TR Forums

So, there's an impending glut of video cards about to hit the market. What's everyone going to do with all those extra video cards? Is there a chance that if enough people just buy into SLI or CrossFire cheaply that game developers would make it worthwhile?

https://hexus.net/tech/news/graphics/12 ... e-brigade/

I mean, it could, though it depends on the devs these days I think.

The fact that there's API support for multiple GPUs means it's not something Nvidia or AMD can claim as their own thing. It'll be a feature of the game instead of a feature of the GPU. SLI and Crossfire didn't die because the technology went away. It died because the marketing campaign behind it went away.

Redocbew wrote:

The fact that there's API support for multiple GPUs means it's not something Nvidia or AMD can claim as their own thing. It'll be a feature of the game instead of a feature of the GPU. SLI and Crossfire didn't die because the technology went away. It died because the marketing campaign behind it went away.

SLI/Crossfire died because the burden of optimization shifted from the IHV driver team to the game devs. No game development studio wants to expend the resources to optimize the game for 1% of the market (or less).

...and yet the developers of the studio are the ones who have the best chance of getting it right without causing hiccups and stutters in the rendering pipeline. Making SLI/Crossfire work from the driver was always going to be a losing battle.

If it's cool, and people want to do it anyway, then that's almost the definition of a feature driven by marketing instead of technology. If it's not cool, then it's just garbage that gets in the way. The nerd in me still thinks multi-GPU gaming rigs are cool, but thanks to the testing of frame times the engineer in me now thinks they're not cool enough to be worth it.

IIRC, they also died out because testers (including this august website) found that games ran more smoothly with one monster graphics card than with two cards, even if the two card setup theoretically had higher performance. (i.e. a 1080TI will give you better 99th percentile performance than a pair of 1070TIs would, even if the 1070TIs may deliver more peak FPS).

Multi-GPU renderings will come back once the industry gets a handle on the next generation of APIs (DX12, Vulkan) and move away from AFR schemes (they are terrible for frame-timing). Ray-tracing works very well with multi-GPU setups.

Krogoth wrote:

Ray-tracing works very well with multi-GPU setups.

We might even go back to the PhysX days of having separate cards for shaders and rays.

yeah sure if:

1. Consoles go totally-not-crossfire mGPU
2. the totally-not-crossfire optimizations make their way to PC
3. NVidia's fastest cards don't beat these totally-not-crossfire results

In that case Nvidia will push SLI to work for many large games.

I guess alternatively Intel could decide they could spend a few billion dollars on it for their dGPU launch?

(I don't think the consoles would call it crossfire, but... well. It would be crossfire)

Krogoth wrote:

Multi-GPU renderings will come back once the industry gets a handle on the next generation of APIs (DX12, Vulkan) and move away from AFR schemes (they are terrible for frame-timing). Ray-tracing works very well with multi-GPU setups.

Pretty much this.

In addition, most of the engine developers already have SLI support implemented. That gap of marketing came from the transition to low-overhead APIs (DX12, Vulcan), because the engine developers and then game developers have to get it working.

Fortunately it looks like they're pretty much there now.

Further, multi-GPU is for more performance, and while that's not really needed much for today's raster loads and common resolutions, stuff like 4k120+ and VR will absolutely need more; ray tracing is going to be a bear on top of that too.

Expect to see greater pushes at 7nm (10nm Intel) and beyond.