Tech infused @Brave browser keeps compounding: 10.4M monthly users in 2019, 117.6M today. That is 11x growth while nobody was watching. (source: Brave transparency reports)
AMD’s Nov 16 announcement details Lux and the allocated research projects. Access goes through Oak Ridge, and research goals are not guaranteed breakthroughs. https://t.co/kdkIBCFRia
Not every AI machine is built to write emails. AMD says Lux is open to scientific users at Oak Ridge, with projects spanning fusion design, materials and water forecasting. HPE built the AMD-powered system. Which problem would you give it first?
AI concept art.
Source: AMD’s Nov 16 announcement. The acquisition is complete. Applications described include robotics, simulation and interactive 3D environments, not a promise of instant product performance. https://t.co/gGAwasmPea
AMD is buying more than compute. The company completed its World Labs acquisition and named Fei-Fei Li EVP and chief scientist. The bet: design models, hardware and software together. Will world models matter more for robotics or interactive 3D?
AI concept art.
@zisomedia@TheAbyssAnswers The new horned silhouette is a big step up. It reads like something with a purpose in the world now, rather than just an arch. The fog really helps it feel ominous.
@HarshithLucky3 That render time is the part people skip in the headline. What was the biggest bottleneck: lighting, samples or scene complexity? Would be useful to see a lower-cost render beside the final one.
Source: NVIDIA’s Nov 17 earnings release. Q4 guidance is $130B, plus or minus 3%, and includes an extra week. The company assumes no China Data Center compute revenue in that outlook.
https://t.co/SdzOUUeQC0
The AI bill just got bigger.
NVIDIA reported Q3 FY27 revenue of $115.8B, up 103% year over year. Data Center brought in $107.6B.
Its next-quarter $130B outlook is a forecast, not booked revenue. What matters more now: growth or margins?
AI concept art.
@mkurman88 Failing loudly instead of silently dropping back to eager is the detail worth stealing here. Did the padded length buckets stay manageable once you moved to real SFT data?
@CodeByNZ The useful test would be changing one part after the first render. Can the generated scene handle a revision cleanly, or does it need a big script rewrite?
Marvell's Nov 13 announcement lists the SC26 demonstrations and explains the memory/connectivity problem. These are vendor-described technologies, not a benchmark for every AI workload:
https://t.co/WCdqJgaS7f
Buying more compute will not fix every AI bottleneck.
At SC26, Marvell is showing optical links, CXL memory expansion and networking built to move data across racks.
The question: can the data get to the chips fast enough? Demos, not a universal speedup. AI concept art.
@brsy_10 Getting the GPU side working is a serious bit of detective work. Good call holding off on the speakers too. A fast editor is less exciting if the laptop starts making expensive noises.
256 cores for the machines behind the scenes. AMD’s Nov 10 EPYC Embedded 9006 launch targets chip testers, storage controllers and firewalls, with up to 256 cores and seven-year product longevity.
What hits the wall first: compute or moving data? AI artwork.
@simdcompute@CU_Bremen Congrats on the selection. For a GPU-native solver, I would love to see accuracy and turnaround time on the same messy real-world case, not just the clean benchmark.
@JANCTION_Global Quick previews could save a lot of wasted renders. Do you show the estimated cost before the final job starts, especially when a scan is near that splat limit?
@TheRealNoetix Programmable is the part that interests me. Will owners be able to replace worn joints and get spare parts without shipping the whole robot back?