The official GT300/Fermi Thread

[Helloworld_98]

Crysis 2 was built with consoles in mind so i'd assume it would be less demanding

Crysis 2 has two teams behind it, one for PC's, the other for consoles.

Crysis 2 was already demo'd on the 5870's in EyeInfinity so if that's any indication, it might not be that demanding and/or its more optimized

Yes but the 5870 six or whatever it's called seems to do quite a bit better in AMD's show off session than the standard 5870 does in reviewers and consumers PC's.

[netkas]

it does support Dx11 right??

technicaly it doesnt.

gt300 has no hardware tessellation unit.

[Luka_Aveiro]

technicaly it doesnt.

gt300 has no hardware tesselation unit.

So if it has no dedicated hardware to do tesselation, it doesn't support DX11 games.

[Helloworld_98]

So if it has no dedicated hardware to do tesselation, it doesn't support DX11 games.

fail, just because it says it's DX11 doesn't mean it has all of DX11's features, so as long as the game doesn't use tesselation then you'll be fine.

[Luka_Aveiro]

fail, just because it says it's DX11 doesn't mean it has all of DX11's features, so as long as the game doesn't use tesselation then you'll be fine.

Do you even know what tesselation is?

[zalbard]

I'm sure GT300 will support tessellation. Otherwise it will be an absolutely terrible flop.

[netkas]

I'm sure GT300 will support tessellation. Otherwise it will be an absolutely terrible flop.

it will. but via additional load on "cuda" cores.

So, nvidia cards will have bigger fps drop with tesselation than radeons, imho.

[Luka_Aveiro]

it will. but via additional load on "cuda" cores.

So, nvidia cards will have bigger fps drop with tesselation than radeons, imho.

In a percentil kind of way, I agree. Radeons may even have no fps drop if the tesselation hardware is fast enough for it's job.

[orangekiwii]

How do you people know the hardware tessalation instructinos are not physically built into the cuda cores? If that's the case, then gt300 would be faster. I can't imagine nvidia would make it software tessalation. Its porbably hard ware level just in a different way. Since the whole purpose of fermi appears to be multi-capable... I suspect the cuda cores can natively handle tessalation commands without another software layer.

[Chumbucket843]

both ATi and nvidia use software tessellation for dx11. it is superior to hardware tessellation because of performance per transistor. so technically it does support d3d11. fixed function is only useful for ROPs.

[zalbard]

both ATi and nvidia use software tessellation for dx11. it is superior to hardware tessellation because of performance per transistor. so technically it does support d3d11.

Doesn't look it.

[Helloworld_98]

both ATi and nvidia use software tessellation for dx11. it is superior to hardware tessellation because of performance per transistor. so technically it does support d3d11. fixed function is only useful for ROPs.

how is spreading performance over a chip which is already doing work superior to having an individual chip for that job?

[SKYMTL]

technicaly it doesnt.

gt300 has no hardware tessellation unit.

To put this to rest; a hardware tessellation unit isn't needed.

However, I would like to add this little tidbit from an Nvidia DX11 presentation:

[Luka_Aveiro]

how is spreading performance over a chip which is already doing work superior to having an individual chip for that job?

it is not, of course, but you can use idling units that are waiting for other units to finish their job to do some calculations in between.

Why do you think a program like furmark (<2MB) can put a higher load on your GPU than Crysis? because with crysis not all units are at work at the sime time and with furmark they actually are.

So, with a good driver optimization, you are able to redirect tesselation calculations to idling units, at least theoretically.

I don't know if that's how it's going to be done with nVidia's Fermi architecture however, but it's a way of doing it, I think.
Other way would be detect if tesselation is going to be necessary with an app, and if so, address a group of SPs to do the job.

[Farinorco]

both ATi and nvidia use software tessellation for dx11. it is superior to hardware tessellation because of performance per transistor. so technically it does support d3d11. fixed function is only useful for ROPs.

AFAIK, Radeon HD5000 series cards have a hardware tesselation unit which makes all the fixed function tesselation work. They started to develope this as soon as they knew it was to be a main feature of DirectX (formerly DX10, then delayed until DX11), when they were designing the XBox 360 Xenos chip. Of course, the final DX11 version is wider than the previous one implemented by ATI (AFAIK, the DX11 tesselation is a superset of ATI tesselation), so previous ATI cards are NOT compatible with the new DX11 tesselation.

