Go Back   nV News Forums > Linux Support Forums > NVIDIA Linux

Newegg Daily Deals

Reply
 
Thread Tools
Old 12-03-07, 04:07 PM   #25
rootkit
Registered User
 
Join Date: Dec 2005
Posts: 23
Default Re: hardware accelerated xrender ?

Quick comparision
Attached Files
File Type: pdf bench.pdf (21.9 KB, 1282 views)
rootkit is offline   Reply With Quote
Old 12-03-07, 06:29 PM   #26
kmare
Registered User
 
Join Date: Jul 2003
Location: greece
Posts: 180
Send a message via ICQ to kmare
Default Re: hardware accelerated xrender ?

thanx for sharing your benchs.. it really looks like a few things got hardware accellerated in RENDER in the latest 169.04 drivers!
kmare is offline   Reply With Quote
Old 12-04-07, 08:59 AM   #27
Linuxhippy
Registered User
 
Join Date: Feb 2004
Posts: 585
Default Re: hardware accelerated xrender ?

I wrote both to the Cairo mailing lists as well as the xorg list and linked your benchmark-results.

This is really great news ... looking at my oprofile information very often explained why ApplicationX/Y behaved so bad - a lot of time was spent in software loops. Great that now this is solved.

I hope I find some time to replace some of Java's software-rendering loops with XRender functionality - which means thousands of java-programs can benefit from a few source changes *yippie*

lg Clemens
Linuxhippy is offline   Reply With Quote
Old 12-04-07, 02:54 PM   #28
zbiggy
Registered User
 
Join Date: Sep 2002
Posts: 623
Default Re: hardware accelerated xrender ?

The 169.04 release shows that Nvidia learned how to do right 2D acceleration. That is nice and increases comparative advantage.
I hope the most remaining not accelerated XRender and Xlib functions will eventually get accelerated some day (like windows GDI+). Most Linux workstations are used to run desktop and GUI professional apps. Better Xlib and XRender acceleration increases work comfort.

Nvidia must have full library of accelerated functions for Geforce to accelerate Windows GDI/GDI+. It is just a matter of code copy-paste from Windows driver source to Linux tree and add some fine tuning specific for X architecture. Simply GPU commands for drawing a line or something else is the same for any system because GPU drawing commands are specific to GPU not OS. That is why I could not understand why Nvidia took so long to port 2D drawing code from Windows to Linux driver.

Nvidia advertises it uses Unified Driver architecture by sharing most code between windows and linux. So why so much accelerations and features are not present in Linux?

Linux and Windows drawing libraries have different names but work in very same way:
Windows: X.org:
GDI ------- Xlib
GDI+ ----- XRender
OpenGL --- OpenGL
WGL ------ GLX
DXVA,Purevideo ----- XvMC,Xv,vaAPI
zbiggy is offline   Reply With Quote
Old 12-04-07, 03:59 PM   #29
Linuxhippy
Registered User
 
Join Date: Feb 2004
Posts: 585
Default Re: hardware accelerated xrender ?

Quote:
I hope the most remaining not accelerated XRender and Xlib functions will eventually get accelerated some day (like windows GDI+). Most Linux workstations are used to run desktop and GUI professional apps. Better Xlib and XRender acceleration increases work comfort.
I have to agree, absolutly.

Quote:
Nvidia must have full library of accelerated functions for Geforce to accelerate Windows GDI/GDI+. It is just a matter of code copy-paste from Windows driver source to Linux tree and add some fine tuning specific for X architecture. Simply GPU commands for drawing a line or something else is the same for any system because GPU drawing commands are specific to GPU not OS. That is why I could not understand why Nvidia took so long to port 2D drawing code from Windows to Linux driver.
Well the situation is very different. GDI+ on WindowsXP was mostly non accalerated. On Windows-Vista accaleration is done through Direct3D, which means completly different code-paths and not much opportunity to share anything. XRender is a 2D only API whereas 2D drawing is only a small sub-system of Direct3D.

Thanks a lot NVidia

lg Clemens
Linuxhippy is offline   Reply With Quote
Old 12-06-07, 05:44 AM   #30
tier
Registered User
 
Join Date: Sep 2007
Posts: 118
Default Re: hardware accelerated xrender ?

