The R420 may not be as big of a difference architecturally to the R300 as the NV40 is to the NV30, however that does not mean the NV40 is better then the R420. Anyone with any knowledge will tell you the R300 was a big step ahead of the NV30; the NV40 corrects many of these issues. In order for the NV40 to do this nvidia had to drastically rehaul their architecture. ATI had it right the first go around, so naturally they didn't have to try as hard to achieve similar results to that of the NV40.

I do not want to cast judgement on any of the architectures (except for my personal usage, as that is the only one, i may consider myself the authority for).
I also do not say, that any of those two Chips, nV40 and R420, is better or worse than the other. They were designed just with different emphasis' on what was to be achieved.
You say, R300 was a big step ahead of NV30. Performance-wise you're correct. Feature-wise one could argue, which features are more important than other.
Additionally, these two were also designed under different circumstances and ideas as to what DirectX9 would eventually become. nV was proposing FP16 as "full precision" (with FP32 as an added bonus for the Pro-User doing scientific stuff in offline-Rendering, hence the "speed" penalty) with a fall-back option to INT12 as what is known as partial-precision. ATi finally got their idea of FP24 as "full-precision" cemented in DX9 and such nV30 was forced to use their FP32 intended for offline-rendering to render every DX9-Shader which did not include pp-hints. Also, the delay in nV30 did not help much.
In the following year ATi did wipe the floor with everything "green" DX9-wise.

Today the situation has changed, nV40s final design phase was late enough to be adapted to the final DX9-Specs and as such there's no more 250% performance delta both Chips could be designed to an already given spec and of course nV had to do their homework more thoroughly than ATi, who already had a very fast and efficient DX9-architecture.

Maybe ATi was not sure, if their 10% lead out of a 30% fillrate advantage would not convince enough consumers to buy their products - i don't know. But that's maybe why they decided to stretch that lead artificially to 30% or more - welcome to the wonders of undefined texture filtering definitions (blame M$ - openGL uses a formula for trilinear filtering).

If you compare the NV40 to the R300 it's nothing more then speed enhancments as well along with SM3 support.

If you compare the R300 to the R200 it's nothing more than speed enhancments as well along with SM2.0 support.
edit:
Damn, Bjorn beat me to this....

It is also unfair to ATI to overlook the fact that both the R420 and R300 are smaller and consume less power then nvidia's offerings. In fact the X800 XT, while being superior to the R300 speedwise as well as offering new features such as 3dc, TAA and SM2.0b, draws less power then the 9800 XT. The same can hardly be said for nvidia.

Well, according to my measurements, ATI did indeed a wonderful job at keeping the power consumption and thus heat generation low. In fact, they did such a good job in this regard, that i did not believe it, when i first measured power consumption on R420-Chips and went to the store to get my test-device checked.

But, to an extent, the same can be said about nVidia. Of course not in comparison with the R420 line of chips, but compared to FX5800U and FX5950, the 6800u as a whole uses less power in 3D-Applications, albeit having almost doubled on transistor count, added SM3, RGAA, FP-Texture Filtering, Tone-Mapping via RAMDAC, quadrupled the number of pipelines etc.

 

[Pete]

jvd, I agree R420 seems to sport some architectural advances over R300, as evidenced by CB.de's "no opt.s" tests that show R420 outperforming RV350 at quadruple the res. I just don't think those advances can be qualified as "major." (BTW, TAA doesn't seem to be something you can describe as new on the hardware side, given that R300 can use it.)

Anyone with any knowledge will tell you the R300 was a big step ahead of the NV30; the NV40 corrects many of these issues.
Performance-wise as it relates to gaming, yes. Feature-wise, I'm not sure it's that clear-cut; in fact, PS2.a would seem to argue in NV30's favor.

If you compare the NV40 to the R300 it's nothing more then speed enhancments as well along with SM3 support.
That's a pretty tough sell, IMO. SM3 doesn't seem to be an easy add-on, as ATI has showed by not "tacking it on" to R420.

