Geforce 3 Titanium TI 500

JUST LIKE WE PROMISED A LONG TIME BACK, we will continue to bring new features into the INQUIRER. And here is an in house review, courtesy of our little lab out in the wilds of Bosnia.

You will see more reviews on the INQUIRER Web site over the next weeks and months. First, we'd like to thank all the companies who helped us in this effort and believed in us to the extent of helping us with review equipment and the like.

After a few days of testing we are ready to give you our judgement about the GeForce 3 TI 500 graphics card.

Code name Titanium
Nvidia decided to change the name of its Geforce 3 family of cards. In our little chat with Graphzilla, we learned that next Geforce 3 is not going to have Ultra or PRO and that there will be no cards with MX in the name.

That's how Titanium was born. I like to think about that that Nvidia marketing really wants to keep things quiet and lot of sites follow some rumors and skate on thin ice since these names and especially dates can be changed in a sec.

Just few weeks ago Detonator 4 become XP and even lot of Nvidia Europe haven't knew about this. At least that's what they told us.

There will be three versions of this Titanium family Geforce 3 Ti 500, Geforce 3 Ti 200 and Geforce 3 Ti. Nvidia claims that they are still on a six month cycle and we believe them. Every six month they get refreshed card such TI but new chips come after 12 months. We like to call it a 12 month cycle for new chips. Even if you are Graphzilla you have problems if you are developing such a wide range of products. Graphic chips- a lot of them, chipsets, Xbox chips etc and as far as we know they have six developer teams.

So the Titanium is refreshed Geforce 3 and can be understood as Geforce 3 Ultra with slight difference in chip itself. Titanium has few new features while Geforce 2 ultra was just overclocked version of chip.

So as we said the memory speed is the same old 3.8 ns but now memory is able to reach 250 MHz. To remind you, Geforce 3 had only 230 MHz memory [466 MHz DDR Ed]. The same situation goes with Geforce 3 -- the chip clock is 200 MHz and now it runs on 240 MHZ.

GeForce3 Ti 500 uses the same architecture that Nvidia productized with GeForce3. However, Nvidia has spent a lot of time and engineering effort to make GeForce3 Ti 500 faster, including the following - as we reported yesterday

1) Nvidia worked strategically with TSMC to move the GeForce3 die into a "high performance" (meaning faster!) version of the 0.15u process technology. TSMC doesn't do this for just anybody.....they did it for us because Nvidia drive *so much* business through their fabs.

2) Nvidia re-designed the board. New board has an 8-layer design for the PCB [Printed circuit board] and a new power supply design

3) Nvidia worked strategically with the memory vendors (Elite and UMC) to fine tune the binning parameters on the memory, to allow us to clock 3.8ns memories at 250MHz, whereas with the original GeForce3 the 3.8ns memories were clocked at 230MHz.

This is how things go if you are rich and selling a LOT of boards.

This enthusiast card TI 500 is six months older Geforce 3 targeted to once again take 3D clown from their old card unless ATI surprises us with some new revision of driver that is able to make their 12 GB/s capable Radeon to defeat Titanium. Card is still equipped with 64 MB of Ram simple because 128 MB of this DDR 3.8 NS would cost fortune. Second question is do we really need that.

Geforce 3 Ti500 will be made using enhanced .15u TSMC Process while chip itself will have 57M Transistors just like original Geforce 3. at first I thought that Graphzilla just enable some features in driver but that chip is the same like Geforce 3 chip but I was wrong.

Nvidia did some hardware modifications of the chip so we are talking about new chip. Memory is same old 3,8 ns the same as the one we find on Geforce 3 cards but this time clock is little higher - 500 MHz. Chip can do 960 Billion Operation Per Second and can do 16 AA samples/clock. Card is able to process 3.8 Billion AA pixels/sec.

Geforce 3 TI 200 will be performance and features of Geforce 3 at half the price followed with 700 Billion Operation per Second, 64 MB 400 MHz memory subsystem with 16 AA samples/clock - 2.8 Billion AA sub-pixels/sec.

Ti 200 -- the last but not the least -- is mainstream card that should offer performance of Geforce 3 Ultra at a Mainstream Price. Since this is raw marketing material we would like to keep our opinions until we see these cards. TI 200 will be clocked on 250 MHz with 64MB 400MHz memory subsystem and will be able to run 1.0 Giga Pixels per second. We like to think this card is some kind of slower Geforce 3.

