NVIDIA GeForce 4XX Series Discussion

[Easy Rhino]

im looking forward to some nice midrange dx11 offerings by nvidia! ill buy three!

 

[Solaris17]

Some more news snippets on the new chip:

The chip supports GDDR5 memory, has billions of transistors and it should be bigger and faster than Radeon HD 5870. The GX2 dual version of card is also in the pipes. It's well worth noting that this is the biggest architectural change since G80 times as Nvidia wanted to add a lot of instructions for better parallel computing.

Read the rest at Fudzilla.

wanna know what gets be about most of this? its probably the most true thing iv heard all year out of fud....if you read bta's news article on how nvidia is stepping up and helping microsoft wiht parallel drivers it makes sense...why would nvidia all of a sudden want to spend billions more of the $$ they like to keep to help microsoft with making parrallel drivers? i mean this close to GT300 launch?....with that die shot saying their will be more shaders? GDDR5? last i remember mem wasnt that important in folding but bandwidth counts...so that makes sense...and what about what fud said about added instruction sets? id bealive for better folding in a heartbeat...it just all seems to add up..its almost like nvidia is changing its direction a little and this time aroun d targeting the folding community just as hard the as the gamers...because they know people absolutely dedicated to folding will jump on these cards in a heartbeat..

 

[erocker]

We have yet to see how folding is going to work with these new generation cards.

 

[Easy Rhino]

wanna know what gets be about most of this? its probably the most true thing iv heard all year out of fud....if you read bta's news article on how nvidia is stepping up and helping microsoft wiht parallel drivers it makes sense...why would nvidia all of a sudden want to spend billions more of the $$ they like to keep to help microsoft with making parrallel drivers? i mean this close to GT300 launch?....with that die shot saying their will be more shaders? GDDR5? last i remember mem wasnt that important in folding but bandwidth counts...so that makes sense...and what about what fud said about added instruction sets? id bealive for better folding in a heartbeat...it just all seems to add up..its almost like nvidia is changing its direction a little and this time aroun d targeting the folding community just as hard the as the gamers...because they know people absolutely dedicated to folding will jump on these cards in a heartbeat..

im expecting a blowout by nvidia this time around.

 

[Solaris17]

We have yet to see how folding is going to work with these new generation cards.

imo it doesnt matter. im sure they can get them to fold just as well as this generation and if thats the case anything in the 9 series can already fold on par or out pace a 5870 so personally i think that the GT300 series with the added shaders will easily pull 8-10k ppd i mean their is a chance the program wont work with GT300 at fist but thats only if nvidia changed the arch from scaler to super scaler like ATI which i seriously doubt...i mean sure they may have changed the fab etc but as long as the shaders work the same way it doesnt matter if its 45 32 or 22nm i mean if fud is right and they added micro code then ya the shaders technically act diff...however at that point its like running a 32bit os on a 64bit proc...it just wont use the new optimizations...in which case it should work just find..maybe even faster given the general SP increase.

 

[Binge]

...and BiNGE asked in deep prayer for intellectual conversations regarding the GT300 series. His prayers were answered shortly after. Oh joyous day

Thanks guys

 

[erocker]

imo it doesnt matter. im sure they can get them to fold just as well as this generation and if thats the case anything in the 9 series can already fold on par or out pace a 5870 so personally i think that the GT300 series with the added shaders will easily pull 8-10k ppd i mean their is a chance the program wont work with GT300 at fist but thats only if nvidia changed the arch from scaler to super scaler like ATI which i seriously doubt...i mean sure they may have changed the fab etc but as long as the shaders work the same way it doesnt matter if its 45 32 or 22nm i mean if fud is right and they added micro code then ya the shaders technically act diff...however at that point its like running a 32bit os on a 64bit proc...it just wont use the new optimizations...in which case it should work just find..maybe even faster given the general SP increase.

