PPC Developers Board

[HenryCase]

downix wrote:
Have my own host, thankfully, just need to get off my butt and get it all together

Cool beans.

downix wrote:
Rather than stcking to old standards, which eat up silicon and slow down the system, it is a whole-cloth design change.  The video display system, for instance, can generate a windowing GUI in a fraction of the RAM than a vga-based display system, and context switching or even resolution switching is also much simpler.  

I have no idea how you are achieving this, but well done anyway!

downix wrote:
The basic concept is, rather than add modern features such as memory control, DMA, and bus mastering to an older (XT) based system, you build it from the ground up around these concepts.  You get a performance boost for general functionality and can grow the system with less conflicts as well.  In addition, it is flexible in nature, so rather than having a massve GPU that sits idle while you crunch spreadsheets, you have a massive DSP which can run your spreadsheet, or as a GPU for when you run that must-have game.  I'm in the middle of re-designing this DSP to avoid several patents I ran across.

Could you achieve a similar result by designing a hardware template that included a Tesla GPU as standard, running an OS built using CUDA? More info here:
http://www.nvidia.com/object/tesla_computing_solutions.html

[downix]

HenryCase wrote:

downix wrote:
Rather than stcking to old standards, which eat up silicon and slow down the system, it is a whole-cloth design change.  The video display system, for instance, can generate a windowing GUI in a fraction of the RAM than a vga-based display system, and context switching or even resolution switching is also much simpler.  

I have no idea how you are achieving this, but well done anyway!

Well, for example, the classic RAMDAC model, where your GPU, once you've finished rendering the 3D scene, NOW has to put the frame together.  I split this function away from the 3D section, slimming both cores while being more flexible.  Also I focused a lot more attention on memory control and bus utilization, two areas that really lack in normal PC's but are focused on heavily in the heavy duty workstation market.

downix wrote:
The basic concept is, rather than add modern features such as memory control, DMA, and bus mastering to an older (XT) based system, you build it from the ground up around these concepts.  You get a performance boost for general functionality and can grow the system with less conflicts as well.  In addition, it is flexible in nature, so rather than having a massve GPU that sits idle while you crunch spreadsheets, you have a massive DSP which can run your spreadsheet, or as a GPU for when you run that must-have game.  I'm in the middle of re-designing this DSP to avoid several patents I ran across.

Could you achieve a similar result by designing a hardware template that included a Tesla GPU as standard, running an OS built using CUDA? More info here:
http://www.nvidia.com/object/tesla_computing_solutions.html

Tesla is unable to be purchased without the display engine nor the memory control system, so would be wasting a lot of silicon for no gain.  Conversely, I could utilize an Ageia Phys-X as a GPU much more efficiently, for this reason.

[AJCopland]

@Downix
That reminds me, a while ago I found this analysis of the Physx patents.

Worth noting that the first page is meaningless introductory drivel but the second and third pages contains the goods! Erm, enjoy!

Andy

[kolla]

HenryCase wrote:

kolla wrote:

kolla wrote:
What I meant is that OS4 doesnt need MiniMig :-)

Kolla, OS4 definitely needs more hardware capable of running it. Minimig could be one of those solutions.

OK?
Let's say you've made a MiniMig with PPC - now what?

After a PPC Minimig was created, we'd probably need to do some work to get OS4 working on it. After this, the OS4 h/w problem is solved.

So, let's ignore that there is no PCI or Zorro bus for gfx cards, no networking, no disk controller, and only glorious OCS for OS4 to display its 24bit icons on. Yay!

I'd say you have _a lot_ of work to do still. And after that again you have to make OS4 work on it. Why not just skip directly to "make OS4 work on it" and replace the hardware with something that exists already?

Grab an efika or something, it's a heck lot better starting point than the MiniMig!

More OS4 h/w means more users of OS4, which means more OS4 development. How hard is that to understand?

The hard bit is that I dont see the point in wasting time with OS4 for MiniMig. The point of MiniMig is to be able to run old amiga software. The point of OS4 was to move to PPC and get away from legacy hardware!

OS4 for classic is a nice treat for those of us with capable PPC cards in our amigas, but that's really all it is!

[HenryCase]

kolla wrote:
So, let's ignore that there is no PCI or Zorro bus for gfx cards, no networking, no disk controller, and only glorious OCS for OS4 to display its 24bit icons on. Yay!

I'm working on an AGA solution for Minimig too, and I'm not the only one. OS4 will run with AGA controlling the display stuff.

kolla wrote:
Grab an efika or something, it's a heck lot better starting point than the MiniMig!

Efika is closed source. Minimig is open source. Open source is the only way Amiga can continue to develop and thrive, as it doesn't rely on other companies that may fail us in the future.

kolla wrote:
The hard bit is that I dont see the point in wasting time with OS4 for MiniMig. The point of MiniMig is to be able to run old amiga software. The point of OS4 was to move to PPC and get away from legacy hardware!

Minimig = new Amiga hardware. As far as I can see, you see it as near enough the finished article. I see it as a development platform. Got to have some ambitions otherwise you never get anywhere, even if those ambitions require a lot of work.