Jump to content

Post-GDC Wrapup



I'm back home now, relaxing after a long week away. And what a week it was!

I got my lecture out of the way on Monday. I felt like I was pretty well prepared, but some acoustical problems in the hall really made it very difficult to concentrate, plus I was pretty nervous about talking in front of about 600 people. I did it though, and hope to do another one next year. The recorded lecture should be available in the GDC vault.

During my talk, I described my experience as a PC developer, moving into mobile, and making a renderer to be cross-platform and scalable. I also showed off a deferred renderer running on the iPad. A couple of interesting things came out of this research. FIrst, I figured out a few details on how to render to a texture. This forms the basis of project shadows and post-effects on mobile. More importantly, I discovered that mobile hardware is better than PC hardware at rendering to textures! This is what the performance of a deferred and forward renderer looks like on the PC:


A deferred renderer scales better, but takes an initial hit on performance, which is why Leadwerks 2 has such heavy hardware requirements, and the framerate never goes above 300 or so. It turns out, PowerVR graphics hardware for mobile can render to a texture at no cost. We get the scalability of a deferred renderer, without that initial hit on performance:


I talked to some engineers with Mali, and they told me their graphics chips work the same way. The implications of this are pretty staggering. It means that per unit processing horsepower, mobile graphics hardware is actually more capable than PC graphics hardware, and AAA graphics on mobile are right around the corner.

We set up our booth on Tuesday before the show. This was my first time going behind the scenes, so it was a totally different view of the conference. Before the expo floor opened, I asked the Oculus Rift guys if I could try their new toy out, and they let me demo it. I was only disappointed that they used a Mech-style game and the cockpit took up most of my peripheral vision, which eliminated any reason to look around and experience the full VR effect.

I had a lot of meetings with different graphics hardware manufacturers, got some good technical info, and scored some free hardware for development and testing.

The response we got from the attendees was really good. When we explained what Leadwerks 3 was, everyone got it right away, and were enthusiastic about what they saw. Here's some footage from the show:

It was a push to get Leadwerks 3 out in time for the GDC, but well worth it. We got the word out about Leadwerks 3, got a lot of positive feedback and a better idea of what our target market needs, and put Leadwerks on the radar of some very big companies in the industry.

Starting Monday, bug fixes have top priority, and I will start adding some graphical enhancements that will make a big enhancement to the renderer.

  • Like 1


Recommended Comments

A pity I can't access the vault without having to pay 500 dollar for a membership.


Seems like the GDC was a good choice. Looking forward to the future development of leadwerks.

Share this comment

Link to comment
Guest Red Ocktober


where can we take a look at the lecture...



Share this comment

Link to comment

This is pretty cool

Looking forward to pain my iOS and Android devices a bit with this :)




But actually I think the conclussion is not complete.


PowerVR and Mali end up like that because they have tile based multi gpus. The 3 GPUs of the iPad4 for example render tile by tile independently (grid of Y x Y tiles that make up the screen), so if 2/3 of the screen actually does not require power (as in the deferred rendering movie), they can focus all their power on the remaining 1/3. With deferred this can be great as it can achieve a lot of quality with a design focused to make use of it (Real Racing 3 is a prime example), yet on the other hand it can also be absolutely devastating if too much work ends in a single tile as all other gpus will simply run out of work and enter idle, tiles can not be subdivided to share the workload etc.

Mali and PowerVR also have unified shader units which help them a lot.

Also their early z-out pipeline is much stronger, doing z discards before processing shaders for a pixel at all (thats the reason why alpha mask is much more expensive than alpha blend on them vs desktop where its the other way round)



But thats also what makes this technology troublesome potentially at least for Android devs due to Tegra.

Neither is Tegra tile based, nor does Tegra even offer unified shader units (Tegra4 changes the later, but only partially. And it will still not be tile based)

Tegra 2 / 3 basically uses a desktop gpu back from GF6-7 days neutred enough to run on the mobile energy constraints, while avoiding to use all those expected and smart optimizations to make it use less power or offer more graphical fidelity at the same power draw.

Share this comment

Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Add a comment...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Create New...