Nvidia Interivew questions
- Thread starter Rookie
- Start date
Joe DeFuria
Legend
Ask them what the max number of pixels that contain color values can be written to the frame buffer per clock, on the GeForce FX 5800/Ultra products.
And if it is more than 4, then what do they blame for the FX's inability, in any pixel rate benchmark to date, to produce more than 4 per clock.
And if it is more than 4, then what do they blame for the FX's inability, in any pixel rate benchmark to date, to produce more than 4 per clock.
But this time, PLEASE insist that we don't want PR answers... If you can't say something, then you shouldn't answer it at all and answer another question instead, IMO.
I'll try to make my questions possible to answer: I really don't think asking them the number of clocks FP32 take is reasonable...
Also, I'll make an effort to ask questions which have never been asked before. It would be good if everyone made such an effort too
Three questions:
Q1: It was said before that the NV3x products, including the ones for the mainstream market, will support the CineFX specification. That is rather vague. Will all future NV3x products support CineFX as defined by both PDFs at http://developer.nvidia.com/view.asp?IO=IO_20020721_4820 ? This includes, for example, 1024 PS instructions, Dynamic Flow Control in the VS, Max Temp Storage of 64 in the PS, 16 Temps in the VS and FP32 support.
And in case that's too long ( although I'd still prefer that one to be asked ) :
Q1, Short Version: nVidia already said all NV3x based products will support CineFX. But will it support is as defined by both PDFs at http://developer.nvidia.com/view.asp?IO=IO_20020721_4820 ? That includes 1024 Instructions & 64 Temps in the PS, FP32 and much more.
Q2: It seems the NV25 only supported FP32 in the VS. Can the NV30 use other precision modes, such as FP16, in order to gain performance in cases such as lighting, where FP32 is often overkill? Are there API limitations to this?
Q3: David Kirk seem to have implied in a launch article at extremetch that the GFFX is capable of working on 32 pixels at the same time, although it is ( in optimal conditions ) able to output only 8 pixels/clock ( and often less ) - is this correct? And how many TMUs, color writes/clock is the GFFX capable of doing? Do you think this is balanced with the number of pixels the GFFX is working on at the same time?
LOL, yeah, the 3rd question is a really lame try
Uttar
I'll try to make my questions possible to answer: I really don't think asking them the number of clocks FP32 take is reasonable...
Also, I'll make an effort to ask questions which have never been asked before. It would be good if everyone made such an effort too
Three questions:
Q1: It was said before that the NV3x products, including the ones for the mainstream market, will support the CineFX specification. That is rather vague. Will all future NV3x products support CineFX as defined by both PDFs at http://developer.nvidia.com/view.asp?IO=IO_20020721_4820 ? This includes, for example, 1024 PS instructions, Dynamic Flow Control in the VS, Max Temp Storage of 64 in the PS, 16 Temps in the VS and FP32 support.
And in case that's too long ( although I'd still prefer that one to be asked ) :
Q1, Short Version: nVidia already said all NV3x based products will support CineFX. But will it support is as defined by both PDFs at http://developer.nvidia.com/view.asp?IO=IO_20020721_4820 ? That includes 1024 Instructions & 64 Temps in the PS, FP32 and much more.
Q2: It seems the NV25 only supported FP32 in the VS. Can the NV30 use other precision modes, such as FP16, in order to gain performance in cases such as lighting, where FP32 is often overkill? Are there API limitations to this?
Q3: David Kirk seem to have implied in a launch article at extremetch that the GFFX is capable of working on 32 pixels at the same time, although it is ( in optimal conditions ) able to output only 8 pixels/clock ( and often less ) - is this correct? And how many TMUs, color writes/clock is the GFFX capable of doing? Do you think this is balanced with the number of pixels the GFFX is working on at the same time?
LOL, yeah, the 3rd question is a really lame try
Uttar
