Overview Features Coding ApolloOS Performance Forum Downloads Products Order Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.
Please visit our Apollo-Discord Server for support.



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
Documentation about the Vampire hardware

Vampire Compatible Hardwarepage  1 2 3 4 5 6 7 

Gunnar von Boehn
(Apollo Team Member)
Posts 6207
09 Feb 2017 18:20


Marcus Gerards wrote:

why should I bother the Vampire's CPU if a cheap card like the Prisma can deal with the audio?

What you say would "in theory" by a good idea.

The Prisma could help the systems if
a) the DSP would be more powerful than APOLLO - which it is not!

b) Or the Prisma could do its work on its own.
With own DMA and then not needing the main CPU.
But it can not do this and heeds the help of the 680x0 System CPU. And this way its slowing down the whole system.




Michal Warzecha

Posts 209
09 Feb 2017 20:12


Gunnar von Boehn wrote:

  b) Or the Prisma could do its work on its own.
  With own DMA and then not needing the main CPU.
  But it can not do this and heeds the help of the 680x0 System CPU. And this way its slowing down the whole system.
 
 


Probably You're right, but:

a) How much CPU power need DSP to play MP4/FLAC
b) How much power must use Apollo CPU to do this same? AMMX probably helps a lot here, but CPU must do this even it's now support some multimedia-friendly instructions.

Of course probably DSP can be integrated inside Apollo core with much more power to completly "relax" main CPU but if it's 1% of CPU usage, it's not worth to do this.



Gunnar von Boehn
(Apollo Team Member)
Posts 6207
09 Feb 2017 20:25


Michal Warzecha wrote:

Gunnar von Boehn wrote:

 
  b) Or the Prisma could do its work on its own.
  With own DMA and then not needing the main CPU.
  But it can not do this and heeds the help of the 680x0 System CPU. And this way its slowing down the whole system.
   
 
 

  Probably You're right, but:
 
  a) How much CPU power need DSP to play MP4/FLAC
  b) How much power must use Apollo CPU to do this same? AMMX probably helps a lot here, but CPU must do this even it's now support some multimedia-friendly instructions.
 
  Of course probably DSP can be integrated inside Apollo core with much more power to completly "relax" main CPU but if it's 1% of CPU usage, it's not worth to do this.
 

The point is the bottleneck to connect the PRISMA.

Lets look for example at the "clock-port" bus.
Such cards can not pull data over he bus by themselves.
They do need the help of the CPU doing this.
The clock-port bus allows MAX to transfer 1.7 MB/sec over it.
As you can see the Bus is such slow that the Apollo needs to wait about 50 clock per byte copied.

Per clock the Apollo can execute up to 4 instructions.
This means instead copying 1 byte to a card connected over the bus. Apollo could execute up to 200 Instructions ...

So the CPU is "BURNED" on doing this transfer to the card.



Wawa T

Posts 695
10 Feb 2017 00:03


from what i gather prisma is a specified one purpose chip connected to the system without dma and in case of so called "clockport" that has neven been an user expasion port, which means extremly limited data transefer width (speed) and possibility to damage the system because of wrong polarization montage.

given that you can hear mp3s or os on a 5eur thumb device i wonder why is this considered a selling point in comparison to all purpose cpu like apollo. that sould be seen a completely different philosophy.

as gunnar writes the problem is throughput of a bus providd by the genuine amiga hardware. nothing much can be done about it.


Roman S.

Posts 149
10 Feb 2017 06:10


Contrary to the Vampire, I can order Prisma and connect it to my A1200 as soon as it is delivered (typically within a week). Vampire for my hardware will be available... some day... probably.

Prisma can also play more formats (like FLAC or Vorbis) than software the Vampire can currently run (and all the formats can be played using nice GUI, not the quick&dirty FFPlay one), it can possibly output higher audio quality than Paula (it certainly supports 16 bit).


Michal Warzecha

Posts 209
10 Feb 2017 19:50


So, if You have no Vampire to connect Prisma to it- what is that discssion for? :)
  Next thing- probably when V1200 will be available, also audio trough DIGITAL-VIDEO will be working (I hope). Then- You don't need Paula/Prisma anymore. There will be much better audio I think. And- less addictional hardware/running drivers etc- it's better.
  And even If you decide to run Prisma anyway, then it should work, I think Vampire will not disable A1200 clock port.


