Why was Far Cry 1.2 really recalled?

[KakimotoR]

Firstly open the Farcry Console using the ` key.
For SM2.0b path type (without quotes) “\r_sm2bpath 1” and hit return.
For Instancing support type (again without quotes) “\r_GeomInstancing 1”
Of course if your enabling Instancing you’re going to want to bump up the vegetation detail, to do this you need to change your Farcry settings from (e_vegetation_sprites_distance_rati o,1.000000) to (e_vegetation_sprites_distance_rati o,100.0000). This change removes spite vegetation and replaces it with animated vegetation throughout the scene.

With that /r_sm2bpath 1 , does that actually improve Image Quality for NVIDIA cards or decrease?

Wat's the original path?

 

[ZenOps]

24-bit is actually very standard, even in 2D. All Jpegs are 24-bit integer.

Why the obsession with Base2? It doesn't have a dithered performance hit as long as the bus is larger than the bitdepth you are trying to use. IE: a 40 or 48 bit colour precision scanner is no less valid than a 32 or 64 bit scanner, its an acceptable happy-medium in visual quality vs performance. Just because NVidia cannot figure out a non-base-2 visual system does not mean everyone else can't.

Everthing we see is base3, visual reality is base3 RGB colour (unless you are colourblind, of which approximately 20 percent of males are) Life is 3D which is X,Y,Z. Microsofts Xbox2 will use a triple core CPU.

I would like to see future PVS have support for 48-bit precision (for CAD/CAE) even more than 32-bit or 64-bit.

PS: Base3 is easier for humans to use and understand, Base2 is easier for computers to use and understand. I think the computers should be conforming to our way of perceptions and needs, we should not be conforming to what the computer wants.

 

[Lord of Shadows]

Well, I would like to think base10 being more the norm than base 3, and 8 bits per color is fine, and thats what most games use, but its nice to have an extra 8 bits for alpha intensity. But untill I get a hold of a good OpenGL book and write some shader code I wouldnt consider myself an authority on these things. (Just pre-empting some one who really knows thier stuff from calling me an idiot =) )

 

[NEVERLIFT]

NeVeRLiFt hands everyone a shovel since its got so deep in here with all the BS.

 

[R1ckCa1n]

If you guys dont think money is changing hands all over the place then your sorely mistaken lol. Nither one of them give a shit about any of you. All they care about is making money and being faster then the other.

This might be the root of all evil in gaming, but is necessary I guess

nVidia HAS made a big step forward this round though by giving the ability to disable all Tri and AF optimizations.

I would say HUGE. The 6800 is doing to ATI what Nvidia said the 5900 would do to the 9700.

We aren't talking about thousands of dollars here. Were talking about MILLIONS and MILLIONS. And it all rests on the performance in these games. If you think nVidia or ATI ither one are an honest company your smoking something that sounds like alot of fun lol.

Yep... what did ATI pay for HL2 in a box? 6 million. OUCH

 

[R1ckCa1n]

And you are getting *tad* bit over excited burningrave? There is only 2.8 FPS between the 6800GT and X800XT @ High quality 1600X1200 4AA 8AF? and only 7.2 FPS diffrence with the 6800U?
SO that is quite posable for ATI to make them numbers up.And they dont constitute (in my eyes at least) the hammering some of you people are trying to spread.Its the X800Pro that isnt fairing so nice?

You are comparing Nvidia's middle offering against ATI's top offering. That is not good.

 

[R1ckCa1n]

What? This is funny... not able to run full precision SM3.0 code?

Proof, please?

While it might not be 100% true, it's not 100% false. I'm telling you, something big is about to happen in the vidoe card areana and it sure will be interesting.

Here you go: LINK

"Nvidia encouraged Crytek to use partial precision in the shaders wherever possible. This means that many of the shaders will run in 16-bit precision, not the 32-bit precision (a requirement of SM3.0) that they are touting to the press. Ironically, while promoting the 32-bit precision of SM3.0 as a "must have", Nvidia is asking developers to use less precision than was available in SM1.0 - that is, all the way back in DirectX8! Even more ironically, ATI hardware will run most of the FarCry shaders more accurately (ATI hardware runs all shaders in 24-bit precision). Microsoft, the owner of DirectX, defines 24-bit and greater as "full precision" and 16-bit as "partial precision"

 

[R1ckCa1n]

Look at the performance of the X800Pro vs the 6800GT. The only thing that was keeping the PE even in it was turning on 8xAF. I would of liked to of seen 1600x1200 w/ no AA and no AF lol.

nVidia will likely increase their OpenGL performance further also. These are still just raw drivers for a new NV40 core. The X800XT PE is living off of older more stable drivers. nVidia has an excellent driver team and performance will continue to increase with each release.

