PCIE express bandwidth with 2 GPU?

[ianm_ozzy]

Hello

I have a 5800X3d CPU, with 32GB, and an ASUS b550-E motherboard.

My GPU is an RTX 3080.

I am looking to ad a second GPU to put into the second slot.

It would mean both cards running at pcie x8 speeds.

I am looking to maybe buy an inexpensive GTX 1060 3GB for the second card. They are very inexpensive.

The RTX 3080 is pcie 4
The GTX 1060 is pcie 3.

The gtx 1060 3GB will run with pcie x8 version 3 bandwidth. That is fine.


What about the RTX 3080?
If it gets pcie x8 version 4 bandwidth, then that is fine.
If it gets dragged down to pcie x8 version 3, then that is a problem.

What will the RTX 3080 run at in this situation?

 

 

I want to do this for cuda based applications.

 

For gaming dedicated GPU for cuda based gibs and fluids - killing floor 2.

Also for gaming,  a dedicated card for frame generation using the lossless scaling  application.  It can get stuttery with a maxed out gpu.

 

It would be a much much less expensive option that buying one of the  RTX 4000/5000 cards.

 

 

Useful information is appreciated.

Thanks

 

[Hinjima]

1 minute ago, ianm_ozzy said:

Hello

I have a 5800X3d CPU, with 32GB, and an ASUS b550-E motherboard.

My GPU is an RTX 3080.

I am looking to ad a second GPU to put into the second slot.

It would mean both cards running at pcie x8 speeds.

I am looking to maybe buy an inexpensive GTX 1060 3GB for the second card.

The RTX 3080 is pcie 4
The GTX 1060 is pcie 3.

The gtx 1060 3GB will run with pcie x8 version 3 bandwidth. That is fine.


What about the RTX 3080?
If it gets pcie x8 version 4 bandwidth, then that is fine.
If it gets dragged down to pcie x8 version 3, then that is a problem.

What will the RTX 3080 run at in this situation?

Thanks

 

Any particular reason you want to do this?

 

[ianm_ozzy]

5 minutes ago, Hinjima said:

Any particular reason you want to do this?

 

Edited the post to explain.

[Alex Atkin UK]

10 minutes ago, ianm_ozzy said:

Hello

I have a 5800X3d CPU, with 32GB, and an ASUS b550-E motherboard.

My GPU is an RTX 3080.

I am looking to ad a second GPU to put into the second slot.

It would mean both cards running at pcie x8 speeds.

I am looking to maybe buy an inexpensive GTX 1060 3GB for the second card. They are very inexpensive.

The RTX 3080 is pcie 4
The GTX 1060 is pcie 3.

The gtx 1060 3GB will run with pcie x8 version 3 bandwidth. That is fine.


What about the RTX 3080?
If it gets pcie x8 version 4 bandwidth, then that is fine.
If it gets dragged down to pcie x8 version 3, then that is a problem.

What will the RTX 3080 run at in this situation?

 

 

I want to do this for cud based applications.

 

For gaming dedicated GPU for cuda based gibs and fluids - killing floor 2.

Also for gaming,  a dedicated card for frame generation using the lossless scaling  application.  It can get stuttery with a maxed out gpu.

 

It would be a much much less expensive option that buying one of the  RTX 4000/5000 cards.

 

 

Useful information is appreciated.

Thanks

 

If you're messing with the frame buffer content on the main card, I'd expect this is make performance worse not better, if its even possible at all.  Moving data between two cards is not efficient.

Do the games even have the option to run the compute on a different GPU to what is rendering the graphics?

[ianm_ozzy]

1 minute ago, Alex Atkin UK said:

If you're messing with the frame buffer content on the main card, I'd expect this is make performance worse not better, if its even possible at all.  Moving data between two cards is not efficient.

  Maybe.

It does not answer my question though.

 

I will be  using  the lossless scaling forums  to search for  such information.

 

I am posting here to get information about the PCIE bandwidth issues.

 

If it just helps with  the killing floor 2 game, it is still worth buying for me.

 

 

[SorryBella]

2 hours ago, ianm_ozzy said:

I want to do this for cuda based applications.

 

For gaming dedicated GPU for cuda based gibs and fluids - killing floor 2.

Also for gaming,  a dedicated card for frame generation using the lossless scaling  application.  It can get stuttery with a maxed out gpu.

This... Doesnt explain anything. I think youre conflating CUDA (Nvidia general computing system) and just general nvidia gaming features like PhysX. Those two things are not interchangeable.

[ianm_ozzy]

1 hour ago, SorryBella said:

This... Doesnt explain anything. I think youre conflating CUDA (Nvidia general computing system) and just general nvidia gaming features like PhysX. Those two things are not interchangeable.

