Testing performances of my A 4000 os4.1 Classic

A forum for Classic Amiga specific support issues.
Post Reply
tlosmx
Beta Tester
Beta Tester
Posts: 195
Joined: Sat Feb 09, 2013 9:15 am

Testing performances of my A 4000 os4.1 Classic

Post by tlosmx »

I dont understand how is possible ... but i sadly test my A 4000 D with os4.1.6 with one of the best benchmark for amiga ... Quakegl / awinquakewos... and...
My amiga have the 375 mhz 604e downclocked to 366 and the radeon 9200 256.
Quake gl in 640x480@60 16bit gave the result of 26 fps
Awinquakewos in 320x240 8 bit gave me the result of 51 fps (os 3.9)

When 10 years ago i had the 604e 233 and cybervisionppc the quakegl result on 3.9 was 26.7 fps
on morphos 0.4 beta on the same machine waswith 400x300 awinquakeppc was 57.4 fps
http://www.amigazette.org/numero_09/morphos.html

On a guy on amibay the 1200 with 060 100 mhz and voodoo 2000 16mb Quakegl 68k gave a result of 25 fps
http://www.amibay.com/showthread.php?t=38703&page=8


...there is a way for optimize the machine with os 4.1 for be more faster? how can is possible have less performances with a better and 2x faster cpu?
AmigaOne Xe 933 mhz Radeon 9000pro Samsung 840 SSD
Pegasos2 1266mhz Radeon 9800pro Verbatib Sata3 SSD
Sam 460ex Lite 2gb Ram , Radeon 6670,Toshiba 500gb Sata/HDD
Amiga Cd32 / Amiga 600 / Amiga 500 / Amiga CDtv/ Amiga 4000 / Amiga 1200 /Amiga 2000 and many many many z3 Boards
User avatar
Karlos
AmigaOS Core Developer
AmigaOS Core Developer
Posts: 84
Joined: Sun Jun 19, 2011 11:42 am
Location: United Kingdom of England and anybody else that wishes to remain.

Re: Testing performances of my A 4000 os4.1 Classic

Post by Karlos »

Assuming everything else is configured correctly*, it might be down to bus speed.

I would not be in the least surprised to learn that the CVisionPPC connected directly to the CSPPC has a much faster CPU to VRAM write performance than anything going via the Mediator. I'm told the A4000 version is better, but as an example, on my Mediator1200/Voodoo3000 system, writing to VRAM peaks at around 10MiB/s. Conversely, I get around 16MiB/s on my BVisionPPC with the same accelerator. Please note, these figures were the highest rates I could get for an 040@25MHz using different methods. IIRC, move16 on the BVision, but regular move.l on the Mediator1200, on which move16 was slower. In both cases, these are rather lower than the speed at which the CPU can perform the same writes to regular RAM and so the bus is ultimately the bottleneck. I would expect this observation to be equally true for PPC.

For software-rendered graphics, this bus throughput equates to frames/second since you can only display frames as quickly as you can write the pixels. If you are creating 8-bit pixels, you aren't going to be writing them straight to the framebuffer, you'll collect them in some local fast RAM region and use properly aligned bus transfers. It might be that you have an entire framebuffer in fast ram which is copied to you offscreen buffer, or if you are doing some DooM style columnar rendering, perhaps a smaller buffer. Either way, you make 32-bit writes to VRAM. A 400x300 display requires ~ 56% more pixels than a 320x240 one, which means you need to be able to shove 56% more data over the bus per frame to reach the same frame rate as 320x240. Which, coincidentally enough, corresponds to the difference in bandwidth I get comparing my Voodoo/BVision.

For hardware-rasterized graphics, we aren't copying pixel data, we're sending geometry data. Here, the bus speed factor equates to polygons/sec which has a less obvious effect on frames/second as the visible polygons/frame varies from one frame to the next. However, for resolutions where fill rate is not the determining factor, again, a faster bus means more polygons/sec which, all things being equal, equates to more frames/second.

If bus speed is a bottleneck for you, no amount of amping up the CPU will make much difference.

*not a given. For example, are you using multitexture mode or 2 pass in your OS4 GLQuake?
tlosmx
Beta Tester
Beta Tester
Posts: 195
Joined: Sat Feb 09, 2013 9:15 am

Re: Testing performances of my A 4000 os4.1 Classic

Post by tlosmx »

I think the machine is configured in right way :P
with all option OFF(multitextaturing ecc..) i have 22.7 fps in timedemo demo 2

note on my old sam 440ap 599,99 mhz with M9 64mb on board
the result was 640x480 16bit this 36.8 sec. 26.3 fps
AmigaOne Xe 933 mhz Radeon 9000pro Samsung 840 SSD
Pegasos2 1266mhz Radeon 9800pro Verbatib Sata3 SSD
Sam 460ex Lite 2gb Ram , Radeon 6670,Toshiba 500gb Sata/HDD
Amiga Cd32 / Amiga 600 / Amiga 500 / Amiga CDtv/ Amiga 4000 / Amiga 1200 /Amiga 2000 and many many many z3 Boards
User avatar
Karlos
AmigaOS Core Developer
AmigaOS Core Developer
Posts: 84
Joined: Sun Jun 19, 2011 11:42 am
Location: United Kingdom of England and anybody else that wishes to remain.

