QuadroFx1000 ( NV30GL ) news and pic

[T2k]

EDIT: Just wanted to add that having 16X FSAA support got *nothing* to do with having 256MB of RAM. Some Wildcats got 256MB, some even 384MB, and can only do 4X FSAA ( and 16X Line AA, as I said earlier )

Educate yourself first...

 

[T2k]

Hmmm...

Let say 1600x1200 MSAA... that's 1,920,000 x 16 = 30,720,000 per channel

So look at this w/ 16bit per channel: 30,720,000 x 8 = 245,760,000 ~ 246MB!!!

And it's just 16bit only! Where is the geometry, texture?

It's absolutely useless this way... or am I wrong? :?

 

[pcchen]

Using lossy compression you don't need 16X memory for 16X FSAA.
For example, using Z3 you can perform reasonable 16X FSAA with only 30 bytes per pixel, that's less than 4 times of memory.

 

[Arun]

Err, what? I fail to see what I said which was incorrect.
Yes, having 1600x1200 @ 16X AA is not even possible with 256MB of RAM.
But that's not the point. I've got two points:
Point 1 is that it's not because you got more memory you can do 16X FSAA.
Here's the link: http://www.amazoninternational.com/html/benchmarks/graphicCards/wildcat/wildcat4/wildcat4.asp

The 7210 got 384MB of RAM ( 128MB frame buffer, 256 MB texture ) and can't do more than 4X FSAA.
Okay, so with 128MB frame buffer you can't do a lot more. But 6X or 8X should be possible at many resolutions if you've got all of it reserved for the frame buffer. Could be wrong on that however, didn't do the calculations.

Point 2 is that it's not because you can do 16X FSAA than you've got 256MB. Of course, you won't be able to use 16X FSAA at much more than 800x600 if you've got 128MB of RAM, but it could still be available.
Proof: The Bjorn3D preview indicates both the FX 1000 and the FX 2000 got 128MB of DDR-II RAM. Seems strange the FX2000 doesn't have 256MB, wondering if that preview is accurate. But who knows...

The 2048 Pixel Shader instructions bit is interesting. Sounds like nVidia NV3x architecture *is* very flexible; they never did such a thing for the NV25GL or another product.


Uttar

 

[mboeller]

Using lossy compression you don't need 16X memory for 16X FSAA.
For example, using Z3 you can perform reasonable 16X FSAA with only 30 bytes per pixel, that's less than 4 times of memory.

Unfortunately you need 16 X the memory. We had the same discussion about the R300 months ago. The result was, that you cannot guarantee that every frame can be compressed at all, so you have to use or reserve all of the memory for Frame- and Z-buffer as without compression. The compression only lessens the demand of bandwidth.

 

[Dave Baumann]

The 7210 got 384MB of RAM ( 128MB frame buffer, 256 MB texture ) and can't do more than 4X FSAA.

What are you talking about? Wildcat’s (non-VP’s) have SuperScene antialiasing cabled of rand sample 16X MSAA. SuperScene also has a fundamentally different method of managing the memory under MSAA.

http://www.beyond3d.com/articles/wildcatiii/

The 16X mode is likely not more than the mixed modes we’ve already seen. I’ve just come back from the Quadro FX UK launch and I’ll have a look in the documentation.

 

[Arun]

The 7210 got 384MB of RAM ( 128MB frame buffer, 256 MB texture ) and can't do more than 4X FSAA.

What are you talking about? Wildcat’s (non-VP’s) have SuperScene antialiasing cabled of rand sample 16X MSAA.

Really? Sorry for the mistake, I was basing myself on:
http://www.amazoninternational.com/html/benchmarks/graphicCards/wildcat/wildcatVP/wildcatVP.asp

Line Antialiasing (up to 16 samples)
Full Scene AA : Off, 2X, Quincunx and 4X

That page clearly states the Wildcat VP is not able to do more than 4X FSAA. So, that's not correct? Thanks for correcting me. Couldn't guess that page was wrong...


Uttar

 

[Dave Baumann]

a.) thats a completely different link to your earlier one. b.) Thats a completely different card!

You were talking about the Wildcat4 7120, with 384MB of RAM before, which is capable of 'SuperScene' AA to 16X. The second link is to a completely different board, with Wildcat VP, with only up to 128MB of RAM.

 

[T2k]

Err, what? I fail to see what I said which was incorrect.
Yes, having 1600x1200 @ 16X AA is not even possible with 256MB of RAM.

