My notes from Las Vegas Launch Event

DemoCoder

Veteran
GeForceFX name comes from Geforce + 3dfx
NV30 is 125M transistors
NV30 runs at 500 Mhz core clock
NV30 uses 1Ghz DDR2

runs
Quake3 @ 2048x1536 max quality @ 173fps
Doom3 @ 1280x1024 @ 49.8fps (vs 20.9 for 4600)
UT2k3 4XFSAA 8xANISO @ 108fps vs 37.3 for 4600.
Triple performance in nature benchmark in 3dmark2001 with NV30 vs 4600
Runs Viewperf: 1830 on NV30GL vs 690 on Quadro900XL

A01 silicon was demoed
Uses new thermal cooling technology to pump air in and out through the back of the card
Uses new "Silent Running" technique so that when the card is not running at its full capability (browsing, etc) it is silent.

Josh @ Penstar asked about Deferred Rendering and T-Buffer. Gary said any effect done by T-Buffer can be done by pixel shaders and multisample buffer in NV30.

Someone else answered that the NV30 is NOT a deferred renderer or tiler, but instead uses 3rd generation hierarchical Z, etc

Tim Sweeney demoed a next-gen engine that looked Doom3-ish but with soft shadows!

EverQuest2 was shown on NV30, it rocked!

NVidia hired a local offline 3D firm and took one of their animations and rendered it in real time on the NV30. It was a dancing ogre.

NVidia demoed a Toy Story demo in real time that used real-time depth-of-field effects to simulate a video camera.

NVidia demoed an old non-shiny pickup-truck with rust and wood shaders.

EA demeod Command and Conquer: Generals on the NV30. Definately I am going to buy this game. :)

Jen-sen is a really good, down-to-earth presenter. You feel like he is "one of the guys", engineering Bill Gates type, rather than a slick, CEO marketing type.

Jen-Sen called David Kirk, Gary T, and others on the stage and publically presented them to the audience to be thanked for advancing graphics forward.


Outside, about 10 OEMs of NVidia had desktops with NV30's in them running demos from Unreal2 to Splinter Cell, to No One Lives Forever 2.


Thats all for now. I have tons of pictures and recorded live video from the event that I will be uploading Wednesday after I get back to my desktop .

-DC out
 
runs
Quake3 @ 2048x1536 max quality @ 173fps
Doom3 @ 1280x1024 @ 49.8fps (vs 20.9 for 4600)
UT2k3 4XFSAA 8xANISO @ 108fps vs 37.3 for 4600.
Triple performance in nature benchmark in 3dmark2001 with NV30 vs 4600
Runs Viewperf: 1830 on NV30GL vs 690 on Quadro900XL

Did they demonstrate those benchmarks, or just supply the data? (What CPU was UT2K3, and what demo...one of the fly-bys, and at what resolution?)
 
runs
UT2k3 4XFSAA 8xANISO @ 108fps vs 37.3 for 4600.
Did they demonstrate those benchmarks, or just supply the data? (What CPU was UT2K3, and what demo...one of the fly-bys, and at what resolution?)

The 37fps score of the 4600 resembles the 4600 [Edit: Asbestos] score in 1280x1024 here:

http://www.anandtech.com/showdoc.html?i=1686&p=9

Anand probably used 4600 8x Aniso versus 9700 16x Aniso.

If only someone would have reliable benchmarks with documented settings..
 
DemoCoder said:
Uses new thermal cooling technology to pump air in and out through the back of the card

Uses new "Silent Running" technique so that when the card is not running at its full capability (browsing, etc) it is silent.

Call me an idiot but this is by far the most exciting detail I have heard so far. A high end video card with a noise level that won't drive me insane. That's groundbreaking.
 
At the UK launch no bechmark numbers were given out, although they did have a few running. However, were they running at full speed...?
 
Nexus said:
runs
UT2k3 4XFSAA 8xANISO @ 108fps vs 37.3 for 4600.
Did they demonstrate those benchmarks, or just supply the data? (What CPU was UT2K3, and what demo...one of the fly-bys, and at what resolution?)

The 37fps score of the 4600 resembles the 4600 [Edit: Asbestos] score in 1280x1024 here:

http://www.anandtech.com/showdoc.html?i=1686&p=9

Anand probably used 4600 8x Aniso versus 9700 16x Aniso.

If only someone would have reliable benchmarks with documented settings..