Hardware isn't going to matter. The software side is what is going to matter, and since the GT300 is going to be a very different chip from GT200, I'm quite sure things are going to work differently. With DX11 compatability will Nvidia still need to use their CUDA? I would think that any new software for F@H would take advantage of compute shaders. The F@H client will have to be redone.

 

[Solaris17]

Hardware isn't going to matter. The software side is what is going to matter, and since the GT300 is going to be a very different chip from GT200, I'm quite sure things are going to work differently. With DX11 compatability will Nvidia still need to use their CUDA? I would think that any new software for F@H would take advantage of compute shaders. The F@H client will have to be redone.

things will work diffirently im sure but F@H has nothing to do with the directX api's so thats not going to matter in the slightest.as for cuda im going to bank on yes as DX has to be conformed to by any company that wishes to use it. that being the case they simply arent allowed to impliment a software technology base that favors one company it cant be done no matter how much $$ nvidia wants to pump into anything DX is talked about and standardized by multiple companies including ATI and im sure they would have a serious problem with microsoft leaning heavy on nvidia. That being said i think your right in saying that F@H in general will need to be totally reworked but imo it woulod only need to be done to accomidate and utilize nvidias updated microcode. but if they didnt i honestly dont see why F@H wouldnt run i think honestly it would do so just minus the extra acceleration....that being said nvidia imo will still need to use cuda as neither F@H or microsoft can single handedly make the decision to favor one manufacturer sure nvidia cards fold better than ATI but thats just because ATI cant do physx thats not F@H taking a side..thats F@H utilizing what already existed.

 

[erocker]

things will work diffirently im sure but F@H has nothing to do with the directX api's so thats not going to matter in the slightest.as for cuda im going to bank on yes as DX has to be conformed to by any company that wishes to use it. that being the case they simply arent allowed to impliment a software technology base that favors one company it cant be done no matter how much $$ nvidia wants to pump into anything DX is talked about and standardized by multiple companies including ATI and im sure they would have a serious problem with microsoft leaning heavy on nvidia. That being said i think your right in saying that F@H in general will need to be totally reworked but imo it woulod only need to be done to accomidate and utilize nvidias updated microcode. but if they didnt i honestly dont see why F@H wouldnt run i think honestly it would do so just minus the extra acceleration....that being said nvidia imo will still need to use cuda as neither F@H or microsoft can single handedly make the decision to favor one manufacturer sure nvidia cards fold better than ATI but thats just because ATI cant do physx thats not F@H taking a side..thats F@H utilizing what already existed.

Well, I guess what I want to know then, is will there be PhysX with Nvidia's new cards? Are we going to continue using proprietary features on video cards that divide the industry? I hope not.

 

[Solaris17]

Well, I guess what I want to know then, is will there be PhysX with Nvidia's new cards? Are we going to continue using proprietary features on video cards that divide the industry? I hope not.

well like the rest of us i can only go by speculation and history but history tends to repeat itself and given what very little we know in rumor and given the current news about nvidia wanting to step up to parallel computing it only seems to make sense ...i mean everything just comes togeher too perfectly for us to overlook it the next part i am not responding directly to you this is just general so dont think im disagreeing or starting to argue but lets all look at some facts and you tell me were i can be flawed if you wish but i might have some good points

-their is news that nvidia is in bed with microsoft with developing really good parelell computing drivers

-lets face it the F@H community houses hardcore ppl that will blow big $$ on their belifes or simply to be able to say im #1 on the list in ppd

-that being said lets not blow too much money..lets keep cuda it works and with microsoft helping us it can only get better...but this also means we will still be devided unfortunetely

-the new arch is going to have more shaders...i mean we can compare die's and speculate but lets be honest when was the last time nvidia came out with a new card that DIDNT have more SP's?

-now lets take into consideration the folders...they blow big $$ in which case folding is getting ALOT more popular lets face it atleast 80% of the ppl reading this thread contribute..alot of the folders here run farms...when i feel like fiding power cables i will too honestly...that being the case...this would be like striking oil for nvidia...a totally new community that wont be afraid to drop lots of $$ and boost nvidias sales...they can only see us folders as a relatively untapped resource...so what better than to target us?