Excuse me?

That was proven above: you CAN do it. Even w/ MSAA just pointless.

But that's not the point. I've got two points:
Point 1 is that it's not because you got more memory you can do 16X FSAA.

Aaarghhh... mister, check what I wrote, OK? MSAA - MultiSample

So now, why you don't need the memory?

The 7210 got 384MB of RAM ( 128MB frame buffer, 256 MB texture ) and can't do more than 4X FSAA.
Okay, so with 128MB frame buffer you can't do a lot more. But 6X or 8X should be possible at many resolutions if you've got all of it reserved for the frame buffer. Could be wrong on that however, didn't do the calculations.

Yes, you're wrong. First of all: that's SSAA - SuperSample.
The rest is above...

Point 2 is that it's not because you can do 16X FSAA than you've got 256MB. Of course, you won't be able to use 16X FSAA at much more than 800x600 if you've got 128MB of RAM, but it could still be available.

I couldn't understand this, sorry...

Proof: The Bjorn3D preview indicates both the FX 1000 and the FX 2000 got 128MB of DDR-II RAM. Seems strange the FX2000 doesn't have 256MB, wondering if that preview is accurate. But who knows...

That means it couldn't be MSAA 'cause it's useless - this was my goal: to prove that. Thx for your confirmation.

 

[Arun]

Wavey: Damn, sorry... Didn't notice the other cards had 128MB. Supposed they had the same amount of memory. Stupid me

T2k: Err, what most people consider as "MSAA" is a form of FSAA. And both MSAA & SSAA take as much memory. So I don't see why you want to correct me between FSAA & MSAA & SSAA, since we're talking about memory...

Yes, 1600x1200 @ 16X FSAA ( or SSAA or MSAA ) is possible, but not practically due to textures and stuff. I supposed that was obvious...

If I'm understanding you correctly, you're saying the GFFX 16x FSAA is useless? Yes, it's useless at 1600x1200
But at 1024x768, if you had 256MB, 16X AA might be theorically possible.
But practically, you couldn't have much textures. So it's useless in about 75% of cases I'd say.


Uttar

 

[T2k]

OK Doc - but you wrote this earlier: EDIT: Just wanted to add that having 16X FSAA support got *nothing* to do with having 256MB of RAM.
Simply: it's not true and I've corrected. That's all.

 

[Dave Baumann]

FYI. Clocks are 300/300 for FX 1000 and 400/400 for FX 2000. They are clocked down from GeForce FX to ensure reliability in the workstation environment.

 

[LittlePenny]

[edit and cut]It appears I have trouble reading things properly[/edit]

 

[Joe DeFuria]

FYI. Clocks are 300/300 for FX 1000 and 400/400 for FX 2000. They are clocked down from GeForce FX to ensure reliability in the workstation environment.

Lol...To hell with ensuring reliability in the gaming consumer market then?

Speed binning of a single chip to this extent: from 300 Mhz to 500 Mhz must be some sort of record....

 

[T2k]

10,000 to make it sound better than 9700.

They're targeting against the X1, not the 9700.

 

[Joe DeFuria]

...now I am just guessing they are using 10,000 to make it sound better than 9700.

That's 1,000, not 10,000.

 

[T2k]

...now I am just guessing they are using 10,000 to make it sound better than 9700.

That's 1,000, not 10,000.

Exactly what I forgot to mention.

 

[Psikotiko]

FYI. Clocks are 300/300 for FX 1000 and 400/400 for FX 2000. They are clocked down from GeForce FX to ensure reliability in the workstation environment.

I don't remenber where, but i saw a video ( i think it was german ) with the GFFX giving 2 blue screens.

I hope it was a driver's problem.

 

[Xmas]

Using lossy compression you don't need 16X memory for 16X FSAA.
For example, using Z3 you can perform reasonable 16X FSAA with only 30 bytes per pixel, that's less than 4 times of memory.

Unfortunately you need 16 X the memory. We had the same discussion about the R300 months ago. The result was, that you cannot guarantee that every frame can be compressed at all, so you have to use or reserve all of the memory for Frame- and Z-buffer as without compression. The compression only lessens the demand of bandwidth.

That's with lossless compression. pcchen was talking about lossy compression with a fixed compression rate (which Z3 effectively is).

 

[Dave Baumann]

http://www.beyond3d.com/forum/viewtopic.php?t=3956