But at the cost of how much IQ? Once they show me they can release drivers that don't render at lower percision or fudge things to increase speed I will buy it.

Forget Doom3. That is one game that will release on the engine this year. DX9 and DX10 games are today and the future. The NV40 isn't as fast as the X800 card in DX9 games and that is a given even though the NV40 renders at lower persicion most of the time. What do you think will happen when a NV40 has to render at full persicion all the time?

Like I said, just watch what is going to happen once Crytek re-releases the patch.

 

[R1ckCa1n]

Heh, you guys are a bit off topic, oh well. SM3 is not a "bust" as R1ckCa1n put it.

I don't think it's a bust. I don't think it's ready for the current hardware / software. Nvidia is only blowing their horns on two or three features. Why? From what I have read it's because they don't have the horsepower to do it in full percision in realtime.

They made huge jumps in perfomance over their last card, but still do things like render part of the screen in 16 bit and others in 32 bit just to boost performance. If they do that on SM2.0, what makes us think running full 32 bit in SM3.0 is gonna work well?

 

[Tigerblade]

It was removed because Nvidia was, how can i say this, 'battered' at its PS3.0 'feature' by PS2.0b an the XT PE and didn't like it....my 2 pennys worth

Anmyone wanna argue against whats black and white go ahead....but its there for all to see......

what makes me laugh is when the Ultra is ahead in one game, ATI are crap, but when Nvidia are being whooped in everything else, they are 'just behind marginally' 10fps + isnt marginal in my book, but then this does some to be a very Nvious forum.......

 

[lopoetve]

While it might not be 100% true, it's not 100% false. I'm telling you, something big is about to happen in the vidoe card areana and it sure will be interesting.

Here you go: LINK

"Nvidia encouraged Crytek to use partial precision in the shaders wherever possible. This means that many of the shaders will run in 16-bit precision, not the 32-bit precision (a requirement of SM3.0) that they are touting to the press. Ironically, while promoting the 32-bit precision of SM3.0 as a "must have", Nvidia is asking developers to use less precision than was available in SM1.0 - that is, all the way back in DirectX8! Even more ironically, ATI hardware will run most of the FarCry shaders more accurately (ATI hardware runs all shaders in 24-bit precision). Microsoft, the owner of DirectX, defines 24-bit and greater as "full precision" and 16-bit as "partial precision"

I want proof, not an editorial from a relatively new site that so far hasn't proven even IT'S credibility.

So, got proof? I highly doubt what you're saying, and you obviously haven't read what Crytek themselves said about SM3.0 and FP16 shaders. Namely, that they used them ONLY when it would not compromise image quality at all, and then when it did, they used the full 32bit FULL precision shaders.

 

[botreaper10]

I have yet to see my 6800U whooped in anything yet. Personally, i think optimising is a great thing. Only having to do partial precision when i can't see the object is a smart thing. why waste energy rendering everything the same when the image quality is the same. The same is true for Ati and Nv's optimisations for trilinear. If i cant' tell a difference, i really don't give a crap.

 

[lopoetve]

But at the cost of how much IQ? Once they show me they can release drivers that don't render at lower percision or fudge things to increase speed I will buy it.

Forget Doom3. That is one game that will release on the engine this year. DX9 and DX10 games are today and the future. The NV40 isn't as fast as the X800 card in DX9 games and that is a given even though the NV40 renders at lower persicion most of the time. What do you think will happen when a NV40 has to render at full persicion all the time?

Like I said, just watch what is going to happen once Crytek re-releases the patch.

When Crytek re-releases the patch, it will run faster on Nv40. Nothing else

And I still don't see what you're talking about on IQ. There is no IQ drop from using the partial precision shaders, as Crytek (the bloody DEVELOPERS) said.

 

[lopoetve]

It was removed because Nvidia was, how can i say this, 'battered' at its PS3.0 'feature' by PS2.0b an the XT PE and didn't like it....my 2 pennys worth

Anmyone wanna argue against whats black and white go ahead....but its there for all to see......

what makes me laugh is when the Ultra is ahead in one game, ATI are crap, but when Nvidia are being whooped in everything else, they are 'just behind marginally' 10fps + isnt marginal in my book, but then this does some to be a very Nvious forum.......

What? What the fuck are you talking about?

Oh, and 80% of this forum used ATi in the 9700 era, so your point there is baseless and merely flamebait.

And Far Cry ran fine with the 1.2 patch. I just played it, and things were faster and smoother. Crytek doesn't care if it makes ATi win or what not, just that the game plays, which is why they pulled it (it didn't work).

 

[lopoetve]

I don't think it's a bust. I don't think it's ready for the current hardware / software. Nvidia is only blowing their horns on two or three features. Why? From what I have read it's because they don't have the horsepower to do it in full percision in realtime.