A dedicated hw is always more efficient in doing the speciallized task it is built for than more generic, flexible and programmable hw. You are going to use more transistors to do a task if those transistors are from a generic processor than if they are dedicated (especially designed/optimized to) to do that task. The advantage of using more generic hw, is that it can be used for different things (so if there's no need to do that task, there are no wasted transistors), and that you may balance the work load the way you want.

But a dedicated, specialized hardware for a task, is always going to be more efficient at doing it.

The question is... ¿has Fermi also a dedicated hardware to solve the fixed function tesselation, or it hasn't? If not, how big of an impact that means when the chip has to resolve tesselation, when compared with a chip with dedicated hardware to it (Radeons)?

[JohnZS]

Hmm here is hoping that eVGA's announcement this time next week is actually for the GT300
It does sound like it is going to be a monster GPU (in areas of size, performance, power consumption and price), however it WILL be the fastest Single GPU card out there and that is all that matters to me.
I do not give two hoots about SLi or Crossfire GPU's as they have woeful minimum framerates, rely on praying to the driver Gods and also are usually loud and extremely power hungry.
John

[cx-ray]

I do not give two hoots about SLi or Crossfire GPU's as they have woeful minimum framerates, rely on praying to the driver Gods and also are usually loud and extremely power hungry.
John

I suggest you try SLI with two GPUs on an X58 system. Almost all games of the past two years play better in high res in SLI than a single card. Of the 20+ games I got only two come to mind that have real problems with SLI; Brothers in Arms Hell's Highway (framerate feels like 15fps) and Gears of War (problems with graphics flickering in rain levels).

[003]

technicaly it doesnt.

gt300 has no hardware tessellation unit.

lol

[JohnZS]

I suggest you try SLI with two GPUs on an X58 system. Almost all games of the past two years play better in high res in SLI than a single card. Of the 20+ games I got only two come to mind that have real problems with SLI; Brothers in Arms Hell's Highway (framerate feels like 15fps) and Gears of War (problems with graphics flickering in rain levels).

I have yet to have any experience with the i7 platform
My only opinions of SLi were gathered from a nforce 790 based board and 2 Geforce8 GTS 512 cards Yes the 3dmark scores were high, yes the games did have high and average high frame rates... however they did have lower minimum frame rates and stutter

As for Crossfire... awful, just awful mind you this was back in the day of HD3850.

John

[jfromeo]

Hmm here is hoping that eVGA's announcement this time next week is actually for the GT300

eVGA already claimed the announcement to be a GTX275 (GPU) + GTS250 (PhysX) single-pcb card.

[Andrea deluxe]

To put this to rest; a hardware tessellation unit isn't needed.

However, I would like to add this little tidbit from an Nvidia DX11 presentation:

ok perfect!

fixed function= no problem for fermi and tassellation.......

[Hornet331]

AFAIK, Radeon HD5000 series cards have a hardware tesselation unit which makes all the fixed function tesselation work. They started to develope this as soon as they knew it was to be a main feature of DirectX (formerly DX10, then delayed until DX11), when they were designing the XBox 360 Xenos chip. Of course, the final DX11 version is wider than the previous one implemented by ATI (AFAIK, the DX11 tesselation is a superset of ATI tesselation), so previous ATI cards are NOT compatible with the new DX11 tesselation.

A dedicated hw is always more efficient in doing the speciallized task it is built for than more generic, flexible and programmable hw. You are going to use more transistors to do a task if those transistors are from a generic processor than if they are dedicated (especially designed/optimized to) to do that task. The advantage of using more generic hw, is that it can be used for different things (so if there's no need to do that task, there are no wasted transistors), and that you may balance the work load the way you want.

But a dedicated, specialized hardware for a task, is always going to be more efficient at doing it.

The question is... ¿has Fermi also a dedicated hardware to solve the fixed function tesselation, or it hasn't? If not, how big of an impact that means when the chip has to resolve tesselation, when compared with a chip with dedicated hardware to it (Radeons)?

ATI had a tesslator way before dx10, even before dx9. They just needed to adjust it to the dx10 specs.

[zalbard]

ATI had a tesslator way before dx10, even before dx9. They just needed to adjust it to the dx10 specs.

I wonder if this is actually one of the reasons why Fermi is being late... because ATI engineers have a ton more experience with tessellation, and it's the biggest change from DX10.

[N19h7m4r3]

Sure didn't the old Radeon 8500 feature a Tesselator?

[zalbard]

Sure didn't the old Radeon 8500 feature a Tesselator?

Yep. They had that TruForm thing also. For smoothing surfaces etc. There were even patches for a few games.