Titanium technical details
Basically there are few new things in this Titanium core

The card has all capabilities of Geforce 3 featuring some new:

- 3D Textures

- Shadow Buffers

These features will become part of Graphzilla's nfiniteFX Engine that have

- Vertex Shaders

- Pixel Shaders

This nfiniteFX with these 4 riders of apocalypse are finally ready to be used in new games. We heard that games using this features are close - Aquanox will soon go gold.

Card as Geforce 3 have High-Resolution Antialising (HRAA) with Quincunx, Lightspeed Memory Architecture support for WinXP and NEW OpenGL 1.3 ICD

3D textures
This is one really necessary feature that had its implementation in ATI Radeon series of cards. Radeon was really feature rich card. Nvidia is now following that example even they decide to use Polynomial surface while ATI use N patches for higher order surfaces. To be fast, 3D texture contains information in three dimensions as opposed to two. The addition of the third dimension allows developers access to a depth component of texture information, as well as width and height. 3D textures enable developer to think in new ways when texturing objects.

Nvidia presents its solution as the first usable one, which is really brave since ATI uses this features for quite a long but we will try to compare these technologies sometimes in future.

Shadow buffers
This is a new technique in lighting for PC graphics. Nvidia's new GPUs are the first to bring hardware acceleration for realistic shadows to consumer PCs. Special hardware features in the texture unit of the GeForce3 Ti family enable these GPUs to create and use shadow buffers for real-time shadows. The technique of using shadow buffers has been field proven for such animated films as Columbia Pictures computer generated (CG) motion picture Final Fantasy: The Sprits Within. Until now, shadow buffering was only available on professional rendering equipment outside of the consumers' reach (like SGI's Octane workstation). TI shadow buffer technology closes the gap between PC gaming and cinematic style special effects for realistic, real-time shadows. This Technical Brief outlines the techniques used to create and utilize shadow buffers. It also describes the unique hardware features in the GeForce3 Ti GPU's that enable this capability and its benefits for end users. Doom 3 is going to use these benefits once when it's done.

We will keep you posted on this but at this time we where really short notice with this test but we where able to run quite a few benchmarks.

Geforce 3 TI 500
The card uses Detonator XP driver's version 21. 85 and it have same good old Geforce 3 reference design - green heatsinks with green coolers. Card have VGA, DVI and S video out in this reference board. Nvidia re-designed the board. New board has an 8-layer design for the PCB [Printed circuit board] and a new power supply design. TI 500 GPU is clocked on 240 while memory runs on 500 MHz.

How we did the tests
We set up our little test boat that has:
AMD ATHLON C 1400 MHZ
Gigabyte GA-7DX with AMD 760
2 x 128 DDR PC2100 CL 2.5
Hard drive WDC WD400AB 49 GB UDMA 100
Toshiba SD M-1502 16X 48X DVD
ATI TV-wonder TV tuner card
Windows 98 SE

Even though I have a load of cards to test I decide to compare 500 Titanium with MSI Geforce MSI 8822 aka GF3 Pro VT64D. This was done simply because luck of time to do some more comparison and fact that I haven't had opportunity to test Radeon 8500 that is only product with potential treat to TI 500 cards.

What and why we use for testing?
I decided to run both cards on Windows 98 SE since I don't find Millennium so friendly from some reasons.
Followed tests were running in past few days here:
Quake 3 Timedemo 1 the ultimate test for cards - no comments required since if you haven't seen this game you don't know what do you missing. We where using version 1.11 since upgrade to 1.29 remove my favorite Timedemo 1 test.
3D mark 2001 - the only test that left me without breath - especially Nature demo. Test is pure classic since persons from Remedy really know their job. Just for the record, the same company made Max Payne.
Evolva - good DirectX7 game with ability to test Bump mapping.

Unreal tournament - good old game that and benchmark. Running UT demo
DMZG - Dagoth More Zoological Gardens old Nvidia test to show you capabilities of accelerators with T&L
Special test settings were used while testing FSAA methods
We run FSAA in Quake 3, Unreal tournament and 3Dmark 2001 with 2 samples, 4 samples and Quincunx.
Benchmarks
3Dmark 2001 results

TI 500 was able to go over the limit of 7000 on this machine. We used default settings and we see that TI 500 on 1280x1024 is faster than Geforce 3 on1024x760. As we expected strength of this card is in higher resolutions. Due technical reason we was unable to run any of benchmarks on 1600x1200x32 even we wanted to do so. Test monitor simply do not support it! At 1280x1024x32 TI is 18.5 per cent faster that Geforce 3.