They made huge jumps in perfomance over their last card, but still do things like render part of the screen in 16 bit and others in 32 bit just to boost performance. If they do that on SM2.0, what makes us think running full 32 bit in SM3.0 is gonna work well?

Dude, precision in the shaders isn't like rendering colors. You have no proof that they don't have the horsepower to run SM3.0 because such proof doesn't exist. You have only speculation because one developer chose to OPTOMIZE by running partial precision at some times. Whether it was to help performance, or just to optomize (if you were a programmer, you'd know about optomizing, you ALWAYS do it) is not known, but you are making an assumption that I would consider baseless.

DH just has an EDITORIAL on it. They have no proof that I've seen.

 

[ZenOps]

In Nvidia's defense, using 16bitFP instead of 24-bitFP will probably be just about as noticable as ATI's optimized trilinear is to textbook trilinear - Which is that 99 percent of the time, noone will notice the difference or could tell which one is better. Heck, Nvidia was even talking 12bitINT precision (it would work perfectly fine in some situations)

Only one case I could think of where precision and textbook trilinear may be a more important factor, would be if you were to zoom-in with something like a sniper scope. In both cases, you might see a small difference.

I'd hate to see too many precision optimizations though, ATi could probably double performance if they only did a theoretical 12bitFP (comparable but inferior to a 32-bitINT)

 

[tornadotsunamilife]

In Nvidia's defense, using 16bitFP instead of 24-bitFP will probably be just about as noticable as ATI's optimized trilinear is to textbook trilinear - Which is that 99 percent of the time, noone will notice the difference or could tell which one is better.

precisely, it will probably be only by side-by-side comparisons in which we will see marginal differences.

and to whoever said people are saying that the x800 series is faster than the 6800 series by 10 fps in dx (with high af enabled that is, otherwise they are pretty neck and neck). then that is pretty good for the x800 but not great for a card that is in its optimized architecture API, while the 6800GT is beating the x800pro in doom 3 with high quality, 1600x1200 4xaa 8xaf 14.8 fps, and that wil probably increase with more matured drivers.

 

[R1ckCa1n]

What? What the fuck are you talking about?

Nice way to talk to someone on this forum. The mod ban for less these days.

 

[tranCendenZ]

Driverheaven Fantasyland:

While it might not be 100% true, it's not 100% false. I'm telling you, something big is about to happen in the vidoe card areana and it sure will be interesting.

Here you go: LINK

"Nvidia encouraged Crytek to use partial precision in the shaders wherever possible. This means that many of the shaders will run in 16-bit precision, not the 32-bit precision (a requirement of SM3.0) that they are touting to the press. Ironically, while promoting the 32-bit precision of SM3.0 as a "must have", Nvidia is asking developers to use less precision than was available in SM1.0 - that is, all the way back in DirectX8! Even more ironically, ATI hardware will run most of the FarCry shaders more accurately (ATI hardware runs all shaders in 24-bit precision). Microsoft, the owner of DirectX, defines 24-bit and greater as "full precision" and 16-bit as "partial precision"

Reality: DriverHeaven is full of BS (as usual)

Microsoft, the owner of DirectX, defines 24-bit and greater as "full precision" and 16-bit as "partial precision"

*FP24 is only "full precision" as defined by Microsoft in DX9 under Shader Model 2.0. FP24 is considered partial precision under Shader Model 3.0, which requires FP32 minimum as full precision. If any precision is "full precision" for DX9 currently, it is FP32, not FP24, as FP32 is full precision using the most advanced DX9 Shader Model. Therefore this statement by Driverheaven is inaccurate, as 24-bit is not defined as "full precision" by Microsoft under the most advanced shader model of DX9.

Even more ironically, ATI hardware will run most of the FarCry shaders more accurately

* Crytek themselves stated that FP16 shaders were used only to increase performance without IQ loss weeks before DriverHeaven's article was published. If a shader does not need more than FP16 precision, there will be zero IQ benefit from using FP24 or FP32 precision. Therefore this statement by Driverheaven is also BS.

Nvidia is asking developers to use less precision than was available in SM1.0 - that is, all the way back in DirectX8

* Future graphics cards, such as NV50 and R520, will utilize both FP16 and FP32, simply because there are many shaders that do not need FP32 precision. Running these shaders as FP32 instead of FP16 would yield no IQ gain and large performance losses. FP16/FP32 is going to be the future standard of shaders, not FP24. FP24 is now an outdated standard for SM2.0, FP16/FP32 is the future with SM3.0+. Even Unreal 3, which is coming out in 2006, is being programmed with FP16 calls. Nvidia is leading the pack in terms of precisions for the future. More propaganda by Driverheaven.