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
10 Feb 2017 20:43


Michal Warzecha wrote:

And even If you decide to run Prisma anyway, then it should work, I think Vampire will not disable A1200 clock port.

Yes you are right.
Prisma works fine with Vampire, no problem!



Marcus Gerards

Posts 58
11 Feb 2017 11:37


And being burned or not - the CPU overhead for Zorro transfers to the Prisma is practically NIL and not much more with clockport.

I doubt we'll  reach a point where it would be beneficial to let the Apollo do the deconding instead of pushing the data into the Prisma. ;)


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
11 Feb 2017 11:52


Marcus Gerards wrote:

And being burned or not - the CPU overhead for Zorro transfers to the Prisma is practically NIL and not much more with clockport.
 
  I doubt we'll  reach a point where it would be beneficial to let the Apollo do the deconding instead of pushing the data into the Prisma. ;)

Sorry my friend, but what you say is wrong!

Decoding video-audio over PRISMA is slower than decoding it on APOLLO directly.

The overhead for copying data over clockport is HUGE.
Don't you understand this?


Roman S.

Posts 149
11 Feb 2017 12:28


Michal Warzecha wrote:

So, if You have no Vampire to connect Prisma to it- what is that discssion for? :)

For me there is no discussion - In my opinion this is currently the easiest option to play MP3/FLAC/OGG on A1200 (and probably the Zorro equipped Amigas too).

Michal Warzecha wrote:

Next thing- probably when V1200 will be available, also audio trough DIGITAL-VIDEO will be working (I hope). Then- You don't need Paula/Prisma anymore. There will be much better audio I think.

"Probably", "will be available", "I hope", "I think", etc. Prisma is available at this moment. For sure :)


Mr Niding

Posts 459
11 Feb 2017 13:00


I do own a Prisma on my A1200, and I really enjoy the soundquality, and the fact that it plays any format without any hassle.
With V600 I have run into soundfiles that just refuses to play, and the quality of the Paula output isnt brilliant.
So Ive been asking questions regarding the Prisma and Vampire combo, and noone has really answered it in plain english so a non-coder/hardware guy can understand it, until BigGun pretty much laid it out in plain language, like above;

"Decoding video-audio over PRISMA is slower than decoding it on APOLLO directly.
The overhead for copying data over clockport is HUGE.
Don't you understand this?"

The data overhead issue is probably a missing part of the puzzle for people to understand, so thanks BigGun for just spelling it out.
I guess what could be workable is a way to connect the Prisma to the Vampire down the road, but if the Vampire core gets an addition that produces a much higher quality sound than Paula, with little overhead, Prisma will have very little meaning in that enviroment.

I still love Prisma on my A1200 tho (until the V1200 arrives I guess).


Marcus Gerards

Posts 58
11 Feb 2017 14:54


Gunnar von Boehn wrote:

 
  Sorry my friend, but what you say is wrong!
 
  Decoding video-audio over PRISMA is slower than decoding it on APOLLO directly.
 
 
  The overhead for copying data over clockport is HUGE.
  Don't you understand this?
 

 
  One of our testing machines during Prisma driver development has been a bare A1200 (14 Mhz, no fast ram). It can play FLAC via Prisma/clockport - stutter free and the machine is still usable.
 
  Maybe we're talking about different things - I'm just talking about pure audio playback. So your assumption is, that playing FLAC (or let's say MP3 with >300kbit/s) would cost a Vampire equipped A1200 less CPU cycles if decoding & playing is handled by the CPU instead of clockport/Prisma? That's would be very impressive indeed. Would that also be valid for a Zorro-connection?
 


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
11 Feb 2017 15:33


Yes - You can calculate this very easily yourself.
 
Lets make an easy example.
 
For example I run here my Vampire at 100Mhz
One busaccess - on the clockport takes 560 ns.
This means to write 16bit to a device over the clockport
The CPU will need to wait 1120 ns.

On the other hand a CPU clock for the APollo is 10 ns.
And in 10ns the 68080 CPU can execute up to 4 instructions.

 
1120 ns = 112 clock for the APOLLO CPU.

In a single clock cycle the Apollo CPU can do
4 x 16bit Multiplications!
 
So we can choose either to cope
1 WORD of 16bit to the Prisma

Or we do 4x112 = 448 Multiplications of computations in the Apollo CPU in the same time!
 
