Tseng Wen-Wei
1.2K posts


@tor1457174 @TeksEdge Yeah, at this low memory bandwidth it really struggles to run dense models. (our lab has a DGX spark and we met this bitter truth)
English

@TeksEdge @sosiristseng I have the 395 version, with LPDDR5 8000MT/s.
It's really unusable for anything really serious -- maybe for a toy run?
So I bought another PC with two 7900 GPU -- this is something really usable, though it only have 64GB VRAM.
English

🆕 Localmaxxers News: Gorgon Halo (Ryzen AI MAX 400) Latest Updates!
AMD’s next high-end APU refresh is taking shape.
✅ Confirmed from Recent Leaks:
• Ryzen AI Max+ PRO 495 flagship spotted in PassMark
• 16 Zen 5 cores / 32 threads
• Up to 5.2 GHz CPU boost
• Radeon 8065S iGPU (40 CUs, up to 3.0 GHz)
• 192GB unified LPDDR5X memory support (major upgrade from 128GB)
🔑 Key Improvements:
• CPU & GPU clocks than current Strix Halo
• Significantly higher memory capacity for AI & heavy workloads
• Still based on Zen 5 + RDNA 3.5 (not a full new generation)
Expected Launch: Later 2026 or early 2027
This looks like a focused mid-cycle refresh with memory and efficiency as the main priorities. More details expected soon. 🔥

English

@TeksEdge Thanks! So it's on par with the DGX Spark (also 273 GB/s)
English

@sosiristseng Just a little with new LPDDR5-8533 up from 8000 👉 273 GB/s a ~6.6% uplift.
English

@alphaomegavan @HaruhiismOtaku Not true i guess. Black people aging very well too and i don't think we avoid the sun lol
English


@ChibiReviews The first lady of France, Brigitte Macron, is 25 years older than Emmanuel Macron.
English

@hgds0 "In the grim darkness of the far future, there is only war."
English

@TeksEdge Some comparisons there
phoronix.com/review/intel-a…
English

Will Performance or Value win the home inferencing game? 🔥
Intel Arc Pro B70 drops 32GB VRAM for ~$950
NVIDIA wants 2-3x more for similar memory 💸
B50 gives 16GB for $349 beating entry-level pro cards on memory-per-dollar.
Local AI (LLMs, image/video gen) is hungry for VRAM.
Intel bets accessibility + multi-GPU scaling wins. NVIDIA bets CUDA ecosystem.
Raw speed 🏎️ or Smart value 📈?

English






























