Anand has Farcry 1.2 patch and tests SM3.0

[gokickrocks]

well i wanted to know that if they were comparing the jump from 1.1 to 1.2 or if they were comparing the 2 cards on 1.2...but i took some time to read the review and found my answer

 

[Bjorn]

Just out of curiosity, what about the IQ problems for the NV3X series, anyone knows if that has been fixed ?

And if it has, what about performance ?

 

[Hanners]

Just out of curiosity, what about the IQ problems for the NV3X series, anyone knows if that has been fixed ?

It seems to have been fixed, yes.

 

[ChrisRay]

Ya the IQ problems of the NV3x series are gone on the NV40.

 

[Bjorn]

Ya the IQ problems of the NV3x series are gone on the NV40.

Yes, but are they gone on the NV3X with the new patch.

Hanners implied that it was, but what about performance then ?

 

[jb]

Any shots of that one level that showed the banding (the one with the pic of the floor tiles and the "glass cage" with the blue light)?

 

[Bjorn]

Any shots of that one level that showed the banding (the one with the pic of the floor tiles and the "glass cage" with the blue light)?

NVNews has some pictures:

http://www.nvnews.net/vbulletin/showthread.php?t=31104

That's not with a NV3X though.

 

[Bad_Boy]

that banding seems to be fixed with the 1.2 patch

 

[Neeyik]

A general question for anyone who knows: in the X800 review, Dave suggested that a new HLSL profile was to be developed in order to fully expose all of the hardware limitations (aka PS2.b vs. PS2.a for the NV3x). Does DX9.0c offer this profile?

I'm just thinking that with such support available, would it be possible for CryTek to single pass the lighting shaders for the R4xx too? Obviously the geometry instancing doohickey wouldn't be available but surely the lighting shader rework would help.

More random thinking. I can't recall seeing any particularly long shaders in the pak file for FC, so I presume that the earlier comment in this thread about the instruction limit in PS2.0 is due to the number of lights being used, and not the shader length, or am I barking up the wrong tree?

 

[John Reynolds]

I have a PCIe 6800 Ultra so if I can lay hands on this 1.2 patch over the weekend I'll do some testing. I need to finish up some other benchmarking first before I'll install SP2 RC2 onto my Intel test rig (just in case it does a horrible BSOD on me).

 

[Bjorn]

I'm just thinking that with such support available, would it be possible for CryTek to single pass the lighting shaders for the R4xx too? Obviously the geometry instancing doohickey wouldn't be available but surely the lighting shader rework would help.

From Firingsquad review:

The second addition Crytek has made involves lighting. Crytek has used PS 3.0’s increased number of interpolated registers (8 in PS 2.0a and PS2.0b versus 10 in PS 3.0) to collapse multiple lights into one pass. We were told that instead of an individual pass per light with PS 2.0, up to three or four lights can be combined into one pass using PS 3.0. This should bring performance improvements in scenes with multiple light sources.

Seems that 2.0b won't help in this case. Though perhaps people with more knowledge will disagree

 

[Smurfie]

Hmm, I don't think this is conclusive of the 6800's SM3.0 superiority yet.

Quite a couple of questions.
1. Does running SM2.0's codepath for the 6800 improve on Farcry v1.2? If yes, did they change the codes for the SM2.0 then?
2. Does running SM2.0's codepath for the X800 change on Farcry v1.2?
Again, if yes, did they change the codes for SM2.0 then?


If SM2.0's codepaths have different results from v1.2 and v1.1, then this isn't just a SM3.0 codepath addition for the 6800's benefit. It might be along the lines of them changing some codes, and we won't know the full details of it all. Did Crytek cripple/improve some SM2.0 codes? I personally doubt it, but it's still fuel for conspiracy theorists.

I'm not quite convinced yet of the Geforce 6800's SM3.0 capabilities as of now. This patch merely adds questions, and might even prove to be code changes to benefit the 6800. New SM3.0 features would have been nice. I can wish though.

 

[Bjorn]

If SM2.0's codepaths have different results from v1.2 and v1.1, then this isn't just a SM3.0 codepath addition for the 6800's benefit. It might be along the lines of them changing some codes, and we won't know the full details of it all. Did Crytek cripple/improve some SM2.0 codes? I personally doubt it, but it's still fuel for conspiracy theorists.

I've only checked this on one of the levels/charts but the X800 had the exact same framerates as without the patch.

 

[Ruined]

Overall good showing for SM3.0, looks like a 5-25% increase in framerate on parts that used SM3.0 effects, I'm impressed. From Anand's benchmarks, SM3.0 even launches the $400 6800GT past the $500 x800xt in all the timedemos with sm30 effects.

timedemos with sm30 effects on anandtech:
research - 21% speed gain w/ sm3.0 over sm2.0 (same patch/driver/demo)
regulator - 5% speed gain w/ sm3.0 over sm2.0 (same patch/driver/demo)

training - 8% speed gain w/ sm3.0 over sm2.0 (same patch/driver/demo)

volcano - 20% speed gain w/ sm3.0 over sm2.0 (same patch/driver/demo)


Not bad. Looks like Nvidia made a wise decision incorporating SM3.0. Imagine how much performance increase there would be if a game was designed with SM3.0 from the start, if one that is simply patched gives this kind of increase? Could be quite a bit.

 

[Galduta]

http://www.firingsquad.com/hardware/far_cry_sm30/

nv 40 with AF optimizations OFF ?

nv 40 with AF optimizations ON

 

[AlNom]

Now I wait for a game using SM3.0 across the board
Actually, would that even be practical? What I mean is, are there effects in PS1.X or 2.X that can be done equally as fast or mathematically or practically coded so that earlier SMs would never really phase out?

 

[Ruined]

nv 40 with AF optimizations ON

We should pay more attention to this since AF optimizations can't be disabled in the X800's control panel.

 

[Smurfie]

http://www.firingsquad.com/hardware/far_cry_sm30/

nv 40 with AF optimizations OFF

nv 40 with AF optimizations ON

You quoted 2 different reviewer's benchmarks. How is that a valid comparison?

FiringSquad used
Athlon 64 3800+
1gb RAM
DX 9.0b <- ??
WinXP SP1

Anandtech used
Athlon 64 3400+
Unknown amount of RAM
DX 9.0c
WinXP SP2 RC2

Big difference, if you ask me.

Edit: And, where did Firingsquad say that they disabled AF optimizations?

 

[Ruined]

FiringSquad used
Athlon 64 3800+
1gb RAM
DX 9.0b <- ??
WinXP SP1

Anandtech used
Athlon 64 3400+
Unknown amount of RAM
DX 9.0c
WinXP SP2 RC2

Big difference, if you ask me.

Firing Squad used dx9.0b? I thought you cant get the advantages of SM3.0 without dx9.0c?

 

[Hanners]

Firing Squad used dx9.0b? I thought you cant get the advantages of SM3.0 without dx9.0c?

I noticed that.

Which begs the question as to how/why 6800 boards are getting increased performance in the 3.0 shader path over the 2.0 path, despite (assumedly) not being able to make sure of those 3.0 shader features...