Yeah it got faster... but compared to my technically slower x3100 from intel it is still slow as hell. I hope you are working on more improvements :-)

Great work so far!
tier is offline   Reply With Quote
Old 12-06-07, 06:04 AM   #31
Linuxhippy
Registered User
 
Join Date: Feb 2004
Posts: 585
Default Re: hardware accelerated xrender ?

Quote:
but compared to my technically slower x3100 from intel it is still slow as hell.
Do you have any benchmarks? Because as far as I know intel's EXA driver isn't in a very good shape.

lg Clemens
Linuxhippy is offline   Reply With Quote
Old 12-06-07, 08:59 AM   #32
tier
Registered User
 
Join Date: Sep 2007
Posts: 118
Default Re: hardware accelerated xrender ?

Quote:
Originally Posted by Linuxhippy
Do you have any benchmarks? Because as far as I know intel's EXA driver isn't in a very good shape.

lg Clemens
Benchmarks?
I can SEE that the screen of my slower Thinkpad with x3100 is drawn much faster and the system stays more responsive when opening complex websites, for instance.

Modechanging is much much faster, too.
tier is offline   Reply With Quote

Old 12-06-07, 09:41 AM   #33
Linuxhippy
Registered User
 
Join Date: Feb 2004
Posts: 585
Default Re: hardware accelerated xrender ?

The subjective faster painting could also come from the fact that the X3100 is s hared-memory solution, and therefor the driver is able to optimize many things which can't be done so well for boards with seperated VRAM.

So far the intel driver isn't any better (even in 2D), it just benefits from the cheap hardware-design

lg Clemens
Linuxhippy is offline   Reply With Quote
Old 12-06-07, 10:40 AM   #34
tier
Registered User
 
Join Date: Sep 2007
Posts: 118
Default Re: hardware accelerated xrender ?

Quote:
Originally Posted by Linuxhippy
The subjective faster painting could also come from the fact that the X3100 is s hared-memory solution, and therefor the driver is able to optimize many things which can't be done so well for boards with seperated VRAM.

So far the intel driver isn't any better (even in 2D), it just benefits from the cheap hardware-design

lg Clemens
That's paradox.
Why shall a cheap hardware design be much faster than an expensive solution with own vram?
Why should it be able to optimize things much more?
Imo that's nonsense
tier is offline   Reply With Quote
Old 12-06-07, 11:01 AM   #35
Linuxhippy
Registered User
 
Join Date: Feb 2004
Posts: 585
Default Re: hardware accelerated xrender ?

A weak driver very often falls back to software rendering - however for software rendering the source and the destination surface have to reside in system-ram (so that the processor can access it).

So a typical case is that the destination surface has to be pulled from VRAM into system RAM, the the software rendering occurs, and later the result is uploaded to VRAM again.
Very often those copy-operations dominate the "rendering" time, not the actual rendering done by the CPU - this was also the case for NVidia old subpixel-font-rendering which was done by the CPU. When the switched from AGP to PCIe latencies were much higher and many people complained - however the rendering of the small glyphs was never the problem.
Thats also the reason why pixman (the software rendering library used by Xorg) is not extremly optimized - because most time is lost during those transfers.

Now imagine a shared memory system, everything is already in RAM - the driver maybe has to do some management stuff to allow the CPU to access it, or re-map at - but after all you don't have expensive bus-communication, nor don't you copy mb's of data.
Thats why shared memory systems _are_ faster for operations which are not accalerated by the driver.
Thats also the reson why a GF6150 outperformed a 8800GTX with the unaccalerated drivers in renderbench.

I own a GMA950 powered laptop, and when stressing e.g. FireFox I see ~15-20% of total CPU time spent in software rendering loops. (with XAA). So its not the great hardware, nore the superior driver - its just the fact that the GMA950 system is a shared memory architecture.

lg Clemens
Linuxhippy is offline   Reply With Quote
Old 12-06-07, 11:06 AM   #36
tier
Registered User
 
Join Date: Sep 2007
Posts: 118
Default Re: hardware accelerated xrender ?

Okay, good explanation :-) Thank you.
I think i made a mistake... i tested rendering of some websites without obeying that the laptop with the nvidia chip had plugins like flash activated and the other one didn't.

