multi gpu vs single gpu

When you say the card isnt running as fast as it could im assuming you mean that a single card is running faster clocks etc ? If this is the case then my understanding of it is that they throttle them back to releave the heat issues that you will get with two cards running side by side at full tilt. This is what i beleive the CEO was getting at, as you can cool a single better than you can a dual card solution then it can run faster. A dual core CPU is differant to a card like the X2 which has two seperate chips, They are still basically a crossfire set up and as such still rely on driver profiles which is also a reason for arguing that a single card is always better.
Mactronix

Exactly, you cannot tune/crank everything up full tilt to a GPU's fullest capability. You have to make some concessions to get everything to fit and work together on 1 card.

But you guys still didn't answer my question. Why wouldn't it be easier to cool a dual (or even quad) core gpu? Intel and AMD discovered that it was more efficient to have one processor with multiple cores. Why isn't the gpu industry doing the same thing?

its b/c Nvidia [whoever has the final decision] is being an A-hole right now, they are fooling themselves into believing something that simply isn't true.

granted their products are better clock per clock, stream processor for stream processor, compared to ati, but it seems they might just pull an AMD, which I hope they don't, and it is all speculation b/c I'm just not sure where they are going for the next gen cards, but time will tell...

Theyre heading in that direction, its just now that gpus have had to do this, as were reaching a point where 1 di cant be shrunk small enough to compensate for heat/cost production. With cpus, they hit that mark a year and a half ago or thereabouts. Larrabee and fusion (Intel and AMD) will sport this new trend, and well see anentirely different structure with our graphics solutions, but that still 2 years out, so for now, we are stuck with wringing out the best from single gpus, or using dual sli/cf types for more performance, tho with terrible scaling issues

I understand, but even ati is doing the same thing. As of now, nvidia is using the 65nm process for their gpus. ATI is using the 55nm process. Why wouldn't ati have incentive to go dual or even quad? With that architecture, I think it would be well within their reach.

Thanks jaydeejohn. That's what I was asking about.

I think at 45 nm youll see it, but probably not before

Why the holdup? I understand that they are having trouble shrinking it, but what makes a gpu so much harder to shrink than a cpu?

What Im somewhat concerned about is, with Intel and M$ getting together working on trying to get multithreaded aps/programs et al , Im hoping the future of graphics isnt influenced by the way its meant to be played, Intel style

Its not the smaller process, its working with multithreaded games, which there are few of, and those that are are barely doing it. Its a software solution first, so that the hardware can take advantage of it

How would the gpu industry be influenced by something that mostly involves the cpu (ex. multithreaded apps)?

Its all code. Just look at CPUs. How many apps or programs actually use 4 cores? I know that gpus are parallel, but even so, theres some things that cant be done, thus the need for a cpu. Raytracing is an example. No need for a gfx card at all.

But I think that any attempts that intel and micro$oft make to multithread their apps will encourage the gpu industry to get on its feet and deliver the next generation of cards.

Thats true, but what if they lean towards a completely cpu direction? Like raytracing? Forgetting some good things such as tesselation? If youre going to lead the way, why not go in the direction where youre already set?

Err... too many people in this thread do not seem to understand how a GPU works. For argument's sake, a GPU is already a multicore CPU. Each ROP, TMU, and SPU is essentially another core. You do not need to add multiple GPU cores to make it more powerful. Simply increasing the ROP, TMU, and SPU could would yield a greater result as you would not have to deal with the issues related to SLI and multi GPU setups in general. If they were to make a true "9800" series card that simply had the combined power of the 256 SPUs, 32ROPs, and 128 TMUs, with a 1GB memory pool, and 512-bit bus, we'd see a much faster card than Nvidia is currently releasing.

True, but then how large would that be?

My point is, we are now reaching the limits of how many transistors we can put on a di that can carry all that you said, and still be cool enougn and efficient enuff, and by linking smaller gpu cores that that solves the heat/production cost issues. The OP was asking why dont we do this yet? Im just telling him that we will eventually, and its already in the wind

yeah that or with the die shrinks getting smaller, they might simply be able to fit it all on one die, kind of like what heyyou27 was talking about.

And besides, CPU's are not like GPU's, so making them into dual or quads is not necessary and isn't feasible at this point, or Dee de dee, they would be doing it right now...

The most cost effective solution for each tier of the consumer market will be addressed in this manner, and in this manner only!

So, NO SOUP FOR YOU!!!

 
 
 

OK OK But I did say 2 years out, allowing for smaller processing/dishrink. Now wheres my soup?

no.. the Op doesn't get any soup....      

LOL so sad.....