 

[DegustatoR]

trylinear = brilinear for some textures + a couple of other tricks which DEGRADE visual quality.

As for painkiller -- i'm failing to any difference between FX5900 and X800Pro. Well, no, when i'm starting to think about it, i'd say that FX5900 has slightly better aniso in Painkiller

 

[CarstenS]

jvd, I agree R420 seems to sport some architectural advances over R300, as evidenced by CB.de's "no opt.s" tests that show R420 outperforming RV350 at quadruple the res. I just don't think those advances can be qualified as "major."

Please keep in mind, that the X800 although features quadruple the memory bandwidth, which in den CB-Setting with 4xAA enabled in 800x600 still is a factor. Although hierarchicalZ does not work with only one pixelprocessor enabled in R9600 (this is good for about 10% performance also).

 

[Xmas]

The M$ comment, while made in an nV40 context, cleary stated that M$ had not yet updated the DX rasterizer for any new-gen gpu/vpu.

Even if you repeat it endlessly, (and I pointet this out to you before) it did not.

Ditto, yourself... That is exactly what the unnamed M$ employee is quoted as saying.

I wonder, if that is exactly what the unnamed MS employee is quoted as saying, why can't I find it in the quote?

Where in this statement is R420 excluded from [...]

Nowhere. It isn't even mentioned. But why do you think that's what I was arguing?

the unnamed M$ employee was asked by TR why the DX rasterizer results in a later set of Forcenators was inferior to the result obtained with the rasterizer in an earlier set of Forcenators,

He was asked why there is such a difference between NV40 and the reference rasterizer. The driver (or the hardware, for that matter, besides the CPU FP precision) has no influence whatsoever on what the reference rasterizer does.

(I am making no comment on the believability of the statement whatever.) I do find it odd that the M$ employee wrote in run-on sentences--two of them back to back--which would at least superficially indicate the author was a bit challenged in English punctuation, perhaps. I am ever skeptical of unattributed quotes...

I can assure you that the person is absolutely correct in stating that the LOD selection algorithm in NV40 is better than the one used in the refrast, no matter who he is. And that it is better than the one used in R420, which is very similar to the one used in refrast.

 

[Pete]

Quasar, you mean double the bandwidth, right? Or did you change the default clocks (4P@500/300 vs. 16P@525/560MHz = 4:1 core, 2:1 mem)? I figured improvements in HyperZ and bandwidth compression contributed to the X800XT's 32% advantage at 16x12 AF and 43% advantage at 16x12 AA+AF.

 

[MuD]

Trylinear means that the driver tries to determine if a particular mipmap needs "optimizations" or not (which ATI does). If the transitions are too harsh, it'll revert to legacy or "true" trilinear filtering (no optimizations). The optimizations can range from "closer to full trilinear" to "near bilinear" in quality.

Sorry, but this is not right.

There are only two setting. "optimizations" ON and "optimizations" OFF.

The procedure that the make the choice is very simple. If you know how it works you can even show the optimizations with colored Mipmaps. And in this case you have "harsh transitions" that do not get full trilinear.

On the other hand the procedure have such a high threshold that for a normal game texture you will never get the optimazation off.

They only thing that the procedure can do is detect a "strange" use of mipmaps and as I said before even this works not very well.

Well, no offense, but I believe you're not up to date with the facts. Drivers nowadays can do some pretty clever stuff, INCLUDING changing optimizations on the fly. That's why they called it TRYlinear (the driver is trying to guess how harsh the mipmap transitions are and thereby adjusting the filtering ("optimizations") on a frame by frame scenario. Look at it this way:

- If the driver determines that transitions are too harsh, "optimizations" will not occur and full tri-linear filtering will be applied.

