Can someone who knows more than me comment on this?

[GwymWeepa]

Well there's been a lot of talk of the x360's GPU, and whether it would utilize unified shaders or not. Well, my question is, could a hybrid system work at all? What if a certain percentage of transistors were vertex shaders, another percentage were pixel shaders, and another percentage were unified. This could allow some flexibility for developers who may want a boost in pixel or vertex performance depending on the scene. Would this work? Would it be a good idea, or does something limit the GPU to being either one or the other configuration?

 

[jvd]

Isn't that what the whole point of unified shaders are ?

You have x amount of shaders that can be used for vertex or pixel operations . the developer than has control over how much time power is devoted to what .


Or am i missing something ?

 

[GwymWeepa]

Isn't that what the whole point of unified shaders are ?

You have x amount of shaders that can be used for vertex or pixel operations . the developer than has control over how much time power is devoted to what .


Or am i missing something ?

I mean x amount of transistors were dedicated vertex shaders, y amount would be dedicated pixel shaders, then the rest would be flexible. Usually the discussion is whether they would be all dedicated or all totally flexible/unified.

 

[jvd]

well surely them all being unified would be the best as you may still end up in a situation where you need more vertex power but you only have the hardwired pixel shader alus not being fully used .

I'm no programer of course , but i would think that a whole unified chip would be better than a partialy

 

[GwymWeepa]

well surely them all being unified would be the best as you may still end up in a situation where you need more vertex power but you only have the hardwired pixel shader alus not being fully used .

I'm no programer of course , but i would think that a whole unified chip would be better than a partialy

Well the argument I've read against 100% unified is that you can't beat a dedicated pixel shading unit with a flexible one in terms of efficiency...well why not have a base number of efficient pixel and vertex shaders, and then a certain amount of unified shaders that could augment the pixel or vertex shaders depending on the situation at hand. Seems like a good compromise. Though, then again, if a GPU is just monstrously fast, then all unified would be best, since then developers would have total flexibility and control over the amount or vertex or pixel operations they may want in a given situation, even if it costs them some efficiency.

 

[London Geezer]

Isn't that what the whole point of unified shaders are ?

You have x amount of shaders that can be used for vertex or pixel operations . the developer than has control over how much time power is devoted to what .


Or am i missing something ?

I thought Dave said that it's not as simple as that. There are some function that just work better in vertex shaders, and would be a pain to try and process them on pixel shaders, and vice versa. So, a unified shader would either be better at vertex or pixel shading, depending on the situation.

But i really think we need some expert word on this.

 

[Titanio]

I think GwymWeepa's scenario is plausible. Most games are going to require a base minimum of both types of functionality, so it makes sense to go dedicated up to a point, probably. Yes, you're taking away ultimate flexibility, but it's flexibility 99.9% of devs aren't going to need.

The other scenario being suggested is that the shaders would be unified, but biased towards either pixel or vertex shading. So still not necessarily as efficient as dedicated hardware, but this would account for more typical usage. However, they couldn't be so biased as to make processing of the other type of workload infeasible (pixel shading on vertex-biased unified shaders, and vice versa)...

 

[Iron Tiger]

I think GwymWeepa's scenario is plausible. Most games are going to require a base minimum of both types of functionality, so it makes sense to go dedicated up to a point, probably. Yes, you're taking away ultimate flexibility, but it's flexibility 99.9% of devs aren't going to need.

Kinda like how 99.9% of developers didn't feel they needed to use the EE's vector units to make games. And we ended up with uninspired graphics and a sea of people pissed off because the PS2 was "overhyped", when it was really underutilized.