My point is clear now, right?


Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
18 Feb 2017 00:16


Clear as bleach. Are you saying that Vampire struggles with slow PIO? The Prisma is just a device you throw encoded data on, that it is slow doesn't really matter. You say 1.7MB/sec is max... well, how many seconds of encoded audio is 1.7MB worth?


Gunnar von Boehn
(Apollo Team Member)
Posts 6207
18 Feb 2017 11:11


Kolbjørn Barmen wrote:

    Are you saying that Vampire struggles with slow PIO?
   

   
Its not a problem of the vampire.
Its a general problem of the slow bus.
 
   
   
Kolbjørn Barmen wrote:

The Prisma is just a device you throw encoded data on, that it is slow doesn't really matter. You say 1.7MB/sec is max... well, how many seconds of encoded audio is 1.7MB worth?
   

You ask the wrong question.
 
The point is that with this slow bus - compared to the very fast CPU. Can you get an reasonable acceleration?
 
The answer is: NO
 
 
Lets make it simple and lets look at real numbers

 
A MUL on the 7 Mhz 68000 takes    5600 nano seconds
A MUL on the 14 Mhz 68020 takes  1900 nano seconds
A MUL on the 80 Mhz 68080 takes      3 nano seconds
One clockport write takes          700 nano seconds  (on all systems)

 
You can easily see that some systems do calculate slow,
and for them writing to clockport instead calculation themselves is a win.

And you can easily see that one system calculates so fast, that its faster for it to calculate itself.


Wawa T

Posts 695
18 Feb 2017 12:43


One of our testing machines during Prisma driver development has been a bare A1200 (14 Mhz, no fast ram). It can play FLAC via Prisma/clockport - stutter free and the machine is still usable.

usable as in, you can move a window around (which is done by amiga chipset dma)?

clockport is to my knowledge a bus that is 8 bit wide. so wider data needs to be transferred sequentially, which conversion/process need to be taken care of by cpu.

prisma has only very tiny buffer on the card which needs to be constantly fed to gaurantee continous playback. the transfer is pio.

developers actually communicated that this was a problem to overcome.

taking this into acount, do you really want to say, feeding data to prisma costs no cpu time, even though a 68020 may be enough for it in most cases, which in comparison to what apollo provides is a fraction not to be bothered with?

now, it remains to be seen what de4coding speed and features would be provided by apollo core in future, but certainly it wouldnt suffer from narrow and slow data bus to output.
 


Kolbjørn Barmen
(Needs Verification)
Posts 219/ 2
18 Feb 2017 18:20



  One clockport write takes          700 nano seconds  (on all systems)
 

  Yeah, and so? Why is that a problem? The Prisma is not a coprocessor, you only throw data to it, with the added bonus of having audio output. Just like the MAS player, or various USB devices that have similar functionality.


Roger Shimada

Posts 30
18 Feb 2017 18:51


My understanding is that the Prisma has a processor, so is doing work (decoding audio).

I believe what Gunnar is saying is:

Getting data through the clockport is so slow on a Vampire that the work is being done on a Prisma might was well be done in software.

This would mean using the on-board audio, which is convenient but will have limitations.


Wawa T

Posts 695
18 Feb 2017 22:57


all that is a similar idea to use sonnet ppc card alongside vampire for hardware acceleration. the drawback is twofold:

1. the hardware in both cases needs specialized binaries, prisma doesnt let itself to be accessed via amiga standard apis like ahi, instead it needs players with precompiled support, similarly sonnet needs either genuine warpos, to be patched on the fly, or binaries adapted to a modified warpos api. so both are not useful for legacy software.

2. the hardware in both cases lies behind a pio driven bus with a limited transfer rate. sonnet behind a mediator with its 7mb/s via zoroo3, and prisma either behind zorro or behind clockport with even less throughput.

one word: pointless..


Thierry Atheist

Posts 644
19 Feb 2017 00:42


wawa t wrote:
one word: pointless..

I concur.

Isn't the BEST option, making the AGA custom chipset (Paula), which will become 16 bit (S-AGA), then converting all of your MP3s into AMIGA audio files (you play those), then very little transformation is necessary because it goes straight through our S-AGA?

All the old hardware is a bottleneck.

Looking VERY MUCH forward to getting that standalone.

posts 121page  1 2 3 4 5 6 7