nVidia and why SLI Technology is behind.

[tweakboy]

Very simple. How long has nvidia been providing SLI ? 10 years ?

real SLI is ,, doing it with different video cards. Like have a 580 and throw a 670 and SLI them. Nope you gotta get identical video card. If they can open up this technology and make it happen look how popular SLI would be.

 

[VulgarDisplay]

Very simple. How long has nvidia been providing SLI ? 10 years ?

real SLI is ,, doing it with different video cards. Like have a 580 and throw a 670 and SLI them. Nope you gotta get identical video card. If they can open up this technology and make it happen look how popular SLI would be.

The problem with your theory is that you would make the faster card operate at the speeds of the slower card. So while it would be cool to pair your gtx670 up with a gtx 570 you just turned your $400 speed demon into a 2 year old GPU.

Crossfire does allow you to mix cards of the same number generation. I believe it means you can do 7970 + 7950, or 7870 + 7850, etc. It will however only operate at the slowest cards speed, and with the amount of memory available on the smallest card.

Not all it's cracked up to be.

 

[truckerCLOCK]

Wasn't there something a couple years ago called Hydra I believe that let you pair different cards...

 

[Final8ty]

The problem with your theory is that you would make the faster card operate at the speeds of the slower card. So while it would be cool to pair your gtx670 up with a gtx 570 you just turned your $400 speed demon into a 2 year old GPU.

Crossfire does allow you to mix cards of the same number generation. I believe it means you can do 7970 + 7950, or 7870 + 7850, etc. It will however only operate at the slowest cards speed, and with the amount of memory available on the smallest card.

Not all it's cracked up to be.

And again, CF does not run the same speed as the slower card.

 

[imaheadcase]

Oh heavens no you have to use the same 2 cards! Whatever shall we do!

Look if you can fork over money for a system and even have the option to get a 2nd card, you can afford the 2nd same card. If not, then SLI is not for you in the first place.

 

[boxleitnerb]

Very simple. How long has nvidia been providing SLI ? 10 years ?

real SLI is ,, doing it with different video cards. Like have a 580 and throw a 670 and SLI them. Nope you gotta get identical video card. If they can open up this technology and make it happen look how popular SLI would be.

What kind of silly thread is that? SLI is behind? Behind what?

First, you cannot pair up cards with different architectures.
Second, if you pair up cards that perform differently, this will make microstutter worse. Nvidias framemetering can combat this to some extent, but in general, it is NOT a good idea.

 

[SlowSpyder]

And again, CF does not run the same speed as the slower card.

People do still post that a lot, my guess is because I believe that is what SLI does. I don't know if CF ever limited the clock speeds, I remember a long time ago when Appopin was running a 2900XT and a 2900 Pro 256bit together, even.

You have to wonder, though, with cards of different speeds, that will change the even odd fram rendering time that much more. I wonder if that hurts smoothness in a set up that already gets plenty of criticism for microstutter.

 

[digitaldurandal]

It doesn't matter. Read any review comparing CF and SLI and you will see the same thing. I have seen it on WSGF, H, and others. Even at a lower FPS SLI feels smoother than CF, and graphs show a lesser amount of microstutter.

So until AMD spends the time correcting the problem I will continue to choose an Nvidia solution.

Also Hydra was a huge fail. You actually lost performance in most cases on lucid's solution.

CF will allow you to use different cards to a point, like 7950 + 7970. But I do not really think that changes the market for the product much. The market for hybrid cards is really like someone getting a 6550 + 7670, that would explode the multiple card market. Unfortunately I believe it offers a lot of hurdles to be overcome before you would see a real benefit.

 

[Final8ty]

People do still post that a lot, my guess is because I believe that is what SLI does. I don't know if CF ever limited the clock speeds, I remember a long time ago when Appopin was running a 2900XT and a 2900 Pro 256bit together, even.

You have to wonder, though, with cards of different speeds, that will change the even odd fram rendering time that much more. I wonder if that hurts smoothness in a set up that already gets plenty of criticism for microstutter.

If the difference is too big it will it will cause problems, i ran one GPU at 500Mhz to see what would happen and in bf3 it was a flickering strode affair, but in other games it made little difference, i just lost some fps.

