Founder - HAVEN8, Spatial computing, VR, AR CEO - RiseXI Digital
Texas, USA
Joined February 2009
- Tweets13.1K
- Following380
- Followers1.4K
- Likes6.9K
Positano, Italy - Almalfi Coast - Apple Vision Pro
Another 2 million triangles added to the scene… this M2 is GETTING FRIED!
I’m starting to find out that the m2 GPU is very good at geometry. I can probably run 10-12M triangles without a problem… it’s the shaders that hit bottlenecks really fast.
Dexter Talbert - HAVEN8 ᯅ retweeted
Saturday project: turn the coffee table into a jungle. Before and after, made in Create.
apps.apple.com/app/id6825845…
Dexter Talbert - HAVEN8 ᯅ retweeted
Thinking of starting a community for people who want to learn hardware, work in hardware, or just enjoy tinkering with things in general.
Reply to the tweet if you want in.
Positano, Italy - Almalfi Coast - Apple Vision Pro…..
Sheesh it’s been a long day. 5.5 MILLION TRIANGLES running at full resolution….
I’ll give some more updates tomorrow I’m tired 😪
I’ll try my best to get this out before Thanksgiving so you all can wow your families with your Vision Pro 🥱🤪
Positano Italy - Apple Vision Pro
So far most geometry is done before we start optimizing.
Most texture work is complete.
The foliage isn’t really great yet so working on that now.
Afterwards we’ll make some cuts to get framerate up to 70-80fps
Alright we’re in round 2 for Positano Italy on Apple Vision Pro. Excuse the low res textures, clipping, and stuttering.. just dropped in about 5M triangles of buildings and mountains and terrain to see how much the m2 can handle.
Hopefully by EOD we will have all the models cleaned up and layered with hi-res textures, shadows, and lighting.
It’s going to take some very clever deferred rendering to pull this off. Apple did a hell of a job with the Environments they built. Granted, their environments are smaller than this one… but still.
Dexter Talbert - HAVEN8 ᯅ retweeted
I've taken matters into my own hands with @Apple's Metal. 🔥
I'm building my own custom Metal kernel for TensorFold, starting with 4-bit inference on M5.
It'll set me back a little while I get this built and tested properly, but what I've achieved locally with (mlx_tf) already feels like a real breakthrough.
The M3 Ultra has 819 GB/s of memory bandwidth. The M5 Ultra has 1.2 TB/s. I look at that hardware and think there's still a lot more decode performance to unlock.
Why should the Sparks have all the fun... Watch this space 👀
Here’s today’s latest update on Positano Italy environment on Apple Vision Pro…
Lots of progress.. we have boats and yachts in the water, about 40 seagulls flying, reworked water physics, and reworked sand. We put a dock in the water and currently working on the city that’s not in view.
This scene is currently running at about 65fps at maximum resolution, and about 85fps with adaptive resolution on. 4X multisample AA.
This is all running on the M2
