View Full Version : NV40 64 bit?
In the new maximum PC they say that nv 40 will be 64 bit. Seeing to the fact that it will feature either four 16 bit alu's per pipeline(combineing into two 32 bit pipes) or two 32 bit alu's(I think it will be four 16 bit alu's so that they still can put it in 16 bit mode for aperformance boost), do you they will be able to combine either two 32's or four 16's to do this? I think its a waste for gaming.
Razor04
12-12-03, 06:23 PM
Isn't that sort of similiar to what they are doing now? Someone that has more experience in this area can elaborate but I was under the impression that the current crop from NV (i.e. the NV3X) did this.
IIRC, the nv40 will utilize fx16 as well as fp16 and 32...Perhaps they are merely mentioning 64bit in relations to the fx aspect of the nv40 design (r16g16b16a16?).
Dunno, with the way nvidias marketing is i wouldn't be suprised if they just wanted it to *sound* like it is 64 bit capable.
LiquidX
12-13-03, 11:54 AM
Believe it! The January issue of Maximum PC comfirms NV40 will include full 64-bit precision. Nvidia is not playing around this time boys, there going to throw everything at us to get back to the top and in the game (as if they ever left).:)
Top end and in game? i dont see how 32-bit precision helped them do this with the last generation.. so why would 64-bit help this time? Sure it's a high number ;p but as seen with NV3x thats not always a good thing.. will NV4x be powerful enough to use 64-bit precision? theoretically it would have to be more than twice as powerfull as NV3x to achieve the title of 'being powerful enough for its own good'.
Personally i dont see it happening.. Nvidia better just hope their FP32 works well enough because i dont see any developers using FP64 anytime soon :p
The Baron
12-13-03, 12:39 PM
yeah, only not. they won't use FP64. maybe 64-bit internal precision in places (but hey, didn't they have that already?).
-=DVS=-
12-13-03, 02:13 PM
its just another PR scam it woun't be usefull in games anyways :rolleyes:
Perhaps its there because of the shared technology with the high end quadro cards.. 64bit precision may be useful for movie makers, who dont really care about performance..
Originally posted by Nutty
Perhaps its there because of the shared technology with the high end quadro cards.. 64bit precision may be useful for movie makers, who dont really care about performance..
Which was part of why fp32 was in the fx series to begin with... I would beleive it, but who cares, they will not use it for anything and who would expect them to? I mean 32>24, no need for 64 to say we have more precision than ATI
Razor04
12-13-03, 07:07 PM
*sigh* it is all marketing guys. Think of it this way...remember during the PR hype phase of the NV30 when they were talking about 128bit color...and how their competitor only had 96 bit color?
16x4 = 64
24X4 = 96
32X4 = 128
Also don't count ATI out of the game remember that the R3XX is only FP24 in 2 stages of the pipeline...the rest is FP32. Who do you think it going to be better long term? I would say ATI as they have already demonstrated that their current implementation is fast...but lets look at another issue. Is FP64 needed? I don't think so...not even close to it. Hell people have a hard time showing issues with the current formats so what good is FP64 going to do? All I see is a company being driven by a marketing dept. and not sound engineering. I hope that NV does well in the next generation but I find it hard to believe that they can pull it off with all the stuff we have seen or heard about over the past few months.
EDITl: sxotty if they didn't include FP32 they wouldn't be DX9 compliant...end of story.
Also don't count ATI out of the game remember that the R3XX is only FP24 in 2 stages of the pipeline...the rest is FP32. Who do you think it going to be better long term? I would say ATI as they have already demonstrated that their current implementation is fast...but lets look at another issue. Is FP64 needed? I don't think so...not even close to it. Hell people have a hard time showing issues with the current formats so what good is FP64 going to do? All I see is a company being driven by a marketing dept. and not sound engineering. I hope that NV does well in the next generation but I find it hard to believe that they can pull it off with all the stuff we have seen or heard about over the past few months.
IIRC, the r3xx may indeed be fp24 in only 2 of its stages, but in general because of errors in fp arithmetic, you need to have higher internal precision to guarantee accurate results. That is the only reason that internally the r3xx is fp32; if the r4xx was to have fp32, internally it would need more than 32 bits to represent the larger mantissa (maybe on the order of 40bits total as x86 cpus use 80 bits internally to represent 64bit doubles).
