Originally Posted by medion
You didn't debate any of the facts that I mentioned. But anyway, the original PSP only had 16MB of system RAM, and 2MB of video RAM. The newer PSP models have 32MB of system RAM, but that's for extra functionality, and no game is allowed to require it.
Also, memory bandwidth is used to shuttle data in and out from video memory to system memory as needed, usually used for texture swapping when there isn't enough video memory available. The Zune HD uses a unified memory architecture, so memory bandwidth won't be a limitation for this. Also, having double the memory of the Xbox means that more data can be store in local memory without having to be shuttled off storage.
The Xbox used a 733mhz Celeron with a Coppermine core, which was originally released on the PC side in March 2000. The Tegra uses an Arm 11 600mhz MPCore. While no official specs have been released about it's performance in terms of MIPS/FLOPS, etc, handheld devices today are far more powerful than they were from almost 10 years ago. To just assume that the older tech is better because it was made for a desktop is a bit premature.
However, we can draw some conclusions here. According to Intel, their 1.2ghz Atom CPU is as powerful as a Celeron from 2003-2004. According to nVidia, their 700mhz Tegra 650 is equal to or slightly more powerful than an Atom at 1.6ghz. So, it's a solid guess that the Tegra APX 2600, which is 100mhz slower than the 650, could be slightly slower than a Celeron from 2003-2004. Do you think that a Celeron from 2000 is going to be faster than one made 3-4 years later?
Slightly understated, the NV2A was actually based off the GF3, but used some tech from the GF4. Like you said, it was very powerful for it's time. Most devs have likened the original iPhone's graphics hardware to that of the Dreamcast, while the PSP's GPU is exactly half the power of the GPU within the PS2. The Zune HD's Tegra is more powerful than both.
While it would be premature to say that the Tegra is more powerful than a GF 3.5, it only needs to be 35-45% as powerful to pull off Halo-esque visuals at a resolution of 480x272.
The Xbox used 6:1 texture compression. If MS is uses XNA, then they are using elements of DirectX, which means the same or better texture compression. Also, as stated above, memory bandwidth doesn't mean squat in unified memory architectures, which is one reason why some consoles lately get away with far lower memory bandwidth than PCs.
Ok, since I am a mobile app developer who is working with the Tegra 2 Chipset, I can shed some light on how the APX2600 works. (Some of my work might be present on a few Android Platforms who use Tegra 2.
For starters- it is about 780MP/s and Texture sits at about 980MT/s. Vertex is variable but under ideal conditions about 75-80MV/s. Texture is very limited due to low bandwidth. So its got to be very efficient at its texture compression.
Now as for the CPU- It has 2 technically. The ARM11 is for App Processing, which includes Video Processing. The ARM7 does Audio and assists with Video when needed. Texture Compression sits at 6:1 for the GeForce GPU within the chipset. (At least thats what I use at most since 8:1 don't work when its forced to accept it).
The ARM11 CPU is a RISC CPU, its actually about equal to a 1.2Ghz Pentium III from what I can tell.