does Memory bandwidth doubles in sli ?

[notty22]

As posted above. Actually nothing is doubled in 2 card SLI.
It's as correct to look at as you have 2 cards to do the same amount of work. 1/2 the work load each.
We measure performance in FPS.
Memory bandwidth is also a specification quoted per /second.
So getting 60 fps requires each card to process 30frames, never saturating memory bandwidth. Where one card outputting 60fps might.

Memory amount needed is never lessened because each card is dong work on the same amount of pixels/resolution needed for those alternating frames in a continuous scene.

 

[aigomorla]

Memory bandwidth doubles with dual gpu. Memory framebuffer does not.

This is where most people get confused.

+1

the bandwith doubles.. the frame buffer doesnt.
Becuase each card has a set buffer... and ur still using the same buffers.
But only difference is your using both buffers at the same time.

Its like u have a 2 x 3ton truck and a 6 ton truck. (looking ONLY at memory buffer)

But that's not what I said.

Again, only examples:

1. Benchmarks. Two GTX 650 Ti are faster than one GTX 680.

2. One GTX 680 is not twice as fast as GTX 650. One GTX 650 Ti is not twice the performance of a GTX 650 (mines the clock differences).

and more such example. How do you explain that?

because now u have bandwiths coming into play and not buffer.
As we said.. having 2 cards double bandwith.. hence why SLI is faster... u got 2 cards rendering half what 1 card normally would do at the same time.
But the Memory buffer used in both cards are still cap'd and the same.

Your confusing SLI as multi core..
Its not the same... a multicore shares uniform resorces...
SLI is putting 2 systems together and bridging it... You still have physical limitations when looking at the machines individually, however you have more bandwith or work being able to process because you have 2 systems splitting the work.

The GTX680 -> GMC Typhoon pickup truck...
GTX650 in SLI -> 2 Ford Escapes....

Obviously ur gonna get more work done in the 2 ford escapes if they took off at the same time vs the single typhoon.
However if the work needed to transport is small, the GMC typhoon will beat the 2 fusions due it being faster.

It has to do with how the program is setup to handle SLI... as well as the cards being SLI'd.


SLI isnt all thats cracked up to be.
Because you got 2 cards rendering simultaneously... sometimes they dont SYNC when rendering.
One card will get Tasks which are IMBA compared to the other card.
This is where u see what people complain of SLI Micro shuttering.
Its when 1 SLI card is not in complete sync with the other causing whatever that card is rending to render slowly and u get ripping / desyncing.

 

[sxr7171]

It does but isn't it also consumed in double?

 

[bystander36]

+1

the bandwith doubles.. the frame buffer doesnt.
Becuase each card has a set buffer... and ur still using the same buffers.
But only difference is your using both buffers at the same time.

Its like u have a 2 x 3ton truck and a 6 ton truck. (looking at memory buffer)

That depends on how you look at it. For each frame rendered, the bandwidth is the same, as it is only 1 GPU that renders it. If you look at it in groups of 2 frames, then each two frames has 2 GPU's rendering it, with each frame having its own bandwidth.

 

[stateofmind]

+1

the bandwith doubles.. the frame buffer doesnt.
Becuase each card has a set buffer... and ur still using the same buffers.
But only difference is your using both buffers at the same time.

Its like u have a 2 x 3ton truck and a 6 ton truck. (looking ONLY at memory buffer)



because now u have bandwiths coming into play and not buffer.
As we said.. having 2 cards double bandwith.. hence why SLI is faster... u got 2 cards rendering half what 1 card normally would do at the same time.
But the Memory buffer used in both cards are still cap'd and the same.

Your confusing SLI as multi core..
Its not the same... a multicore shares uniform resorces...
SLI is putting 2 systems together and bridging it... You still have physical limitations when looking at the machines individually, however you have more bandwith or work being able to process because you have 2 systems splitting the work.


SLI isnt all thats cracked up to be.
Because you got 2 cards rendering simultaneously... sometimes they dont SYNC when rendering.
One card will get Tasks which are IMBA compared to the other card.
This is where u see what people complain of SLI Micro shuttering.
Its when 1 SLI card is not in complete sync with the other causing whatever that card is rending to render slowly and u get ripping / desyncing.

I agree with that. I was against the simple statement that the bandwidth stays the same.

 

[aigomorla]

That depends on how you look at it. For each frame rendered, the bandwidth is the same, as it is only 1 GPU that renders it. If you look at it in groups of 2 frames, then each two frames has 2 GPU's rendering it, with each frame having its own bandwidth.

actually no..

its like comaring a PC which is single core PC at 100mhz... then u got 2 pc's at 75mhz...

the 2 pc's at 75mhz if on the same platform will obviously be faster in work which was designed to be parallel's.

And as Shintai stated... that is what makes it no equal.

If the 2 PC's were at 50mhz... and 1 PC at 100mhz was held... then i still dont think it will be equal.

Its because the 2 cards we scaled is not half the speed of the card were trying to match, which is why the SLI setup typically wins in scenarios where a single card would actually struggle.

That depends on how you look at it. For each frame rendered, the bandwidth is the same, as it is only 1 GPU that renders it. If you look at it in groups of 2 frames, then each two frames has 2 GPU's rendering it, with each frame having its own bandwidth.

again.. no because of what shintai stated.
The 2 frames being rendered are not on the same speed or scale as the 1 frame being rendered.
The fact the 2 frames being rendered are about 3/4th the time as the 1 frame being rendered... the end product is where u see result.

Shintai's arguement takes it here.
The time that 1 card has already rendered a single frame..
The 2 card setup has only done 3/4ths the work of both frames... there is no complete work.

 

[bystander36]

again.. no because of what shintai stated.
The 2 frames being rendered are not on the same speed or scale as the 1 frame being rendered.
The fact the 2 frames being rendered are about 3/4th the time as the 1 frame being rendered... the end product is where u see result.

Shintai's arguement takes it here.
The time that 1 card has already rendered a single frame..
The 2 card setup has only done 3/4ths the work of both frames... there is no complete work.

You quoted the same quote for two different arguments...was that intentional?

As far as the 2 cards doing 3/4th the work of both frames, that is not correct. It may appear that way in the end at times, but in some games, scaling is 100%, and 2 cards does double performance, yet in other cases it does not. That is because there are other aspects in play, such as the CPU which has to prep the frames for both cards.

Anyways, each individual frame takes the same time to be rendered regardless of SLI or not. SLI just lets two frames to be rendered at once, or rather offset by half a frame time. When 100% scaling isn't achieved (most the time, but not all), it is due to the prep time happening on the rest of the computer.

Whether you view it as doubling the bandwidth depends on how you want to view things. More bandwidth can be viewed if you consider groups of two frames, but not if you consider how the affects of a single frame.