-taking the above facts and theorys into consideration..lets talk about prioce....no doubt given history nvidias cards are usually more expensive..his means we should be expecting this...their is already a disposition...the problem is that like the tech idk if you have noticed but proces have also grown expodentially...how would nvidia make this work? easy..target a new bunch of people (folders) that will buy many of these cards and not care about proce...then ..dump a shit ton of money into microsoft to help them that way it looks like a worthwhile investement...i mean lets be honest i see this parelell computing partnership as a way to cover their ass if gamers dont wan to blow the $$


-more SP's mean more heat...and everyone knows the GTX200 series wasnt as well known as the 9 series for OC's gamers like to OC and so do enthusiests..so if it doesnt clock they will loose a little profit and general revenue...this would also lead them to seek a new adience..lets face it their is no reason for a dad and 5 kids to buy a GT300 to put in their family dell. and it probably wont happen...were the community and they know it.

-being the 2 main communities its going to have to perform..so SP numbers are almost garenteed to be above what we see now it just makes sense...lets face bigger numbers arent always better..but were overclockers we have a pre disposition for bigger numbers anyware.

GDDR5 wanna know why this makes sense? DX11 games and apps...its going to like bandwidth..thats just what happens with a new DX which is also why were going to see more SP's that and ATI has GDDR5 ..whens the last time you saw nvidia NOT do somthing ati did? and i dont mean DX10.1 thats software thats useless. i mean it may be programmed into the GPU but at the end of the day its just microcode...but GDDR5 on the otherhand..its a physical object..and last i checked nvidia doesnt like to be out played...because in their mind remember "nvidia..the way its ment to be played" and in order to keep up with their slogan they have to match them part for part

my 2c

 

[Binge]

if nV's GT300 still supports CUDA then I am sure they could patch their own drivers to support better F@H performance with the new generation of cards. Yes, it's that flexible.

 

[erocker]

-their is news that nvidia is in bed with microsoft with developing really good parelell computing drivers

Excellent. From what we have learned ATi is also in bed with Microsoft meaning MS is hopefully leading the way and keeping things compatible throughout the hardware manufacturers. The more things can stay on even ground the easier it is for a game developer to work within parameters that can benefit different hardware. I love new computer hardware and better performance as much as the next guy, but it's the advancements of the games we play and applications we use that I want more.

 

[Binge]

... From what we have learned ATi is also in bed with Microsoft...

The plot thickens

 

[Solaris17]

Excellent. From what we have learned ATi is also in bed with Microsoft meaning MS is hopefully leading the way and keeping things compatible throughout the hardware manufacturers. The more things can stay on even ground the easier it is for a game developer to work within parameters that can benefit different hardware. I love new computer hardware and better performance as much as the next guy, but it's the advancements of the games we play and applications we use that I want more.

too true indeed i mean faster processors more bandwidth its all awsome really and im excited for the new gen...but its not going to be anything really diffirent...if you think about it it would just be more of the same...everyone loves to see better stuff....but i mean it does get boring..look ooooooo 1ghz core...$100 dollars more..it happens every release..software is what needs to make a real improvement..and if they can play 100% catchup with the software and with the new hardware this gen..well sit back boys we are in for a very very exciting next couple of years. I mean if you sit back and really think about it we have made incredable and i mean absolutely incredable hardware changes...from core 2 to i7 architecture to graphics cards fast enough to throughput more than a unix farm on their own with a new software codebase just released and more and more being co-developed...i mean damn boys i remember the apple II's and stuff and like many of you i have seen alot of change..but right now ..late 2009 to be a computer enthusiest is a damn good time to be alive.

