View Full Version : G80 to be "hybrid" DX10 design
http://www.xbitlabs.com/news/video/display/20060220100915.html
According to that info, it will not have unified vertex/pixel shaders.
I hope they don't do that.
superklye
02-22-06, 11:25 PM
Why? Is anything (besides Vista) coming out any time soon that's going to be DX10 (or WGF 1.0...whatever the hell they're calling it now)? By the time DX10 games are out, there will be new lines of GF and XX1800000XXXX XTXXXTX cards out that are fully DX10 compatible.
I hope they don't do that.
And the benefit of a unified design is.... ?
Just because ati is doing it doesnt meat nvidia has to follow suit.
It will still work with DX 10 the cards doesn't have to be unified to be DX 10 Compliant at all.
Once unified pipes become more efficient they will be forced to make the switch, or they'll get owned.
And the benefit of a unified design is.... ?
Just because ati is doing it doesnt meat nvidia has to follow suit.
Oh, I thought it was going to be required at some point. I don't know of any benefit and it really doesn't matter to me that much. It's just the perception. I do think companies are better perceived when getting there first. Just like Nvidia was thought of as forward thinking and better prepared by supporting SM 3.0 first. Then there's the whole "what, your card doesn't have unified shaders yet?" type comments. So, no really good reasons, I was just hoping they'd have it by then. If it's not a requirement to go in that direction, then it doesn't really matter to me at all. I thought it was.
Crysis will be DX10 :)
Halo 2 could be aswell.
It better! For a game to require Vista it better have some graphical lovin :D
nutball
02-23-06, 04:11 AM
Oh, I thought it was going to be required at some point. I don't know of any benefit and it really doesn't matter to me that much. It's just the perception. I do think companies are better perceived when getting there first. Just like Nvidia was thought of as forward thinking and better prepared by supporting SM 3.0 first. Then there's the whole "what, your card doesn't have unified shaders yet?" type comments. So, no really good reasons, I was just hoping they'd have it by then. If it's not a requirement to go in that direction, then it doesn't really matter to me at all. I thought it was.
To repeat what has already been said :rolleyes: the requirement is that the driver present a unified shader API to the application. It says nothing about the underlying hardware. This is a distinction which seems to be lost on people.
So long as NVIDIA are able to deliver performance-competitive DX10-compliant parts based a non-unified hardware architecture, why should people care what the underlying hardware is?
(Oh, silly me, of course the reason is that they're worried that the Red Teams fanbois might rag on their favourite Green Team hardware and score some cheap points).
DAvid kirk already told ..something similar..
If anyone can point ot the interview.. he told that MICrosoft Directx10 while unified is their new API .. the hardware doesnt need to be "unified" to be be 100% compliant. they explained that. BUt seems to be a lot of misunderstanding about UNified hardware . People seems to think that "unified hardware" automatically means = " more advanced " or "FASTER" with "better graphics". while this is not necessarily the case. just because something is "unified" does not means is better, even more without knowing anything about real world performance in diferent games.. for example QUake4 plays very choppy and slow in the supposely more advanced unified graphics hardware. in the xbox360.. (while the same game rocks in any medium PC) that its hardware with "unified" hardware. G80 could be dedicated directxt10 processor instead of unified. like Nvidia told.. JAck of all trades master on none.. :)
Quake 4 is Opengl so I don't think its a good example to judge a unified shader architecture which is clearly DX oriented.
To repeat what has already been said :rolleyes: the requirement is that the driver present a unified shader API to the application. It says nothing about the underlying hardware. This is a distinction which seems to be lost on people.
So long as NVIDIA are able to deliver performance-competitive DX10-compliant parts based a non-unified hardware architecture, why should people care what the underlying hardware is?
(Oh, silly me, of course the reason is that they're worried that the Red Teams fanbois might rag on their favourite Green Team hardware and score some cheap points).
A few things...
Like I already mentioned, I don't really care unless it's a requirement. So, what I was trying to say is this. If it's a requirement/standard, I want them to support it. If it's not, I don't care if they do or not.
Like I already mentioned, I didn't know for sure. I don't follow this closely enough to be sure and that's why my first post was vague and my 2nd post said "Oh, I thought it was going to be required at some point".
You don't post here very often so I can understand why it wouldn't bother you or wouldn't care. This is the only site I frequent though, but regardless of that fact, I like when things are calm and civil (at any site). What I was afraid of is this, if unified shaders were a requirement and nvidia didn't have full support with the G80, the site would be flooded with people reminding us of that fact in every post. Sites go through some turmoil with every new release but if Nvidia had been missing some requirement, it would have given them even more incentive. That in turn would also mean taking much longer for things to return to normal.
Anyway, it's not like I'm going to apologize or feel bad for wanting things civil. I really don't see anything wrong with that point of view so there's no reason to make your comment more discreet by using a smaller font-size.