 

So what. If you do  not understand, why.,  it is fine. Why I am doing it is not important for the purposes of the information I seek.   Please read the last line of my post.

 

[starsmine]

1 hour ago, ianm_ozzy said:

 

So what. If you do  not understand, why.,  it is fine. Why I am doing it is not important for the purposes of the information I seek.   Please read the last line of my post.

 

No we understand what you got confused on, physX (cuda) on a seperate card is a thing, but its not anymore and hasnt been for about 10 years. And using a slower GPU for computing physX actually slows down your game. 

notice the GTX 580 and the 8600GTS physX card results.

My comment is on the video still

I never did go back to having a dedicated physX card because a GTX 760 would slow down my RTX 3060 as shown by the chart using GT 730as physX cards. and for all of a couple of games? a GTX 1060 will compute the physX SLOWER then the RTX 3080 can compute the render and the physX. meaning the 1060 will be your bottleneck. 

I also dont get free electricity anymore so I dont keep the second GPU in the PC to inefficiently fold@home either so its not even there to mess with.

If you want to dedicate a GPU to frame generation, you need one with Tensor cores, a GTX 1060 will not work for your wants.

Anyways, if you go forward with the experiment, just try it, 50% chance it works the way you want it to

[ianm_ozzy]

2 hours ago, starsmine said:

No we understand what you got confused on, physX (cuda) on a seperate card is a thing, but its not anymore and hasnt been for about 10 years. And using a slower GPU for computing physX actually slows down your game. 

notice the GTX 580 and the 8600GTS physX card results.

My comment is on the video still

I never did go back to having a dedicated physX card because a GTX 760 would slow down my RTX 3060 as shown by the chart using GT 730as physX cards. and for all of a couple of games? a GTX 1060 will compute the physX SLOWER then the RTX 3080 can compute the render and the physX. meaning the 1060 will be your bottleneck. 

I also dont get free electricity anymore so I dont keep the second GPU in the PC to inefficiently fold@home either so its not even there to mess with.

If you want to dedicate a GPU to frame generation, you need one with Tensor cores, a GTX 1060 will not work for your wants.

Anyways, if you go forward with the experiment, just try it, 50% chance it works the way you want it to

 

 

A gtx 1060 could be used  for only frame generation, with another card for render.

I suggest you become familiar with the lossless scaling application.

How well it works It all depends on settings & the game.

 

Also  I am referring to  flex, which is not the same as physx.

Also if ever using physx, would just use the same card for render & physx.

 

None of what I typed directly above is relevant to what I wish to know though.

 

The motherboard diagram and  table is useful.

My ASUS b550-E motherboard has a very similar setup.  Both GPUs would  connected to the CPU, I assume with a chip to split the lanes.

 

What  I do want to know:

 

If:

A pcie 16x  version 4 is in  the top slot -       (1 in the chart).

&

a pcie 16x  version 3 is in the second slot -     (3 in the chart),

 

will the top card run at pcie x8 version4 spped  (About 16GB/s),

or,

will the first card  (1)  be dragged down to  the second card speed (About 8GB/s).

 

If the speed of the card in slot 1 speed s dragged down, it is a problem.

 

 

I have tried to  explain  what I wish to know as clearly as I can!

 

 

 

 

Thank you.

 

 

 

 

 

 

 

 

[starsmine]

3 minutes ago, ianm_ozzy said:

 

 

A gtx 1060 could be used  for only frame generation, with another card for render.

I suggest you become familiar with the lossless scaling application.

How well it works It all depends on settings & the game.

 

Also  I am referring to  flex, which is not the same as physx.

Also if ever using physx, would just use the same card for render & physx.

 

None of what I typed directly above is relevant to what I wish to know though.

 

The motherboard diagram and  table is useful.

My ASUS b550-E motherboard has a very similar setup.  Both GPUs would  connected to the CPU, I assume with a chip to split the lanes.

 

What  I do want to know:

 

If:

A pcie 16x  version 4 is in  the top slot -       (1 in the chart).

&

a pcie 16x  version 3 is in the second slot -     (3 in the chart),

 

will the top card run at pcie x8 version4 spped  (About 16GB/s),

or,

will the first card  (1)  be dragged down to  the second card speed (About 8GB/s).

 

If the speed of the card in slot 1 speed s dragged down, it is a problem.

 

 

I have tried to  explain  what I wish to know as clearly as I can!

 

 

 

 

Thank you.

 

 

 

 

 

 

 

 

Bottom slot can only ever run in x4, it is not wired to go any higher.

Slot 1 and 2 will both be in x8 mode. What I do not have clarification on is if they will be running the same PCIe version, I doubt they would, but its not clear.

That diagram IS your motherboard's diagram, not just similar.