As the argument goes, (I dont remember where) Uttar mentioned some time ago that the nv40 would have approximately 2x the fp32 shading power as the nv3x had fp16 power. This leads me to believe there are twice as many fp units in the nv40 and that the registers are natively 32 bits rather than needing a combiner to make a 32 bit register. If nvidia wishes, maybe they can make their fp32 hardware link up (since there is twice as much) into one 64 bit shading unit. I am not really sure if there are many instances now where that precision is needed for graphics, but hell, nVidia loves to boast things that arent practical.
hey guys.. dont underestimate the technology that we need or not.
already there are game developers that have in their wish list Fp64.. and Nvidia is aware of that.. fp64 may not be needed now in games ,same with Fp32,but in 2-3 years is a diferent story.. and thats the time a directx9 game will take in development at the least . so Fp64 hardware so early in 2004 could mean only that developers will be able to target that for games that will be released in 2-3 more years. right now UNreal3 is being developed by Epic with FP32 precision -as minimun- and probably with complex long PS shaders.. so we need to thanks to Nvidia for that. yes probably the game will no be playable with decent speed in any first generation of NV3x products .. but by the time the game is released like 2004-2005 the speed will be there. i wish the FP64 rumor is true ,that only will mean that game developers will be able to play with that more earlier than later. i see FP64 used first for professional aplications and later in games. because the games that we are playing today ,are not designed around today hardware ,but around the features/speed of hardware made 2-3 years ago.that is geforce3/and radeon8500 ..something to think about.
The Baron
12-13-03, 09:01 PM
Originally posted by Nutty
Perhaps its there because of the shared technology with the high end quadro cards.. 64bit precision may be useful for movie makers, who dont really care about performance..
this might be the reason, if it's there at all.
Razor04
12-14-03, 01:10 AM
Originally posted by LS32
As the argument goes, (I dont remember where) Uttar mentioned some time ago that the nv40 would have approximately 2x the fp32 shading power as the nv3x had fp16 power. This leads me to believe there are twice as many fp units in the nv40 and that the registers are natively 32 bits rather than needing a combiner to make a 32 bit register.
Now wouldn't that bring it to roughly the level of R3XX hardware? (Speed Wise) From what I have heard the whole reason (or most of) that the NV3X is slow is because everytime it runs in FP32 it only has half the number of units available to work with as it does in FP16. Well it would be great to have a faster card...is it any good if NV goes down the same path they are taking now? ATI isn't going to sit around and use the same design again and again letting NV overtake them and NV's current plan of "optimizing" and whining to get their way (*cough* changes to the DX9 standard after it was released *cough*) isn't going to work much longer. Eventually people other than the enthusiast market will wake up to what they are doing.
Originally posted by NV40
hey guys.. dont underestimate the technology that we need or not.
already there are game developers that have in their wish list Fp64.. and Nvidia is aware of that.. fp64 may not be needed now in games ,same with Fp32,but in 2-3 years is a diferent story.. and thats the time a directx9 game will take in development at the least . so Fp64 hardware so early in 2004 could mean only that developers will be able to target that for games that will be released in 2-3 more years. right now UNreal3 is being developed by Epic with FP32 precision -as minimun- and probably with complex long PS shaders.. so we need to thanks to Nvidia for that. yes probably the game will no be playable with decent speed in any first generation of NV3x products .. but by the time the game is released like 2004-2005 the speed will be there. i wish the FP64 rumor is true ,that only will mean that game developers will be able to play with that more earlier than later. i see FP64 used first for professional aplications and later in games. because the games that we are playing today ,are not designed around today hardware ,but around the features/speed of hardware made 2-3 years ago.that is geforce3/and radeon8500 ..something to think about.
Ah...where to start with this one...
1) Understimating Technology: True it is never a good idea to underestimate technology but one must ask whether it is needed now. I personally do not see the point of adding something like FP64 when it is not needed. Maybe down the road it will be needed but is their any point in it being added now when it won't be usable? Usable means you can actually play a game with it. Use the transistors that you would need for an unusable feature to either implement something that will be usable or speed up the current stuff. I know NV loves to play the feature card but unusable features are well unusable.
2) Unreal 3: The two things you list that we should thank NV for are once again unusable. FP32 is barely usable (if you can even call it that) on the current crop of cards and long shaders are unusable on any card (long as in longer than today shaders and this does include the R3XX). I also do not see anything being released with a minimum precision of FP32 being needed. The devs might quote that but are they honestly going to turn down a significant market segment?
