Is NVIDIA GeForce RTX 3050 8GB worth for experimenting real time ray tracing (on the GPU) in Unity?

I am thinkg to buy new GPU and I was wondering if I should buy NVIDIA GeForce RTX 3050 8GB the desktop version which has RTX and DLSS support, or just go with slighlty cheaper one, NVIDIA GeForce GTX 1660 Super which doesn't have RTX and DLSS support?

I am buying mainly for gaming, the later one would suffice, but since I am experimenting with games programming, is it worth spending few more denars and get the RTX enabled card?

Or if I want to experiment with RTX and DLSS those more expensive cards are better, meaning, don't bother with the entry RTX card and get the basic GTX 1660 card?

 

Neither. Save a bit more and at least get a 3060. If not try and see prices for a rx 6600 from amd

 

Neither. 3050 is supposed to be underpowered, and 1660 is quite old. I advise to save more money and buy a better GPU.

 

Neither. In fact I wouldn't buy any graphics cards right now. NVIDIA has a new series that should be launching soon but it's not the new series that is important. It's the existing cards that have not been able to sell. Companies will dump their stock of new cards on eBay for a discount.

For example back when the RTX 20 series launched I was able to pick up a GTX 1080 for $400 which was about $200 below MSRP and $100 below the RTX 2060's MSRP when it finally launched half a year later.

 

Eh, the game has changed, 4000 series could release, sell out, and then increase the price of the 3000 series

 

A few months ago I would have agreed with you, but that was when cryptocurrency was going strong. Worst case even if you are correct what you described takes time to happen. A new series doesn't just come out, sell out, and then everything else becomes more expensive in five minutes. If it does that means the sell out didn't cause it.

 

I wondering how much crypto really played a roll in that, just because ethereum is supposed to be moving away from mining very very soon, so if you were buying a 3000 series card for that, you were pretty much guaranteed to lose money

 

Ethereum isn't the only currency. One of the Discord communities I'm in is largely using NiceHash. I don't remember what it covers but it covers a few different ones.

 

If you want to make any money it is. The next best thing to mine is some unknown S***coin

 

They've been moving away from mining for three years and it literally doesn't matter regardless. The reason the prices are equalizing is because crypto has been in a sustained downward turn since November, with many currencies having dropped in value 60-70% since then.

 

It's definitely more complicated than just "cuz crypto". Console shortages and gaming in general getting way more popular is to blame too. The gaming market started to slow way down as gpu prices started coming down at the same time, nvidia lost 7% off their stock price when they announced that in may, after have a crazy boom for a few years.

From the ceo of nvidia in 2021
"So in order to talk about supply, we first have to discuss the demand. We did have an exceptional overall holiday season. Gaming demand is off the charts. Our overall Ampere architecture and ray tracing are really a true success. This demand has remained stronger for longer. Okay? So, supply does remain tight at this time."

 

I guess a RTX 3050 is fine for gaming. It's not the strongest card out there but for the price it has it's fine.

 

To be honest I've heard that it is possible to actually contact a factory and have a GPU ordered at non-insane price. Currently wondering if it is worth attempting or not.

3050 is listed as "not recommended for VR". Which is one of the reason to get anything higher than that.

 

Hmm ok, I am not intending on experimenting with VR or either mine bit coins. I was just wondering if utilizing the ray tracing power of the GPU is possible in Unity i.e playing with 3D, HDRP to give myself some reason to buy it.

Or those a magnitude more expensive cards with ray tracing are more worth it for developing?

But then, I don't want to spend so much money. Will wait for better times when the ray tracing tech becomes more affordable to develop with.

Judging by the recent comments, nobody knows the answer.

Specifically i just want to play Doom Eternal, needs 3GB VRAM min. to run smoothly.
But I already have GTX 1050 2GB, so kind of not worth buying GTX 1050 TI with 4GB VRAM

SLI no longer a practical solution, but even if it was, would need a motherboard with SLI. Currently I have AMD Crossfire enabled board.

 

Raytracing is useful if you're modeling/rendering in blender, as it is used to speed up Blender's renderer.

When it comes to gaming, I'm of opinion that raytracing is largely a gimmick.

Personally, I'm interested in having larger amount of VRAM, because that would allow me to run more complex neural networks, but then again, I do not do that often.

