Any Articles looking at NV40's SM3 performance and capabillities in depth?

[otispunkmeyer]

are there any articles actually analysing NV40's SM3.0 performance, ive been reading around and one thing that stuck out like a sore thumb was that people are saying the SM3 performance of the NV40 isnt really all that good. which is why there are very little games using sm3. 1) because they dont wanna bog down current SM3 hardware and 2) coz they dont wanna exclude the rest of the market (since only nvidia have sm3 support)

ive seen one place liken nvidias current SM3 performance to their SM2 performance off the FX series.....is that a fair assumption?

dont get me wrong SM3 is a good feature to have, and im glad ive got it. but you cant really ignore the performance decrease when enabling such features. ie SC:CT, FarCrys HDR, and to an extent the soft shadows in Riddick (altho that is 2++ its meant for 6800 only users and its completely unusable at any decent resolution)

when SM3 games begin to creep out, and i mean games that use the full range of SM3 capabilities, can i expect my 6800GT to be able to cut the mustard? or is the current SM3 just there for show, since its not al that fast?

 

[Koyanisquatsi]

i played thru all of riddick with 2++ enabled at 16x12. so i disagree with you as to it being completely unusable.

 

[otispunkmeyer]

Originally posted by: Koyanisquatsi
i played thru all of riddick with 2++ enabled at 16x12. so i disagree with you as to it being completely unusable.

i tried to play 2++ 12x10 and 2xaa 8xaf

11Fps is unusable to me

your BFG card is quite abit faster than mine also since i dont really overclock

 

[Genx87]

When Stalker comes out take a look at the performance of the 6800s. My initial guess is they will be avg SM3 parts at best. The first run of any product is more or less for development and testing purposes. I would expect the next generation card to be a good performer in SM3.

 

[otispunkmeyer]

thanks for the response Genx87, i expected it wouldnt be too quick, but i hope its not downright useless due lack of speed, its a nice luxury to have. ive had Ati and NV cards, usually i buy whats best for the money i have, but ive been really pleased with my 6800gt in this round of cards. and i prefer the NV drivers to the ATi ones anyday

heres a question then...

seen as Nvidia have had a SM3 part to play with out in the open for nearly 1 year, in the next round when both companies have SM3 cards out on the shelves who do you thinks gonna have the best performance?

at the moment id like to think Nvidia, i mean theyve had a years worth of real world testing, that should help. But on the other hand i know first hand that ATi can make a pixel shader fly. what i really hope is that the next round will be a toss up. you pays your money and takes your choice coz you just cant lose with either comapny. its almost like that now.

 

[Matthias99]

Originally posted by: otispunkmeyer
thanks for the response Genx87, i expected it wouldnt be too quick, but i hope its not downright useless due lack of speed, its a nice luxury to have. ive had Ati and NV cards, usually i buy whats best for the money i have, but ive been really pleased with my 6800gt in this round of cards. and i prefer the NV drivers to the ATi ones anyday

heres a question then...

seen as Nvidia have had a SM3 part to play with out in the open for nearly 1 year, in the next round when both companies have SM3 cards out on the shelves who do you thinks gonna have the best performance?

at the moment id like to think Nvidia, i mean theyve had a years worth of real world testing, that should help. But on the other hand i know first hand that ATi can make a pixel shader fly. what i really hope is that the next round will be a toss up. you pays your money and takes your choice coz you just cant lose with either comapny. its almost like that now.

There's not all that much to "do" when you implement SM3.0 over SM2.0. I mean, you need dynamic branching in the pixel shaders (not trivial, but pretty straightforward), and you need memory access from the vertex shader (again, fairly straightforward). And you need full 32-bit precision (also straightforward, but would be a fairly major overhaul for ATI). The trick is in getting it to still run at a good clock rate with the new features. But this does not appear to be a tremendously difficult task.

How fast it runs is going to be determined almost entirely by how quickly the architecture can execute low-level PS/VS operations, and by how well-optimized the drivers are. Frankly, I don't think NV4X cards (except for maybe a 6800GT/6800U SLI setup) have enough shader power to push really advanced PS3.0 shaders (the kind of things they're using in Chronicles Of Riddick to do their shadowing, for instance). And the big VS3.0 IQ feature, displacement mapping, puts a *huge* strain on not only the vertex shaders, but the pixel shaders and fixed-function pipeline as well (since something has to render all those newly-generated triangles).

 

[imported_X]

In this review, they found that even with an FX55 and a 6800GT, SM 3.0 slowed things down so much that it was unusable:

http://www.bit-tech.net/review/394

 

[otispunkmeyer]

Originally posted by: Matthias99

Originally posted by: otispunkmeyer
thanks for the response Genx87, i expected it wouldnt be too quick, but i hope its not downright useless due lack of speed, its a nice luxury to have. ive had Ati and NV cards, usually i buy whats best for the money i have, but ive been really pleased with my 6800gt in this round of cards. and i prefer the NV drivers to the ATi ones anyday

heres a question then...

seen as Nvidia have had a SM3 part to play with out in the open for nearly 1 year, in the next round when both companies have SM3 cards out on the shelves who do you thinks gonna have the best performance?

at the moment id like to think Nvidia, i mean theyve had a years worth of real world testing, that should help. But on the other hand i know first hand that ATi can make a pixel shader fly. what i really hope is that the next round will be a toss up. you pays your money and takes your choice coz you just cant lose with either comapny. its almost like that now.