Game 1 Car chase low details
Again TI faster than Geforce 3 and we see that Geforce 3 starts do drop marginally on 86.1 FPS while TI is over 20 FPS slower. TI is 19 % faster than Geforce 3 on 1280x1024

Game 1 Car chase high details

On high details performance difference is really small. TI 500 is just 2.5 per cent on 1280x1024. On this heavy load games performance is not so big since this is really hard to get some marginally better performance. This game gets the best from your card. It gives it heavy load.

Game 2 Dragothic low details

On second game and higher resolution Titanium is 23.5 per cent faster then GF3. As we could expect TI 500 is really faster just on higher resolution since it can take benefits of its faster memory and clock speed that Geforce 3. On 1024x768x32 TI 500 is 11.1 faster than Geforce 3 and we could expect even better performance from higher resolution such as 1600x1200 but as we said earlier we could not try it.

Game 2 Dragothic high details

Things changed with this game since 10 FPS more is really a lot of frames if you have "just" 50 FPS. If you calculate that you need 60 FPS this demo would not be playable but I can say it is really nice and smooth playback. TI is 18 per cent faster than GF3 is respectable speed especially on this “high resolution”. Same thing as on low details on 1024 TI is 11 per cent faster.

Game 3 Lobby low details

Max Payne is a really good game - I finished it a few weeks ago. This is really great approximation of direction where 3D graphic should go in future - photo realism. TI 500 is 17 per cent faster than GF 3 on 1280. Geforce 3 sure suffers here from high resolutions.

Game 3 Lobby high details

TI is 13 per cent faster than GF3 and this is first time that GF 3 1024 was able to beat TI on 1280. Ti is sure strong card.

Game 4 nature
This demo is one of the few that can really use all features from your DirectX 8 card. At this point just Geforce 3, TI and Radeon 8500 can do this test. The trees, grass and butterflies are animated, transformed and lit using vertex shader. The movements of the fisherman are made by morphing, also using a vertex shader. The surface of the lake uses vertex shader, pixel shader and cube maps. Due to the pixel shader use, this test cannot be run without DX8 compatible graphics hardware. The scene only has a high detail run and affects the 3DMark score accordingly.

Technical details:

· The leaf and grass movement is done using a vertex shader.
· The fisherman animation uses morphing done with a vertex shader.
· The butterfly flight and wing movements is done with a vertex shader.
· The lake surface uses a pixel shader for per-pixel reflection and cube mapping (it uses 4 texture render states, which corresponds to 4 texture layers)
· The small river has two separate water surfaces, both using two texture layers (color map and light map).
· The landscape, the house and the bridge has two texture layers (color map and light map).
· The leaves, grass and the fisherman have a single texture layer.

Rendered textures per frame with 32 bit textures (min/avg/max): 28.4/33.5/40.0 MB

Nature test is one beautiful and great demo. I just can say it is beautiful to be in 3D graphics now and we hope to see these kind games in future. It is just breath taking. TI 500 is 1/3 better that Geforce 3. We can say full 33 per cent faster than GF 3 which is sure a lot since we are talking about almost the same chip just tweaked to be faster. Not much to add.

Fill rate Single texturing
Fill rate is a measurement of how fast the graphics card is capable of drawing textures onto 3D objects
Single-Texturing: There are 64 surfaces with one texture each. This means that the graphics hardware fill each of these objects separately, no matter how many texture layers that card is capable of drawing in a single pass.

1280 single texturing TI where able to be having 22.4 per cent faster performance than GF 3. On 1024 TI had 14.5 faster performances than GF3. AS we said earlier TI can show its mussels on higher resolutions. TI on heavy load is able to do 869.1 M texels or to bee more detailed 869.1 million texels (texture elements, pixels in the source texture) drawn per second (MTexels/s).