* Take everything Driverheaven says about Nvidia with a huge block of salt. They have already written two articles which accused Nvidia of NV40 driver optimizations, and they were proven wrong both times. This is another account of Driverheaven's inflammatory biased article writing when they obviously know little about the subject they are writing about. Proof of their lack of knowledge in this case?

This means that many of the shaders will run in 16-bit precision, not the 32-bit precision... ATI hardware runs all shaders in 24-bit precision

Besides what I pointed out, read the above passage - Driverheaven has no clue what they are talking about. FP32 is 128-bit, not 32-bit. Also, FP24 is 96-bit, not 24-bit. Lastly, FP16 is 64-bit, not 16-bit. They really should do some research before they make fools of themselves in front of people who know what they are talking about.

 

[R1ckCa1n]

Driverheaven Fantasyland:

Reality: DriverHeaven is full of BS (as usual)

I trust them much more than Toms!

* 24-bit is only "full precision" as defined by Microsoft in DX9 under Shader Model 2.0. 24-bit is partial precision under Shader Model 3.0, which requires FP32 minimum as full precision. If any precision is "full precision" for DX9 currently, it is FP32, not FP24, as FP32 is full precision using the most advanced DX9 Shader Model.

It's not ATI saying they can do full 32 bit........ Isn't full 32 bit precision required for SM3.0? Who says they are SM3.0 compliant?

Like I said, something big is about to happen and someone got nervous......

 

[tranCendenZ]

It's not ATI saying they can do full 32 bit........ Isn't 32 bit precision required for SM3.0? Who says they are SM3.0 compliant?

Heh, no. ATI can't do FP32/128-bit (not 32-bit), which SM3.0 requires. Driverheaven doesn't even know simple shader terminology.

FP16 = 64-bit = SM2.0 Partial Precision - SM3.0 Partial Precision
FP24 = 96-bit = SM2.0 Full Precision - SM3.0 Partial Precision
FP32 = 128-bit = SM2.0 Full Precision - SM3.0 Full Precision

Like I said, something big is about to happen and someone got nervous......

If you are talking about some Crytek - Nvidia conspiracy, that is simply silly. Driverheaven cooked up BS against Nvidia. If you don't believe me, I can link you to two other stories Driverheaven came to similar conclusions that Nvidia was using lower quality or optimizing with NV40, etc, and then were proven wrong, by Microsoft no less. Ubisoft could potentially lose tons of customers over their problems with this patch (just look at their forums). That is FAR more of a concern to them than any money Nvidia would pony up.

 

[R1ckCa1n]

Heh, no. Driverheaven doesn't even know simple shader terminology.

FP16 = 64-bit = SM2.0 Partial Precision - SM3.0 Partial Precision
FP24 = 96-bit = SM2.0 Full Precision - SM3.0 Partial Precision
FP32 = 128-bit = SM2.0 Full Precision - SM3.0 Full Precision

You need to make up your mind. Here is a quote from you today:

"SM3.0 is supposed to be a mix for FP16/FP32. That is how the spec works because doing full FP32 is extremely resource heavy and a waste of performance when FP16 offers no IQ loss in many scenarios. R520/NV50 will both be FP16/FP32 mix. Crytek themselves said that FP16 was used only when FP32 was not useful."

What you fail to say is FP32 is required for SM3.0. Check the bjorn3d forum, HERE to see that this is true. Make sure you read the last post.

So what your saying is running true SM3.0 would be too slow yet Nvidia is marketing it......

 

[tranCendenZ]

You need to make up your mind. Here is a quote from you today:

"SM3.0 is supposed to be a mix for FP16/FP32. That is how the spec works because doing full FP32 is extremely resource heavy and a waste of performance when FP16 offers no IQ loss in many scenarios. R520/NV50 will both be FP16/FP32 mix. Crytek themselves said that FP16 was used only when FP32 was not useful."

lol reread my post on the last page, I think you are misunderstanding me.

The point is Driverheaven called FP32 "32-bit." FP32 is 128-bit color precision, not 32-bit. Also FP16 is 64-bit, not 16-bit, and FP24 is 96-bit, not 24-bit. Driverheaven can't get simple shader precision facts in order. This is aside from the fact that that Driverheaven article is biased BS as you can see from what I pointed out in the last page.

In summary, FP16/FP32 is the future of shader precisions whether Driverheaven likes it or not. FP24 is done with, this is the last generation you will see it used.

What you fail to say is FP32 is required for SM3.0. Check the bjorn3d forum, HERE to see that this is true. Make sure you read the last post.

According to the DX9 spec, FP32 is the required minimum full precision for SM3.0,but FP16 may be used in partial precision calls. In other words, for your card to be SM3.0 compliant it must support FP32 precision, but developers can still choose to utilize a lower precision for better performance if the higher precision is not necessary for a shader.