Quake 4 is Opengl so I don't think its a good example to judge a unified shader architecture which is clearly DX oriented.
I though XBox360 doesnt support "openGL". either case.. which ever API was used.. QUake4 plays smooth in the PC at 1280x1024 in a single P3 ghz in previous generations geforce6/7 or in a X1800xt. So something in the "unified" magic is not working well. For example one of the arguments of Nvidia by not using Unified hardware is precisely performance ,they dont believe it will as efficient for every case. G80 will be 100% Directx10 complaint ..there is nothing to worry..they never had problems with required features checklist in the past with any DIrectx API .The FX problems was not in features but with fp precision performance.
I guess it may give NV a performance lead in legacy games win compatibility mode. But they said it's much harder to do unified sdaer and vertexes. So are they saying ATi a smaller company can do it but they can't yet?
walterman
02-23-06, 06:30 PM
I'm not sure about what 'unified shaders' could mean at hardware level, but perhaps, it could mean that the chip could have a lot of ALUs in the chip, & it could assign each one to a different task (vertex shader, pixel shader ...) dinamically. It could mean more vertex shader power when needed, or more pixel shader power when needed, ... A good idea, but, who knows if i'm right.
Red_Shift
02-23-06, 10:46 PM
The thing is companies do testing and they refresh their VS/PS ratio with every new generation. Some years from now it might be really important but not for now. I still don't get why ppl are so worried about that, or maybe I know, it's the ppl that doesn't even know what that technically really means but it's a fancy word so they think it's a must have; it's like "ehehe P4 is better than AMD64, it has Hyper Threading".
Unified shaders is not even a feature, it's just making for each non-ROP shader an ALU* with every instructions needed for vertex and pixel processing; well it's not just that, a good scheduler for dynamic load balancing is needed too. It obviously has it's advantages as with a well made architecture shaders would be less time resting, and also I think vertex and pixel shaders have some instructions in common so eliminating one of each of them would shorten the pipeline.
I guess it may give NV a performance lead in legacy games win compatibility mode. But they said it's much harder to do unified sdaer and vertexes. So are they saying ATi a smaller company can do it but they can't yet?
Their architecture philosophies are not the same, it's easy to understand why two companies take different paths, something would be wrong if it wasn't like that.
*i know a pipeline has more than ALUs, I'm just generalizing for an easier understanding.
nutball
02-24-06, 03:28 AM
Unified shaders is not even a feature, it's just making for each non-ROP shader an ALU* with every instructions needed for vertex and pixel processing; well it's not just that, a good scheduler for dynamic load balancing is needed too.
Surely a scheduler is only needed if you're doing a truly unified hardware architecture, ie. the vertex shaders and pixel shaders are sharing the same resources (ALUs as you put them) on the chip?
The alternative approach (which I think NVIDIA are taking at the moment from what Kirk has said) is to extend the vertex shader units and pixel shader units so that they both have a common instruction set, but to keep them as distinct functional units on the GPU, like they are currently. From a software perspective they appear unified, from a hardware perspective they aren't. But the point is you don't need a complex scheduler load-balance... just like you don't need a scheduler to load-balance between VS & PS on current generation GPUs.
Unification can be partial, for instance it doesn't look very hard for a PS unit to run a VS shader, assuming the programmable shading unit's output can be sent to a ROP or to the rasterizer. With such an architecture, you can keep a fixed number of VS units and dynamically enrolls PS units to do VS work... (and then there's a need for a scheduler, but even if a relatively simple one might do).
Redeemed
02-24-06, 03:44 PM
Wow, true geek-dom at its best. Dudes, ya'll last me a while back with this unified load balancer crap.
The geek gods- awesome! :-D
smallSHEEP
02-24-06, 03:47 PM
I though XBox360 doesnt support "openGL". either case.. which ever API was used.. QUake4 plays smooth in the PC at 1280x1024 in a single P3 ghz in previous generations geforce6/7 or in a X1800xt. So something in the "unified" magic is not working well. For example one of the arguments of Nvidia by not using Unified hardware is precisely performance ,they dont believe it will as efficient for every case. G80 will be 100% Directx10 complaint ..there is nothing to worry..they never had problems with required features checklist in the past with any DIrectx API .The FX problems was not in features but with fp precision performance.
There are way too many variables for this assumption you make to hold water. Ports are generally botch jobs and never run as well on their adopted platform. Take Halo for example, I can only just now run it at acceptable frame rates on my 1900xt, complete madness.
Redeemed
02-24-06, 04:29 PM
Funny, I'm running Halo just fine at 16 x 12 with everything maxed including AA/AF. And that is on a 6800GT which is with a s754 3700. Runs beautifully.
Now you can, when Halo was released it was a mess. The game looked horrible, and ran horrible on my top of the line (at the time) 9800 Pro. I was so pissed off at the time that I sold it :D
vBulletin® v3.7.1, Copyright ©2000-2012, Jelsoft Enterprises Ltd.