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
Running Games and Apps.

Quake 3 On Amiga (68k)page  1 2 

Hugo Pereira

Posts 72
10 Dec 2017 23:15


What do you guys think about this?
 
  EXTERNAL LINK 
Do you think V4 achieves this performance?


Daniel Sevo

Posts 299
11 Dec 2017 00:21


Remove the Voodoo 3 and redo the test and you might get a decent answer. ;-)

Without knowing any details about this particular port, it looks like its using the Voodoo3 to do a lot of the "heavy lifting" and running on some kind of hardware acceleration. (Textures look filtered etc).
AFAIK Quake 2 was the last Quake game that could be run entirely on a software renderer wehereas Q3 required a 3d graphics card. (At the time, it was those who came before the first geforce, like Voodoo, nVidia TNT etc.. They still let the CPU do the vertex pre-processing and transforms then the GPU would "do the rest" (like primitives, fragments, draw pixels at a fast speed (fill rate)..

Using WAZP3d carefully optimized for AMMX2 on 68080 could probably speed up alot of those things comapared to a CPU with no SIMD but it can't compete with Voodoo3 just yet.

I think for now, the focus should be to get a nice Quake 2 port going and all those mods that are out there. Its a bit painful but with patience one can still find all the tools required to make new Q2 maps etc..

In the future however, who knows what else they'll fit into the Cyclone V.




Mr-Z EdgeOfPanic

Posts 189
11 Dec 2017 06:18


What i understand from the EAB thread it's a WOS version running on a Sonnet PPC card in this case, running Q3 on a 68060 would be incredible slow I'm afraid.

Running this game decent on a Vampire would require a really fast FPGA like Cyclone 10 GX or ARRIA and a GPU to offload the processor.

Link to EAB thread: EXTERNAL LINK


Vojin Vidanovic

Posts 770
11 Dec 2017 16:57


Daniel Sevo wrote:

  Using WAZP3d carefully optimized for AMMX2 on 68080 could probably speed up alot of those things comapared to a CPU with no SIMD but it can't compete with Voodoo3 just yet.

Well, as ANY 3D I have high hopes of waZp3D as of *any* Warp3D compatibility layer.

My suggestion is that *IF FEASABLE* sometime in future v4 models and beyond should have some kind of Voodo3D in FPGA offering fast 3D. That would make it complete and modern system, Glide compatibile too, accompaniying but not nulling SAGA.
And would make advanced 3D titles of Q3+ era possible (my fav RTCW).


Daniel Sevo

Posts 299
11 Dec 2017 18:46


Vojin Vidanovic wrote:

 
  My suggestion is that *IF FEASABLE* sometime in future v4 models and beyond should have some kind of Voodo3D in FPGA offering fast 3D. That would make it complete and modern system, Glide compatibile too, accompaniying but not nulling SAGA.
  And would make advanced 3D titles of Q3+ era possible (my fav RTCW).

Since you did not specifically mention Voodoo3 I guess, some sort of implementation of the first Voodoo would be achievable to a degree.. But for example, Voodoo 3 transistor count is >50 million.. In comparison, the 68060 is 2,5 million. This tells me you need a really big and ridiculously expensive FPGA compared to what you get out of it.



Jan Vonka

Posts 60
12 Dec 2017 00:00


You still can have Voodoo3 in your Mediator with Vampire 1200 and the bigbox one.


Gregthe Canuck

Posts 274
12 Dec 2017 07:49


There is a lot of maturity that still needs to happen in the toolchain - the compilers, the developers and the Apollo cores themselves (i.e. fpu debugging).
 
  Getting Quake 1/2 running at a reasonable frame rate (close to 30fps) would be extremely cool. Quake 3 is a whole new level. Even a 500MHz PPC only managed frame rates from the low teens up to almost 30 FPS (depending on scene). Expecting a <100MHz processor to keep up with the PPC + VooDoo3 combo is a bit of a stretch. Maybe the V5?
 
For now I think the focus should be on 2D performance before everyone goes "nuts" on 3D. By that I mean basic graphics handling, datatypes, RTG optimization (P96), etc...
 
 
 


Nick Fellows

Posts 176
12 Dec 2017 08:19


Surely re-implementing Voodoo on FPGA is a nuts proposition. Would it not be simpler to include an actual GPU thats still manufactured on board?