- If the driver determines that the transitions are smooth enough (mipmaps that don't differ that much from each other), more "optimizations" are done, meaning the driver will dynamically determine how much filtering (including "closer to tri-linear" or "near bilinear") should be applied. But because the driver makes all these calculations so fast, you won't notice a difference. This has been confirmed by ATI themselves. Look up the chatlog that was held a few weeks ago about this issue.

Did you know that drivers can even detect when a screenshot is taken and output a higher quality image than what is actually displayed on the monitor? Every day drivers are getting smarter (sneakier I would say) and it's very difficult to give an apples to apples comparison, if that's even possible.

 

[Demirug]

Well, no offense, but I believe you're not up to date with the facts. Drivers nowadays can do some pretty clever stuff, INCLUDING changing optimizations on the fly. That's why they called it TRYlinear (the driver is trying to guess how harsh the mipmap transitions are and thereby adjusting the filtering ("optimizations") on a frame by frame scenario. Look at it this way:

- If the driver determines that transitions are too harsh, "optimizations" will not occur and full tri-linear filtering will be applied.

- If the driver determines that the transitions are smooth enough (mipmaps that don't differ that much from each other), more "optimizations" are done, meaning the driver will dynamically determine how much filtering (including "closer to tri-linear" or "near bilinear") should be applied. But because the driver makes all these calculations so fast, you won't notice a difference. This has been confirmed by ATI themselves. Look up the chatlog that was held a few weeks ago about this issue.

Did you know that drivers can even detect when a screenshot is taken and output a higher quality image than what is actually displayed on the monitor? Every day drivers are getting smarter (sneakier I would say) and it's very difficult to give an apples to apples comparison, if that's even possible.

MuD, I know that a driver can do many thing. I have write some driver by myself. Not for graphic adapters but I even know this part of the DDK/LDK.

But this will not stop me to ask who gives you they information about how the detection works? I ask this because I want to know who have lie to you.

Perhaps you or somebody else will ask now from where I take this insolence? The answer is stupid simple. I was the person who wrote they special benchmark programm that have analyzed ATI "TRYlinear". The results of many different test runs allows my coworker and me to show the "optimisations" even with colored mipmaps that have harsh transitions.

They only thing that "TRYlinear" really tries is to hide itself from detection. But as I said and show before it even fail on this.

But as always I am open for new aspects in this story. Therefore as a request to all: Please show me a texture that use something between full trilinear and that default optimisation level.

 

[jimmyjames123]

Some people have their blinders on, no matter what you try to tell them. MuD is just quoting marketing-speak anyway.

These large corporations do not act like saints. Their primary goal is to increase marketshare, top their competition, and make money for their stockholders.

The fact of the matter is that, in this graphics marketplace, speed sells. So each company will have major incentive to boost speed as much as possible using both hardware and software optimizations.

 

[FUDie]

But this will not stop me to ask who gives you they information about how the detection works? I ask this because I want to know who have lie to you.

Perhaps you or somebody else will ask now from where I take this insolence? The answer is stupid simple. I was the person who wrote they special benchmark programm that have analyzed ATI "TRYlinear". The results of many different test runs allows my coworker and me to show the "optimisations" even with colored mipmaps that have harsh transitions.

They only thing that "TRYlinear" really tries is to hide itself from detection. But as I said and show before it even fail on this.

So because you wrote some "special benchmark program" you are somehow more qualified to speak about what ATI is doing than ATI themselves? That's rich!

But as always I am open for new aspects in this story.

It sure doesn't sound like you are open at all. You've already made up your mind, judging by your posts.

-FUDie

 

[ANova]

And the R300 is nothing more then a R200 with MSAA and SM2.0 support ?

I never claimed anything more did I? But to be fair, SM2 was a bigger upgrade from SM1 then SM3 is to SM2. SM3 is more akin to a small update.

TAA isn't exlusive to the X800. And most reviews haven't overlooked the "less power consuming" part. And the fact that the X800 Pro is actually available now compared the the 6800, which "supposedly" is going to ship to the stores in a couple of days.