 

[Solaris17]

i dont mean to double post but i would like to extend my thanks to erocker for having one of thee most engaging conversations with me this very fine evening. I mean damn thats was just straight forward and good its amazing the kind of shit you can hammer out when you just talk to soeone about it. good show lads...intelligent people all around this is why i love TPU

 

[Benetanegia]

Comparing those two die shots, the alleged GT300 die is holding 384 shaders, not 512. Look at those shader core blocks (red rectangle you marked on the GT200). There are 10 of those on the GT200. So 240/10 = 24. They bear a strong resemblance to the ones on the alleged GT300 die shot. There are 16 of those on the die. 16 x 24 = 384.

I'm stupid, stupid, stupid!!! Thanks a lot you made that comment and made me look at the pictures again. It's fake! Absolutely fake!!

The image is rotated 90ª in respect to the original one.

Image rotating, some tone/saturation and maybe hue variation applied, and cut and paste applied everywhere...

I love so much speculating about things instead of waiting for the info, that I'm sometimes so blind. Sure that die shots have repetitive structures and all, but this faked shot is so badly executed that I feel so embarrased that I don't know if do a Yakuza and cut one of my fingers. Being a guy who uses PS almost everyday...

 

[btarunr]

I'll link your post to the IT168 guys. It will be so 'in your face' lol

 

[wolf]

My current GT300 thoughts.

It has to have more shaders or shader power, from 128 to 240 was almost double, if we are sticking with traditional-ish shader units, id hope for 480-512 or more, or maybe in the realm of ~384 but clocked off their nuts (like 2-2.5ghz)

GDDR5 is a must now, and my speculation is they will keep 512-bit bus and go for BULK bandwidth, really I don't see anything less than 200 gb/s of bandwidth coming from GT300.

ROPS are an interesting thought, 32 is already a lot, and I really don't have much to offer in terms of speculation on where they will go next. 64 would be tits tho.

SLi, I have the feeling SLi might get another tweak, in terms of being able to dual/tri/quad more of the newer cards together. This thought stems from Nvidias bulk-talk about parallelism.

sheer computing power; touting MIMD is something on nvidias cards too, and given ATi can now churn over ~2.7 TFLOPS with a current single GPU, Nvidia will really need some serious chip to best that.

using a future model of their 'current' architecture, to get ~3 TFLOPS they have to have..

512 shaders @ ~1940mhz = ~2.9 TFLOPS
384 shaders @ ~2600mhz = ~2.9 TFLOPS

or a crazy 768 shaders @ a more down to earth 1300mhz for 2.9 TFLOPS

this is all given their architecture changes, just not that drastically, thoughts?

 

[Bo_Fox]

Well, I guess what I want to know then, is will there be PhysX with Nvidia's new cards? Are we going to continue using proprietary features on video cards that divide the industry? I hope not.

Yeah, it certainly looks like as if there will be even more PhysX games released next year than this year. However, one big news is that with Windows 7, there will no longer be support for a dedicated PhysX card, be it Ageia or Geforce.

My current GT300 thoughts.

It has to have more shaders or shader power, from 128 to 240 was almost double, if we are sticking with traditional-ish shader units, id hope for 480-512 or more, or maybe in the realm of ~384 but clocked off their nuts (like 2-2.5ghz)

GDDR5 is a must now, and my speculation is they will keep 512-bit bus and go for BULK bandwidth, really I don't see anything less than 200 gb/s of bandwidth coming from GT300.

ROPS are an interesting thought, 32 is already a lot, and I really don't have much to offer in terms of speculation on where they will go next. 64 would be tits tho.

SLi, I have the feeling SLi might get another tweak, in terms of being able to dual/tri/quad more of the newer cards together. This thought stems from Nvidias bulk-talk about parallelism.

sheer computing power; touting MIMD is something on nvidias cards too, and given ATi can now churn over ~2.7 TFLOPS with a current single GPU, Nvidia will really need some serious chip to best that.

using a future model of their 'current' architecture, to get ~3 TFLOPS they have to have..

512 shaders @ ~1940mhz = ~2.9 TFLOPS
384 shaders @ ~2600mhz = ~2.9 TFLOPS

or a crazy 768 shaders @ a more down to earth 1300mhz for 2.9 TFLOPS

this is all given their architecture changes, just not that drastically, thoughts?

