Overview Features Coding Performance Forum Downloads Products OrderV4 Contact

Welcome to the Apollo Forum

This forum is for people interested in the APOLLO CPU.
Please read the forum usage manual.



All TopicsNewsPerformanceGamesDemosApolloVampireAROSWorkbenchATARIReleases
The team will post updates and news about our project here

RTG Mode In 32bits

Philippe Flype
(Apollo Team Member)
Posts 299
22 Mar 2016 08:17


Interesting progress on core, 32bits mode is now implemented in last beta. Works perfect. Core and Driver supports 8/15/16/24(wip)/32 and YUV422 modes.

EXTERNAL LINK 


Wawa T

Posts 695
22 Mar 2016 11:21


very cool. i have heard there is a penalty in access to ram when using higher resolutions an probably deeper color modes, because the rtg and the main ram is using the same bus in parallel. the higher the amount of memory needs to be accessed at refresh the higher the penalty.

is that the case?
i heard the suggestion to use different/opposite flanks of the clock signal to trigger access main ram (fast?) and rtg ram (chip? vram?), which would virtually decouple bot kind of ram access from each other without the need of using higher frequencies, which is impossible i guess. 


Nixus Minimax

Posts 416
22 Mar 2016 16:01


wawa t wrote:
i have heard there is a penalty in access to ram when using higher resolutions an probably deeper color modes, because the rtg and the main ram is using the same bus in parallel. the higher the amount of memory needs to be accessed at refresh the higher the penalty.
 
  is that the case?

Yes, it is. The vampire has a unified memory approach just like the original Amiga. The main difference is that the RAM of the vampire is very fast when compared to the Amiga's chipmem.

i heard the suggestion to use different/opposite flanks of the clock signal to trigger access main ram (fast?) and rtg ram (chip? vram?), which would virtually decouple bot kind of ram access from each other without the need of using higher frequencies, which is impossible i guess.

Regardless of how it would be done, one could, of course, add "fastmem" to the vampire, i.e. RAM that is exclusive to the CPU. However, it will have much less effect on the overall performance than with a classic Amiga which is mainly due to the CPU caches (meaning that the CPU can live without memory in many situations) and the relatively lower fraction of mem bandwidth taken by the screen DMA. Hence, it is not really worth it. It would be much better to use the same approach (extra pins or higher transfer clock or whatever) to make the unified RAM faster.



Gregthe Canuck

Posts 274
23 Mar 2016 01:04



Agreed - a UM architecture makes sense. This is where AMD is going with their Zen cores for good reason.

I fully expect future generations of the Apollo core to adopt a 64 bit memory interface and at higher clockrates. But that is a ways down the road yet.

posts 4