At the moment, can you enable TAA on nvidia hardware? Yes review sites don't overlook the less power consuming aspect, but most people that argue in favor of the 6800 do. Also, forgive me for being skeptical of 6800s making it to stores in significant quantities anytime soon.

You say, R300 was a big step ahead of NV30. Performance-wise you're correct. Feature-wise one could argue, which features are more important than other.

I'm curious, what features do you think the NV30 had over the R300?

Additionally, these two were also designed under different circumstances and ideas as to what DirectX9 would eventually become. nV was proposing FP16 as "full precision" (with FP32 as an added bonus for the Pro-User doing scientific stuff in offline-Rendering, hence the "speed" penalty) with a fall-back option to INT12 as what is known as partial-precision. ATi finally got their idea of FP24 as "full-precision" cemented in DX9 and such nV30 was forced to use their FP32 intended for offline-rendering to render every DX9-Shader which did not include pp-hints.

Nvidia has no one to blaim but themselves for this. They assumed MS would accept their proposal of DX9 no questions asked, well...they assumed wrong. Besides, FP24 is better then FP16 so the consumer was the beneficiary.

But, to an extent, the same can be said about nVidia. Of course not in comparison with the R420 line of chips, but compared to FX5800U and FX5950, the 6800u as a whole uses less power in 3D-Applications, albeit having almost doubled on transistor count, added SM3, RGAA, FP-Texture Filtering, Tone-Mapping via RAMDAC, quadrupled the number of pipelines etc.

From what I've seen the 6800 Ultra's power consumption is a fair amount higher then the 5950. Why do you think it needs two molex connectors? The 5950 doesn't.

Performance-wise as it relates to gaming, yes. Feature-wise, I'm not sure it's that clear-cut; in fact, PS2.a would seem to argue in NV30's favor.

It's obvious that it doesn't from looking at the NV30's performance and image quality.

That's a pretty tough sell, IMO. SM3 doesn't seem to be an easy add-on, as ATI has showed by not "tacking it on" to R420.

You make it sound as if ATI was unable to implement it, which isn't the case. ATI simply didn't see any benefit as the costs outweighted the advantages. It also would have effected ATI's yields as we are seeing with nvidia.

 

[Pete]

It's obvious that it doesn't from looking at the NV30's performance and image quality.
I explicitly separated gaming performance from features.

You make it sound as if ATI was unable to implement it, which isn't the case. ATI simply didn't see any benefit as the costs outweighted the advantages. It also would have effected ATI's yields as we are seeing with nvidia.
Yep, because that's exactly what I believe happened. I believe ATi had engineers dedicated to too many other projects (Xbox 2, Gamecube 2) and was sitting on an already-class-leading architecture, so they stuck with SM2.0 for two reasons: one, b/c they didn't have enough resources to release a SM3.0 part on time, and two, b/c they wanted to cut the legs out from under SM3.0 by limiting their parts to 2.0, thus forcing the market to aim for the lowest common denominator, thus rendering SM3.0 mostly moot for this generation.

I'm not sure why my speculation is less valid than yours, though. You're sure ATi didn't see any benefit to SM3.0? Did they see a benefit to hiding trylinear, or to releasing the 8500 only to see it eclipsed by the GF4 in a matter of weeks? ATI is not all-powerful, and I believe their hand was somewhat forced, and they played it as best they could ATM (which is pretty well, considering they were in the position of power WRT mindshare). The fact is that we have gone from ATi being a tech generation ahead in features with the 8500, to nV being a tech generation ahead with the 6800. Whether things will play out the same way as 8500->9700 and GF4->GF FX is hard to say, particularly with the seemingly increasing process limitations. But I don't think you can say ATi had the power to implement SM3.0 and chose not to solely based on profit margins. They'd be setting themselves up to make even more money if they'd kept the tech lead for two generations in a row, rather than ceding it to nVidia after just one generation clearly on top.

