MunkyMark

[Zstream]

Originally posted by: munky

Originally posted by: Zstream
61

14

11

2



3800+ @ 2.45ghz
1900XTX @ 670/782

http://img146.imageshack.us/my.php?image=munkytestie4.jpg

You didnt hit "Enter" to run the benchmark, did you?

I was just testing you

 

[Munky]

Ok, I just made some updates, and uploaded the newer version. Hopefully it will fix some of the issues people were having with SM2 cards not running and hanging while closing the app. Let me know how it works.

 

[happy medium]

This time I hit the enter button
5.11
5.00
5.87
1.14

 

[xtknight]

Congrats, it works nicely in wine. That's some potential!

GeForce 7800 GT/PCI/SSE2
PS3, single pass, dynamic rendering
iterations: 255
fps: 11
SM3=6.11, SM2=6.00, FBO=6.10, CPU=1.56

Not a big deal, but in VMware it still crashes with this. I'm not sure why (if the capabilities are unavailable there should be a graceful exit though). (DEP is only on for essential services.)

The exception unknown software exception (0xc0000005) occurred in the application at location 0x00000000.

Click on OK to terminate the program
Click on CANCEL to debug the program

 

[Zstream]

Originally posted by: xtknight
Congrats, it works nicely in wine. That's some potential!

GeForce 7800 GT/PCI/SSE2
PS3, single pass, dynamic rendering
iterations: 255
fps: 11
SM3=6.11, SM2=6.00, FBO=6.10, CPU=1.56

Not a big deal, but in VMware it still crashes with this. I'm not sure why (if the capabilities are unavailable there should be a graceful exit though). (DEP is only on for essential services.)

The exception unknown software exception (0xc0000005) occurred in the application at location 0x00000000.

Click on OK to terminate the program
Click on CANCEL to debug the program

So does that mean if my SM3 score is 24 and yours is 6 my 1900XTX is four times faster?

 

[schneiderguy]

Originally posted by: Zstream

Originally posted by: xtknight
Congrats, it works nicely in wine. That's some potential!

GeForce 7800 GT/PCI/SSE2
PS3, single pass, dynamic rendering
iterations: 255
fps: 11
SM3=6.11, SM2=6.00, FBO=6.10, CPU=1.56

Not a big deal, but in VMware it still crashes with this. I'm not sure why (if the capabilities are unavailable there should be a graceful exit though). (DEP is only on for essential services.)

The exception unknown software exception (0xc0000005) occurred in the application at location 0x00000000.

Click on OK to terminate the program
Click on CANCEL to debug the program

So does that mean if my SM3 score is 24 and yours is 6 my 1900XTX is four times faster?

no

edit: well in this benchmark, yes. but not in real games

 

[Munky]

Originally posted by: schneiderguy

Originally posted by: Zstream

Originally posted by: xtknight
Congrats, it works nicely in wine. That's some potential!

GeForce 7800 GT/PCI/SSE2
PS3, single pass, dynamic rendering
iterations: 255
fps: 11
SM3=6.11, SM2=6.00, FBO=6.10, CPU=1.56

Not a big deal, but in VMware it still crashes with this. I'm not sure why (if the capabilities are unavailable there should be a graceful exit though). (DEP is only on for essential services.)

The exception unknown software exception (0xc0000005) occurred in the application at location 0x00000000.

Click on OK to terminate the program
Click on CANCEL to debug the program

So does that mean if my SM3 score is 24 and yours is 6 my 1900XTX is four times faster?

no

edit: well in this benchmark, yes. but not in real games

The SM3 method uses per-pixel dynamic branching. Depending on the hardware, it may or may not provide an increase in performance. Factors such as cost of dynamic branching as well as how many pixels per batch the gpu works with will affect the results. For comparison, use the single pass SM2 method to measure brute PS performance without the benefits/penalties of dynamic branching.