Samuel Crow

Posts 424
12 Dec 2017 08:38


nick fellows wrote:

Surely re-implementing Voodoo on FPGA is a nuts proposition. Would it not be simpler to include an actual GPU thats still manufactured on board?

I would think that would probably be harder than implementing an obsolete one due to IP restrictions like patents.  Let's walk before we run and cross that bridge when we come to it.


Gregthe Canuck

Posts 274
12 Dec 2017 08:39


Hi Nick -
 
I suppose it boils down to where the teams sees the Apollo core going in the long run.
 
Is this planned to be a "full" SOC chip in future? To me that means built-in 3D and video decoding... something you see in ARM cores.
 
Or... is the roadmap to offload that functionality into something external... over a PCIe bus? But then that opens a large can or worms.
 
Or... does the team license a 3D core from a 3rd party?
 
Lots of options... but at the moment I think they are beyond the short to medium term scope of the project. Of course who knows what the team has cooking in their secret underground lair... ;)
 


Mr Niding

Posts 459
12 Dec 2017 09:09


Wont 3rd party board access to the system open the floodgates to driverdemand nightmare?
 
  The Vampire is cool from the perspective its very clearcut what hardware you are dealing with, and developers not have to consider the myriad of boards you can add to a PCI-E slot for example.
 
  Ive been reading AW.net, and there are pages upon pages of "Ive tested 20 different 3d cards, and still havent found one that works with my *insert Amiga version*.
 
  Hans and Daniel is doing their best to provide a good Amiga NG graphics expirience, but the sheer workload required to add such options make it seem like a bad idea.
 
  Isnt the RELATIVE simplicity, yet high performance, one of the sellingpoints of the Vampire?


Gregthe Canuck

Posts 274
12 Dec 2017 10:25


Mr Niding

I did mention the "large can of worms" on PCIe. Just tossing out all the obvious options for 3D... good or bad. :)




Gunnar von Boehn
(Apollo Team Member)
Posts 6207
12 Dec 2017 10:40


gregthe canuck wrote:

  Hi Nick -
   
  I suppose it boils down to where the teams sees the Apollo core going in the long run.
   
  Is this planned to be a "full" SOC chip in future? To me that means built-in 3D and video decoding...
 

 
Yes, APOLLO has build in 3D and Video acceleration instructions.
These instructions give a major performance boost.
You can see their effect already in our recent work in progress rasterizer video. The performance impact is big, they allow us to run the texture-rastercode 10 times faster than a 68060 CPU.

 


Tim D

Posts 9
12 Dec 2017 14:27


I guess porting Q3 to 68k may be the bigger issue here, wouldn't it? The Q3 shown is a PPC build and I doubt a 68k version exists.


Thierry Atheist

Posts 644
12 Dec 2017 14:43


Aren't there really just a small amount of instructions on 3D cards, they just have upwards of 400 units dedicated to each one?

We would have, on the V4 (V2 has all the LE's used up), just say, 8 or 16 of each unit, instead? I have no idea how all the extra bandwith would enter/exit the FPGA. The CPU, FPU, etc. would slow down when it was in use, I would think.


Markus Horbach

Posts 35
12 Dec 2017 16:04


>Let's walk before we run and cross that bridge when we come to it.
good Point !

a Software renderer optimised for ApolloCore Features to achieve  a platform for Quake1 like games is a good goal without requiring any Hardware changes to the existing Vampire boards.

Using ancient 3Dfx Cards is a bad choice. It will end to a ebay hunt like now 68060 accelerators with insane pricing and a dead end in sight already today.
Also no fast hardware Interface to Vampire Cards is available today.

Adapting to a modern 3d accelerator (PCI-E) will degrade the Apollo core to a better Keyboard Controller because an actual gfx Card has more calculating power.
E.g. a Raspi Zero for under 20 bucks used as a RTG device for the Vampire would mean the 68080 sends the command "start Linux based OGL Version of Quake3" and the Vampire acts as a mouse and Keyboard Controller again. Is this a real Goal for the Amiga ?

There has to be a balance between CPU and GPU, just my 2 Cents.


Marlon Beijer

Posts 182
12 Dec 2017 16:31


Tim D wrote:

  I guess porting Q3 to 68k may be the bigger issue here, wouldn't it? The Q3 shown is a PPC build and I doubt a 68k version exists.
 