So what your saying is running true SM3.0 would be too slow yet Nvidia is marketing it......

Nvidia is currently running true SM3.0. ATI will be marketing the same shader precisions next year, unless you expect ATI cards to run SM2.0 forever. Running full FP32 all the time is a waste of resources when shaders don't need more than 64-bit, and this is why the best programmers in the industry (Crytek, id, Epic, etc) are all using FP16/FP32. Unless ATI wants to lose in the performance ring while gaining zero IQ advantage, they will have to adopt FP16/FP32 shader precisions to compete with Nvidia with their future SM3.0 card. Of course, Driverheaven never would say this because they are extremely ATI biased (I consider Rage3D's articles less ATI biased than Driverheaven's, because at least Rage3D doesn't pretend to be unbiased and write BS uneducated articles)

 

[R1ckCa1n]

lol reread my post on the last page, I think you are misunderstanding me.

I understood ya, just rubbing you!

According to the DX9 spec, FP32 is the required minimum full precision for SM3.0,but FP16 may be used in partial precision calls. In other words, for your card to be SM3.0 compliant it must support FP32 precision, but developers can still choose to utilize a lower precision for better performance if the higher precision is not necessary for a shader.

This is my bone of contention with Nvidia.... Nvidia is marketing to everyone in the freaking world how they are compliant yet instruct TWIMTBP partners to mix FP16/32. What this is telling me is they rushed half assed support for SM3.0 knowing they couldn't fully support the minimum requirements without turning into a slide show.

I like ATI's approach to run full FP32 internaly and output FP24 (or whatever) but at least be honest about it. I'm more than confident that if ATI thought they could run full FP32 internally and externally on the R420, they would have supported SM3.0.

Now back to my original point of this thread, I think we are going to see SM3.0's relevance to Far Cry (and most other upcoming DX9.0 games) go out the window very soon.

 

[tranCendenZ]

This is my bone of contention with Nvidia.... Nvidia is marketing to everyone in the freaking world how they are compliant yet instruct TWIMTBP partners to mix FP16/32. What this is telling me is they rushed half assed support for SM3.0 knowing they couldn't fully support the minimum requirements without turning into a slide show.

What are you talking about?

First of all, Nvidia meets the minimum requirements for SM3.0 and is fully SM3.0 compliant - if you read the DX9 spec, it specifies partial precisions such as FP16 are permitted as long as the card is capable of FP32. Reason being because FP32 is much more resource hungry and might offer no benefit over a lower shader precision in some cases, so giving the developers the option to use FP16 or FP32 is more productive - it allows for more shaders to be used with the same performance and IQ compared to full FP32.

Second, you can see from 3dmark03 scores which runs Game Test 4 in FP32 precision that FP32 is about 15-20% slower, but hardly a slideshow in comparison, to FP24 or FP16.

I like ATI's approach to run full FP32 internaly and output FP24 (or whatever) but at least be honest about it. I'm more than confident that if ATI thought they could run full FP32 internally and externally on the R420, they would have supported SM3.0.

ATI is going to support FP16 partial precision in R520. If they don't, they will lose to Nvidia and won't look any different losing, because plenty of shaders don't need more than FP16 (in fact most in FarCry don't), and using FP32 on a shader that only needs FP16 is like displaying a 256 color image in 16.7 million color mode when 65k color mode can display the same image 20% faster. Again, even the most advanced DX9 game in development, Unreal 3, is going to be using FP16 partial precision calls for SM3.0 along with full fp32 shaders where they are needed.

Now back to my original point of this thread, I think we are going to see SM3.0's relevance to Far Cry go out the window very soon.

Or, we will see it increase in relevance. Crytek has stated that more will be added in the 1.3 patch, right now SM3.0 support is still in beta form. Nevermind SM3.0, IMO, the big draw for Nvidia cards for FarCry will really be the FP16 blended HDR Lighting, another spot where ATI is behind the curve.

 

[gordon151]

The specification for SM3 doesn't say anything about FP16. Microsoft only outlines under SM3 what they consider full precision (FP32) and what they consider partial (anything under FP32). The fact that nVidia's cards support a mix of FP32/16 has little to do with SM3 (well they do want to say they are fully compliant with the spec), but the fact that FP32 is entirely unnecessary for the gross majority of calculations and they consider it simply wasteful to be doing full precision all the time with their current hardware (they actually like their cards being fast, go figure).

Also I'd be surprised if a year or two from now nVidia is still pushing FP32/16 as probably by then they will have hardware that can fully handle doing all calculations with 128bit precision. When that happens they'll probably tout anything less as a thing of the past.

 

[tranCendenZ]

The specification for SM3 doesn't say anything about FP16. Microsoft only outlines under SM3 what they consider full precision (FP32) and what they
consider partial (anything under FP32).

