[TheSandman]

We have SLI and crossfire, they work fine. But they are lacking in an area that could be implemented, with a little more work from both camps. Now you have an 8600GTS and a 8800GTX per say, just to say it ok. Now they can't work in SLI can they? The anwser is no, the reason is the 8800GTX would become an 8600GTS in the traditional way this is done. But what if Nvidia and ATI worked on the code and instead made it possible to mix and match like this, as long as feature set is the the same. Now the 8600GTS wouldnt do half the work, but what if it took care of say drawing the sky, or doing say 20-25% of the work, leaving the majority to the faster card, a load balancing, scheme, would work also if cards have diffrent memory ect, the faster gets more load to balance out the pair. Now that 8800GTX + 8600GTS would do better than say a stock 8800GTX would it not. What is keeping them from the next most obvious step in this technology.

 

[Bazmecc]

isn't that what the Lucid Hydra will/does do?

 

[7hm]

Because it'd be a waste of their time and you'd see very minimal gain? and by the way its per se and it doesn't make sense in the context you used it.

 

[TheSandman]

yes but do you see it implemented, also it could be done in software, PC's are fast enough this code could be done in Software, just like lowerend cards can SLI without a bridge

 

[Afrodisiac]

ATI already does this. CFX FTW.

 

[TheSandman]

Quote:

Originally Posted by 7hm Because it'd be a waste of their time and you'd see very minimal gain? and by the way its per se and it doesn't make sense in the context you used it.

except mine was an extreme, what if you wanted to pair your 8800GTX and GTX280, that would be a good gain. Or pair your 3870 with an HD4670? there are places this would be quite nice

Quote:

Originally Posted by Afrodisiac ATI already does this. CFX FTW.

its limited in its aspects though, can I pair a 3870 with a 3650, or a 4670?

same base features, but no you cant

 

[7hm]

Quote:

Originally Posted by TheSandman yes but do you see it implemented, also it could be done in software, PC's are fast enough this code could be done in Software, just like lowerend cards can SLI without a bridge

Quote:

Originally Posted by Bazmecc isn't that what the Lucid Hydra will/does do?

yes

http://techgage.com/article/lucid_hy...pu_technology/

 

[TheSandman]

a chip thats out there, but isnt in common use doesnt mean its in use, the tech for it may exist but I dont see anyone besides them adveristing that tech

 

[TurboTurtle]

Couple of points here:

1) This would actually be detrimental to the industry. Why? If ATI and Nvidia worked on something like together, we'd lose the element of competition. The entire reason we even have cards like the GTX 295 or the 4870x2 is because of the relatively fierce competition between the two companies. If it were just one or the other - we'd probably be seeing something like the GeForce 8 series still being the highest end card, but with a much heavier price tag. Why push out more expensive research if you have no reason to make your own product obsolete? It's all marketing.

2) You'd have terrible tearing issues if you matched an (using your example) 8800 with an 8600. The card's don't know what they're rendering - it's all just pixels being shaded different colors. How would the programmers be able to code "Object X, being more detailed than Object Y, will be rendered by the 8800, while Object Y will be rendered by the 8600"? Don't see that happening.

Quite simply you'd get some objects being rendered long before others - and at that the quality of the textures would be dramatically different. For example, the 8800 might be able to handle textures in certain areas of WoW at very high detail - the 8600 quite frankly, can't. You'd be limited to the performance of the lower card.

3) Scalability. What I mean by this is (again using your example) a single 8800 (depending on the model) can be better than two 8600's in SLI. It would like throwing a cyclist on the back of Lance Armstrong during the Tour de France. Sure, technically, there's more power there from the presence of another GPU (or cyclist as the case may be) but the addition is actually a burden.

 

[TEntel]

ATi has implemented, to an extent, what you are talking about.

http://ati.amd.com/technology/crossfire/charts.html

4870x2 + 4830 FTW.

 

[TheSandman]

Quote:

Originally Posted by TurboTurtle Couple of points here: 1) This would actually be detrimental to the industry. Why? If ATI and Nvidia worked on something like together, we'd lose the element of competition. The entire reason we even have cards like the GTX 295 or the 4870x2 is because of the relatively fierce competition between the two companies. If it were just one or the other - we'd probably be seeing something like the GeForce 8 series still being the highest end card, but with a much heavier price tag. Why push out more expensive research if you have no reason to make your own product obsolete? It's all marketing. 2) You'd have terrible tearing issues if you matched an (using your example) 8800 with an 8600. The card's don't know what they're rendering - it's all just pixels being shaded different colors. How would the programmers be able to code "Object X, being more detailed than Object Y, will be rendered by the 8800, while Object Y will be rendered by the 8600"? Don't see that happening. Quite simply you'd get some objects being rendered long before others - and at that the quality of the textures would be dramatically different. For example, the 8800 might be able to handle textures in certain areas of WoW at very high detail - the 8600 quite frankly, can't. You'd be limited to the performance of the lower card. 3) Scalability. What I mean by this is (again using your example) a single 8800 (depending on the model) can be better than two 8600's in SLI. It would like throwing a cyclist on the back of Lance Armstrong during the Tour de France. Sure, technically, there's more power there from the presence of another GPU (or cyclist as the case may be) but the addition is actually a burden.

