PCPer Mailbag #57 - How Many Rays Would a Ray Tracer Trace If a Ray Tracer Could Trace Rays?

Subject: Editorial | September 7, 2018 - 10:03 PM |
Tagged: video, Ryan Shrout, pcper mailbag

It's time for the PCPer Mailbag, our weekly show where Ryan and the team answer your questions about the tech industry, the latest and greatest GPUs, the process of running a tech review website, and more!

On today's show:

00:51 - NVIDIA 7nm GPU plans?
03:57 - NVIDIA vs. AMD for older CPUs?
06:33 - Threadripper workstation boards?
08:43 - Jensen Huang?
10:19 - Ray tracing performance drop?
12:50 - Ray tracing quality vs. performance slider?
15:10 - Updates for RTX AI breaking games?
17:13 - PCPer interviews with game developers?

Want to have your question answered on a future Mailbag? Leave a comment on this post or in the YouTube comments for the latest video. Check out new Mailbag videos each Friday!

Be sure to subscribe to our YouTube Channel to make sure you never miss our weekly reviews and podcasts, and please consider supporting PC Perspective via Patreon to help us keep videos like our weekly mailbag coming!

Source: YouTube

September 8, 2018 | 04:22 AM - Posted by Anony mouse (not verified)

Did you guys enable RTX on your videos? Ryans forehead and nose are shiny

September 9, 2018 | 01:43 PM - Posted by ThemsSomeNewRenderPassPipelinesForSure (not verified)

Nvidia better have some Extra ROPs available on their RTX series, or some more effcient ROPs on Turing! That and more efficient Shaders and higher clocks. Because otherwise they will not be able to show much improvement on any current/older games that are not tuned to make use of the RTX Ray Tracing cores and that New Tensor Cores based denoising AI that's going to allow Nvidia's Ray Tracing output to be useful for gaming.

Folks need to be do the math and divide that Ray Tracing cores GigaRays/second calculation metric down and have that devided by 1000(1000 milliseconds per second) and then take that Rays per millisecond figure multiplied by the Frame Time at aorund 16.67 milliseconds per frame at 60 FPS, and 33.33 milliseconds at 30 FPS.

So that's 8 GigaRays for the RTX 2080 divided by 1000 that eguals 8 million rays per millisecond multiplied by the number of milliseconds per frame. So for a 60 FPS Frame Rate that's 8 million ray calculations per millisecond times 16.67(Milleseconds Frame Rate) for a total of 133.36 million Rays calculations Available Per a single 16.67 milliseconds frame time interval. And those 133.36 million Ray Calculations per 16.67 milliseconds have to be fruther divided by the number of distinct Completed Ray Paths traced and that's a variable amount also depending on each ray's fully traced path through the scene. Some Rays are going to be reflected and refracted as they bounce around, off, and through many objects/object-parts(opaque, semi-transparent, and transparent) and with each interaction they will pick up some of the object's materal value/s along the way including any refractive index values and other values also, throw in the global illumination/other values also.

So the amount of actually completed Ray Paths is going to be much smaller than 133.36 million Ray Calculations per 16.67 milliseconds can accomplish and that output will be very grainy if Nvidia trys to randomly distribute the rays output across the entire scene more evenly. The special sauce is definitely the AI/Tensor Cores and their Denoising AI algorithm that's also going to be constrained by that 16.67 frame time for any 60 FPS gaming, for example.

Nvidia/Games makers will need to tweak the games to only ray trace the needed parts of the scene/objects and maybe also allow for enough Rays available for any global Light(illumination) and Shadow Casting done via Ray Tracing that may also have to be limited to only the closer to the view port meshes. So Limit the Rays To Only the needed Refraction, Reflection and Shadow calculations nearest to the view port. So that's a very Limited number of Rays for Ambient Occlusion and Subsurface scattering using Ray Tracing calculations in addition to Global light/shadow Ray casting calculations.