Which is why I said partial precisions such as FP16.

The fact that nVidia's cards support a mix of FP32/16 has little to do with SM3 (well they do want to say they are fully compliant with the spec), but the fact that FP32 is entirely unnecessary for the gross majority of calculations and they consider it simply wasteful to be doing full precision all the time with their current hardware (they actually like their cards being fast, go figure).

Right

Also I'd be surprised if a year or two from now nVidia is still pushing FP32/16 as probably by then they will have hardware that can fully handle doing all calculations with 128bit precision. When that happens they'll probably tout anything less as a thing of the past.

I doubt that we will see that hardware for a while. If NV40 can't run something like 3dmark03 Game Test 4 in full FP32 at the same speed as FP24, IMO it won't be until at least NV60/R600 that we will see cards that can run FP32 the same speed as a FP16/FP32 mix in a game as complex as Unreal 3. Plus with the way NV40 is designed, Nvidia really isn't losing any FP32 performance by offering FP16, as the FP16 shader hardware is combined to get FP32. If NV50 is designed the same way and R520 isn't, I think we will see Nvidia take the lead in intensive DX9 apps like Unreal 3.

 

[^eMpTy^]

I understood ya, just rubbing you!

This is my bone of contention with Nvidia.... Nvidia is marketing to everyone in the freaking world how they are compliant yet instruct TWIMTBP partners to mix FP16/32. What this is telling me is they rushed half assed support for SM3.0 knowing they couldn't fully support the minimum requirements without turning into a slide show.

I like ATI's approach to run full FP32 internaly and output FP24 (or whatever) but at least be honest about it. I'm more than confident that if ATI thought they could run full FP32 internally and externally on the R420, they would have supported SM3.0.

Now back to my original point of this thread, I think we are going to see SM3.0's relevance to Far Cry (and most other upcoming DX9.0 games) go out the window very soon.

I think you're missing the point here. Starting with the NV3x Nvidia has supported the 2 modes, FP32 and FP16...they are touting thier FP32 ability because ATi is still straight FP24...as well they should be...FP16 is the performance option, FP32 is the quality option and you pick one based on what the specific shader needs to simultaneously provide higher quality than FP24 or better speed than FP24, depending on what you need.

It's about options...Nvidia has them, ATi doesn't...and in more cases than just this one.

There's nothing wrong with Nvidia simultaneously pushing for FP32 saying that it's better quality and pushing developers to use FP16 because it's faster...this is how the cards are designed to work...this is how the r500 is going to work...

At this point if this doesn't make sense to you, then I suggest you take another look at this tomorrow...cuz it all makes sense, I promise...

 

[R1ckCa1n]

I think you're missing the point here. Starting with the NV3x Nvidia has supported the 2 modes, FP32 and FP16...they are touting thier FP32 ability because ATi is still straight FP24...as well they should be...FP16 is the performance option, FP32 is the quality option and you pick one based on what the specific shader needs to simultaneously provide higher quality than FP24 or better speed than FP24, depending on what you need.

It's about options...Nvidia has them, ATi doesn't...and in more cases than just this one.

There's nothing wrong with Nvidia simultaneously pushing for FP32 saying that it's better quality and pushing developers to use FP16 because it's faster...this is how the cards are designed to work...this is how the r500 is going to work...

At this point if this doesn't make sense to you, then I suggest you take another look at this tomorrow...cuz it all makes sense, I promise...

It makes sense to people trying to justify their purchase, but I'd feel lied too.

The fact still remains it can not support running FP32 by it self. Why bother pushing something so big only to down grade the output because it can't render it at a reasonable frame rate? Maybe the marketing should go as such: "We support "partial" SM3.0".

Like I said on the first post, something big is going to happen very soon.

 

[KakimotoR]

so how many paths are there instead of "r_sm2bpath 1" ?
Wat does it do when u put "r_sm2bpath 2" ? or even "r_sm2bpath 3" ?
Wat are the difference and wat's best image quality path I can use?

 

[uwackme]

The "1" denotes..."ON", not the first in a series of paths.

BTW, I installed patch1.2 and am using Cat 4.6 on a 9800pro and it works just fine. What is all the fuss about?

Buess I shouldnt move to Cat 4.7 till the storm clears ?

 

[tranCendenZ]

It makes sense to people trying to justify their purchase, but I'd feel lied too.

Um, if you feel lied to you didn't do your research. Nvidia's FP16/FP32 shader precisions have been around since the 5800 Ultra first came out, and it was widely covered in the reviews of the 6800 that it was using the same design in terms of precision. Likewise, if you had researched the SM3.0 spec you would see that partial precision is a part of it - namely because it makes sense. It's already been stated by numerous developers, and we have seen with our own eyes, that a FP32 is not necessary on all shaders; this is proved by the excellent IQ of the FarCry 1.2 patch, which uses FP16/FP32.

