• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Bring back "SLI"?

SLI and Crossfire died when they moved the responsibility of optimizing the tech to the game developers. It started with DX12 and its mGPU implementation and then NVIDIA and AMD both stopping driver level profiles for games to recognize multi GPU solutions.

Game developers will not spend resources they don’t have on tech that almost no one uses. They are already overworked enough and crunch time is a serious issue in that profession. If it’s only going to benefit less than 1% of the gaming population they just aren’t going to devote resources to it.

The only way the tech is going to make a comeback is if NVIDIA and AMD start implementing driver level optimizations and profiles for it to work.
 
CraptacularOne said:
SLI and Crossfire died when they moved the responsibility of optimizing the tech to the game developers. It started with DX12 and its mGPU implementation and then NVIDIA and AMD both stopping driver level profiles for games to recognize multi GPU solutions.

Game developers will not spend resources they don’t have on tech that almost no one uses. They are already overworked enough and crunch time is a serious issue in that profession. If it’s only going to benefit less than 1% of the gaming population they just aren’t going to devote resources to it.

The only way the tech is going to make a comeback is if NVIDIA and AMD start implementing driver level optimizations and profiles for it to work.
Click to expand...
This is not SLI....at all. This is using 2 GPUs to do frame gen.
 
1.1.2.3.5... said:
This is not SLI....at all. This is using 2 GPUs to do frame gen.
Click to expand...
I know but I was addressing bringing back SLI that’s in the topic of the thread.

And I far from hate SLI or Crossfire as I was an avid user of both and had SLI or CFX on virtually every generation it was reasonably supported.
 
Krypton said:
While everyone is playing the generational GPU swap game and Nvidia / AMD are raking in great profits with minor incremental updates that always leave you wanting more and a little frustrated...


This guy did a "DIY SLI" with interesting results....


View: https://www.youtube.com/watch?v=PFebYAW6YsM&ab_channel=Lecctron
Click to expand...

I've seen a number of people configure this. It's a painful and nasty process that works for some stuff and not at all for other stuff. Pretty cool, regardless. Just a total PITA to configure.
 
Brackle said:
please no more sli/crossfire. I would rather have 1 single powerful GPU.
Click to expand...
SLI was great when you could combine two sub 100 dollar video cards together and receive top tier performance. I had gaming rigs that could easily handle fluid FPS with maximum settings or nearly maximum settings for a number of years. Problem is, NVIDIA stopped offering SLI on lower end boards and that killed off the value segment. The last sweet setup I had was a pair of GTX970's in SLI and it was impressive. After that I moved to the 1080Ti and agree with you on the singular, powerful, graphics card. However, paying 700 bucks for top tier performance no longer exists.

Paid a Grand for my EVGA 2080Ti Black, and had to RMA that one 3 times before I got one that worked without flaws.

Paid a Grand for my 7900XTX

You can't get a top tier GPU for a 1,000 dollars anymore.

If SLI was easy to support (or this example of frame gen for that matter) I would be all over it. No one (except possibly Intel and that will only remain a value so long as they are the underdog) makes affordable graphics cards anymore. Even AMD's 9070XT's are still pretty salty since you're not getting a top tier card and you can't even get them at MSRP anymore.

The Current State of top tier GPUs is well beyond my reach financially unless I want to starve for several months.

I'm curious to see what Intel unveils in the next couple days. Rumors on the B770 and more.
 
I'd much rather see someone hack the nvidia drivers to get MFG on the 40-Seires cards, which is entirely possible on the new Transformer model as the OFA would not be a "limitation". I'm sure those with 30 series cards would also love to see just normal FG on their cards as well with the new model that no longer uses the OFA. Would really help extend the life of those cards considering how nutty the GPU market is right now.
 
III_Slyflyer_III said:
I'd much rather see someone hack the nvidia drivers to get MFG on the 40-Seires cards, which is entirely possible on the new Transformer model as the OFA would not be a "limitation". I'm sure those with 30 series cards would also love to see just normal FG on their cards as well with the new model that no longer uses the OFA. Would really help extend the life of those cards considering how nutty the GPU market is right now.
Click to expand...
The transformer model is the algorithm they are using for DLSS for a clearer image as opposed to CNN it doesn’t have anything to do with frame generation. As I currently understand it, the Optical Flow Accelerator is still used on the 4000 series cards for frame generation.