 

[Munky]

Originally posted by: Sentry2
SM3 = 51.5
SM2 = 31
FBO = 28.66
CPU = 3.72 @ 3.6GHz

Updated

Hey, it even works with CF!
Anyone wanna test SLI?

 

[GundamSonicZeroX]

Did you code this yourself?

 

[Munky]

Yep. I got some ideas from looking at other similar apps, but I made this myself. Mainly I wanted to see just how much faster a GPU can crunch math than a CPU.

 

[GundamSonicZeroX]

How come some people are getting 1 digit results while others are getting high 20's?

 

[xtknight]

When multiple threads are used for cpu rendering, each thread renders an interlaced image - meaning that for 2 threads each thread renders every other line. With 4 threads each one does every 4th line and so on. The number of rendering threads is equal to the number of processors detected.

Would you mind explaining how this is? Aren't 3D apps rendered using triangle/vertex arrays and line strips/etc? Or are you doing it in a way that is slower but multithreaded, allowing it to do one line at a time?

 

[Munky]

Originally posted by: GundamSonicZeroX
How come some people are getting 1 digit results while others are getting high 20's?

The x1k cards from Ati have much more efficient dynamic branching performance than other cards. Dynamic branching is used in the SM3 mode to break out of a loop when the result is found, instead of running all 255 iteration for each pixel.

 

[Munky]

Originally posted by: xtknight

When multiple threads are used for cpu rendering, each thread renders an interlaced image - meaning that for 2 threads each thread renders every other line. With 4 threads each one does every 4th line and so on. The number of rendering threads is equal to the number of processors detected.

Would you mind explaining how this is? Aren't 3D apps rendered using triangle/vertex arrays and line strips/etc? Or are you doing it in a way that is slower but multithreaded, allowing it to do one line at a time?

The cpu method still uses OpenGL for the actual drawing on the screen, by simply reading from a buffer that the cpu writes to. The drawing is not multithreaded, it just draws pixels based on the data in the buffer, without any triangles or vertices. But the cpu does all the math, and if there 2 threads, each thread calculates the values for every other line: so thread 1 does even-numbered lines, and thread 2 does odd-numbered lines. Both threads write to different parts of the same buffer. When both threads are finished, OpenGL reads the buffer and draws the result.

 

[JRW]

Originally posted by: x80064
I just use escape to exit the program, works fine everytime

Yea I did read the readme.txt lol , escape doesnt work for me even in the updated version.

My scores seem low after reading through this thread? using latest official nvidia driver (91.47).

Results: http://img127.imageshack.us/img127/8761/clip2ic9.jpg

 

[Munky]

Originally posted by: JRW

Originally posted by: x80064
I just use escape to exit the program, works fine everytime

Yea I did read the readme.txt lol , escape doesnt work for me even in the updated version.

My scores seem low after reading through this thread? using latest official nvidia driver (91.47).

Results: http://img127.imageshack.us/img127/8761/clip2ic9.jpg

Your SM2 score does seem too low, even compared to other 7-series users. Not sure what the issue is, but your cpu score and FBO score is almost the same as mine. I'll have to try some more ideas on the refusing to exit issue. Do the F-keys work for you, at least?

 

[xtknight]

Originally posted by: munky

Originally posted by: xtknight

When multiple threads are used for cpu rendering, each thread renders an interlaced image - meaning that for 2 threads each thread renders every other line. With 4 threads each one does every 4th line and so on. The number of rendering threads is equal to the number of processors detected.

Would you mind explaining how this is? Aren't 3D apps rendered using triangle/vertex arrays and line strips/etc? Or are you doing it in a way that is slower but multithreaded, allowing it to do one line at a time?

The cpu method still uses OpenGL for the actual drawing on the screen, by simply reading from a buffer that the cpu writes to. The drawing is not multithreaded, it just draws pixels based on the data in the buffer, without any triangles or vertices. But the cpu does all the math, and if there 2 threads, each thread calculates the values for every other line: so thread 1 does even-numbered lines, and thread 2 does odd-numbered lines. Both threads write to different parts of the same buffer. When both threads are finished, OpenGL reads the buffer and draws the result.

