ATI 9800 VS NV FX 5950

[micb]

I did a small comparision of a stock FX 5950 vs a stock ATI 9800 non pro.

The FX is running 52.16 and the ATI is CAT 3.8's.

This shows that the FX 5950 looks to be 15-20% slower when doing PS 1.1, PS 1.4 and PS 2.0 using partial precision (FP16).

When using proper full precision PS 2.0 (FP32) the FX is a massive 50% slower VS ATI 9800 non pro (Full precision FP24).

The fillrate on the FX 5950 looks to be lower then the ATI 9800 by three precent.

As the ATI 9800 Pro and 9800 XT are even faster it make you think doesn't it.

 

[Pete]

Hi, BV.

What's interesting is that the 5950's pure and single-texture fillrate numbers are higher than its core clock would lead me to assume. 475 * 4 = 1900, yet the 5950 scores higher than that in both cases.

The rest of the numbers make sense from what I've learned here.

 

[zsouthboy]

Remember that NV3x is not exactly 4x2, but not exactly 8x1 either. That's all I can think of as the reason.

 

[Dave Baumann]

Under texturing situations it is exactly a 4x2

 

[zsouthboy]

Under texturing situations it is exactly a 4x2

Well there goes my explanation.

 

[Pete]

So what am I missing, Dave? Shouldn't the 5950 be at or under 1900 for "pure" fillrate and any texturing? Come to think of it, the Z rate shouldn't be faster than 3800, either. I'm guessing this is an OC'ed 5950, as many have said they OC quite well.

zs, I don't think GPU architectures are that fluid. AFAIK, the NV38 is either 4x2 or 8x0, depending on the instructions its given. As these are very specific synthetic tests, I wouldn't think they'd test more than one "path" in a GPU.

 

[Tridam]

This is an overclocked FX5950.

Here are the results at 475/475 MHz

Fillrate Tester
--------------------------
Display adapter: NVIDIA GeForce FX 5950 Ultra
Driver version: 6.14.10.5216
Display mode: 1280x1024 A8R8G8B8 60Hz
Z-Buffer format: D24S8
--------------------------

FFP - Pure fillrate - 1866.379395M pixels/sec
FFP - Z pixel rate - 3537.976318M pixels/sec
FFP - Single texture - 1789.346802M pixels/sec
FFP - Dual texture - 1665.091919M pixels/sec
FFP - Triple texture - 849.811768M pixels/sec
FFP - Quad texture - 566.553589M pixels/sec
PS 1.1 - Simple - 939.009460M pixels/sec
PS 1.4 - Simple - 886.052612M pixels/sec
PS 2.0 - Simple - 598.241638M pixels/sec
PS 2.0 PP - Simple - 890.335876M pixels/sec
PS 2.0 - Longer - 256.549927M pixels/sec
PS 2.0 PP - Longer - 450.940369M pixels/sec
PS 2.0 - Longer 4 Registers - 255.341461M pixels/sec
PS 2.0 PP - Longer 4 Registers - 595.507751M pixels/sec
PS 2.0 - Per Pixel Lighting - 102.980316M pixels/sec
PS 2.0 PP - Per Pixel Lighting - 128.135620M pixels/sec

FX5950 is never 8x1. It's always 4x2. The only exception is the "Quad 8 Mode". The chip works on two quads in the same clock cycle. The back end interpolates the depth (8 per cycle or 2 per pipeline) and stores 4 results on the depth data output and the 4 other results on the color data output. After, the combiners pass color and depth output unchanged to the ROPs. This is not 8x1. This is 8x0. No color calculations (shader or texturing) can be done in this mode as the color output is used for the 4 more depth datas.

 

[PeterAce]

For consistancy both cards were set to the Display mode: 1024x768 A8R8G8B8 85Hz.

(I'm Micb's brother).

 

[zsouthboy]

zs, I don't think GPU architectures are that fluid. AFAIK, the NV38 is either 4x2 or 8x0, depending on the instructions its given. As these are very specific synthetic tests, I wouldn't think they'd test more than one "path" in a GPU.