Having said that I agree that I would love to see what MFG looks like on other cards since there is no real technical reason it can’t be implemented on them now that NVIDIA has a frame generation tech that runs on the Tensor cores.

I suspect we won’t see anything official for quite some time if at all because then there would be even less of a reason than there is now for anyone to move from a 4000 series card to a 5000 series card. NVIDIA is going to try to hold that as close to their chest as possible for as long as possible.
 
This method really isn't so difficult to set up. I did it using a 6600xt and my 7900xtx. The offloading of upscaling and frame gen to 6600xt enables the 7900xtx to concentrate solely on the renders and really reduces the power used by 7900. I was trying to get it to work for VR but no luck there. Maybe a future update will fix it for VR as that would ideal for my case.
 
Legendary Gamer said:
SLI was great when you could combine two sub 100 dollar video cards together and receive top tier performance. I had gaming rigs that could easily handle fluid FPS with maximum settings or nearly maximum settings for a number of years. Problem is, NVIDIA stopped offering SLI on lower end boards and that killed off the value segment. The last sweet setup I had was a pair of GTX970's in SLI and it was impressive. After that I moved to the 1080Ti and agree with you on the singular, powerful, graphics card. However, paying 700 bucks for top tier performance no longer exists.
Click to expand...
Until you realize that you are at the mercy of Nvidia or AMD to release new profiles for games. It isn't fun having a good SLI system to see 1 GPU utilization at 1-2% while your other card is at 99% ......Then praying there isn't any microstutter.

As someone who has been using SLI since the 3dfx days. I am VERY happy its dead.
 
Brackle said:
Until you realize that you are at the mercy of Nvidia or AMD to release new profiles for games. It isn't fun having a good SLI system to see 1 GPU utilization at 1-2% while your other card is at 99% ......Then praying there isn't any microstutter.

As someone who has been using SLI since the 3dfx days. I am VERY happy its dead.
Click to expand...
I can't argue with that. I have had it since the 3dfx days of scan line interleave as well. Just wish the GPU market wasn't a crapshoot and massively overpriced. I think this past generation was the first I really couldn't afford to upgrade into. Granted, that's on AMD for not releasing a top tier card. Because I refuse to spend more than a grand to play the small amount of games I keep going back to. ... I'm still intermittently playing Sword of the Stars, which debuted in 2006. Got it running stable into hundreds of turns, can't turn on grand menaces... or it crashes around 129 turns. Nothing quite like it on the market.
 
Brackle said:
Until you realize that you are at the mercy of Nvidia or AMD to release new profiles for games. It isn't fun having a good SLI system to see 1 GPU utilization at 1-2% while your other card is at 99% ......Then praying there isn't any microstutter.

As someone who has been using SLI since the 3dfx days. I am VERY happy its dead.
Click to expand...
Similar,

As tempted as I was to get 2 cheaper lower / mid range cards, the support for games, the glitchy issues..I am all for pushing back on game dev's to be the ones responsible to clean up their poorly optimized code to run better vs getting hardware makers to try and do it for them.
 
MrGuvernment said:
Similar,

As tempted as I was to get 2 cheaper lower / mid range cards, the support for games, the glitchy issues..I am all for pushing back on game dev's to be the ones responsible to clean up their poorly optimized code to run better vs getting hardware makers to try and do it for them.
Click to expand...
Problem with that is that you needed to have the same GPU card for sli/crossfire to work. Could be from different vendors, but you needed lets say 2 970 GTX's. You couldn't match a 970 GTX with a 960 GTX.

mGPU did change that, but you would have to rely on the developer. Well shit developers can't even make their games run properly let alone making sure mGPU works lol..
 
Never again SLI, it doesn't fit with Nvidia's philosophy: the more you buy, the more you save.
With SLI you could get more performance for less money than a single high-end GPU, but in the end no one accepted it, it never really caught on in terms of gaming.
 