  The WarpOS sources also has some work done to run on 68k. But it still requires mini/tinyGL and a GFX card that supports OpenGL.

EXTERNAL LINK


Daniel Sevo

Posts 299
12 Dec 2017 18:31


Marlon Beijer wrote:

 
Tim D wrote:

    I guess porting Q3 to 68k may be the bigger issue here, wouldn't it? The Q3 shown is a PPC build and I doubt a 68k version exists.
   

    The WarpOS sources also has some work done to run on 68k. But it still requires mini/tinyGL and a GFX card that supports OpenGL.
   
    EXTERNAL LINK   

 
  The whole point of Mini GL back in the day was to get Quake running on hardware that did not support Open GL. So I guess, in theory, you could add precisely that limited set of instructions to AMMX and make a port of Mini GL to use AMMX.
 
  Anyway, like I said previously, Q3 is probably not a sane target. IMHO its much better to work on Quake 2 as the "default" 3d engine. Plenty of nice mods for Q2 and its in every way a much more realistic target specs-wise...
 


Marlon Beijer

Posts 182
12 Dec 2017 20:20


Daniel Sevo wrote:

Marlon Beijer wrote:

   
Tim D wrote:

      I guess porting Q3 to 68k may be the bigger issue here, wouldn't it? The Q3 shown is a PPC build and I doubt a 68k version exists.
     

      The WarpOS sources also has some work done to run on 68k. But it still requires mini/tinyGL and a GFX card that supports OpenGL.
   
    EXTERNAL LINK   

   
    The whole point of Mini GL back in the day was to get Quake running on hardware that did not support Open GL. So I guess, in theory, you could add precisely that limited set of instructions to AMMX and make a port of Mini GL to use AMMX.
   
    Anyway, like I said previously, Q3 is probably not a sane target. IMHO its much better to work on Quake 2 as the "default" 3d engine. Plenty of nice mods for Q2 and its in every way a much more realistic target specs-wise...
   

Yes, of course. I'm just saying that the code has been worked on for 68k support, as response to Tim D doubting it existed.

Sometime in the future (wether it be in 1 year or 10) it could work on a Vampire setup. Maybe Vampire V5 or V6? Who knows. The sources are there when it's ready.


Vojin Vidanovic

Posts 770
12 Dec 2017 20:44


Daniel Sevo wrote:

    So I guess, in theory, you could add precisely that limited set of instructions to AMMX and make a port of Mini GL to use AMMX.
   

   
    Warp3D is kind-of-miniGL.
    Its software version (wazp3d) could be utilized to use ApolloFPU/AMMX, but again, that wouldnt reach performance of real gfx card, one of old Warp3D types + burden the CPU.
   
   
nick fellows wrote:

  Surely re-implementing Voodoo on FPGA is a nuts proposition. Would it not be simpler to include an actual GPU thats still manufactured on board?
 

 
  My comment on old Voodos is just because our Warp3D fully supports these. It also supports radeon R100/R200/R280 chips, Parmedia and Virge3D. None of them is still manufactured. Going above them would need Goa3D backport from MorphOS, and that would support a bit newer chips. AEON has OS4 RadeonHD 3D drivers, but that is too OS4 centric.
 
  We need to start where Amiga Classics left off, we cant just skip to present (like Pi`s include 3D chip).
   
   
gregthe canuck wrote:

  . Expecting a <100MHz processor to keep up with the PPC + VooDoo3 combo is a bit of a stretch. Maybe the V5?
   
  For now I think the focus should be on 2D performance before everyone goes "nuts" on 3D. By that I mean basic graphics handling, datatypes, RTG optimization (P96), etc... 
 

 
    That is exactly why Voodo 1/2 or Radeon 9200 is a proper scale GPU combo. 2D optimisations are ongoing, and 3D has to come to table on day. Warp3D is last area of 68k compatibility Vamp hasnt reached yet and could be starting point.
 
Gunnar von Boehn wrote:

  Yes, APOLLO has build in 3D and Video acceleration instructions.
  These instructions give a major performance boost.
  You can see their effect already in our recent work in progress rasterizer video. The performance impact is big, they allow us to run the texture-rastercode 10 times faster than a 68060 CPU.

AMMX is nice speed up to CPU alone, but cannot substitute real 3D chip.

posts 36page  1 2