GPU drivers @imaginationtech | Prev. @Qualcomm @TheForge_FX and @Sony India | I make Razix Engine | Opinions are my own.
Hyderabad, India
Joined November 2015
- Tweets953
- Following1K
- Followers162
- Likes4.3K
Pinned Tweet
Wow! unity thanks for giving a free marketing run to other open source engines, this is why I think having a monopoly of engines could be bad especially that indies depend on.
If you're looking for some custom engines you can try my github.com/Pikachuxxxx/Razix Razix Engine. #unity3d
Phani retweeted
Off-screen light not left in the dark: our latest research intros a lightweight model for real-time indirect illumination.
💡 Uses reflective shadow maps to capture off-screen geometry.
🧠 Generalizes scenes never seen during ML training.
More info ⬇️
gpuopen.com/learn/lightweigh…
Phani retweeted
CJP’s Demand Rooster Has Been Updated:
1. Dharmedra Pradhan must resign - ✅
2. Rs 1 Cr to the families of all students who died by suicide - Pending
3. No action against any of the student protesters - Pending
4. Public Apology From RAF & Delhi Police - Pending
Jantar Mantar & Entire Country Is Still Waiting On The Rest Of The Demands.
Phani retweeted
A historic victory for democracy, the people, and the youth of this nation.
Your contribution to this movement, Sonam Wangchuk Sir, has been immense. The country will never forget the role you played in this fight for justice.
Phani retweeted
IT'S A VICTORY OF DEMOCRACY
direct democracy... straight from the streets.
It's a victory of peace, patience & persévérance.
Congratulations CJP, Gen Z of the nation and thank you all citizens for shedding fear and the fear of fear and rising up from every corner of the nation.
FROM ACCOUNTABILITY, NOW TO REFORMS – at Gurgaon, India
Phani retweeted
Today at #siggraph2026, Disney Animation has four more talks! Here's a mini-thread about our four talks today, including three from the Hyperion rendering team about Disney Animation rendering tech:
(1/10)
disneyanimation.com/events/?…
Phani retweeted
There's a new paper out in JCGT that proposes a super simple modification of existing microfacet models to implement a retroreflection model. What's interesting is that despite being insanely simple, the model fits measured data better than you'd think!
jcgt.org/published/0015/01/0…
Phani retweeted
Neural motion synthesis in Threejs/WebGPU,
even runs on my 7y old Iphone!
try it here: motionsynth.sweriko.com
sauce: github.com/sweriko/ai4anim-w…
Phani retweeted
We now have new demos in AI4AnimationPy for stylized biped and quadruped locomotion control. Besides PC you can also run them in your browser or on your phone. Try it out here🎮:
Web-Demo: paulstarke-ai4animationpy.hf…
GitHub: github.com/facebookresearch/…
Video: youtube.com/watch?v=qifmLpNz…
Phani retweeted
🚨 Someone just built a fully open-source mocap system that works with any camera.
It's called FreeMoCap, a markerless 3D tracking system that runs on ordinary webcams. It turns multiple camera feeds into research-grade skeletal data automatically.
100% Open Source.
Phani retweeted
I hacked the engine out of comfyUI's image-to-3d to utilize llama.cpp and now it can run quantized models and this workflow is down to 4g of VRAM.
You're paying $400 a month in meshy credits for something you can run on a GPU from 2018.
Phani retweeted
Implemented Google's TurboQuant paper on Gemma 3 4B with a custom Triton kernel for fused quantized attention.
It's real.
Results on RTX 4090:
2-bit FUSED: character-for-character identical to fp16 baseline. On every prompt. At 16x theoretical compression.
The Triton kernel reads uint8 key indices directly — never materializes fp16 keys. Pre-rotate query once (R is orthogonal so ⟨q, Rᵀ·centroids[idx]⟩ = ⟨R·q, centroids[idx]⟩), then per-position work is just a table lookup + dot.
Speed (avg tok/s across 3 prompts):
→ fp16 baseline: 17.7
→ 4-bit fused: 16.5 (-7%)
→ 2-bit fused: 17.7 (0% — matches baseline)
VRAM (KV cache delta):
→ fp16: 26 MB
→ 4-bit fused: 4 MB
→ 2-bit fused: 7 MB
The paper's theoretical guarantees hold up completely in practice. Zero accuracy loss, zero speed loss, fraction of the memory.
Paper: arxiv.org/abs/2504.19874
Introducing TurboQuant: Our new compression algorithm that reduces LLM key-value cache memory by at least 6x and delivers up to 8x speedup, all with zero accuracy loss, redefining AI efficiency. Read the blog to learn how it achieves these results: goo.gle/4bsq2qI
Phani retweeted
Wine 11 rewrites how Linux runs Windows games at the kernel level, and the speed gains are massive.
NTSYNC support uses a new Linux kernel module to handle Windows-style thread synchronization directly at the kernel level.
Notable benchmark gains (compared to basic upstream Wine without prior optimizations):
- Dirt 3: around 110 FPS to over 860 FPS
- Tiny Tina's Wonderlands: 130 FPS to 360 FPS
- Resident Evil 2: 26 FPS to 77 FPS
- Call of Juarez: around 100 FPS to 224 FPS
Phani retweeted
A Texture Lookup Approach to Bézier Curve Evaluation on the GPU
Muhammad Anas, Alan Wolfe
arxiv.org/abs/2603.15447 [𝚌𝚜.𝙶𝚁]
Phani retweeted
We use main branch development practice. There's no PRs and no feature branches. Everybody pushes multiple small commits every day. Having code review per commit would interrupt people all the time. Code goes behind feature flags (disabled) until it is ready...
github.com/Pikachuxxxx/tanu-…
Hmmm... tanu is idk what it's doing...
P.S.(qwen2.5:0.5B and gemini and Claude trying to be like her...)
176