CraptacularOne said:
The transformer model is the algorithm they are using for DLSS for a clearer image as opposed to CNN it doesn’t have anything to do with frame generation. As I currently understand it, the Optical Flow Accelerator is still used on the 4000 series cards for frame generation.

Having said that I agree that I would love to see what MFG looks like on other cards since there is no real technical reason it can’t be implemented on them now that NVIDIA has a frame generation tech that runs on the Tensor cores.

I suspect we won’t see anything official for quite some time if at all because then there would be even less of a reason than there is now for anyone to move from a 4000 series card to a 5000 series card. NVIDIA is going to try to hold that as close to their chest as possible for as long as possible.
Click to expand...
I do not believe this is correct. I'll have to find the article, but I believe even Nvidia themselves stated that the new transformer model does not use the OFA anymore for frame generation...
 
III_Slyflyer_III said:
I do not believe this is correct. I'll have to find the article, but I believe even Nvidia themselves stated that the new transformer model does not use the OFA anymore for frame generation...
Click to expand...
For the new 5000 series yes they do not use OFA anymore. It’s a model that runs on the tensor cores, however that does not apply to the previous generation. They only stated that they are “exploring” the option to expand this capability to other model series cards.

Edit: just to be clear here: we are talking about frame generation right? You keep referring to the “Transformer model” that’s for DLSS upscaling and not frame generation.
 
Last edited:
There is many issue for this to return and be attractive

) If you want a lot of frame at the cost of latency, so so support and a bit of trouble, frame generation could do it for free
) Lot of temporal based effect, a lot of full screen effects, make for a need for a lot of data to be shared or at least it is really less obvious how to split the work versus the 90s

As for the thing in the video, the cost of interpolating frame by the gpu, slowing it down, seem easier to beef up the part of the gpu doing the interpolation than making a whole different card for it, it is relatively cheap when "well made", going from generating 1 frame to 3 frame on a 5060ti cost 7 ms at max for cyberpunnk at 1440p native, that not a lot versus the time it take to wait for the extra frame. Hogwart legacy with a strong gpu it is 4.5 ms for 3 time the amount of generated frame.


Hardware unboxed with a 5090 (they could be faster at frame generation too):

View: https://youtu.be/B_fGlVqKs1k?t=1288

They go from 34ms (at 78 fps) to 40ms at 2x and 44ms at 4x, a 6ms to 10 ms cost
 
Watched this video and I don't like such clickbait content.
His setup made zero sense.

As for SLI it was like FG but with more issues and you had to pay twice for your GPU and pay for it with electricity bills and ear plugs.
Besides today's GPUs are way too big for SLI to possibly be viable thing. I cannot imagine fitting second 4090 in my PC.

SLI would be cool if it was made to be SFR only and Nvidia worked on SFR.
They found that AFR gives better scaling and went with that - then people who invested in SLI got worse experience than they would get with just single card.
 
Brackle said:
please no more sli/crossfire. I would rather have 1 single powerful GPU.
Click to expand...
The problem with this is that even an RTX 5090 is limited in what it can do. It's still dependent on DLSS and framegen to increase its performance to acceptable or desirable levels in newer games. The idea that one GPU will be enough to max out all current titles and near future titles at high resolution at frame rates good enough for all is a pipe dream. It always has been and probably always will be without a revolutionary step forward in technology. Two of the most powerful GPU's available at any one time will always be more powerful than any single GPU.

The problem is that it really needs to be done at the hardware level so that its not dependent on driver optimizations or game developers to work. Otherwise the technology will continue to be problematic and a bit of a shit show where at least some games are concerned. As stated above, modern GPU's have gotten too big, too hot and too power hungry for this to be viable for most people. That's not to say it can't physically be done today, but its not without massive caveats. Not to mention you'd probably need 1,600watt PSU's to make that work. To do that today would require motherboards and cases be designed more like they were back in the SLi hey day. You might have to run dual PSU's and if you didn't, such a system would be potentially difficult to handle using residential electrical outlets. At that point it would be tough to deal with the heat load in most cases and home environments without going to water cooling.

