speed, (for experts)

Page 1/2
| 2

By PingPong

Prophet (3228)

PingPong's picture

19-10-2016, 12:20

Hi, all. someone know the speed of the amiga blitter ?
how much time take a aligned copy of a 300x300 px from x,y to x', y' assuming aligned transfers and 4bpp ?
is there the same penality doing a normal copy or a copy with transparency emulation?
thx

Login or register to post comments

By Metalion

Paladin (987)

Metalion's picture

19-10-2016, 16:40

Errr ... This is a MSX forum, right ?

By mars2000you

Enlighted (5467)

mars2000you's picture

19-10-2016, 16:43

Maybe PingPong is building an adapter to use this Amiga device on MSX ...

By PingPong

Prophet (3228)

PingPong's picture

19-10-2016, 18:34

mars2000you wrote:

Maybe PingPong is building an adapter to use this Amiga device on MSX ...

only comparing blitter speed with v9938 and v9990 speed.
amiga should be able to copy a 320x200 screen @4bpp almost 110 times x second.
i'm not surprised it was easy to make games for this computer.
it's rougly 2 times faster than v9990 (the v9990 has 8 bit bus however, the amiga 16 bit one)
a conpare with the v9938 looks embarassing.... Tongue

By tfh

Paragon (1705)

tfh's picture

19-10-2016, 19:29

I thought you were working on an Amiga Emulator for the MSX ;-)
It will be hard to make a better one then the one we released in 1992. It worked flawlesly!

By Louthrax

Paragon (2029)

Louthrax's picture

19-10-2016, 19:36

Mmmm... We should now compare that to modern PCs with the latest NVidia/whatever killer video card Smile (still using 320x240 resolution of course).

By igal

Master (211)

igal's picture

20-10-2016, 09:29

@PingPong: Msx2 digitizing in real time in 255 Colors and Transparency in résolution high résolution 256X212(interleaved)!
0 SCREEN 8,,,,,3
10 COPYSCREEN
20 GOTO 10
You can see easely the transparency in the moments of the colors are in border of Black background Wink

See this vidéo (real hardware 8280): (Audio stéréo is the digitizing with two Music Modules)
https://www.youtube.com/watch?v=8yimCUW0Dw0
The Colorbus is copied to the Vram at incredible speed :hannibal:

If you have any doubt about the validity of the full digitization in the video below, I scroller VRAM from page 1 to page 0!
Which proves that during scanning, the [Color Bus] VDP 9938 and / or 9958 transfer its status in the VRAM
https://www.youtube.com/watch?v=mSQiA43x3qc

By PingPong

Prophet (3228)

PingPong's picture

21-10-2016, 20:05

hi, this does not surprise me at all. i know the vdp put 60 times / second the entire vram so the inverse process is perfectly possible. what i mean is the blitter speed. and the speed of v9938 blitter is 2 times x 1 order of magnitude slower than amiga blitter. this mean that v9938 can do about 1/20 of the amiga blitter speed.

By hit9918

Prophet (2844)

hit9918's picture

23-10-2016, 07:44

Quote:

it's rougly 2 times faster than v9990 (the v9990 has 8 bit bus however, the amiga 16 bit one)

but one cant just count databus pins. burst, doublebanking, things can be a factor 4 off from databus pin counting.
further the 9990 is not 8bit but a 128 pins monster with dualported RAMs.
4 vram chips 256kx4bit. makes 16bit of databus.
and dual ported means one got twice of that? then one got 32 bits.
120 nanoseconds

this makes makes a 33 megs per second ballpark. without bursting.
page 2 in the manual,"capable of direct acess from cpu to vram by means of the 16bit bus".
this whole thing smells like 486 vesa local bus and make DOOM with ARM Smile

if the second port of the dual ported RAMs is always available for cpu, and if it is also available for port IO,
then the vram write of port IO could be 7 megs per second.
an 8bit acess to one port of the dualported 120 ns RAMs.
that fills a 256x192 screen of 32768 colors in 60 fps!
well, just speculation, the TurboR touches it at 0.5 megs per second, a magnitude lower.

By PingPong

Prophet (3228)

PingPong's picture

23-10-2016, 09:47

hit9918 wrote:
Quote:

it's rougly 2 times faster than v9990 (the v9990 has 8 bit bus however, the amiga 16 bit one)

but one cant just count databus pins. burst, doublebanking, things can be a factor 4 off from databus pin counting.
further the 9990 is not 8bit but a 128 pins monster with dualported RAMs.
4 vram chips 256kx4bit. makes 16bit of databus.
and dual ported means one got twice of that? then one got 32 bits.
120 nanoseconds

this makes makes a 33 megs per second ballpark. without bursting.
page 2 in the manual,"capable of direct acess from cpu to vram by means of the 16bit bus".
this whole thing smells like 486 vesa local bus and make DOOM with ARM Smile

if the second port of the dual ported RAMs is always available for cpu, and if it is also available for port IO,
then the vram write of port IO could be 7 megs per second.
an 8bit acess to one port of the dualported 120 ns RAMs.
that fills a 256x192 screen of 32768 colors in 60 fps!
well, just speculation, the TurboR touches it at 0.5 megs per second, a magnitude lower.

i didn't know of the 16 bit databus, always sure of 8 bit width.
Well, about the direct vram access is obscure to me how cpu/vdp will synchronize in a situation like command in progress + cpu access. v99x8 handles this gracefully because it's the 'bus master' ;-)

Page 1/2
| 2
My MSX profile