There's not all that much to "do" when you implement SM3.0 over SM2.0. I mean, you need dynamic branching in the pixel shaders (not trivial, but pretty straightforward), and you need memory access from the vertex shader (again, fairly straightforward). And you need full 32-bit precision (also straightforward, but would be a fairly major overhaul for ATI). The trick is in getting it to still run at a good clock rate with the new features. But this does not appear to be a tremendously difficult task.

How fast it runs is going to be determined almost entirely by how quickly the architecture can execute low-level PS/VS operations, and by how well-optimized the drivers are. Frankly, I don't think NV4X cards (except for maybe a 6800GT/6800U SLI setup) have enough shader power to push really advanced PS3.0 shaders (the kind of things they're using in Chronicles Of Riddick to do their shadowing, for instance). And the big VS3.0 IQ feature, displacement mapping, puts a *huge* strain on not only the vertex shaders, but the pixel shaders and fixed-function pipeline as well (since something has to render all those newly-generated triangles).

cheers matthias good explanation, again i am amazed at your knowledge

 

[Matthias99]

Originally posted by: otispunkmeyer
cheers matthias good explanation, again i am amazed at your knowledge

See, I knew wasting my youth playing video games and then getting a Bachelor's in CS was good for something.

 

[otispunkmeyer]

Originally posted by: X
In this review, they found that even with an FX55 and a 6800GT, SM 3.0 slowed things down so much that it was unusable:

http://www.bit-tech.net/review/394

well everyone but Koyanisquatsi seems to agree that 2.0++ isnt a good idea on riddick, i even oc'd my 6800gt to past ultra levels and 12x10 2xaa and 8xaf was still slide show heaven. what they say there about the other SM3.0 title is that there must be a problem with the drivers since SM3 slows down the gameplay in unusual places like the less graphically intense, and that when enabling SM1.1 on the 6800 they didnt really see any performance increase...also bear in mind they are testin the demo, ive jus picked up the full game, and i have to say i think they've ironed a few things out, seems to be smoother...definately smooth enough for some SM3.

what i dont understand is this...how did they enable AA? with SM3? in the game when you select the SM3 shader it blocks out the AA option leaving just the AF.

yet they seem to be using both. and in the fulll game its caches the SM3 shaders before you can actually play. it loads the level, then a big bar appears saying caching SM3 shaders. i would like to know what they've changed, and why AA cant be used in the full game with SM3

edit: im talking about the new splinter cell btw and i have to say the fact that a £300 card cant play SC smoothely at high res with AA and AF is disappointing

 

[hans030390]

just to let you know, SM3.0 supports displacement mapping, where as 2.0 doesnt support it at all (you can make it look similar, but it will kill your card.)

also, riddick does not fully support sm3.0 so don't use that as an example.

games built from the ground up with SM3.0 intented to be used heavily will be better examples of SM3.0 than what we have now. So wait, you can't say Sm3.0 is bad because it does poorly in today games.

 

[otispunkmeyer]

Originally posted by: hans030390
just to let you know, SM3.0 supports displacement mapping, where as 2.0 doesnt support it at all (you can make it look similar, but it will kill your card.)

also, riddick does not fully support sm3.0 so don't use that as an example.

games built from the ground up with SM3.0 intented to be used heavily will be better examples of SM3.0 than what we have now. So wait, you can't say Sm3.0 is bad because it does poorly in today games.

apparently SM3 displacement mapping kills current cards anyway, and yes im fuly aware that riddick isnt a SM3 game, but its features put it beyond SM2, and i did say yes i know its SM2.0++ but its still only for the 6800 owners

you have completely got the wrong end of the stick, im not saying its bad. i know for a fact SM3 is a good thing, and im glad my card can support it.

i was just wanting to know whether or not the NV40's architecture was actually good enough to do the job properly, at the moment id say it isnt based on what we have, far cry revcieves minimal gains in speed, and nothing in IQ, and so far SC:CT has slightly questionable performance under SM3...still annoyed that i cant use AA with SM3 on it tho. but if i follow you then i presume things can only get better

 

[hans030390]

i heard that SM3 and SM1.1 run the same with SC:CT...

and the reason SM2++ runs bad is because it's more or less trying to mimic what SM3 can do, but it has to work alot harder to produce that eye candy, where as SM3 doesnt have to work as hard (not saying it will run great though)

and im sorry if i misunderstood you on SM3...i kinda read through the post fast

 

[otispunkmeyer]

Originally posted by: hans030390
i heard that SM3 and SM1.1 run the same with SC:CT...

and the reason SM2++ runs bad is because it's more or less trying to mimic what SM3 can do, but it has to work alot harder to produce that eye candy, where as SM3 doesnt have to work as hard (not saying it will run great though)

and im sorry if i misunderstood you on SM3...i kinda read through the post fast

ahh cool

i know why u cant use AA and SM3, well not SM3 as such, just the HDR part. they use the same buffers or something so its either one or the other.

and ive posted this in another thread, SC:CT doesnt seem to play nice with FRAPS, my game play became really fluid once i stopped using fraps.

 

[BFG10K]

We already had a lengthy discussion about this in the past and the key point I was getting across is that you don't have to use SM 3.0 for IQ gains, you can just use it for free performance gains.