Two GPU's back in the day pulling 275w or less taking two expansion slots each wasn't a big deal. Dual 600w GPU's that take up four slots each is another matter entirely.
 
One obvious use would be VR when you have to render 2 full viewport, where it could work 100% well and in an easy way, does not even need to be SLI in any way could have VR kit that receive 2 display port input.

SFR sound hard with the amount of full screen effect.

Strange bird said:
With SLI you could get more performance for less money than a single high-end GPU,
Click to expand...
Would still this be the case (if it ever was when we talk about the total performance portfolio and not just the perfect case, i.e. latency, 1% low, handling of very large textures are also form of performance)

Added motherboard needs and mid-end GPU modern price could make that thought, a 5090 is already 2x5080 right now pretty much.

A 5080 is already a bit more than 2 time a 5060ti, both with the massive benefit of monolithic die and memory.

Imo chiplet sound way easier than SLI and they have yet to crack it, will see with the C2C interconnect.... and the moment they cracked it, racking die on the same gpu will probably be the better solution people will do, one lure of SLI was the buy a gpu now, rebuy the same at a low price a year later, thats gone...
 
Last edited:
Strange bird said:
With SLI you could get more performance for less money than a single high-end GPU, but in the end no one accepted it, it never really caught on in terms of gaming.
Click to expand...
Strictly speaking this wasn't necessarily true. The fact is two mid-range GPU's may have been faster than a single high end GPU, but memory bus and total VRAM limitations often caused games to run less smoothly than they do on higher end GPU's despite simple frame rate benchmarks indicating the opposite. People who often talk about his also forget where SLI was often more attractive on the high end. It would allow next generation performance on current generation hardware. It could also leverage graphics options out of reach for even single high end GPU's in games to be available. It made high resolution and multi-monitor resolutions playable where it otherwise wouldn't have been.

In total, I saw more high end SLI GPU configurations than mid-range ones for whatever that's worth. In the early days of the technology I built systems and serviced them professionally and though SLI was rare in general, mid-range SLI seemed even rarer.
 
Last edited:
Brackle said:
Problem with that is that you needed to have the same GPU card for sli/crossfire to work. Could be from different vendors, but you needed lets say 2 970 GTX's. You couldn't match a 970 GTX with a 960 GTX.

mGPU did change that, but you would have to rely on the developer. Well shit developers can't even make their games run properly let alone making sure mGPU works lol..
Click to expand...
I don’t agree the sentiment that developers are “shit” in general. Every dev I spoken too is highly passionate and dedicated to making the best game they can. Almost every game that you see that’s released in a poor state or seems half baked is the result of the higher ups forcing them to release an unfinished unpolished product. It’s almost never the fault of the devs for the state of a release.

It’s unfortunate that the devs themselves are then punished by “shit” decisions made by management and the dev are the first ones that are let go.
 
Dev that speak to people are not necessarily representative and that a passion can go into many ways in term of having a well running game at the end or not.

Is the passion about making a simple game that run well... or high ambition ? You can have a lot of passion without being particularly passionate about writing test and documentation, devs will lie all the time to their superior (that a both way dance that timeline not being met), it is superior jobs to evaluate how much they are being lie too by devs, but it is hard. Of course they would work for pay forever on the game and say it is never finish... that does not mean much.

That said, what is being talked about sound extremelly hard, it is not like Sony Playstation team, AMD or Nvidia pulled it of either and some dev team are just great without any doubt (think the good Call of Duty level one), it would not surprise me if a good proportion of the best developer (optimization wise) are in the gaming industry world
 
Last edited:
No developer wants a game to come out that doesn’t run well just like they don’t want a game to come out that isn’t enjoyable.

I know quite a few dev on various teams and various levels of programming like engine and graphics and none of them want a poorly reviewed product.
 
Bit of a wacky experiment and I have doubts. Once upon a time I also got janky with a cheap second power supply sitting on top of my case just to power HDDs and fans. SO many molex splitters. Ghetto life, baby!
 
