nForce GPU has less bandwidth than a TNT1?

[Alekz]

The "internal AGP 6X interface with 1.5GB/s bandwidth" between GPU and SIA (Single Intelligent Arbiter) is cool, but isn't the traffic between GPU and RAM also through this interface? Even if it isn't, I assume that the frame buffer is not divided between the two memory banks, although textures might well be. That would mean another bottleneck, this time 2.1GB/s.
That fits in well with the claims of NVIDIA that nForce (with 128-bit memory) is 20-30% faster than the MX200, not twice as fast.

Some other peak bandwidths for comparison:
MX200: 1.33 GB/s
TNT1: 1.76 GB/s
Original GF2MX, MX400: 2.66 GB/s
TNT2 Ultra: 2.93 GB/s

BTW, AGP 6X would correspond to 1.6GB/s, not 1.5GB/s.

 

[xtreme2k]

I think integrated graphics is never intended to compete with high end accelerators. However, Nvidia's move shows that it IS feasiable to implement 128bit memory bus on the motherboard without adding significant cost, as well as giving 'decent' gaming performance. It is still WAY better than intel's i815 64bit 133MHz = 1066MB/s bandwidth I must say. Also, the extra bandwidth will decrease the 'effect' of 'stolen' bandwidth from the Graphics Core from the CPU.

 

[TheCorm]

Well, if you think about the mx200 being a value integrated graphics card, it stands quite well next to the old TNT which was a high performence card of it's time.

I was a little suprised that the TNT2 outperforms the MX400 though, interesting....

Anyone know the bandwith for the TNT2 M64???

Corm

 

[Remnant2]

why would the framebuffer need to be only in one bank? The design of the controller is such that that should be fairly transparent. Certainly, if the system memory is split between the controllers, then I see no reason why the GPU won't function similarily.