Nvidia's FP16/FP32 design is superior to ATI's FP24, so if you felt lied too, its a good lie Devs have the option of higher performance at 64-bit precision or highest quality at 128-bit precision. Both are extremely fast, but of course FP16 is faster. Again though, developers have the flexibility with Nvidia's design to pick whatever they'd like to use for their game for each particular shader instead of an in-between compromise for all shaders.

The fact still remains it can not support running FP32 by it self.

Sure it can, 3dmark03 and all the dx9 homebrew demos prove that. 3dmark03 Game Test 4 uses a crapload of shaders, and they are all running FP32 on 6800 cards.

Why bother pushing something so big only to down grade the output because it can't render it at a reasonable frame rate?

You seem to be misunderstanding the concept of shader precision. Many shaders can't benefit from more than FP16. If you render those shaders with FP16, FP24, or FP32, it won't make a difference in visual output. However, it will make a difference in performance.

Maybe the marketing should go as such: "We support "partial" SM3.0".

Nope as that would be incorrect. The 6800 series meets the DX9 SM3.0 specs in full.

Like I said on the first post, something big is going to happen very soon.

Yes, its called Doom3, and we've already seen the benchmarks for that

I'm really not sure why you are reaching here, unless you have some interest in tearing down Nvidia. The facts have already been stated numerous times in this thread. Nvidia supports the best possible shader architecture for SM3.0. It supports FP32/128-bit for shaders that need highest precision, and FP16/64-bit for shaders that don't. It gives the developers the power and flexibility to use either the highest shader precision available on any consumer video card in games, or use a lower precision if such high precision is not necessary - they get to choose. There is no conspiracy, it is simply the *best* way to do SM3.0. And, as I've stated numerous times before, unless you expect ATI to simply give up the graphics race, you will see the next ATI card with FP16/FP32 precision as well. It is the future standard for video cards. If it makes any difference to your thinking its also the precisions used by major Hollywood special effects studios, including Pixar and Industrial Light & Magic. The 6800's OpenEXR HDR Lighting implentation is also the same one used by Industrial Light & Magic.

Nvidia is simply ahead of the curve. They have a more advanced and more flexible shader architecture and the proper precisions to go with it. Since ATI was sticking with SM2.0, it would make no sense to redo their whole architecture for FP16/FP32 yet. But they will have to if they plan to support SM3.0 and be competitive with Nvidia.

 

[^eMpTy^]

Um, if you feel lied to you didn't do your research. Nvidia's FP16/FP32 shader precisions have been around since the 5800 Ultra first came out, and it was widely covered in the reviews of the 6800 that it was using the same design in terms of precision. Likewise, if you had researched the SM3.0 spec you would see that partial precision is a part of it - namely because it makes sense. It's already been stated by numerous developers, and we have seen with our own eyes, that a FP32 is not necessary on all shaders; this is proved by the excellent IQ of the FarCry 1.2 patch, which uses FP16/FP32.

Sure it can, 3dmark03 and all the dx9 homebrew demos prove that. 3dmark03 Game Test 4 uses a crapload of shaders, and they are all running FP32 on 6800 cards.

You seem to be misunderstanding the concept of shader precision. Many shaders can't benefit from more than FP16. If you render those shaders with FP16, FP24, or FP32, it won't make a difference in visual output. However, it will make a difference in performance.

Nope as that would be incorrect. The 6800 series meets the DX9 SM3.0 specs in full.

Yes, its called Doom3

I'm really not sure why you are reaching here, unless you have some interest in tearing down Nvidia. The facts have already been stated numerous times in this thread. Nvidia supports the best possible shader architecture for SM3.0. It supports FP32/128-bit for shaders that need highest precision, and FP16/64-bit for shaders that don't. It gives the developers the power and flexibility to use either the highest shader precision available on any consumer video card in games, or use a lower precision if such high precision is not necessary - they get to choose. There is no conspiracy, it is simply the *best* way to do SM3.0. And, as I've stated numerous times before, unless you expect ATI to simply give up the graphics race, you will see the next ATI card with FP16/FP32 precision as well. It is the future standard for video cards. If it makes any difference to your thinking its also the precisions used by major Hollywood special effects studios, including Pixar and Industrial Light & Magic. The 6800's OpenEXR HDR implentation is also the same one used by Industrial Light & Magic.

Nvidia is simply ahead of the curve. They have a more advanced and more flexible shader architecture and the proper precisions to go with it. Since ATI was sticking with SM2.0, it would make no sense to redo their whole architecture for FP16/FP32 yet. But they will have to if they plan to support SM3.0 and be competitive with Nvidia.