Gotcha.

All F-Keys and Escape work fine here.

Edit: disregard what was asked here (about the multiple images), it was covered in the readme.

 

[JRW]

Originally posted by: munky

Originally posted by: JRW

Originally posted by: x80064
I just use escape to exit the program, works fine everytime

Yea I did read the readme.txt lol , escape doesnt work for me even in the updated version.

My scores seem low after reading through this thread? using latest official nvidia driver (91.47).

Results: http://img127.imageshack.us/img127/8761/clip2ic9.jpg

Your SM2 score does seem too low, even compared to other 7-series users. Not sure what the issue is, but your cpu score and FBO score is almost the same as mine. I'll have to try some more ideas on the refusing to exit issue. Do the F-keys work for you, at least?

Yep the F keys work as well as the other controls (Tab etc.) ,The escape key just doesnt wanna respond and I made sure my ESC key does work outside of the program

 

[Munky]

Ok, all the people who have trouble exiting the app, I made a new version with an optional "-debug" command line switch. The switch forces the app to run in a 640x480 window, so try it and tell me if you still can't exit correctly. That way I'll know if it's related to the fullscreen mode or not.

 

[JRW]

Originally posted by: munky
Ok, all the people who have trouble exiting the app, I made a new version with an optional "-debug" command line switch. The switch forces the app to run in a 640x480 window, so try it and tell me if you still can't exit correctly. That way I'll know if it's related to the fullscreen mode or not.

Escape does work when running it with -debug option.

 

[GundamSonicZeroX]

Originally posted by: munky

Originally posted by: GundamSonicZeroX
How come some people are getting 1 digit results while others are getting high 20's?

The x1k cards from Ati have much more efficient dynamic branching performance than other cards. Dynamic branching is used in the SM3 mode to break out of a loop when the result is found, instead of running all 255 iteration for each pixel.

Call me dumb but are you going to recode it where NV cards get ore reasonable scores?

 

[Munky]

Originally posted by: GundamSonicZeroX

Originally posted by: munky

Originally posted by: GundamSonicZeroX
How come some people are getting 1 digit results while others are getting high 20's?

The x1k cards from Ati have much more efficient dynamic branching performance than other cards. Dynamic branching is used in the SM3 mode to break out of a loop when the result is found, instead of running all 255 iteration for each pixel.

Call me dumb but are you going to recode it where NV cards get ore reasonable scores?

What do you mean? Currently I use the same shaders for both Ati and Nvidia cards, and even more surprisingly, both cards can actually run it without errors. If the next generation of Nvidia cards improve dynamic branching efficiency, then they will get increased SM3 scores without me having to change anything.

 

[Paratus]

Still no joy. I can get the debug window but it still just goes away and never comes back. I still have to kill it with the task manager. (Seems to be using 50% of my P4 w HT so it's only spawning 1 thread)

 

[Munky]

Originally posted by: Paratus
Still no joy. I can get the debug window but it still just goes away and never comes back. I still have to kill it with the task manager. (Seems to be using 50% of my P4 w HT so it's only spawning 1 thread)

Does it at least respond to the F-keys or Enter? Does it even display anything?

 

[Paratus]

Originally posted by: munky

Originally posted by: Paratus
Still no joy. I can get the debug window but it still just goes away and never comes back. I still have to kill it with the task manager. (Seems to be using 50% of my P4 w HT so it's only spawning 1 thread)

Does it at least respond to the F-keys or Enter? Does it even display anything?

Nope without the debug it resets the screen to 1024x768 and then it goes black. None of the F keys work nor ESC. In the task manager it using about 34,000K of memory and 50% of the CPU. I also no longer get the SM3 warning message.

Does it take several minutes to do anything? While I'm on an older machine it's not that old (P4 3.2E 800FSB w HT skt 478 - 2GB Ram, AGP 8x 9600XT -128mb Vram all stock)

I would still like to try it if you feel like doing some more work on it.