Nvidia Ray Tracing

[Billy Pilgrim]

Just wondering if anyone else had the thought that Nvidia could have created a separate Ray-Tracing card that would calculate rays, kind of like with PhysX, then they could merge them at a future date. This would mean that people uninterested in ray tracing would be able to buy a 20 series card that would be more powerful, with a greater focus on power, rather than a feature with extremely low support. that radically decreases frame rates when in use. This would be a win for Nvidia because they could sell an extra  card for enthusiasts and large companies as well as offer a more compelling reason to upgrade.

[Jurrunio]

Maybe they can't do that due to bandwidth requirements for RT cores to communicate with the rest of the GPU. We just don't know atm. PhyX calculations can be done with CUDA cores instead, so it doesnt have this problem.

[thedonheath]

the problum with that is no one may buy the RT card if they put it in the gpu it will become more common there for making games look better in the long run

[Billy Pilgrim]

That's true, but it just seems that ray tracing is not making RTX sell.

[Billy Pilgrim]

1 minute ago, thedonheath said:

the problum with that is no one may buy the RT card if they put it in the gpu it will become more common there for making games look better in the long run

It's kind of a catch-22. You need consumers with ray tracing for devs to update games for it, but you need devs to update games for RTX to get people to buy.

[thedonheath]

you must respect nvida for taking the risk tho no one may like it but its better for eveyone in the long run im al for new tech

[Billy Pilgrim]

Just now, thedonheath said:

you must respect nvida for taking the risk tho no one may like it but its better for eveyone in the long run im al for new tech

Yeah, I guess that's true, though maybe they introduced RTX too early. Maybe it was partly for the 20 naming scheme?

[Arika]

1 minute ago, Billy Pilgrim said:

Yeah, I guess that's true, though maybe they introduced RTX too early. Maybe it was partly for the 20 naming scheme?

not too early, the hardware needs to be there before the software, they just released it at too high of a price

[thedonheath]

1 minute ago, Billy Pilgrim said:

Yeah, I guess that's true, though maybe they introduced RTX too early. Maybe it was partly for the 20 naming scheme?

peaple moan but still buy them haha and ofcourse the 3000 will do it better and so on plus the is ouher cool stuff the rtx brings with it

remeber how long tessellation toke now nealy evey game uses it

[thedonheath]

3 minutes ago, Arika S said:

not too early, the hardware needs to be there before the software, they just released it at too high of a price

this is true they toke the piss with the price

[schwellmo92]

16 minutes ago, Jurrunio said:

Maybe they can't do that due to bandwidth requirements for RT cores to communicate with the rest of the GPU. We just don't know atm. PhyX calculations can be done with CUDA cores instead, so it doesnt have this problem.

I doubt it very much, NVLINK can do 100GB/s.

 

I think the real reason is if they don't push it out so aggressively it wont get the market penetration and then adoption it needs.

[Jurrunio]

2 minutes ago, schwellmo92 said:

I doubt it very much, NVLINK can do 100GB/s.

 

I think the real reason is if they don't push it out so aggressively it wont get the market penetration and then adoption it needs.

100GB/s is nothing compared to how fast data can travel within the same GPU die. Also latency, just milliseconds there can mean everything.

[schwellmo92]

2 minutes ago, Jurrunio said:

100GB/s is nothing compared to how fast data can travel within the same GPU die. Also latency, just milliseconds there can mean everything.

People who professionally render cgi (incl ray tracing) normally use 2-4 GPU's and it scales almost perfectly, so I think it would've been fine to offload ray tracing to an add-in card.

[trag1c]

6 minutes ago, Billy Pilgrim said:

Just wondering if anyone else had the thought that Nvidia could have created a separate Ray-Tracing card that would calculate rays, kind of like with PhysX, then they could merge them at a future date. This would mean that people uninterested in ray tracing would be able to buy a 20 series card that would be more powerful, with a greater focus on power, rather than a feature with extremely low support. that radically decreases frame rates when in use. This would be a win for Nvidia because they could sell an extra  card for enthusiasts and large companies as well as offer a more compelling reason to upgrade.

Besides being a horrible business decision it would be a massive pain in the ass for the dev to have another card to deal with. You have to remember that the scene is only partially ray traced so the scene is going to be rendered before hand then essentially post processed by RTX then you have your traditional post processing after that. Because the scene is not preexisting on the RTX card it has to be rendered first then sent to the rtx card then sent back to the GPU. The memory allocations and copies associated with moving that data around is going to slow everything down big time. Memory allocations and copies are hands down the most expensive operation you can perform for a real time application.

 

1 minute ago, schwellmo92 said:

People who professionally render cgi (incl ray tracing) normally use 2-4 GPU's and it scales almost perfectly, so I think it would've been fine to offload ray tracing to an add-in card.

 

5 minutes ago, schwellmo92 said:

I doubt it very much, NVLINK can do 100GB/s.