But how CF works out allocation can get quite complex.

http://developer.amd.com/gpu_assets/Harnessing the Performance of CrossFireX.pdf

 

[Nemesis 1]

Very simple. How long has nvidia been providing SLI ? 10 years ?

real SLI is ,, doing it with different video cards. Like have a 580 and throw a 670 and SLI them. Nope you gotta get identical video card. If they can open up this technology and make it happen look how popular SLI would be.

I think NV is llooking at the big picture. They know that the desk top market is going the way of DoDo. Nv is into bringing a complete compute package now . pouring money into a dead end would be foolish on NVs part. They have a great future in Compute

 

[aaksheytalwar]

Why are people feeding the troll? sli is far better than cf anyway.

 

[exar333]

SLI tech was designed from the ground-up to utilize 2 exactly the same GPUs. They would need to create a new tech to link cards together to support different archs/speeds/etc. with how rare SLI/CF is, and how much of the pain drives are already with 2 or 3 exactly identical cards, I think this would probably make It worse.

 

[BD231]

Very simple. How long has nvidia been providing SLI ? 10 years ?

real SLI is ,, doing it with different video cards. Like have a 580 and throw a 670 and SLI them. Nope you gotta get identical video card. If they can open up this technology and make it happen look how popular SLI would be.

Never mind you can us the 570 as a dedicated phys-x card, you fail.

 

[OCGuy]

Is using the same two cards really a problem? If they are alternating frames, you want them to be the same speed.

Failthread.

 

[BallaTheFeared]

There is an SLI hack which allows you to run two different cards.

http://www.youtube.com/watch?v=XH46uLDxxHY

The problem is at least with this hack, and as logic would assume based on how Nvidia does SLI, the faster card operates at lower usage, instead of 99/99 you see 99 for the slower card, and 68 from the faster card. Basically you now have two slower cards in SLI.

It really doesn't make a lot of sense to allow it, never mind support it.

 

[Keysplayr]

And again, CF does not run the same speed as the slower card.

But it still has to wait for the slowest card to complete a frame and render it before it can release it's own much more quickly rendered frame.

So, the faster card, while it won't slow it's clock speed to the slowest card, is going to sit around and wait while the slowest card completes it's work anyway.
This is with AFR, Tile based rendering. The only way this would be beneficial is if using Split screen rendering. Faster GPU takes care of a larger percentage of the workload (70% say) while the slower GPU handles (30% say). And this is load balanced in the driver due to changing loads in scenes.

In your example though, the 950MHz Card (if indeed a more powerful card so lets say all other charectaristics of the 4 cards are the same save clock speed.), would wait for the 900MHz card, the 900MHz card would wait for the 850MHz card, and the 850MHz card would wait for the 800MHz card. So, theoretically, all four of these cards are STILL only as fast as the slowest card.

 

[Keysplayr]

There is an SLI hack which allows you to run two different cards.

http://www.youtube.com/watch?v=XH46uLDxxHY

The problem is at least with this hack, and as logic would assume based on how Nvidia does SLI, the faster card operates at lower usage, instead of 99/99 you see 99 for the slower card, and 68 from the faster card. Basically you now have two slower cards in SLI.

It really doesn't make a lot of sense to allow it, never mind support it.

It does offer a bit more flexibility. For example, I could use a 4870 and a 4850 together without much difference in performance between the two, and still get better framerate benefit from two or more cards. So, it's not "worthless". It does have some value. But the slowest card in the bunch dictates how fast the frames come.

 

[MrK6]

But it still has to wait for the slowest card to complete a frame and render it before it can release it's own much more quickly rendered frame.

So, the faster card, while it won't slow it's clock speed to the slowest card, is going to sit around and wait while the slowest card completes it's work anyway.
This is with AFR, Tile based rendering. The only way this would be beneficial is if using Split screen rendering. Faster GPU takes care of a larger percentage of the workload (70% say) while the slower GPU handles (30% say). And this is load balanced in the driver due to changing loads in scenes.

In your example though, the 950MHz Card (if indeed a more powerful card so lets say all other charectaristics of the 4 cards are the same save clock speed.), would wait for the 900MHz card, the 900MHz card would wait for the 850MHz card, and the 850MHz card would wait for the 800MHz card. So, theoretically, all four of these cards are STILL only as fast as the slowest card.

That's not how AFR works and your entire post is incorrect. You shouldn't spread misinformation if you don't know how the technology you're discussing works.

 

[cmdrdredd]

That's not how AFR works and your entire post is incorrect. You shouldn't spread misinformation if you don't know how the technology you're discussing works.

