What's the maximum amount of Dual GPU cards that you can SLI or Crossfire

[Tech_Dreamer]

I kinda know that the max amount of Dual GPU being set to support only 2 , But recently read a post about how maingear guys didi a quad Titan Z buildfor a total of 8 GPU in a single PC , How is that possible? or was this possible from the begining taken that "only 2 dual GPU being supported" being a myth....

 

Also , NVIDIA grid servers use like 12-16 GPU in series , presumabily sli'd ,So i was wondering how is this possible , do they use a seperate driver to acheive this feat or is it something else? or is it due to mobo limitation?

 

[MRE Teck]

2

[thekeemo]

2 but they can be used in rendering without SLI

[SeanAngelo]

pretty sure its 2.

[dadasmithywinkle]

2 in SLI, but more with software-based linking.

[TheRetiredSlave]

SLI/CF only support 4 GPUs for gaming (either 4 single cards or two dual-GPU cards).

 

AFAIK, you can have as many GPUs as you want, provided that the application you're running has support for it. Things like coin mining and other GPU-compute tools often support large arrays of GPUs (if you have space to fit them).

[Flight1sim]

You can use dedicated software to give each pair of Cards (with 2GPU's each) to have it's own operations (like each card does like 1/8th the workload)

Also, i know this is true with compute cards, but you can also set them up to work together

[Thebman712]

you can use 2 dual gpus/4 gpu cores in sli but you can use more non sli, if you really need to

[Enderman]

2 in SLI/CFX

 

you can have MORE than 2 in a system, they just cant be in SLI/CFX

they will still work, it just means that you cant game with the combined power of all of them

some programs for rendering or 3d modeling can use GPUs that are not in SLI/CFX because its not gaming

[ForsakenLive]

You can use them without SLI just fine. It's on games that you need it to work together on the same process.

However parallel tasks can be configured different. You can have a GPU just to hook up more monitors, do rendering on the background, etc.

SLI is for games.

[luke_2bossly]

in servers alot.... but in normal systems 4 max 

[Aniallation]

@Tech_Dreamer

 

Two in SLI. However if not in SLI and you use each card independently you can have as many as you want.

[QueenDemetria]

In the Maingear system, only two of the cards are being used for SLI while the other 2 cards can be used for Cuda or other professional applications. For nVidia grid or any other GPU based server/workstation, SLI is a waste of time and resources, so all cards operate independently in terms of how they are connected, but they can still work together.

[Tech_Dreamer]

2 in SLI, but more with software-based linking.

 

 

SLI/CF only support 4 GPUs for gaming (either 4 single cards or two dual-GPU cards).

 

AFAIK, you can have as many GPUs as you want, provided that the application you're running has support for it. Things like coin mining and other GPU-compute tools often support large arrays of GPUs (if you have space to fit them).

 

 

You can use dedicated software to give each pair of Cards (with 2GPU's each) to have it's own operations (like each card does like 1/8th the workload)

Also, i know this is true with compute cards, but you can also set them up to work together

 

 

2 in SLI/CFX

 

you can have MORE than 2 in a system, they just cant be in SLI/CFX

they will still work, it just means that you cant game with the combined power of all of them

some programs for rendering or 3d modeling can use GPUs that are not in SLI/CFX because its not gaming

 

 

You can use them without SLI just fine. It's on games that you need it to work together on the same process.

However parallel tasks can be configured different. You can have a GPU just to hook up more monitors, do rendering on the background, etc.

SLI is for games.

 

 

So Each card has to be identified seperately & the work load can be split & spread over it by the software , i have a few more question, can you please answer them to the best of your abilities?

1. So does this use like the  GPU driver of a different calliber to get the job done or use the same drivers as everyone else uses?
2. Does this cut off the memory to pretty much half or less (as in SLI) when on a threaded workload like that like that?
3. Any ideas on why games cannot utilize this?
4. Any current mainstream software support this form of rendering like Adobe?

[Flight1sim]

 

So Each card has to be identified seperately & the work load can be split & spread over it by the software , i have a few more question, can you please answer them to the best of your abilities?

1. So does this use like the  GPU driver of a different calliber to get the job done or use the same drivers as everyone else uses?
2. Does this cut off the memory to pretty much half or less (as in SLI) when on a threaded workload like that like that?
3. Any ideas on why games cannot utilize this?
4. Any current mainstream software support this form of rendering like Adobe?

 

1. Yes, of course. They also don't want people using the software for some reason.

3. I'm not 100% sure about memory, but if the workload is split, im guessing each set will run in SLI (of some sort) and only get the memory of one card, meaning only 50% of the memory of all cards.

3. Games already suck at using SLI, i can't imagine games using this

4. I'll take a shot and say movie editors etc. have "modded" versions of Premiere etc?

[ForsakenLive]

 

So Each card has to be identified seperately & the work load can be split & spread over it by the software , i have a few more question, can you please answer them to the best of your abilities?

1. So does this use like the  GPU driver of a different calliber to get the job done or use the same drivers as everyone else uses?
2. Does this cut off the memory to pretty much half or less (as in SLI) when on a threaded workload like that like that?
3. Any ideas on why games cannot utilize this?
4. Any current mainstream software support this form of rendering like Adobe?

 

 

1- The driver will allow to have extra cards up to some extent, but usually it requires a lot of fiddling to get several GPUs working properly. So you do need extra software.

2- Usually each card can use it's whole vram amount. But it can vary if it's processing the same thing.

3- Even if the API could sopport it (they don't), you would only generate more and more issues. Also you need the game makers to have proper scaling...

4- Most of them do, there are some small addons and patches that can be added for improved functionality.

[TheRetiredSlave]

 

So Each card has to be identified seperately & the work load can be split & spread over it by the software , i have a few more question, can you please answer them to the best of your abilities?

1. So does this use like the  GPU driver of a different calliber to get the job done or use the same drivers as everyone else uses?
2. Does this cut off the memory to pretty much half or less (as in SLI) when on a threaded workload like that like that?
3. Any ideas on why games cannot utilize this?
4. Any current mainstream software support this form of rendering like Adobe?

 

 

Questions 1, 2 and 4 can only really be answered by saying "it depends on the software".. AFAIK, people with those ridiculous 20 R9-290X bitcoin(and similar) mining rigs only needed to install normal drivers and the mining software. Professional CAD tools generally require Quadro//Tesla/Firepro drivers (which of course require the user to have a Quadro, Tesla, or Firepro card).. I'm sure that 3D animation companies like Pixar have plenty of their own software specifically optimized for the hardware that they're using.

If the software is using the different GPUs for different tasks, the GPUs will be able to have different things in their memory which would allow for the VRAM to "stack", but of course, this still all depends on whether or not the software can make an intelligent decision about that.. Last time I played around with password cracking software (Hashcat, which as a funny side note, can support up to 128 GPUs), my VRAM usage only was around 400-500 MB on each of my GPUs, which of course is low enough that it doesn't really matter.

 

Games don't really utilize it because it becomes impractical for the game to communicate with tons of different GPUs. Game performance generally doesn't scale very well above 2 GPUs (just look at some 4-way SLI/CF gaming benchmarks), so creating games that can "use" 8 GPUs is just a waste of time. Most people have 1 or 2 GPU rigs, so it doesn't really make a ton of sense for developers to spend development resources on insane multi-GPU rigs.