I'm not sure I understand what you are saying... I had gotten the idea that sometimes the NV3x is either 4x2 or 8x1(instead of 0), and it just depended on the situation. In actuallity, it is a 4x2 + 4x0, right?

 

[Tridam]

For consistancy both cards were set to the Display mode: 1024x768 A8R8G8B8 85Hz.

(I'm Micb's brother).

Why do you make this comparison with an overclocked 5950 ?

 

[micb]

For consistancy both cards were set to the Display mode: 1024x768 A8R8G8B8 85Hz.

(I'm Micb's brother).

Why do you make this comparison with an overclocked 5950 ?

We did not, the FX 5950 card was at stock clock.

You have been running a higher resolution 1280x1024 @ 60.

In our test the FX 5950 was stock clock running 1024x768 @ 85 the same res as my stock ATI 9800.

 

[Tridam]

I'm not sure I understand what you are saying... I had gotten the idea that sometimes the NV3x is either 4x2 or 8x1(instead of 0), and it just depended on the situation. In actuallity, it is a 4x2 + 4x0, right?

Actually, it's not 4x2 + 4x0. It's 4x2 or 8x0.

8x0 = 8 depth data but 0 color data
4x2 = 4 color data + 4 depth data
8x1 = 8 color data + 8 depth data

 

[Dave Baumann]

Where did you get the 5950 from?

 

[Tridam]

For consistancy both cards were set to the Display mode: 1024x768 A8R8G8B8 85Hz.

(I'm Micb's brother).

Why do you make this comparison with an overclocked 5950 ?

We did not the FX 5950 card was at stock clock.

You have been running a higher resolution 1280x1024 @ 60.

In our test the FX 5950 was stock clock running 1024x768 @ 85 the same res as my stock ATI 9800.

No, it was not the default clocks

Fillrate Tester
--------------------------
Display adapter: NVIDIA GeForce FX 5950 Ultra
Driver version: 6.14.10.5216
Display mode: 1024x768 A8R8G8B8 85Hz
Z-Buffer format: D24S8
--------------------------

FFP - Pure fillrate - 1884.801025M pixels/sec
FFP - Z pixel rate - 3542.484131M pixels/sec
FFP - Single texture - 1726.633667M pixels/sec
FFP - Dual texture - 1490.212158M pixels/sec
FFP - Triple texture - 798.393860M pixels/sec
FFP - Quad texture - 537.607605M pixels/sec
PS 1.1 - Simple - 938.785889M pixels/sec
PS 1.4 - Simple - 885.801453M pixels/sec
PS 2.0 - Simple - 597.972290M pixels/sec
PS 2.0 PP - Simple - 889.860535M pixels/sec
PS 2.0 - Longer - 256.424377M pixels/sec
PS 2.0 PP - Longer - 450.701233M pixels/sec
PS 2.0 - Longer 4 Registers - 255.315384M pixels/sec
PS 2.0 PP - Longer 4 Registers - 595.546570M pixels/sec
PS 2.0 - Per Pixel Lighting - 102.137840M pixels/sec
PS 2.0 PP - Per Pixel Lighting - 128.088409M pixels/sec

...

 

[micb]

The FX 5950 card was in a friends machine in America.

The ATI 9800 is my personal stock clocked card.

I assumed he was running stock clockrates when he sent me the file.

I have sent him a message to check to make sure the card was at stock clocks.

I will reprt back as soon as I get a reply.

 

[zsouthboy]

I'm not sure I understand what you are saying... I had gotten the idea that sometimes the NV3x is either 4x2 or 8x1(instead of 0), and it just depended on the situation. In actuallity, it is a 4x2 + 4x0, right?

Actually, it's not 4x2 + 4x0. It's 4x2 or 8x0.

8x0 = 8 depth data but 0 color data
4x2 = 4 color data + 4 depth data
8x1 = 8 color data + 8 depth data

Now I'm more confused... 4x2 isn't 8 color and depth data? at least some of the time? I was under the impression(once again, fallaciously, I guess) that that is 4 pipes, each with 2 texture units = 8, but only when multitexturing(which is in most games)...?