The "official" rumor now is that GT300 will be using 384-bit bus with GDDR5. Assuming that the GDDR5 will be clocked at 5GHz effective, it will translate into 50% greater bandwidth than a GTX 285. Note how a 5870 has only 23% more bandwidth than a 4890? And how a GTX 285 still has a little more bandwidth than a 5870?!?

With a 384-bit bus, it is highly likely that Nvidia will use 12 memory chips once again with their GT300 like they did with 8800GTX. So, we'd be seeing 1.5GB of memory, which should be plenty for at least a year anyways. It is much more cost-effective than 2GB (or 16 chips) after all.

Also, 512 shaders is extremely likely, which is over 2 times that of a GTX 285 (240 shaders). Some unreliable rumor sites claim that the GT300 will have over 3 times the performance of GT200 but I find this a bit outrageous. It must be in relation to a specific scenario, like a DX11 game, huh? Eh?

I wouldnt be surprised if it has over 3 billion trannies. Perhaps Nvidia will use such a different architecture that it actually yields 3 TFLOPS?? (Keep in mind how the GT200 was so much bigger than a RV770, yet still had lower TFLOPS performance.)

 

[troyrae360]

I wouldnt be surprised if it has over 3 billion trannies.

3 billion trannies, Now I know why I stick with ATI, you might only get one Ruby but hey at least she dosnt have balls!!

 

[Binge]

Futa.... futa ruby?...

 

[wolf]

The "official" rumor now is that GT300 will be using 384-bit bus with GDDR5. Assuming that the GDDR5 will be clocked at 5GHz effective, it will translate into 50% greater bandwidth than a GTX 285. Note how a 5870 has only 33% more bandwidth than a 4890? And how a GTX 285 still has a little more bandwidth than a 5870?!?

With a 384-bit bus, it is highly likely that Nvidia will use 12 memory chips once again with their GT300 like they did with 8800GTX. So, we'd be seeing 1.5GB of memory, which should be plenty for at least a year anyways. It is much more cost-effective than 2GB (or 16 chips) after all.

Also, 512 shaders is extremely likely, which is over 2 times that of a GTX 285 (240 shaders). Some unreliable rumor sites claim that the GT300 will have over 3 times the performance of GT200 but I find this a bit outrageous. It must be in relation to a specific scenario, like a DX11 game, huh? Eh?

I wouldnt be surprised if it has over 3 billion trannies. Perhaps Nvidia will use such a different architecture that it actually yields 3 TFLOPS?? (Keep in mind how the GT200 was so much bigger than a RV770, yet still had lower TFLOPS performance.)

I too noticed the 5870's memory bandwidth being uninspiring compared to even a GTX285, its super disappointing really. IMO it is what is holding the 5870 back, 200+ gb/s of bandwidth and I think they'd do even better, I could be wrong tho.

384-bit GDDR5 with entry level 1.5gb sounds SWEET as dude, plenty of bandwidth and a perfect amount of frame buffer IMO.

the TFLOPS question is a weird one, the 4870's creamed GT200 for GFLOPS and the 5870 is no slouch on paper either, I reckon It's funny to see Nvidia at a 'on paper' disadvantage there but it mean sweet f/a in games

IMO if this new chip pushes over 2 TFLOPS, it will be an outright BEAST.

 

[[I.R.A]_FBi]

id hit ruby


















with my digital pens

 

[Solaris17]

Yeah, it certainly looks like as if there will be even more PhysX games released next year than this year. However, one big news is that with Windows 7, there will no longer be support for a dedicated PhysX card, be it Ageia or Geforce.

what?

 

[DaedalusHelios]

3 billion trannies, Now I know why I stick with ATI, you might only get one Ruby but hey at least she dosnt have balls!!

http://img.techpowerup.org/090928/RubyBG.jpg

Ruby looks like a russian dyke in cosplay gear. I don't even remember what Nvidia uses as its mascot. I know Nvidia used some creepy model at one point though.