 

I think the real reason is if they don't push it out so aggressively it wont get the market penetration and then adoption it needs.

This isn't a problem for SLI because there really isn't much for card communication as they each have a copy of the data and work independently of each other so the only time they transfer data is when memory allocation occurs in which typically is what your loading screen is for. Or you slowly stream data to the GPU each frame but you allocated the memory block before hand.

 

As far as the ray tracing with software such as Blender or Maya they tile the scene so each card gets a tile at a time, the data set can be split among the cards, and the cards are completely independent of each other. The others cards do not need to know about each other where as the RTX standalone would have a massive circular dependency with the GPU which means lots of communication and dependencies which means frame time budgets shrink by a lot.

[Jurrunio]

11 minutes ago, schwellmo92 said:

People who professionally render cgi (incl ray tracing) normally use 2-4 GPU's and it scales almost perfectly, so I think it would've been fine to offload ray tracing to an add-in card.

just because it scales without external RT cores involved doesnt mean it works...

[N1NJ4W4RR10R]

It would not surprise me at all of something like this happened in the future. But for the time being, it needs to be adopted and optimised software side before Nvidia/AMD can/will look to dedicated cards for ray tracing alone.

[Billy Pilgrim]

1 hour ago, N1NJ4W4RR10R said:

It would not surprise me at all of something like this happened in the future. But for the time being, it needs to be adopted and optimised software side before Nvidia/AMD can/will look to dedicated cards for ray tracing alone.

I would think that if this were to happen they would start of as two separate cards and then merge, as the industry tends towards integration. For example, vram was, at one point upgradable and CPU cache was external.

[WereCat]

3 hours ago, schwellmo92 said:

People who professionally render cgi (incl ray tracing) normally use 2-4 GPU's and it scales almost perfectly, so I think it would've been fine to offload ray tracing to an add-in card.

Thats different. 

There is a huge difference between rendering a scene and waiting till it finishes and rendering a scene in real time with which the player interacts with all the time.

[N1NJ4W4RR10R]

4 hours ago, Billy Pilgrim said:

I would think that if this were to happen they would start of as two separate cards and then merge, as the industry tends towards integration. For example, vram was, at one point upgradable and CPU cache was external.

It's a relatively nique sounding thing though, and likely wouldn't be picked up if sold as a seperate unit (at least as much) while software side it's unoptimised as hell.

 

Add onto that I'm sure Nvidia wanted to guarantee their profits, I'd bet they'll only develop dedicated hardware once they know it's worthwhile.

[Sahl]

6 hours ago, Billy Pilgrim said:

Just wondering if anyone else had the thought that Nvidia could have created a separate Ray-Tracing card that would calculate rays, kind of like with PhysX, then they could merge them at a future date. This would mean that people uninterested in ray tracing would be able to buy a 20 series card that would be more powerful, with a greater focus on power, rather than a feature with extremely low support. that radically decreases frame rates when in use. This would be a win for Nvidia because they could sell an extra  card for enthusiasts and large companies as well as offer a more compelling reason to upgrade.

Imagine if we applied this to all new tech, then we would not have SSD, USB, etc.. "to name so newer tech" because people would instead go for full performance focused hardware. Personally I do not have a problem with the new RTX cards, people have the choice to buy GTX cards if they do not want to invest in the new tech. However I am certain that once Ray Tracing is used properly by developers and have been on the market abit longer ti will have alot better performance, but the real exiting thing for me is the DLSS, not Ray Tracing.

 

People also expect to much performance increase, like unrealistic in my opinion. I see people demand 50% improvement for the same price point every year and in my honest opinion that is not going to happen and one might argue have never really happened unless you upgrade your hardware constantly. It would be counter productive for Nvidia to just have upped the performance abit more, but without new tech and even more so when we think about what AMD might do in half a year or so.. I personally think AMD will dominate the budget and avg gamer GPU market during next year. Would I use Ray Tracing in BFV ? most likely not once I get my new rig build, because I want FPS.. Will I put all settings on ultra? no.. why would I ? the gain from that is almost nothing. Now once a few of the none FPS games that supports Ray Tracing comes out next year, I think we will see much better performance and a real difference. This might actually make people want to upgrade to a RTX card.

 

 

[Billy Pilgrim]

7 hours ago, Sahl said:

but the real exiting thing for me is the DLSS, not Ray Tracing.

What exactly is DLSS?

[Sahl]

1 hour ago, Billy Pilgrim said:

What exactly is DLSS?

deep learning super-sampling

 

Or do you actually want me to make a technical post about it ? if so I rather link to the various articles describing it much better than I could ever do.

[Billy Pilgrim]

1 hour ago, Sahl said:

Or do you actually want me to make a technical post about it ? if so I rather link to the various articles describing it much better than I could ever do.

That would be useful. Thanks