Sorry your browser is not supported!

You are using an outdated browser that does not support modern web technologies, in order to use this site please update to a new browser.

Browsers supported include Chrome, FireFox, Safari, Opera, Internet Explorer 10+ or Microsoft Edge.

Dark Physics & Dark A.I. & Dark Dynamix / No perfomance increase with multithreading or ppu. Why?

Author
Message
General Reed
18
Years of Service
User Offline
Joined: 24th Feb 2006
Location:
Posted: 20th Aug 2008 23:49
Link
I decided to see what the actual performance gain was with and without multithreading and ppu support. Here are the results

Single core software mode (phy start 0, 0, 0) - 155 fps
Multi core software mode (phy start 0, 0, 1) - 155 fps
Single core hardware mode (phy start 1, 0, 0) - 155 fps
Multi core hardware mode (phy start 1, 0, 1) - 155 fps

The scene consists of a building made up of lots of rigid bodys (about 150).

I have a few theories,
- Do i need more than one physx scene for multithreading and hardware to be taken into account
- No acceleration provided on rigid bodys, which is odd because the physx mod for ut3, shows a great increase in speed and that mainly uses rigid bodys

There is a definate acceleration in fluid and cloth with ppu/gpu acceleration. But why not rigid bodies?

Any explination would be appreciated

-General Reed

CPU: AMD X2 6000+ 3.0ghz GFX: NVIDIA BFG Geforce 8800GTS 640MB OC-550mhz core RAM: 2048mb

Back to top
Profile PM Email Website
Lukas W
21
Years of Service
User Offline
Joined: 5th Sep 2003
Location: Sweden
Posted: 21st Aug 2008 00:26
Link
Don't you need to setup DP like this to enable mulithreading?



Though I could be wrong.

I allways afraided from a clowns. aww..
Back to top
Profile PM Email
BatVink
Moderator
21
Years of Service
User Offline
Joined: 4th Apr 2003
Location: Gods own County, UK
Posted: 21st Aug 2008 10:15
Link
Lukas is correct (almost). You need to update with a 0 at after your SYNC. This tells DP to get working on the next iteration. Then, you update with a 1 just before you SYNC to lock and transfer the data into a buffer to be used in your simulation.

Back to top
Profile PM Email
Lukas W
21
Years of Service
User Offline
Joined: 5th Sep 2003
Location: Sweden
Posted: 21st Aug 2008 12:07
Link
So something like this:



From the help files:
Quote: "The process to use this involves calling phy update and passing in 0 directly after phy start is called. In the main update call phy update 1 and then phy update 0 before sync. Finally at the end of the program call phy update 1."

That is what I based my code on. But I wouldn't know if that's for multithreading or not

I allways afraided from a clowns. aww..
Back to top
Profile PM Email
General Reed
18
Years of Service
User Offline
Joined: 24th Feb 2006
Location:
Posted: 21st Aug 2008 15:39
Link
Ok thanks, that appears to work tho the perfomance increace would only appear to be about 30fps.

-General Reed

CPU: AMD X2 6000+ 3.0ghz GFX: NVIDIA BFG Geforce 8800GTS 640MB OC-550mhz core RAM: 2048mb

Back to top
Profile PM Email Website
BatVink
Moderator
21
Years of Service
User Offline
Joined: 4th Apr 2003
Location: Gods own County, UK
Posted: 21st Aug 2008 18:32
Link
Maybe the 30FPS increase is the limit of your system. But now if you start adding more physics (or increasing the accuracy) the PPU method should be far better than the decreasing alternative.

Back to top
Profile PM Email
Hoozer
18
Years of Service
User Offline
Joined: 8th Sep 2006
Location: Bremerhaven (Germany)
Posted: 23rd Aug 2008 23:33 Edited at: 24th Aug 2008 00:21
Link
I had the chance to compare the Hardware- and Software-Mode on a "8600m GT" with "512MB DDR-2 VRAM" (CPU: T7300, RAM: 2GB, OS: Vista-32; Forceware: 177.89 (modded INF-file), PhysX-Sys.-Software: 08.08.01) and I saw NO difference in FPS, BUT the processing-load was different!
Software-Mode: CPU ca. 50 to 60%
Hardware-Mode: CPU ca. 20 to 40%

I saw this in my demo-entry as well as the standard-demos (DB-Pro-DP-demos) and also the old C++-demos from the older AGEIA-SDKs!

What is the sense of Hw-Mode when I get no increase in FPS and only some less CPU-load? (Which is of no use in my eyes, when I can't increase the usability/performance of the game/demo/application!)

Is there a trick to time the simulation in a better way or is there something else I'm missing? (I doubt this, cause all other demos showed the same behaviour and they were made by more professionals than I am!)


