Ok, theoretically you could have a situation where a game just wouldnt run with 1 card, but it would with 2
I never seen a case like that, I saw lots of cases where 1 card works, 2 doesn't.
, so your limit is suddenly infinity (a bit more than 8). But with cards that are capable of running a game properly on their own, I'm not sure off the top of my head why you would be able to more than double performance with two cards (will need to check benchmark results, and allow for margin of error).
Again, you said you understand what theoretical max is, but you keep coming back to "How do that get that max" instead of "Is there are way to get over that." Again, it is like I said there is no way to get beyond 800% and you keep saying that "I can't think of a way to get to 800%." If you think 8 is too large, than give me another formula and then try to support it.
Basically 2 times, with the odd anomaly (which we need more input on).
It isn't hard to see the theoretical max is much higher than 2 times. Again, we have benchmarks showing that under varies cases sli/cf scales better than 200%. Each of these benchmarks runs for several minutes and can be reproduced, so it isn't like a off the norm thing. Of course you may argue that the method of which the statistic is computed may be iffy, but we have been using those benchmarks to test v-cards for a while now.
As Throck said, if the processor is the only thing holding the card back (wouldn't be the case, but hypothetically), then that could lead to an increase of performance of up to 100% (= go up to 200% in your terms).
Let say 1x i/o * 1xmem * 1xGPU = 100%, you agreed that, individually, i/o, mem, and gpu can, in theory scale linearly(200%). So assume that they are indenpendent, can't we have a case where each of them scales by 133%, making the formula 1.33*1.33*1.33 = 235%?
suppose i/o doesn't scale, but doubling the mem and cpu individually scales performance by 145%, why not?
The fact that we've split this into 3 sections (memory, processing, I/O) is arbitrary. There could be 10 little components that make up a piece of electronics, so using your method you'd say 2 pieces of euipement could (theoretically) improve performance by a factor of 2 to the power of 10 = 1024. So 2 cards would be up to 1024 times faster than 1 card. That's obviously beyond daft, but hopefully it explains the flaw in your calculation.
I pick 3 components because even you agreed that, individually, they can, in theory scale performance linearly. The number of screws are doubled, but doubling the number of screws does not scale performance. So if you can find a 4th independent factor that does scale performance and is changed by adding a new card, share it. If we know enough, the formula will become very long and complicated as some of the variables are dependent, some are independent, some scale in reverse, and many other types of dependencies.
Effective graphics memory is not doubled on current SLI or Crossfire platforms. Framebuffer data has to be mirrored across the cards(this is what is getting transferred across the bridge). You are also miscalculating performance as others have noted.
What would be a more likely explanation for the superlinear increase is that with a single card some of these games were being bottlenecked to the point that other game-related processing was being held up (most likely due to poor programming).
We can introduce more variables to the table, or we can first manage the ones on the table first. Yes, there is are common data which is required to be send to all v-cards because at initial phase it doesn't know what is or not necessary. Let say you and I both bought the same game. Each of us will need to install the game into our own storage, so if either of the storage doesn't have sufficient space, we can't proceed.
Now suppose the game install fine for both of us and we starts to play, the data generated from the game, enough is the same game, can be different. To top it off, the reminding storage space does impact performance.
Again, I don't understand why it is so complicated. OP stated that it is "Logically impossible" to have 2x performance, and I say, in theory, it can be much higher than that and gave an example. I don't know the specifics within the video card and definitely layman in terms of SLI/CF, so I simply multiply 3 independent factors which, you guys agreed, can in theory scales performance linearly, and then state that it is the theoretical max. That is all.
So far you guys are saying the number is too high, but ain't able to give me a lower theoretical max or find a contradiction to a possibility that it can indeed scale higher than my prediction.
If our goal is to see whether or not it is "Logically possible" to go beyond 2, and you guys continously, and repeatedly come back and say there is no way 8.
Guys, that is actually what I am saying, that it can't go beyond 8 times. You can only say that I am wrong if you argue that it can go beyond 8 because I said it can NOT be done. Keep giving examples of how you can't get beyond 8 only further proving my point.
Just make sure you buy 8 times as much soapy water to professionally clean those cards with.
I'm sorry Ben, but what exactly is your input to this thread? If you want to make yourself useful, report yourself.