I made some renderbenches, which clearly show which one's faster :-)

Nvidia NVS 140m
Code:
Available XRENDER filters:
nearest
bilinear
convolution
fast
good
best
Setup...
*** ROUND 1 ***
---------------------------------------------------------------
Test: Test Xrender doing non-scaled Over blends
Time: 0.578 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing non-scaled Over blends
Time: 0.626 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing non-scaled Over blends
Time: 0.202 sec.
*** ROUND 2 ***
---------------------------------------------------------------
Test: Test Xrender doing 1/2 scaled Over blends
Time: 0.103 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 1/2 scaled Over blends
Time: 0.086 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 1/2 scaled Over blends
Time: 0.096 sec.
*** ROUND 3 ***
---------------------------------------------------------------
Test: Test Xrender doing 2* smooth scaled Over blends
Time: 0.198 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 2* smooth scaled Over blends
Time: 0.213 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 2* smooth scaled Over blends
Time: 2.698 sec.
*** ROUND 4 ***
---------------------------------------------------------------
Test: Test Xrender doing 2* nearest scaled Over blends
Time: 0.191 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 2* nearest scaled Over blends
Time: 0.215 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 2* nearest scaled Over blends
Time: 0.971 sec.
*** ROUND 6 ***
---------------------------------------------------------------
Test: Test Xrender doing general nearest scaled Over blends
Time: 0.391 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing general nearest scaled Over blends
Time: 0.405 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing general nearest scaled Over blends
Time: 1.614 sec.
*** ROUND 7 ***
---------------------------------------------------------------
Test: Test Xrender doing general smooth scaled Over blends
Time: 0.400 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing general smooth scaled Over blends
Time: 0.400 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing general smooth scaled Over blends
Time: 8.718 sec.
Intel X3100
Code:
Available XRENDER filters:
nearest
bilinear
convolution
fast
good
best
Setup...
*** ROUND 1 ***
---------------------------------------------------------------
Test: Test Xrender doing non-scaled Over blends
Time: 2.946 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing non-scaled Over blends
Time: 2.944 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing non-scaled Over blends
Time: 0.228 sec.
*** ROUND 2 ***
---------------------------------------------------------------
Test: Test Xrender doing 1/2 scaled Over blends
Time: 0.978 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 1/2 scaled Over blends
Time: 0.967 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 1/2 scaled Over blends
Time: 0.083 sec.
*** ROUND 3 ***
---------------------------------------------------------------
Test: Test Xrender doing 2* smooth scaled Over blends
Time: 19.743 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 2* smooth scaled Over blends
Time: 19.767 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 2* smooth scaled Over blends
Time: 2.862 sec.
*** ROUND 4 ***
---------------------------------------------------------------
Test: Test Xrender doing 2* nearest scaled Over blends
Time: 13.712 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing 2* nearest scaled Over blends
Time: 13.687 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing 2* nearest scaled Over blends
Time: 0.976 sec.
*** ROUND 6 ***
---------------------------------------------------------------
Test: Test Xrender doing general nearest scaled Over blends
Time: 20.642 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing general nearest scaled Over blends
Time: 20.563 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing general nearest scaled Over blends
Time: 1.646 sec.
*** ROUND 7 ***
---------------------------------------------------------------
Test: Test Xrender doing general smooth scaled Over blends
Time: 33.400 sec.
---------------------------------------------------------------
Test: Test Xrender (offscreen) doing general smooth scaled Over blends
Time: 33.836 sec.
---------------------------------------------------------------
Test: Test Imlib2 doing general smooth scaled Over blends
Time: 13.169 sec.
I'm sorry for my "aggressive" post before..
tier is offline   Reply With Quote
Reply


Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Forum Jump


Similar Threads
Thread Thread Starter Forum Replies Last Post
Performance Boost: troubleshooting hardware, Android toolchains, and tiny computing News Archived News Items 0 06-16-12 11:10 AM
The technologist's guide to troubleshooting hardware News Archived News Items 0 06-13-12 10:20 AM
Podcast: DDN WOS Software on OCP Storage Hardware to Enable Hyperscale Storage Clouds News Archived News Items 0 05-22-12 07:10 AM
PowerDVD Hardware Acceleration adrianhall General Software 11 06-08-10 03:46 PM
hardware T&L nosrema NVIDIA Linux 2 09-14-02 02:35 PM

All times are GMT -5. The time now is 04:21 AM.


Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright 1998 - 2014, nV News.