For example, I wanted to run this experiment:
https://github.com/NVlabs/instant-ngp
, but it requires at least 8GB of VRAM to function.

 

Everything which can make lightmaps obsolete is worth exploring.

 

Yes, but didn't we have a bazillion of realtime gi solutions recently?

After saying "it is a gimmick", I actually went an checked youtube rtx on/of comparison for cyberpunk.

Spoiler

It does make a huge difference in some scenes, but in many scenarios the added effect is nearly imperceptible.

 

If you intentionally use it than fallback can be very different.
Enlighten 4 indirect lighting and raytracing:

https://youtu.be/IAdrwebc2a4?t=49

 

Specifically this video made me think about the RTX enabled one, at 13:03

I just thought its worth investigating in future, but then Linus answers my question
The tech is still in early stages

 

Let's ignore raytracing for the moment. The best reason to choose the RTX 3050 over the GTX 1660 (other than performance) is DLSS. While it's associated with raytracing it's a great feature by itself that enables you to run higher resolutions and/or settings than you might have previously been able to due to budget constraints.

A 4K monitor is not that expensive but a 4K capable card was until they introduced DLSS.

Just be aware that aside from Minecraft the games he chose to showcase it are very poor examples of it and even then he chose poor scenes in Minecraft. If you want to get an idea of what it can bring to a traditional game just look at Cyberpunk which in my opinion is one of the best showcases of it in action.

Spoiler: Cyberpunk

While I'm at it here is Minecraft. If it isn't obvious from the thumbnail the most impressive aspect of Minecraft is not the water but the lighting. If I had to guess he showed the water because lighting would've been a dead giveaway.

Spoiler: Minecraft

Ignoring the problems in his video the conclusion he came to is incorrect. Raytracing as a technology has been in use now for decades. It's not at all in the early stages. It's just that the techniques we use now are likewise not in their early stages.

What he was correct on is the reason why the technology will ultimately be successful. It greatly reduces the load on the developer. Currently advanced lighting involves smoke and mirrors, expensive middle-ware, and many long baking sessions. Raytracing eliminates all of that.

Bakery can take advantage of it too.

https://assetstore.unity.com/packages/tools/level-design/bakery-gpu-lightmapper-122218

 

This actually does not sound right.

The basic raytracing has been around since amiga days (and it couldn't work in real time), but what modern games use isn't that.

Here's amiga raytracing juggler:

Spoiler

And here's realtime raytracing demo from early 2000:

Spoiler

Both use primitives+CSG and rays do not split. Meaning you have hard shadows, no global illumination of any sorts, etc.

This techniques that resemble hardware-accelerated "raytracing" are actually more similar to path tracing:
https://en.wikipedia.org/wiki/Path_tracing

And that one couldn't be made realtime due to insane number of rays it produces. Hence the need of AI denoiser in the modern tech.

 

Pathtracing is a term that I am relatively new to but from my searches on the differences between it and raytracing I'm coming across multiple answers and they're all largely contradictory. Some people say there are no differences, some say one is a more complex form, etc. It's very annoying to try to find information on all of it.

If we want to be technical about it what we're using is a hybrid of tracing and traditional techniques. Regardless in my opinion my statement doesn't change just because the behavior and quantity of the rays changes since either way we're using them to achieve the same end result.

 

Here's the rough idea.

Ray-tracing, original. Scene is created from mathematical boolean primitives. You largely shoot one ray per pixel, it bounces until it hits something. Ray doesn't split. Perfect mirrors are easy, fuzzy reflections are impossible. Default implementation might have no lighting. Number of bounces is limited. It is a homework level assignment, although it is fun for the first time you do it.

When lighting is added, by default it shoots a ray per each nearby light to see if said light is obscured. That results in sharp shadows.

At some point it has been upgraded to support polygonal meshes, but basically that's the sort of renderer we had in 3dsmax version.... 4. From 2000s. Blender internal renderer, when it existed, was also that.

At some point people started tinkering with reflections and glass, and rays started to split into reflective and refractive ray, but that still had nothing to do with what we see in modern "raytraced" games.

At some point, PBR came to life, and people started experimenting with alternative approaches to have better visuals that do not resemble people of "Reboot".

Spoiler

Phong lighting, non-pbr. but with perfect mercury-like reflections reminiscent of T1000

------
...and tried two approaches:
Path Tracing and Photon Mapping. Ther were also hybrids. (Also, does anyone still remember "Radiosity"?)