I got half way through replying to his post when i decided to read the rest of the thread first...I'm glad I did, you took the words right out of my mouth.

This is pretty obvious to everyone else in this thread...I'm not sure why some people are trying to create yet another conspiracy where none exist.

Either way, I don't feel lied to one bit because I fully understood the way FP16 and FP32 work and I think it's a much better way to do business, that's one of the many reasons I just ordered a 6800GT.

But it's your money man, if the 6800 series is too complex for you to comprehend, then I recommend you look at an X800Pro. It's much simpler and the architecture has been around for over 2 years so you've had plenty of time to figure out how it works.

 

[tranCendenZ]

This is pretty obvious to everyone else in this thread...I'm not sure why some people are trying to create yet another conspiracy where none exist.

Watch out for the black helicopters (or are they green?)

 

[^eMpTy^]

Watch out for the black helicopters (or are they green?)

The ones outside my house are red...I think ATi finally found me...

 

[Dillusion]

blach blahc blah blach FP32 blah blah SM3.0 blah blah WAH WAH WAAHHHH ATI IS LOSING blah blah blah blah. cry.

I have a great fucking idea, if doom 3 plays smooth on your computer, that great. If its choppy, go cry for 2 days and buy the opposing video card thats out. Done. You can talk about FP32 and Sm3.0 and Sm2.0b and ati's large bank account for the next 2 years, you know what? no one cares, except those people who have nothing to do at 2 AM, or whatever th ehell time it is at your house.

 

[^eMpTy^]

I have a great fucking idea, if doom 3 plays smooth on your computer, that great. If its choppy, go cry for 2 days and buy the opposing video card thats out. Done. You can talk about FP32 and Sm3.0 and Sm2.0b and ati's large bank account for the next 2 years, you know what? no one cares, except those people who have nothing to do at 2 AM, or whatever th ehell time it is at your house.

As obnoxious as that is...you're totally right...it's like 3am and I'm bored stupid down at my parents house visiting...and i totally care about all this stuff for some unknown reason...

 

[Rizen]

I have a great fucking idea, if doom 3 plays smooth on your computer, that great. If its choppy, go cry for 2 days and buy the opposing video card thats out. Done. You can talk about FP32 and Sm3.0 and Sm2.0b and ati's large bank account for the next 2 years, you know what? no one cares, except those people who have nothing to do at 2 AM, or whatever th ehell time it is at your house.

Or, instead of you posting a paragraph-sized response in a thread that you obviously have little-to-no interest in, next time you could just click the X in the upper right hand corner.

 

[Spare-Flair]

I don't know if it's been posted in this thread already, but has anyone seen this?

http://www.warp2search.net/modules.php?name=News&file=article&sid=19066

In Far Cry 1.2:
Seems that just by:

a. changing the VendorID/DeviceID from X800 to GF 6800
b. Wrapping the 3.0 shaders to 2.0b shaders.

You gain as much as a 20% performance boost in FarCry 1.2!!!

Now what the heck is going on here? Obviously something underhanded in regards to Nvidia optimizations that has nothing to do with your actual hardware!?

 

[Xenozx]

I hope im not going off track here but i want to ask a question. All the benchies we have seen are with nvidias bi/tri optimized filtering off where on the oposite end ATI's optimizations are on corroect? had the Nvidia hardware been benched with its optimizations on it would gain a performance boost correct? (thats the whole point of an optimization). I agree with someone who posted before, as long as i dont notice the difference i really dont care.

if everything in sm3.0 can be done or converted to sm 2.0b what was the point of sm 3.0? Nvidia would have had to know this and if that was the case i dont see why they would push the 3.0 so much in all their marketing.

i think there is still a lot more headroom for nivida hardware as far as performance is concerned because it is a new artitecture and needs to be polished more and more where as ATI's drivers appear to be already touching their full potential. (there wasnt much of a performance increase from 4.6-4.7 was there?)

anyway to close i hope that we get some of these things cleared up before i get my new videocard in hand.

ati's performance seems to be a tad better in D3D with the new x800xt. The main reason i have a BFG 6800 ultra OC on order is because with SM 3.0 it appears to make up that difference. If ATI hardware can pull the same performance increase with 2.0b then ill be really upset about this 500 dollar purchase because from all the marketing i was thinking that ATI couldnt take advantage of this in any way. If it can, then i probably would sway my purchasing decision back on the other side of the fence. I just hope some official word comes out on this in the very near future(something other than the driver heaven article). that way i can make the right decision and be safe to not need to upgrade for the next 8-12 months.

either way both cards are way above anything in the previous generation. This i am happy for and as it has been mentioned before this will benifit all gamers.

anyway as it has been stated earlier, its really late and i think im gonna call it a night

my 2 cents.