3) Features\Speed: You have the features part right...but I think we all know that while these games are developed around an older cards feature set they sure don't play well or at all on them. Look at Doom 3...that is supposedely targetted for the GF256...now I had one of those...and it worked well enough for a couple years but I know for a fact that it wouldn't run Doom 3 at an acceptable rate.
Just seems to me that NV are going to screw up this geneartion too before they get their act together. While I don't want to see them fail or do poorly maybe it is what is needed to get them to wake up. They aren't the fastest card on the planet anymore and they can't sit around resting on their laurels either.
razer04, actually that bring it to twice the level of r350. 16 bit in the nv35 was twice as fast as it's 32 bit so if nv40 is twice that its 16 bit then it is quadruple its 32 bit.than nv35. If it does have 64bit then it will probally be able to use it as the same speed as nv35 's 32 bit. If it is quadruple the 32 bit speed and it is a 8 pipeline architechure, then it has two 32 bit alu's per pipe. If loki is just a 12 pipeline rv350 @ 500 mhz and nv40 is 600 mhz like rumor says than nv40 should be 50% faster at 32-bit processing.
FP32 is barely usable (if you can even call it that) on the current crop of cards and long shaders are unusable on any card (long as in longer than today shaders and this does include the R3XX). I also do not see anything being released with a minimum precision of FP32 being needed. The devs might quote that but are they honestly going to turn down a significant market segment?
and I SEE IT.... you need to stop looking at today..at the present! and jump to the year 2005 ,what we may need in the year 2003 in game ,not necessarily is what we will need in year 2005. Hl2 and Doom3 are revolutionary games that began their development at the times when the latest hardware in the year 2000and earlier was the TnT2u and geforce3 respectively..Epic is using Fp32 as minimun on their their game ,because they need Fp32 precision no more /no less for their next game. at his game begins when directx9 cards are the latest thing .so what are the options they have ? use ATI cards that play everything in Dx9 -today- fast ,but that doesnt support thet precision they need. or use Nvidia latest cards that while you wont get fast speeds in -today- hardware,it does all the JOB in the precision and features they need for his -future-game. ? for game devepment its not neccessary very fast Fps ,at the beguining of a game,like always something faster is on the horizon.. can you imagine how "fast" was doom3 with realtime shadows on carmacks developer machine in the geforce1/geforce3? you can be sure ,that anyone that use the first development cards used for that game will see all the effects of the game ,but will not be very happy with the speeds ,unless he/she lower down the graphics settings.
The devs might quote that but are they honestly going to turn down a significant market segment?
YEs.. because by the time those games are released->+/- 2005 ,there will be cards from ATI and other IHV that will support at the least the precision that Epic needs for their next game..today video cards with no support for FP32 Geforce4's/R3xx will surely be able to play the game too, but not the way it was mean to be played :),developers will not lose money,you can be sure about that,but the gamers that doesnt upgrade in 2years will lose features,because of hardware limitations ANd /OR speed issues ,because the UNreal3 engine takes advantage of the latest in Dx9.. ,so you will need to turn down or disable some effects and IQ .just remember that the game it just in the early stages of its development,so by the time is finished there will be at least -two more entire generations of ATI and Nvidia hardware- ..the series Nv4x..Nv5x/R4xxand R5xx.so supporting Fp64 its not a waste of silicon in the -long term- ,if there are game developers or COmputer artist already asking for it.. (wich already there are).if the features are there ,game developers will use them in the future,if they arent ,they cant use them..in the short time its a diferent story since there are no games out that requires even Fp16.. but in the long term is another story.if Nv40 support FP64 precision or the 100_textures_perpass-that carmack requested to IHVs-,things that are not necessary today!!!!you can be sure that his next BIg game and engine in 4-5 years will show it.
Unless this was added after mid August, the NV40 doesn't support FP64. It supports FP32, FP16 and FX16. That means 64-bit and 128-bit color.
Uttar
LiquidX
12-14-03, 06:07 PM
Originally posted by Uttar
Unless this was added after mid August, the NV40 doesn't support FP64. It supports FP32, FP16 and FX16. That means 64-bit and 128-bit color.
Uttar
Well I guess it was added after mid-August, after you left there labs if we were to take Maximum PC's word for it.:D They also said R420 will support higher precision (a given) but only confirmed it for NV40.
nutball
12-15-03, 06:41 AM
Originally posted by Nv40
..the series Nv4x..Nv5x/R4xxand R5xx.so supporting Fp64 its not a waste of silicon in the -long term- ,if there are game developers or COmputer artist already asking for it.
