Single slot fermi for PhysX

[djlenoir]

How long do you think it will be until a single slow fermi is released? Any thoughts on whether it will be a better dedicated PhysX card then say a 9800GTX? Basically, would all the tessellation power have any impact on PhysX?

 

[Abit667]

Probably about the same time the single slot 5970 comes out.

 

[Mr. Pedantic]

Never, ever. Unless you count a waterblock as single-slot. In which case, eVGA has a water-cooled 480 out already. But air-cooled single slot definitely won't happen. It's just way too hot. Maybe in a few years we'll see graphics cards with the computational power of Fermi on a single slot due to efficiency increases, but it probably won't be a Fermi.

 

[akotlar]

The thing is, Nvidia is differentiation into other areas of research

The recently expanded their "General Purpose GPU": http://www.youtube.com/watch?v=XT5jo7aZzT (skip to 35s)

I'm not sure where their priorities lie.

 

[vjcsmoke]

The thing is, Nvidia is differentiation into other areas of research

The recently expanded their "General Purpose GPU": http://www.youtube.com/watch?v=XT5jo7aZzT (skip to 35s)

I'm not sure where their priorities lie.

They stand to make way more money selling Fermi for scientific processing applications than for the consumer/video game market. Their priority of course is expanding their bank accounts.

Never, ever. Unless you count a waterblock as single-slot. In which case, eVGA has a water-cooled 480 out already. But air-cooled single slot definitely won't happen. It's just way too hot. Maybe in a few years we'll see graphics cards with the computational power of Fermi on a single slot due to efficiency increases, but it probably won't be a Fermi.

Exactly. Without water cooling the only thing a single slot fermi would do is burn itself out in record time.

 

[ClearM4]

Lower end Fermi based cards look a ways off. Maybe early next year?

 

[akotlar]

They stand to make way more money selling Fermi for scientific processing applications than for the consumer/video game market. Their priority of course is expanding their bank accounts.



Exactly. Without water cooling the only thing a single slot fermi would do is burn itself out in record time.

I don't think he watched the video

 

[Mr. Pedantic]

Lower end Fermi based cards look a ways off. Maybe early next year?

Given what's happened (nothing) to the low-end GT200b derivatives we were going to get to replace G92, I wouldn't hold my breath.

 

[coldstorm]

maybe you get single slot fermi performance in 2 or 3 generations for now. other than that it water cooling to get single lot with top end of gpus

 

[450]

Given what's happened (nothing) to the low-end GT200b derivatives we were going to get to replace G92, I wouldn't hold my breath.

Yep. I think 9800GTX is the best we're going to have for a while.

 

[djlenoir]

I understand that a single slot 480 or even 470 is not a realistic (albeit water cooling) option for the fermi. Maybe I am out of touch with Nvidia's plans going forware (the video link did not work above), but I would think that they would want to try to release reduced functionality fermi based cards in order to fit a DirectX 11 based fermi card into each market segment (low --> high) to remain competitive. I am getting from reading this thread that the plan might be to stick with the G92b for the low end and that fermi will always remain high end only? When I was asking this question originally, I was thinking more along the lines of a future 410 or 430, possibly 450 (yes, I just made these names up) level card that would be less demanding but still perform well as a PhysX card. I am still curious, if anyone knows, if tessellation performance translates into PhysX performance?

 

[Mr. Pedantic]

I understand that a single slot 480 or even 470 is not a realistic (albeit water cooling) option for the fermi. Maybe I am out of touch with Nvidia's plans going forware (the video link did not work above), but I would think that they would want to try to release reduced functionality fermi based cards in order to fit a DirectX 11 based fermi card into each market segment (low --> high) to remain competitive. I am getting from reading this thread that the plan might be to stick with the G92b for the low end and that fermi will always remain high end only? When I was asking this question originally, I was thinking more along the lines of a future 410 or 430, possibly 450 (yes, I just made these names up) level card that would be less demanding but still perform well as a PhysX card. I am still curious, if anyone knows, if tessellation performance translates into PhysX performance?