Which doesn't put the NV30 very far ahead of the R300... :devilish:
 
L233 said:
Call me an idiot but this is by far the most exciting detail I have heard so far. A high end video card with a noise level that won't drive me insane. That's groundbreaking.

Count me in on that. My dual fan leadtek gf 4 can drive me crazy because of the high noise level. And (as was mentioned in the one of the previews) who cares about the sound when you're gaming ?
 
UT2k3 4XFSAA 8xANISO @ 108fps vs 37.3 for 4600.

I would be interested in what version of UT2K3 this is (retail, end user version?) as well as the underlying hardware and settings.

I'd also be interested in any comments on the FSAA/Anisotropic filtering on this new card. As it is being compared to the Ti4600, this would suggest a somewhat non-laughable comparison as far as process is concerned.. (although I would speculate or at least hope the NV30 uses RGMS vs OGMS as on the Ti4600 for 4x mode).

But of the most importance- if there have been any compromises in FSAA or anisotropic filtering quality in the effort to improve performance. I'd hope the processes of FSAA and anisotropic filtering would be an upward progression in terms of delivery instead of a backwards step.
 
BoddoZerg said:
Obviously more than 35,000 3dmarks. I mean, the NV30 was doing that high even before it was alpha silicon!

You being serious?

3dmark2001 is getting incredibly cpu limited atm I'd be suprised if any card could get a score like that on current processors.
 
BoddoZerg said:
Nexus said:
The 37fps score of the 4600 resembles the 4600 [Edit: Asbestos] score in 1280x1024 here:

http://www.anandtech.com/showdoc.html?i=1686&p=9

Anand probably used 4600 8x Aniso versus 9700 16x Aniso.

If only someone would have reliable benchmarks with documented settings..

Which doesn't put the NV30 very far ahead of the R300... :devilish:

108 versus 72fps doesn't look so bad for the NV30, if only:

  • it wouldn't be 9700 16x Aniso versus NV30 8x.

    these numbers wouldn't be from August (=old drivers for 9700).

    you could base an opinion on those numbers, they could very well be totally uncomparable.
 
Anand probably used 4600 8x Aniso versus 9700 16x Aniso. .

It's actually worse than that. The numbers from Anandtech "Aniso + AA" are 9700 at 16X aniso, and GeForce4 at 4XAniso....

Here is where the scores were pulled from:

http://www.anandtech.com/showdoc.html?i=1683&p=18

So....based on Anand tech's benches, we have GeForce4 at 37.7 FPS at 4x aniso,, 4X AA, and 9700 getting 71.6 at 16X Aniso, 4X AA.

So, basically, we can't use Anand's benchmarks for ANYHING...
 
Numbers are really rather insignificant right now, if the thing cannot be bought for some 2-3 months...Of course, I want to know them as well, but...

If it's generally believed that review samples won't be sent out until the end of December, they have a good 5 more weeks of development on the drivers, and then another 4-8 until the thing hits retail shelves.

Anyhow, until this thing can be brought into a controlled env., it's impossible to even guess how this thing really stacks up to the 9700..
 
Unfortunately, by the time NV30 is on store shelves, the big question will likely be "Can NV30 stack up to R350"?

Its impossible to know how NV30 will perform, with the very limited information we have now. It may be a huge disappointment, or it may vastly outpace R300. However, neither of those matter for someone building a high end computer for Christmas - you can't get a NV30, but R300 is everywhere!
 
DemoCoder said:
Tim Sweeney demoed a next-gen engine that looked Doom3-ish but with soft shadows!

I'm passing out... maybe it's using shadow buffers?

NVidia hired a local offline 3D firm and took one of their animations and rendered it in real time on the NV30. It was a dancing ogre.

Spellcraft is a german 3d animation studio (www.spellcraft.de).
The original animations have used a plugin renderer for 3ds max called Brazil (www.splutterfish.com). The most notable difference (beyond stuff like colors and antialiasing) is that the renders have been using Global Illumination. I believe the lighting setup was one shadow casting directional light and a skylight (coming from all directions).

The realtime version seems to use only 1 light. Despite the 128 bit precision, it looks desaturated and flat... Nevertheless, it's very impressive, especially the realtime rendering of subdivision surfaces with adaptive tesselation.

But see for yourself:

yeah_boss_posing_hi.jpg


large02.jpg
 
Back
Top