It's all much more complicated than that with each frame taking a variable amount of millisecinds of hopefully between 16.67 and 33.33 milleseconds for average game play with faster than 60 FPS frames rates getting even less than 16.67 milleseconds of total ray casting calculation and denoising time.

Really if the Imagination Technologies Folks where able to pull that off with their mobile PowerVR Wizard Ray Tracing IP then Nvidia should have no problems tweaking the game titles to make a more judicious usage of a limited amount of total Ray Paths calculated per 33.33 to 16.67 Milleseconds frame times gaming between 30FPS and 60FPS gaming. That's also going to require some methods to keep the frame rate variance to a minimum also.

Nvidia's Ray Tracing and AI denoising Piplines and Raster Operations Pipline will have to be ordered properly in order to mix down all this to the final frame buffer. So there will be less time than that, for example, 16.67 Milleseconds frame time depending on what operations can be done in parallel and what operations have to be done sequentially. There will be render passes that can be done in parallel mostly and just where the AI denoising is done will have to wait for the whitepapers to arrive that have been available for the developers for some time.

If Nvidia can pull this all off even with some early in the cycle problems then they have my respect for even trying to get the Limited Ray Tracing in the Games IP Race started on a wider scale. Props go out to the Imagination Technologies PowerVR Wizard Ray Tracing IP hardware/software engineers also.

September 11, 2018 | 03:42 PM - Posted by Paul A. Mitchell (not verified)

TO: PC Mailbag

Is anyone else unable to Map Network Drive
after updating Windows 10 to version 1803?
Did Microsoft disable HOMEGROUP without
sufficient testing? HELP!!

September 11, 2018 | 06:13 PM - Posted by Jeremy Hellstrom

Paul, you're not running Home Edition or something are you?  Workgroup still works on my Prof. Ed.

September 12, 2018 | 02:52 AM - Posted by Jim Tanous

Both existing and new mapped network drives are working on my end with Windows 10 Pro 1803 (17134.228).

Is the problem limited to actually mapping the drive, and can you connect to it manually?

September 11, 2018 | 05:22 PM - Posted by Paul A. Mitchell (not verified)

Re: TR workstation motherboards

September 14, 2018 | 12:22 PM - Posted by Dbsseven

Do you expect the next gen AMD products to have the equivalent of RT cores? Between the DirectX RayTracing announcement and AMD building the Xbox APUs, it seems like they had to know this was coming.

September 16, 2018 | 11:03 AM - Posted by Anonymous234345 (not verified)

On Z370, would there be a tangible difference between 2xSATA SSDs in chipset RAID0 or a single NVME drive? Looking at general windows use and game loading, so your client QD benchmarks may be most appropriate.

In my region 970 EVOs are 2x the $/GB of 860 EVOs, so a SATA configuration would have double the capacity for the same price. If the real-world performance is comparable, the SATA configuration may be preferred.

September 20, 2018 | 08:18 PM - Posted by c0c0c0 (not verified)

How do the new cards do with Premiere Pro and DaVinci Resolve? I didn't see those results in the review. The only games I care about are minesweeper and spider solitaire.

September 21, 2018 | 11:10 PM - Posted by Dbsseven

Do you think game developers will eventually be able to build/distribute DLSS models themselves? Everything I’ve read says nVidia will be training and releasing the DLSS models. Even if nVidia does this for free, I wonder if this might hurt everyone but AAA game developers.

September 25, 2018 | 11:05 AM - Posted by Jonas (not verified)

To the mailbag.

Has there been any updates on the Freesync/HDR demo software that AMD revealed on PC Perspective back in June?

September 26, 2018 | 01:57 PM - Posted by CB

Do the upcoming tariffs on things like video cards apply to the retail price or the manufacturing cost? Considering how high the markup is on many consumer electronics, the difference in the actual tariff amount could be quite large.

October 8, 2018 | 04:59 PM - Posted by Dark_wizzie

Does higher L3 per core increase single thread perf? Does this mean i7 8700k has faster single thread than i5 8600k if freq is same and HT is off?

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

This question is for testing whether you are a human visitor and to prevent automated spam submissions.