 

[radar1200gs]

And the R300 is nothing more then a R200 with MSAA and SM2.0 support ?

I never claimed anything more did I? But to be fair, SM2 was a bigger upgrade from SM1 then SM3 is to SM2. SM3 is more akin to a small update.

TAA isn't exlusive to the X800. And most reviews haven't overlooked the "less power consuming" part. And the fact that the X800 Pro is actually available now compared the the 6800, which "supposedly" is going to ship to the stores in a couple of days.

At the moment, can you enable TAA on nvidia hardware? Yes review sites don't overlook the less power consuming aspect, but most people that argue in favor of the 6800 do. Also, forgive me for being skeptical of 6800s making it to stores in significant quantities anytime soon.

You say, R300 was a big step ahead of NV30. Performance-wise you're correct. Feature-wise one could argue, which features are more important than other.

I'm curious, what features do you think the NV30 had over the R300?

Additionally, these two were also designed under different circumstances and ideas as to what DirectX9 would eventually become. nV was proposing FP16 as "full precision" (with FP32 as an added bonus for the Pro-User doing scientific stuff in offline-Rendering, hence the "speed" penalty) with a fall-back option to INT12 as what is known as partial-precision. ATi finally got their idea of FP24 as "full-precision" cemented in DX9 and such nV30 was forced to use their FP32 intended for offline-rendering to render every DX9-Shader which did not include pp-hints.

Nvidia has no one to blaim but themselves for this. They assumed MS would accept their proposal of DX9 no questions asked, well...they assumed wrong. Besides, FP24 is better then FP16 so the consumer was the beneficiary.

But, to an extent, the same can be said about nVidia. Of course not in comparison with the R420 line of chips, but compared to FX5800U and FX5950, the 6800u as a whole uses less power in 3D-Applications, albeit having almost doubled on transistor count, added SM3, RGAA, FP-Texture Filtering, Tone-Mapping via RAMDAC, quadrupled the number of pipelines etc.

From what I've seen the 6800 Ultra's power consumption is a fair amount higher then the 5950. Why do you think it needs two molex connectors? The 5950 doesn't.

Performance-wise as it relates to gaming, yes. Feature-wise, I'm not sure it's that clear-cut; in fact, PS2.a would seem to argue in NV30's favor.

It's obvious that it doesn't from looking at the NV30's performance and image quality.

That's a pretty tough sell, IMO. SM3 doesn't seem to be an easy add-on, as ATI has showed by not "tacking it on" to R420.

You make it sound as if ATI was unable to implement it, which isn't the case. ATI simply didn't see any benefit as the costs outweighted the advantages. It also would have effected ATI's yields as we are seeing with nvidia.

At the moment, can you enable TAA on nvidia hardware?
Why would you want to? Like most other people I've spent a lot of time money and effort getting as far away from low refresh rates as its possible to get. I have no intention of returning anytime soon.

I'm curious, what features do you think the NV30 had over the R300?
(almost) everything ATi brought to the table with R420 pixel shader and vertex shader wise and more (considerably more on the vertex side) when it comes to flow control and looping. Isn't it funny how nVidia picked up some of ATi's OpenGL extensions with no fuss, yet ATi can't bring themselves to just use SM2.0a, no they have to have 2.0b, just to be different from nVidia.

Besides, FP24 is better then FP16 so the consumer was the beneficiary.
FP16 is good enough for George Lucas and ILM. We still aren't anywhere near cinema quality games yet. And the consumer certainly hasn't benefitted from high full precsion requirements in DX9, in fact this requirement has arguably delayed the massive uptake of DX9 featured games by 2 years.

(SM3) You make it sound as if ATI was unable to implement it, which isn't the case. ATI simply didn't see any benefit as the costs outweighted the advantages. It also would have effected ATI's yields as we are seeing with nvidia.
Don't worry, ATi will be seeing the benefits of SM3.0 all too clearly real soon now...