well its possible and wouldn't be a burden if say the 8600 in this case renders say only 10% of the total screen, that means 90% of the screen it wouldn't render period the end. Its all in load balancing, plain and simple

 

[7hm]

Quote:

Originally Posted by TheSandman well its possible and wouldn't be a burden if say the 8600 in this case renders say only 10% of the total screen, that means 90% of the screen it wouldn't render period the end. Its all in load balancing, plain and simple

Exactly what turbo said, it's a waste you're basically trying to scale different rendering modes for nothing and hindering the performance of the main card. And in any case why would john doe take his 8800 and get another 8600? And if that's not the case why the hell wouldn't he just straight up SLI it since hes going to get more performance that way anyway or he could just sell those 2 cards and get something that performs double.

 

[TheSandman]

what if john started with the 8600 and got an 8800 later after they got cheaper and could SLI them for an extra 10-15% boost at no cost to him?

 

[7hm]

Quote:

Originally Posted by TheSandman what if john started with the 8600 and got an 8800 later after they got cheaper and could SLI them for an extra 10-15% boost at no cost to him?

I doubt hell get a 10-15% gain from rendering something so minute, its like adding an extra card for physx

 

[TheSandman]

but a gain none the less and there are more practical reasons, wouldnt it be nice to SLI that old GTS640 with an 8800GT, wouldnt that be a nice boost? there are plenty of practical applications for this, the compaines are just greedy plain and simple

 

[7hm]

Quote:

Originally Posted by TheSandman but a gain none the less and there are more practical reasons, wouldnt it be nice to SLI that old GTS640 with an 8800GT, wouldnt that be a nice boost? there are plenty of practical applications for this, the compaines are just greedy plain and simple

You seem more angry than intuitive, the reason why we have all the technology we have now is because competition is a huge part of it. And if i had an old gts640 I'd get rid of it or give it to someone who could use it and I'd save up or sell the cards i do have to upgrade to something better. Also when you say gain none the less, how is it a gain if nothing improves? and slows down your main GPU?

 

[TheSandman]

thats what im saying it shouldn't at all, the two GPU's should run independant but together. The 8800GTX runs at its normal stuff, the 8600GTS runs at its normals stuff, neither gets altered period the end, just the 8600GTS is only given 10% of the screen to work on with its lesser specs while the 8800GTX does the rest, you would see a boost here

 

[TurboTurtle]

Quote:

Originally Posted by TheSandman thats what im saying it shouldn't at all, the two GPU's should run independant but together. The 8800GTX runs at its normal stuff, the 8600GTS runs at its normals stuff, neither gets altered period the end, just the 8600GTS is only given 10% of the screen to work on with its lesser specs while the 8800GTX does the rest, you would see a boost here

Except that practical application currently in use proves that this isn't the case.

Say Jimmy has an 8800 folding rig. He then slaps the 8600 in there and begins folding on it too. It's already been proven that this would hinder the total folding ability of the two cards combined. That's even without SLI.

One thing I forgot to mention is that it's a hardware impossibility. The 8600 and 8800 for example use two different GPU cores. If there is incompatibility in the hardware, no matter how well your software is, it simply will not be able to work.

Also, on your point about the GeForce 6 series card - it wouldn't be a boost, at all. In order for SLI to be possible the two cards must be the same. As it is right now, you have to have the same model - I.E. 8800 with an 8800. But for discussion sake let's say that you could match and 8600 with an 8800. This would be overcoming the difference in the GPU cores (G90 vs. G80, if I'm not mistaken?). However, the difference between the 6 series and 8 series is profound. Not a single aspect of the card design is similar.

However, let's still take it a step further and say that you could conceivably match a 6 series and 8 series card. As with the fundamental design with SLI - the more powerful card would downclock to match the less powerful card. In this case, the 8 series would jump back 2 generations of cards to 6 series speeds. That is a profound loss of power.

As for the mention of PhysX - it's already been proven that PhysX only takes up about 10% of a GPU's power, and at that adding a dedicated PhysX card actually hinders performance by causing the more powerful cards to wait for the dedicated slower card to finish processing. So, no - there would be no gain at all from this.