Fill rate Multi - texturing
In this Multi-Texturing test draws 64 texture layers as fast as possible. This means that we take advantage of the fact that modern cards are usually capable of drawing multiple texture layers on a single object as fast as it would draw one single layer. 64 texture layers are distributed so that each surface in use has as many texture layers as that particular card can draw in a single pass. For example, if your card can draw 8 texture layers in a single pass, then there will be 8 objects with 8 texture layers each. If your card is capable of doing 6 texture layers in a pass, there will be 10 objects with 6 layers and an 11th layer with the remaining 4 layers.

Difference in multi texturing on 1280 is a respectable 28 percent.

High polygon count (1 lights)
This test measures the polygon throughput of the graphics hardware. Therefore, we draw a scene with only few textures, which would limit the performance according to the fill rate, but with a large number of polygons. There are over one million polygons in this scene. The polygon throughput decreases, when there are more hardware lights and also when there are shiny surfaces in the scene done using materials with a specular reflection. For a high throughput, we have only diffuse materials and a single directional light in this run.

TI is again 18.1 per cent faster than GF3 in 1280 and 1024 and we can add that 18.8 Millions of triangles are really respectable result.

High polygon count (8 lights)
This test like the one above is measures the polygon throughput of the graphics hardware. With 8 lights: We have the same scene, but the dragons are of a shiny material giving a specular reflection. There are 8 lights all in all; one directional and 7 point lights. This is really difficult to render.

These results looks interesting to us since it show us that TI 500 card is 16 percent faster in theory. We get almost the same number as we saw on the presentation material. If you are aware of fact that not even one game do not use even 10 000 polygons for characters you do not have to be afraid that you will be limited here especially you are not going to have game with fancy effects and so much polygons anytime soon. That's sad but its reality.

Environment Mapped Bump Mapping
This is not such a new technique since it has its roots in DirectX 8. In this type of bump mapping test can do reflective environment mapped surfaces, which are bumped at the same time.

Nice effect but it needs lot of graphic power. As we expected this is where TI is faster - heavy load and fancy features. TI is once again 33 per cent faster than Geforce 3.

Dot Product 3 Bump Mapping (DOT3)
This method of bump mapping was introduced in DirectX 7. It is a mathematically more correct way to do bump mapping compared to the Environmental bump mapping. The basic DOT3 can only be used on diffuse surfaces. This is a basic DOT3 that is demonstrated here.

Dot Product 3 Bump Mapping (DOT3)

Different method but same result 33 % seams to me magical number. This is where TI 500 is actually faster than Geforce 3. Heavy effects on high resolution gaming can give some large difference in result in this test with TI vs. Geforce 3.

DirectX 8 Feature Tests
Vertex Shader
Vertex shader is like DirectX® 7 hardware transformation and lighting, but with a vertex shader the developer can write a custom transformation and lighting model. While this test demonstrates a vertex shader that does skinning, vertex shaders can be used for countless other types of transformation and lighting too. There are 100 skinned characters from game test 3 running around and shooting each other.

This is what can make both cards sweat. TI is again 13.5 percent faster than GF3.

Pixel Shader
Pixel shader is a technique for making custom per pixel lighting or other per pixel operations in a 3D scene. Bump mapping is a kind of per pixel operation, because it makes a surface look bumpy due to modifications of the pixel colors, without touching the scene geometry, or in other words the polygons of the scene. This test demonstrates a water surface drawn using a pixel shader. This test requires DirectX 8 compatible hardware. Geforce 3 as well as TI use pixel shader version 1.1 while ATI Radeon 8500 uses pixel shader 1.4 that have more features.

Pixel shader is really interesting effect since everybody is starting to use it. Here we have record where TI shows 36 per cent faster that GF3. It is not even close to 1, 5 times better performance but more that 1/3 of performance gain is sure something. On TI you will have ability to have more than 60 FPS with this fancy effect even on 12x10.

Point Sprites
Point sprites are a kind of particles which have more limitations than custom particle systems, but are drawn very fast if there is hardware support for them. There are about 500000 particles simultaneously on-screen in this test and they are used for rendering a 3D model. Particles are usually used in games for effects like fire, smoke and explosions. We are using such a large number of particles that we have to use particles the size of a single pixel. If we didn't, the test would become heavily fill rate limited.

TI 500 is 17 percent faster and it is able to draw 17.7 millions of particles in one second. All smoke, fog, explosion effect will run fast on this card.

Quake 3 Arena
Quake 3 Timedemo 1 16 bit color/ 16 bit textures
Here we are talking about raw performance since under these settings the card is not bandwidth limited.

