Absence grafické karty Intel Arc B770 vzbuzuje otázky, jak moc vážně to Intel myslí s Arc GPU. Ten ujišťuje, že jsou velmi důležité. #intelarc
34
18h
Tried the new Qwopus3.6-27B-Coder on the dual Arc Pro B70 setup.
After getting the 36B MoE running at a respectable 20.7 t/s, I had high hopes for this 27B dense coder model.
Result: 4.13 tokens per second.
Not the outcome I was hoping for.
Sometimes the hardware just doesn't want to play nice.
#IntelArc #LocalLLM #Battlemage
64
A gaming handheld chip made in the USA 🎮 Intel Arc G3 runs on 18A — the most advanced US-made logic node. XeSS 3, ray tracing, Wi-Fi 7, all-day battery. The handheld wars just got interesting. #Intel #PCGaming #IntelArc #Semiconductors
1
26
Intel Arc G3世代の携帯ゲーミングPCが、いよいよ本格的に気になる段階に入ってきました💻🎮
Acer Predator Atlas 8とMSI Claw 8 EX AI は、どちらも8インチ級・120Hz系のWindows携帯ゲーミングPCとして注目度高め。
ただ、携帯ゲーミングPCはスペックだけで決めると、価格・重さ・発熱・バッテリー・ドライバー・よく遊ぶゲームの相性で「思ってたのと違う…」が起きがちです。
みはり的には、国内価格、実機レビュー、保証、ストレージ交換、アンチチート対応まで見て判断したいところ。
#携帯ゲーミングPC #Acer #MSI #IntelArc #ガジェット
詳しくはブログでチェック👇
miharimemo.hatenablog.com/en…
174
Jun 12
#Sextou! Que o #finaldesemana seja belo como essa #IntelArc B580!
Converse com quem está #Desde1982 construindo Workstations no Brasil
Sinco.net/ZAP
🔋
19
Jun 12
Intel just entered the handheld gaming arena 🎮 The Arc G3 is purpose-built for handhelds — ray tracing, XeSS 3 AI upscaling, Thunderbolt 4. Acer, MSI & OneXPlayer already on board. PC gaming in your pocket just leveled up 👀 #Intel #IntelArc #PCGaming #Semiconductors
ALT Intel Arc G3 processor badge and family lineup for handheld gaming devices
51
Jun 12
Just got dual Intel Arc Pro B70 working properly.
vLLM on Intel’s own llm-scaler images kept dying during TP=2 init with the classic shm_broadcast cancelled error.
Switched to llama.cpp SYCL and it just worked.
Qwen3.6-35B-A3B (36B MoE) on 2× B70:
- Q4_K_M (20.7 GB loaded)
- Prompt eval (512): 277 t/s
- Generation: 20.7 t/s
No crashes. Full offload. Stable.
llama.cpp SYCL is currently the most reliable way to run tensor-parallel workloads on dual Battlemage cards.
Intel still has work to do on their vLLM XPU stack.
#IntelArc #Battlemage #LocalLLM #llama.cpp
6
8
162
Jun 12
Hermes Agent for the win with setup on this one..
After fighting with vLLM TP=2 instability on Intel’s own llm-scaler-vllm images (still stuck on 0.14.x), I switched to llama.cpp with SYCL.
Setup:
- 2× Intel Arc Pro B70 (64 GB total)
- Qwen2.5-7B-Instruct Q4_K_M
- Full offload (-ngl 99)
Results:
- Prompt eval (512): 655 t/s
- Token generation: 10.5 t/s
No more shm_broadcast crashes. llama.cpp SYCL handled dual discrete Battlemage cards cleanly where the current vLLM XPU stack struggled.
Still waiting on Intel to bring newer vLLM builds to their official images, but for now llama.cpp is the more reliable path on this hardware.
#IntelArc #Battlemage #llama.cpp #LocalLLM
65
🧩 Intel lanza Arc 101.8826 WHQL
🎮 Optimiza Gothic 1 Remake, F1 25 y Fortnite
⚙️ Corrige tearing con XeSS Frame Generation
🖥️ Llega a Arc serie A, serie B y Core Ultra
#IntelArc #Intel #Drivers #FanáticosDelHardware
fanaticosdelhardware.com/int…
81
Jun 11
손안에서 만나는 차세대 게이밍 경험!🎮
강력한 게이밍 성능을 넘어,
몰입감 있는 경험까지 제공하는
Intel Arc G3 Extreme 기반의 MSI Claw 8 EX AI
최신 AAA 게임부터 AI 기능까지
더 자유롭게 즐겨보세요😎
#인텔 #intel #IntelArc #ArcG3 #Gaming #HandheldGaming #IntelGaming
149
Jun 11
Sigh.
Spent the last few hours wrestling with vLLM on dual Intel Arc Pro B70s (Battlemage).
Started with the official intel/vllm:xpu image and hit the usual wall:
- oneCCL failing with “could not open device directory”
- GPUs showing up as generic “Intel(R) Graphics”
- multiprocessing / XPU crashes
- hf_transfer download failures
Tried every combination of --privileged, device mounts, env vars, and workarounds, it still wouldn’t stabilize.
Switched to intel/llm-scaler-vllm:1.4 (the image Intel seems to be using for B-Series cards) and it immediately behaved better
— proper architecture detection, no oneCCL explosions, and it actually progressed into model loading with tensor parallelism.
Still early days, but this feels like the right base image for B70s. I hope! CUDA is certainly more mature and easier to optimize.
If you’re trying to run vLLM on Arc Pro B70/B60 cards in Docker, skip the standard intel/vllm:xpu image and go straight to llm-scaler-vllm.
Will report back once I have a stable TP=2 endpoint running.
#IntelArc #Battlemage #vLLM #LocalLLM #XPU
2
2
153
ふじさんパパ retweeted
Jun 11
2
5
35
962
Jun 10
VULKAN API DISCUSSION: PHYSICAL DEVICE & LOGICAL DEVICE | amzn.to/3fl06CR | #vulkan #vulkanapi #gpu #physicaldevice #logicaldevice #programming #gpuprogramming #parallelprocessing #parallelcomputing #nvidia #amd #rtx #intelarc #intelgpu #gpu #cudaeducation
12