 

[Tridam]

I'm not sure I understand what you are saying... I had gotten the idea that sometimes the NV3x is either 4x2 or 8x1(instead of 0), and it just depended on the situation. In actuallity, it is a 4x2 + 4x0, right?

Actually, it's not 4x2 + 4x0. It's 4x2 or 8x0.

8x0 = 8 depth data but 0 color data
4x2 = 4 color data + 4 depth data
8x1 = 8 color data + 8 depth data

Now I'm more confused... 4x2 isn't 8 color and depth data? at least some of the time? I was under the impression(once again, fallaciously, I guess) that that is 4 pipes, each with 2 texture units = 8, but only when multitexturing(which is in most games)...?

No 4x2 is never '8 something' output. The maximum is 4 color outputs and 4 depth outputs. The 4 color outputs can use 2 textures. But you can't have 8 color outputs.

 

[micb]

I have compaired the scores from Tridam and my ATI 9800 and the Nvidia FX 5950 looks even worse.

It does look more like my mate in America has fed me some tweeked stats, I hope to get a reply from him on this soon.

 

[Doomtrooper]

The FX 5950 card was in a friends machine in America.

The ATI 9800 is my personal stock clocked card.

I assumed he was running stock clockrates when he sent me the file.

I have sent him a message to check to make sure the card was at stock clocks.

I will reprt back as soon as I get a reply.

 

[micb]

He sent me these Shadermark 2.0 scores. Could someone (mabye Tridam) check to see if these match or wether again these scores are too high. (which would surgest his card is clocked higher)

ShaderMark v2.0 - DirectX 9 HLSL Pixel Shader Benchmark - ToMMTi-Systems (http://www.tommti-systems.com)

video mode / device info
(1024x768) X8R8G8B8 (D24X8) vram used 58720256
HAL (pure hw vp): NVIDIA GeForce FX 5950 Ultra (Anti-Detect-Mode: off, gamma correction: DAC)



options

pixel shader version: 2_0
partial precision: off
number of render targets: 1

results:
shader 2 ( Per Pixel Diffuse Lighting): 169 fps 5.9094 mspf 847 rendered frames
shader 3 ( Per Pixel Directional Light Shader (Phong)): 110 fps 9.0642 mspf 552 rendered frames
shader 5 ( Per Pixel Spot Light Shader (Phong)): 97 fps 10.2908 mspf 486 rendered frames
shader 6 ( Per Pixel Anisotropic Lighting): 107 fps 9.3146 mspf 537 rendered frames
shader 7 ( Per Pixel Fresnel Reflections): 103 fps 9.7516 mspf 513 rendered frames
shader 9 ( Per Pixel Car Surface Shader): 51 fps 19.4994 mspf 257 rendered frames
shader 10 ( Per Pixel Environment Mapping): 206 fps 4.8431 mspf 1033 rendered frames
shader 11 ( Per Pixel Environment Bump Mapping): 167 fps 5.9784 mspf 837 rendered frames
shader 12 ( Per Pixel Bump Mapping): 91 fps 10.9457 mspf 457 rendered frames
shader 13 ( Per Pixel Shadowed Bump Mapping): 50 fps 19.8761 mspf 252 rendered frames
shader 14 ( Per Pixel Veined Marble Shader): 68 fps 14.6982 mspf 341 rendered frames
shader 15 ( Per Pixel Wood Shader): 66 fps 15.1830 mspf 330 rendered frames
shader 16 ( Per Pixel Tile Shader): 55 fps 18.3244 mspf 273 rendered frames
shader 17 ( Fur Shader With Anisotropic Lighting): 8 fps 125.2293 mspf 41 rendered frames
shader 18 ( Per Pixel Refraction and Reflection Shader with Phong Lighting): 41 fps 24.4581 mspf 205 rendered frames

Thanks.