We should have fully working DMA now (if MSE isn't stupid) that should also help a lot with streaming things.
A bigger concern than DMA between the SD card interface and Main RAM (as important as that may be), as I view it at any rate, is if the MMSP2+ will be free from internal communication design errors inherent in the MMSP2 (lack of DMA between crucial blocks). If I am not mistaken, a graphical blit to the display required, on the GP2X, that the 920T (Main CPU) copy the image data to be drawn (let's say, a sprite) from one area of Main RAM, to another area of Main RAM that was acting to store a surface to be drawn to the screen. Then the blitter would access this, with data transfer facilitated by the 920T, blit the surface, and again with the aid of the 920T, copy this surface to the framebuffer (which was also in Main RAM) to then be drawn to the display. The requirement of the main CPU to shuttle around data between every component meant it was heavily burdened during any transfer, and use of the hardware blitter made it a middleman. This would indicate using the hardware blitting functionality of the MMSP2 was actually LESS efficient than blitting entirely in software using the 920T alone. If issues like that exist in the MMSP2+, even discrete memory ICs (and paths to said ICs) would only slightly alleviate such issues.
(Such results were repeatably observed in test applications determining the feasibility of the GP2X as a development platform. Use of the hardware blitting functionality provided in the MMSP2, as implemented by the GP2X, resulted in nearly-identical or in many cases dramatically slower performance.)
I'm not sure I've properly communicated this concern. If such an issue still existed, use of 2D/3D acceleration hardware might actually provide NO performance benefit over existing software graphics on the GP2X, meaning the only worthwhile performance gains were essentially that of a higher-clocked processor. Barely any improvement at all, and a lot of wasted money and power. In light of this, the seriousness of such a decision should be evident.
To address comments that my posts are intended to be 'negative' toward the GP2X, Sam Fisher has correctly asserted that my statements are meant with no ill intent and rather to make clear to non-developers and persons involved in the design of any new system shortcomings involved in prior implementations of the MMSP2 in the GP2X design, and how they might be alleviated to produce the best possible result machine. Upon signing up for an account on this forum I did not observe any regulation stating I was not permitted to state that the GP2X or MMSP2 are imperfect designs. As the old wisdom goes, "Those who do not learn from the past are doomed to repeat it". While the GP2X was a very useful machine, it was by no means perfect, and to deny this fact is to state no future device should exist at all, and this thread closed. This does not appear to be a popular opinion.