Thanks for any info!

Hoozer

AMD64X24800+(939);2GB;GF6800LE (@12PS, 6VS; 380 MHz, RAM: 434 MHz)
DP-Sw-Mode-Comp-Entry (updated to V. 1.4):
http://forum.thegamecreators.com/?m=forum_view&t=109846&b=5&p=0
Back to top
Profile PM Email
Benjamin
22
Years of Service
User Offline
Joined: 24th Nov 2002
Location: France
Posted: 24th Aug 2008 04:43
Link
Perhaps your graphics card is the bottleneck.

Multisync - TCP Server/Client Multiplayer Plugin (DBP/DBCe)
Back to top
Profile PM Email
BatVink
Moderator
21
Years of Service
User Offline
Joined: 4th Apr 2003
Location: Gods own County, UK
Posted: 25th Aug 2008 10:38
Link
Quote: "What is the sense of Hw-Mode when I get no increase in FPS and only some less CPU-load?"


Have you tried the same demo with no physics? I am guessing you'll get the same FPS, or similar. As Benjamin says, a PPU isn't going to magically increase the performance of the graphics card.

Back to top
Profile PM Email
Hoozer
18
Years of Service
User Offline
Joined: 8th Sep 2006
Location: Bremerhaven (Germany)
Posted: 25th Aug 2008 15:14 Edited at: 25th Aug 2008 15:18
Link
@ Benjamin, BatVink:
I suppose it might be possible that the "GeFo-8600m GT" is to weak to handle PhysX- and graphics-content at the same time.
But somehow I think there should be some kind of difference besides the CPU-load (if we assume the following):

Sw-Mode: The CPU handles PhysX and the GeFo graphics (if the CPU handles all PhysX then the GPU should be able to do the max for the graphics-calculations) and so the FPS should be at their max possible. (Given that the standard-DB-Pro-DP-demos and the old AGEIA-demos are not that graphics intensive!)

Hw-Mode: CPU handles only the traditional calculations of the demo/game/app. and PhysX and graphics are done by the GPU, so the FPS should fall (if we assume the GPU ran at it's max possible in Sw-Mode and now has to handle PhysX too), cause the CPU has less to do than in Sw-Mode and does none of the graphics-calculations and therefore it shouldn't be able to catch the "stolen" GPU-performance.

But why the FPS stays absolutely the same in both Modes?

Somehow I can't understand that!

(I only had one chance to try it on my brother notebook, so I'm unable to do deeper testing to clarify this.)

Also if someone has supported GeForce-hardware, it would be nice to know what FPS-values they see in the old AGEIA-demos in Sw- and Hw-Mode. (I packed them and attached them to this post!)


Hoozer

AMD64X24800+(939);2GB;GF6800LE (@12PS, 6VS; 380 MHz, RAM: 434 MHz)
DP-Sw-Mode-Comp-Entry (updated to V. 1.4):
http://forum.thegamecreators.com/?m=forum_view&t=109846&b=5&p=0

Attachments

Login to view attachments
Back to top
Profile PM Email
BatVink
Moderator
21
Years of Service
User Offline
Joined: 4th Apr 2003
Location: Gods own County, UK
Posted: 25th Aug 2008 20:54
Link
I think the deciding factor is the transfer of data between the motherboard and graphics card. Once that reaches it's threshold, there's not a lot you can do.

BUT...if the graphics card is managing the physics, then that gives you a much bigger timeslice for calculating other game features. For example, you can make the AI more intelligent because you have freed the CPU of the Physics.

So maybe the way we should think about measuring the effectiveness of offloading the Physics should be disconnected from FPS. An effective test might be to put some redundant code in to simulate a load. Increase the load over time and measure the FPS in software and hardware mode. In other words, simulate making your program more complex.

Back to top
Profile PM Email
Hoozer
18
Years of Service
User Offline
Joined: 8th Sep 2006
Location: Bremerhaven (Germany)
Posted: 25th Aug 2008 23:19
Link
@ BatVink: You might be right there!

I just thought the Hw-Mode effect would be more visible!
I may make some more tests, if I get the chance/notebook again (or I would win a GeFo in the DP-comp, what I doubt)!

Hoozer

AMD64X24800+(939);2GB;GF6800LE (@12PS, 6VS; 380 MHz, RAM: 434 MHz)
DP-Sw-Mode-Comp-Entry (updated to V. 1.4):
http://forum.thegamecreators.com/?m=forum_view&t=109846&b=5&p=0
Back to top
Profile PM Email
Back to top

Login to post a reply

Server time is: 2024-11-24 20:57:40
Your offset time is: 2024-11-24 20:57:40