Re: Testing performances of my A 4000 os4.1 Classic

Post by Karlos »

Turning off multitexturing on R200 will hurt performance in quake style games. The shadow/lighting information is stored as a texture and if the engine has to make 2 passes to render it, that's not good on a number of reasons.

1) You need to render the entire scene twice. The second time, you need to blend against the framebuffer which requires extra reads and writes and hence is much slower for the graphics chip than blending textures in a pipeline and writing the result to the framebuffer. You'll also incur additional Z buffer reads too, since the second pass will still need to respect any depth information written in the first. You will see this more as you increase the resolution and fill rate becomes dominant.

2) Rendering the scene in 2 passes means you basically redraw all the polygons again, which means you transfer the same geometry data over the bus twice. You will see the negative results of this regardless of resolution.

3) Visual quality is often lower, especially if your framebuffer is 16-bit since you lost colour precision in the first pass.

Note that for the Permedia, 2 pass rendering is unavoidable as it only has 1 texture unit.

My question to you is, how does the framerate you are getting vary by resolution?
tlosmx
Beta Tester
Beta Tester
Posts: 195
Joined: Sat Feb 09, 2013 9:15 am

Re: Testing performances of my A 4000 os4.1 Classic

Post by tlosmx »

Hi karlos,
Open quake and push the tab on the keyboard this will make console appair front you, after you have to write
"TIMEDEMO DEMO2" <--usually the demo2 is used for the benchmark
when the demo start running Fast Close the console pushing the tab button and wait when the demo finish (start another one)
hit the tab and you will read the time and the fps


Hope i understand your question was this :P

Luigi
AmigaOne Xe 933 mhz Radeon 9000pro Samsung 840 SSD
Pegasos2 1266mhz Radeon 9800pro Verbatib Sata3 SSD
Sam 460ex Lite 2gb Ram , Radeon 6670,Toshiba 500gb Sata/HDD
Amiga Cd32 / Amiga 600 / Amiga 500 / Amiga CDtv/ Amiga 4000 / Amiga 1200 /Amiga 2000 and many many many z3 Boards
User avatar
Karlos
AmigaOS Core Developer
AmigaOS Core Developer
Posts: 84
Joined: Sun Jun 19, 2011 11:42 am
Location: United Kingdom of England and anybody else that wishes to remain.

Re: Testing performances of my A 4000 os4.1 Classic

Post by Karlos »

Hi,

I understand how to run the timedemo. What I wanted to know is what your actual settings were. Many things affect the speed, not just the resolution. For the OS4 native version of GL quake, you should use multitexture for best performance. Without it, the game renders every frame twice.

Beyond that, perhaps altering some settings for the R200 Warp3D driver will help:

http://wiki.amigaos.net/index.php/UserDoc:Warp3D
tlosmx
Beta Tester
Beta Tester
Posts: 195
Joined: Sat Feb 09, 2013 9:15 am

Re: Testing performances of my A 4000 os4.1 Classic

Post by tlosmx »

Ops sorry ok thanks :)

My actual configuration is

Amiga4000D 2mb chip 16mb fast 3.1 roms buster 11
CyberstormPPC 060 66mhz 604e 375 at 366mhz 128 MB 60ns Ram
Acard Uw->Sata with a Sata SSD 120gb 6gb/sec on cybppc.device
Ide Uide HD and Samsung DVDrom on scsi.device
Mediator 4000 Di MK2
pci0 Spider USB (not working on 4.1 :( )
Pci1 Radeon 9200 256 all the ram is show in Amiga os 4.1.6
Pci2 Rtl 8029
Pci3 Sondblaster 128
PCi4 (core logic)

just tested and made the parameter on the wiki for the Glquake and i have 24.3 fps :/

a small info there is a tips for made the system use more chip ram for the graphic and not the fast ram in the new Picasso96?
because now i have 235 mb of chipram free and 30 of fastram ... in 2013 my amiga have the opposite problem of 15 years ago :P
AmigaOne Xe 933 mhz Radeon 9000pro Samsung 840 SSD
Pegasos2 1266mhz Radeon 9800pro Verbatib Sata3 SSD
Sam 460ex Lite 2gb Ram , Radeon 6670,Toshiba 500gb Sata/HDD
Amiga Cd32 / Amiga 600 / Amiga 500 / Amiga CDtv/ Amiga 4000 / Amiga 1200 /Amiga 2000 and many many many z3 Boards
Post Reply