Difference starts to show just under higher resolutions you get 11.5 per cent faster performance. On small resolutions difference is really small.

Quake 3 Timedemo 1 32 bit color/ 32 bit textures

We used High quality settings and maxed out all settings and changed just resolutions. This is best of Quake 3 and since Quake 3 is not so limited game this days since it is two years old. Still TI 500 is 13.5 per cent faster that Geforce 3 on highest setting we test.

Quake 3 predefined values

Even we where aware that we can not expect any kind of difference in this test we run it anyway. Geforce 3 is actually faster by 0.1 FPS on 640x480. That is 0,048 faster and it does not mean anything. On heavy load TI 500 is king.

Dagoth More Zoological Gardens

This is a T&L test and is sowed us some performance gain on heavy load as we saw in previous tests. On 1280 TI 500 is 14.5 per cent faster than Geforce 3.

Evolva Rolling demo 32 bit

This is one long but good test. I like the fact that you are able to see highest and lowest frame rate. We present you average values with jus one add that TI was able to get as fast as 369 FPS on 6x4 and to have minimum at 108 FPS. Amazing numbers but I must tell that Geforce 3 was also close to that. TI 500 had 25 per cent faster performance than Geforce 3 in real game! On 10x7 TI is just 10 per cent faster. [Note to ED 10x7 is 1024x768]

Evolva bump mapping

With this per pixel based effect performance gain become even wider. TI 500 on 12x10 is 31.5 percent faster again in good looking Direct X 7 based game.

FSAA testing
Quake 3 FSAA
We decided to use this Quake 3 Timedemo 1 with three different ways of FSAA - Full scene anti aliasing - feature that removes jigged edges to show you how this card perform with this nice feature turned on. This is really nightmare for card since it requires as much power as it gets. Even mighty cards as Geforce 3 TI 500 and Geforce 3 can really sweat here.

FSAA 2 Samples

This situation gives us idea that this accelerator is really strong since it can do more that 150 FPS on 800x600x32. Here we see same situation like on 3D mark TI 500 is 13per cent faster than Geforce 3 but notice that this is not heavy load test. This settings use just 2 samples FSAA. Notice that both cards can run 60 FPS on 12x10 that is more than playable.

Quincunx FSAA
This is Nvidia special say on FSAA or how they like to call it High resolution Anti aliasing HRAA.

This test TI shows 28 percent faster performance over Geforce 3. 1280x1024 becomes playable with this card. Quality on 4 samples AA and performance hit on 2 samples AA. That is Quincunx.

FSAA 4 Samples

On this heavy load TI shows 11 per cent faster performance. If we take 60 FPS as limit you can have a good play on 1024x768 but if we be more rational we find that 42.2 FPS is really playable on 1280x1024.

Conclusion
This is the fastest card I had in my hands so far. We can accept this card as tweaked Geforce 3 but no matter what they done to the card it is the faster solution than Geforce 3. We believe that TI 500 will fight the Radeon 8500 to the death.

Approximately 15 to 35 percent faster from Geforce 3 that is already fast card that can be purchased is really great performance gain. Now FSAA gaming is becoming reality. Even with 4 samples FSAA games can be playable. With Quincunx you can run almost every game on planet on at least 10x7 and have good frame rate.

Theoretically Radeon 8500 can be faster than TI 500 since it has 12 GB/s bandwidth while Geforce 3 TI has only 8GB/s. With detonator XP this card can push limits little bit further than previous cards but I would like to see ATI card with final drivers before we draw conclusion who is king of the hill for this autumn.

Radeon Truform is really nice feature and we don't see Nvidia that they going in HOS [Higher order surface] direction as aggressive as ATI. TI 500 as well as Geforce 3 has support for Polynomial surface but developers do not want to remodel characters again as this technique requires. If Truform find its way to game developers Nvidia might be in trouble since this will for the first time going to make 3D graphic different in different cards. It is again in power of persuasion and who will have better arguments [read benefits and profit from it].

Even I am not 100 per cent sure as you can never be in any kind of market but I may say that TI 500 will likely be fastest solution in the next six months. I can freely say it is the fastest card that I ever seen. Great performance with more than 100 FPS in many cases and gaming on higher resolutions and FSAA makes this card looks good.

INQ Award - It's the fastest card in our hands... µ