Broken Silicon 235 (Video): PlayStation 6 AI, Nvidia, AMD Hawk Point, Intel Meteor Lake | Game AI Developer
🕑 Added 2023-12-11 19:41:32 +0000 UTCA Gaming AI Dev joins to discuss what hardware you’ll need to power next gen games!
0:00 Who is our guest? How can interested people get into the AI Industry?
4:38 What is Pygmalion building to change gaming?
10:54 The Next 2D - 3D Moment for Gaming could be Neural Engine AI
20:07 AMD Hawk Point and the Importance of TOPs in APUs
26:53 Intel Meteor Lake’s NPU – Does it matter if it’s weaker than AMD?
31:20 AMD vs Qualcomm Snapdragon Elite X
39:00 Intel's AVX-512 & NPU Adoption Problem with AI...
51:16 Predicting how soon we'll get Next Gen AI in Games
59:00 Can the PS5 run Next Gen AI? …what about the XSS?
1:13:10 How might the PlayStation 6 do AI?
1:24:04 AMD’s Advancing AI Event & ROCm, Nvidia’s AI Advantage
1:45:30 Intel AI – Are they behind? Will RDNA 4 be big for AI?
1:58:27 Will future APUs be as strong as H100? When will the AI bubble pop?
2:10:30 Will AI hurt Gaming long term?
2:28:22 AI Ethics and AI's impact on Artists
$400 RX 6800: https://amzn.to/3uRsLqX
Main Domain (pygmalion.ai is not them): https://pygmalion.chat/
Discord with Active Devs: https://discord.com/invite/pygmalionai
AI engine github: https://github.com/PygmalionAI/aphrodite-engine
Guest’s github (very new): https://github.com/IsaiahGossner
Main github for the project: https://github.com/PygmalionAI
Their hugging face, where actual models are stored: https://huggingface.co/PygmalionAI
https://www.servethehome.com/wp-content/uploads/2023/12/AMD-Instinct-MI300-Launch_Page_50.jpg
https://www.servethehome.com/wp-content/uploads/2023/12/AMD-Instinct-MI300-Launch_Page_49.jpg
Bryan Heemskerk AI Episode: https://youtu.be/NDEka3tBE1g?si=pd6_xNPgMxo7Jltd
https://www.youtube.com/watch?v=mCxHcvtpfAk&ab_channel=Moore%27sLawIsDead
https://en.wikipedia.org/wiki/Von_Neumann_architecture
Comments
capn_hector
@ ~2:03:00 This is true but also Jevons Paradox will kick in, lowering the cost of using this capability and its ready availability will drive much higher total utilization. You will need more TOPS in the future not less, after an initial adoption dip from such kinds of optimizations. Don’t underestimate the overall benefit of building a “solve a computationally intractable problem with a reasonably optimal solution in realtime” unit for your processor. Once the tech is widely deployed there’s going to be enormous uptake in completely unexpected places.
capn_hector
@ ~1:50:00: AMD has disambiguated the MCD from the infinity link PHY, there is no reason they can’t do a 4-PHY or 8-PHY variant of the MCD and scale to enormous sizes. They already are effectively benefiting from a 2x PHY MCD that lets them double the capacity density per wire, vram density has kinda stalled out too, amd engineered around it.