If you average over the entire games market FP32 is a waste of silicon right now. GPUs are not designed to be future proof. NVIDIA don't want to you to have an NV30 in 2005, they don't want you to have an NV40 in 2005. They want you to have an NV30 in 2003, and an NV40 in 2004 and an NV50 in 2005 (or whatever the precise schedules are).
Anything which doesn't make 2004 games go faster in 2004 is a waste of silicon for NV40. Anything that games designers are asking for right now won't be used in games for 18+ months, in some cases 2-3 years.
My bet for the "64-bit" thing is the FP16/FX16, as others have mentioned. FX16 in render targets is actually very useful, especially if it is as fast (or nearly as fast) as the old 8-BPC formats.
Originally posted by Razor04
EDITl: sxotty if they didn't include FP32 they wouldn't be DX9 compliant...end of story.
Sorry, but I think you are incorrect in your statement.
Nvidia was already planning to have fp32 beforehand. They wanted it so they could use it as a selling point for quadro cards. They wanted to get away with fp16 in games b/c then they could have less silicon implemented then ATI for a given speed. LIke I said I would still not be surprised if they actually do have fp64 sometime in the relative near future simply for workstation market, and yes they could put a check mark on a box as well even though it will perform worse than fp32 does today most likely...
Originally posted by nutball
Anything which doesn't make 2004 games go faster in 2004 is a waste of silicon for NV40. Anything that games designers are asking for right now won't be used in games for 18+ months, in some cases 2-3 years.
[/B]
SPeed is not everything ,there are things like ImageQuality that are very popular in reviews today ,that gamers really care.. the days of only fast fps has ended.. ;)
Anything that games designers are asking for right now won't be used in games for 18+ months, in some cases 2-3 years.
so ,you if you dont give developers what they need ->now (not in the future,but right now!!!) for what they want to do in their -today- development ,and wait 2-3 years ,because "its not neccessary today", you will be delaying the creativity of those developers and waiting Twice the time ,that is 2-3 years to finally give the requested features to developers PLUS! 2-3 MORE YEARS to see it in their retail games. game you play today are as advanced in graphics and features as the "not neccesary/wasted features" made 2-3 years ago. not really sure if you are aware of this but - hardware need to comes first- and later comes the games using the new tech. is not the other way.
Razor04
12-15-03, 01:08 PM
Originally posted by sxotty
Sorry, but I think you are incorrect in your statement.
Nvidia was already planning to have fp32 beforehand. They wanted it so they could use it as a selling point for quadro cards. They wanted to get away with fp16 in games b/c then they could have less silicon implemented then ATI for a given speed. LIke I said I would still not be surprised if they actually do have fp64 sometime in the relative near future simply for workstation market, and yes they could put a check mark on a box as well even though it will perform worse than fp32 does today most likely...
They still wouldn't meet DX9 requirements without FP32 being there. It doesn't matter whether it was intended for the professional market or not. Last I checked it was FP24 as the minimum precision at the launch of DX9. It was NV's own damn fault that they left the DX9 negotiations and their minimum wasn't the one chosen. Had they not acted like a bunch of spoiled little brats I would be willing to bet the precision would have been FP16 and a lot of the current debates would have never needed to happen. Oh and like Uttar said FP16 is 64 bit and FP32 is 128 bit.
riiight... as if we are supposed to believe what maximum pc says ?
:)
they have biased/uneducated hardware reviewers... and any one of their vide card related articles will demonstrate this...
Originally posted by Razor04
They still wouldn't meet DX9 requirements without FP32 being there. It doesn't matter whether it was intended for the professional market or not. Had they not acted like a bunch of spoiled little brats .
I never said they would have met the requirements, you simply said that fp32 was just for directx9 end of story. Similarly here you are going off on a company calling it/them spoiled brats. It is really rather amusing. I was saying that you had an overly simplistic view, and you are merely confirming it.
You are correct they would not have met the specs I think we can all understand this widely known fact. One fact does not, however make all the other information and reasons surrounding something irrelevant.
What you said last about fp16->64bit and fp32->128 I realize and if you were implying that perhaps maximum pc said is that they will have 64bit i.e. fp16 that may well be the case. It is likely that they asked Nvida whether fp16 would still be there, and nvidia said something like yeah we will have 64bit precision in the next chip... could easily have confused someone.
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.