Dan_D said:
Strictly speaking this wasn't necessarily true. The fact is two mid-range GPU's may have been faster than a single high end GPU, but memory bus and total VRAM limitations often caused games to run less smoothly than they do on higher end GPU's despite simple frame rate benchmarks indicating the opposite. People who often talk about his also forget where SLI was often more attractive on the high end. It would allow next generation performance on current generation hardware. It could also leverage graphics options out of reach for even single high end GPU's in games to be available. It made high resolution and multimonitor resolutions playable where it otherwise wouldn't have been.

In total, I saw more high end SLI GPU configurations than mid-range ones for whatever that's worth. In the early days of the technology I built systems and serviced them professionally and though SLI was rare in general, mid-range SLI seemed even rarer.
Click to expand...

SLI became less and less beneficial as graphics becamse more and more advanced. Most of the super advanced stuff that make graphics look good does not work at all and can not work at all in SLI.

You cannot do alternate frame rendering because the engine uses data from the previous frame to render the next, you cannot do interlaced or partial frame rendering because information from the surrounding pixels is used.

This is the same reason why DX12 multi GPU never game to fruition. Some developers were super excited for it and tried very very hard to make it work before realizing it's pointless.

There are alternative techniques developers could use to make games run and look better with multiple GPUs but it is not worth the effort at all with how few people have multiple gpus. Even things like the OP posted doing frame gen isn't going to work as well as having an equivalent NVIDIA card do it all.
 
CraptacularOne said:
No developer wants a game to come out that doesn’t run well just like they don’t want a game to come out that isn’t enjoyable.

I know quite a few dev on various teams and various levels of programming like engine and graphics and none of them want a poorly reviewed product.
Click to expand...
And no game producer want a game to come out that does not run well or is not well reviewed (it is even common to give bonus for the game average review score....)

The producer having 100% of the blame (and thus 100% of the credit when it goes well) is removing a lot of responsability-credit to the game dev imo. Is Call of Duty running so well from a PS4/XboxS to a 7900xtx, incredible server with hard to fantom level of computer per 1/100s all about better managers or also a bit because they attract and pay for the best game developer around...
 
Last edited:
sharknice said:
SLI became less and less beneficial as graphics becamse more and more advanced. Most of the super advanced stuff that make graphics look good does not work at all and can not work at all in SLI.

You cannot do alternate frame rendering because the engine uses data from the previous frame to render the next, you cannot do interlaced or partial frame rendering because information from the surrounding pixels is used.

This is the same reason why DX12 multi GPU never game to fruition. Some developers were super excited for it and tried very very hard to make it work before realizing it's pointless.

There are alternative techniques developers could use to make games run and look better with multiple GPUs but it is not worth the effort at all with how few people have multiple gpus. Even things like the OP posted doing frame gen isn't going to work as well as having an equivalent NVIDIA card do it all.
Click to expand...
" Even things like the OP posted doing frame gen isn't going to work as well as having an equivalent NVIDIA card do it all."
Maybe you should test it before making foolish assumptions and posting them on this forum.
 
I seem to recall a product that was slated for release something like 2 years ago that would combine two GPU's at a hardware level, allowing for full access to the 2X Framebuffer and OS would see the GPUs as a single card. Can't find the article. Wonder if someone like Nvidia quietly gobbled up that startup...
 
learners permit said:
" Even things like the OP posted doing frame gen isn't going to work as well as having an equivalent NVIDIA card do it all."
Maybe you should test it before making foolish assumptions and posting them on this forum.
Click to expand...

How about explain how you think I'm wrong instead?

If you watch thee actual OP video he explains how it doesn't work very well or at all except in a couple games, and I can gaurantee the frame gen doesn't work as well as the real frame gen built into games because games with it integrated build it into the rendering pipeline and exclude certain things from using it because they have terrible artifacting.

My assumptions are based on my experience as a developer and learning from developers that specifically work on optimizing games for frame generation. What are yours from?
 
It was neat when it worked and it was horrendous (sometimes even detrimental) when it didn't. I never felt the juice was worth the squeeze considering it almost never gave you 2X performance in spite of requiring 2x the cost. If someone could give us a more seamless experience with true 1-to-1 bang for the buck performance, I'm open to the idea though. If not, I can live with it never coming back.
 
You must log in or register to reply here.
Back
Top