In both cases you're firing a huge number of somethings that accumulate over many frames and generate resulting image.

Basically, original lighting equation involve approximation of infinite number of rays arriving from random directions and both photon mapping and path tracing pretty much bruteforced that to get an acceptable result. "Infinite rays distributed at random? Let's fire infinite number of rays distributed at random."

And the thing is we still do not have enough horsepower to do that in realtime, because it sometimes takes hours to render one frame. That's why there's ai denoiser, who takes the 1st pass noisy frame full of unfinished details, and make a hazard guess of what it is supposed to be, then hallucinates the result.

And THAT part and this particular combination of techniques is very new.

Because of that I'm of opinion that it is not quite right to say that those are old technologies, and on top of that GI is not what raytracing does. The original tech is about those liquid terminator reflections. Those can be seen in cyberpunk video I posted earlier, and in all honesty, they look fairly bad to this day, because they're too perfect.

By the way, in 2008 there were demos being released called "Quake 4 raytraced".

Take a look:

Spoiler

https://www.q4rt.de/

This is pretty much what normal raytraced engine render normally looks like.

Obviously, it doesn't help that "raytracing" is now being used as a marketing term which muddles its meaning...

 

Does anyone have RX6600 for production in Unity and how does it fare in HDRP and raytracing?

 

I don't know about the RX6600 but I know that a RTX 3060ti can't run the HDRP demo scene at the provided max settings with constant 60+ fps in play mode.

 

What's the rest of your system? What resolution? Which release of Unity? Here are my results in Unity 2022.1.5. I'm on an RTX 3070 which is only around 5 to 10% faster than a 3060 Ti, but I also have an AMD 5950X (16C/32T @ 4.9 GHz) w/ DDR4-3600.

 

i5 12600k, 32GB DDR4 3200 RAM, QHD 2560 x 1440 resolution, Unity 2021 LTS

 

Well there went my hope that Unity had improved performance in a newer release...

 

Looks like the opposite is the case xD. Anyway, i will take some screenshots of all settings + performance when I am back from vacation .

 

Actually raytracing is too often used as a shortcut, that's why all definition are mangled:
- raytracing is just the process of intersecting a line/ray with a primitive. It's a technique to detect occlusion.
- Path tracing is an application of raytracing in which we try to "trace" the "path" from the eyes to light emitters, by recursively generating new rays on hits, base on the rendering equation (which is exponential in the naive case).

Everything else is optimization and fidelity concern (physically based rendering, monte carlo, probability distribution fonction, brssdf, denoiser). But the cycle of using raytracing shortcut to define every application of method (acceleration structure, rays distribution, material reaction) is what mangle things, because people tried to shortcut around distinguishing all these form of the raytracing ecosystem, in their own cultural niche that loosely overlap.

So if you learned raytracing by making a simple path tracer (ie using raytracing techniques), you would use raytracing and be adamant that's called raytracing, which isn't technically false, and that's what everyone in your niche would be using anyway, as there is no point distinguishing the two. But that lead to lack of internal consistency like raytraced shadows which doesn't have bounced rays, which what throw people on a loop.

The reason "raytracing" in game are considered new is that acceleration techniques for real time ARE new, more precisely, it used to be understudied except for a few simple limited demo. Technically what game does is path tracing as it try to resolve the rendering equation through bounced rays (generally delayed temporally using gather as scatter method), BUT path tracing has become a prestige rendering shortcut for high quality rendering, so we say we do not do this. Also we mix so many methods and structures, that it's debatable it's just "raytracing", especially when you add a bit of rasterization and/or raymarching which is truly distinct from ray tracing.

It's best to just talk about Global Illumination rather than raytracing. Pure raytracing techniques (aka path tracing) for game is why people make the distinction, all the other supporting methodology (like accumulating light in proxy structure like voxel, and using propagation to apply the contribution, or rasterizing the primary view rays) dilute the contribution of raytracing as a method. So really the graal of pure path tracing is just a purist dream. IMHO the most likely evolution is that prestige path tracing and Real time GI will just converge into a single technique over time, and the debate will be close.

Now the mangling have been done for so long that even respectable source use both interchangeably, we can't go against language, society and culture. I just remember back in the day of imagina competition, it was a bit more clear what was what. Things evolve hey!