I love when someone posts some piece of info that another person claims is wrong and informs the person who posts said information that they shouldn't post anymore WITHOUT countering the previous info with their supposed correct explination.

In short, it's utterly ridiculous to call someone out as wrong and then not provide a counter claim or post the corrections.

 

[MrK6]

I love when someone posts some piece of info that another person claims is wrong and informs the person who posts said information that they shouldn't post anymore WITHOUT countering the previous info with their supposed correct explination.

In short, it's utterly ridiculous to call someone out as wrong and then not provide a counter claim or post the corrections.

Why aren't you as upset at him for not referencing his nonsense? It's literally common knowledge: ATI/AMD has been doing asynchronous clocking with improved performance over identically clocked cards for almost five years (maybe longer, I forget). Literally anyone with a Crossfire setup can run one card overclocked and one at stock and will see a performance improvement over running both cards at stock.

The point is don't attempt to speak with authority when you have absolutely no clue because it misguides people who are coming to these forums for help and advice, not misinformation.

 

[Keysplayr]

I love when someone posts some piece of info that another person claims is wrong and informs the person who posts said information that they shouldn't post anymore WITHOUT countering the previous info with their supposed correct explination.

In short, it's utterly ridiculous to call someone out as wrong and then not provide a counter claim or post the corrections.

TY

GPU1 at 800MHz: "I'm rendering frame 1 of this scene. Almost done, then the floor is yours GPU2!"
GPU2 at 1000MHz: "Gee thanks GPU1. As soon as you're done, I'll release frame 2 that I already have done."
GPU3 at 700MHz: "I'm working on frame 3. Going as fast as I can which is 700MHz."

This, to me, is physics. Can't get around it. AFR = alternate frame rendering means that each GPU is tasked with the subsequent frame. Right?
I mean, I'm always open to be corrected, but MrK6 you are really nasty about it. Over and above. A little civility goes a long way. If it's not in your ability
to do so, than I must ignore.

 

[3DVagabond]

Whichever card is ready to render the next frame does. The 2 cards don't alternate back and forth every other frame.

 

[Keysplayr]

Whichever card is ready to render the next frame does. The 2 cards don't alternate back and forth every other frame.

So whats faster? a 1000MHz GPU rendering 2 frames in a row leaving the 2nd 800MHz GPU to wait? Somewhere, something is waiting for the other to finish it's work. No getting around that.

 

[MrK6]

TY

GPU1 at 800MHz: "I'm rendering frame 1 of this scene. Almost done, then the floor is yours GPU2!"
GPU2 at 1000MHz: "Gee thanks GPU1. As soon as you're done, I'll release frame 2 that I already have done."
GPU3 at 700MHz: "I'm working on frame 3. Going as fast as I can which is 700MHz."

Except that you've reduced the concept of AFR to the point that you're wrong. Because one card is slower than the other doesn't mean that the entire system can't use the extra horsepower. FPS is dynamic, just as is each frame. The system will overall process more FPS with one faster card and one slower card than both cards clocked the same as the slower card. If you're having difficulty understanding those concepts, read up: http://bit.ly/PqH4Vg .

This, to me, is physics. Can't get around it. AFR = alternate frame rendering means that each GPU is tasked with the subsequent frame. Right?

You have a very elementary understanding of physics, which again, is another reason why you should not be giving "advice" if you lack the knowledge/experience to do so.

I mean, I'm always open to be corrected, but MrK6 you are really nasty about it. Over and above. A little civility goes a long way. If it's not in your ability
to do so, than I must ignore.

And you are being corrected. Just because someone tells you you're wrong doesn't mean he is being "nasty."

 

[Keysplayr]

Except that you've reduced the concept of AFR to the point that you're wrong. Because one card is slower than the other doesn't mean that the entire system can't use the extra horsepower. FPS is dynamic, just as is each frame. The system will overall process more FPS with one faster card and one slower card than both cards clocked the same as the slower card. If you're having difficulty understanding those concepts, read up: http://bit.ly/PqH4Vg .
You have a very elementary understanding of physics, which again, is another reason why you should not be giving "advice" if you lack the knowledge/experience to do so.

And you are being corrected. Just because someone tells you you're wrong doesn't mean he is being "nasty."

This was better than an SNL episode. Thanks MrK6. So both GPU's are utilized 100% then?