Yes, that's the idea. But I'm a bit wary of how well that idea's going to translate into practice. NVidia said the same about GT200, and parts never materialized. I think they would have done well to have put in a shrunk-down version of Fermi in for production, it would have meant much better yields and more money in NVidia's pocket. But hey, I don't work for anyone in the industry, so I don't really know.

 

[SinShiva]

me thinks a low end fermi will never beat the cost/performance of a gtx200 series card; if dx11 doesn't take off, nvidia is screwed lol

 

[SicKlown42012]

me thinks a low end fermi will never beat the cost/performance of a gtx200 series card; if dx11 doesn't take off, nvidia is screwed lol

Nvidia can be a distant second for the next 3-4 gens and still be in a better position than AMD/ATI. The real problem for them will be when CPUs come with integrated GPU dies, and we're still a few years from that be the market norm, especially in notebooks.

 

[Kowan]

I do my part. All three of my notebooks have dedicated graphic cards.

 

[Xzyrus]

I understand that a single slot 480 or even 470 is not a realistic (albeit water cooling) option for the fermi. Maybe I am out of touch with Nvidia's plans going forware (the video link did not work above), but I would think that they would want to try to release reduced functionality fermi based cards in order to fit a DirectX 11 based fermi card into each market segment (low --> high) to remain competitive. I am getting from reading this thread that the plan might be to stick with the G92b for the low end and that fermi will always remain high end only? When I was asking this question originally, I was thinking more along the lines of a future 410 or 430, possibly 450 (yes, I just made these names up) level card that would be less demanding but still perform well as a PhysX card. I am still curious, if anyone knows, if tessellation performance translates into PhysX performance?

Well there's this for mid end (430-450):
http://www.fudzilla.com/content/view/18243/34/

Seeing as powercolor has a 5770 that's single slot, your best bet is 430 (since that's the 5770's competition according to fud). However seeing how fermi is hotter than 5770, you'll probably only get a single slot solution at 420 or lower.

I personally find fud's estimates overly optimistic since GTX 460 hasn't been mentioned yet the 430-450s according to the linked article are already crowding up the mid section. I think the following would make more sense than how fud is drawing the parallels:
460 ~ 5830
450 ~ 5770
440 ~ 5750
430 ~ 5670
420 ~ 5570
410 ~ 5450

FYI: Pulled out of my ass of course.

 

[yimingwuzere]

fudzilla wasn't the original source of the GF104 details.

Anyway it'd better meet estimates given that GF104 was rumoured to be between G92 and Cypress in die size.

 

[Xzyrus]

Regardless of the original source, what I'm saying is with the 470 >= 5850 levels, and the claim that 450 would >= 5830, it doesn't leave any room for a GTX 460 which is why I think something is afoul.

 

[Digital Viper-X-]

you guys do realize, that single slot fermi, could very well mean a .. 256 shader part? or 128shader part etc etc
All GTX 480 / 470 = Fermi
but not all Fermi = GTX 480/470

 

[Tudz]

They stand to make way more money selling Fermi for scientific processing applications than for the consumer/video game market. Their priority of course is expanding their bank accounts.

If someone can make a good, reliable CAE program that runs on CUDA, nVidia would be very well off. I cannot justify the cost of fermi for purely gaming, but if it cuts my engineering simulation times from hours to minutes, I'd be all over it like a rash.

 

[edifyingkarma]

nope. just nope

 

[InvisiBill]

If someone can make a good, reliable CAE program that runs on CUDA, nVidia would be very well off. I cannot justify the cost of fermi for purely gaming, but if it cuts my engineering simulation times from hours to minutes, I'd be all over it like a rash.

The problem here is that the maker of the program would be better off using OpenCL (or even DirectCompute) rather than CUDA (i.e. not instantly eliminating a third of their possible customers). If it was OpenCL and could run on ATI or Nvidia, then the GTX400's drawbacks (heat, power, etc.) would probably elicit more consideration than if the app simply required an Nvidia card.

http://techreport.com/discussions.x/18069

Purportedly, some tests run 2.7 times faster on the Radeon HD 5870 than on the dual-GPU GeForce GTX 295.

How many people in your situation would buy the hotter, more expensive GTX480 for the ~10% in games if the 5870 performed exactly the same in the GPGPU app? What if the 5870 were significantly faster in the GPGPU app, but a little slower in games?