家用NAS/Server/Homelab集中討論區(8)

極北鷲

984 回覆
1 Like 1 Dislike
四架坦克 2025-07-07 17:32:55
今日我有部remote 1618都係QC唔到入去
見reddit前兩日討論話reboot就得
咁因為我同LAN內有機可以tunnel到入去所以用LAN內IP去DSM reboot到
但如果本身係依靠QuickConnect無其他方法access的話暫時好似無解
千反田える 2025-07-07 17:51:09
大棍巴 2025-07-07 19:13:59
可能同時係disk有啲多(7*8TB)
點都好,而家即係冷氣焗開
3HongKong 2025-07-07 20:02:35
諗緊買兩隻 hc590 26tb
極北鷲 2025-07-08 01:50:28
https://www.reddit.com/r/Proxmox/comments/1ltxq6q/nvidias_new_vgpu_solution_cracked_rtx_30series/

Recently, Chinese tech enthusiast pdbear successfully cracked NVIDIA's new GPU virtualization defenses, enabling RTX 30-series and 40-series gaming GPUs to unlock the enterprise-grade GRID vGPU features. It is reported that this functionality was previously creaked by tech enthusiast Dualcoder in 2021, with the open-source project vgpu_unlock hosted on GitHub. However, that project only supported up to the 20-series GPUs (with the highest being RTX 2080Ti). Due to NVIDIA's shift to the SR-IOV solution in its new commercial GRID vGPU solution for 30-series professional cards, no one had managed to breach it for four years.

CL9528 2025-07-08 12:14:27
官網咩 promo code 都冇
#Szczesny# 2025-07-08 15:08:59
謝謝你
千反田える 2025-07-08 21:54:37
大棍巴 2025-07-08 22:27:43
幾時搞埋俾5090
大棍巴 2025-07-08 22:32:42
我仲有5隻16TB,可能機箱大(ATX),望過下全部都係40度左右。
嗰set 7*8TB行ITX箱可能係散熱唔夠
極北鷲 2025-07-09 01:10:22
係咪掛部tablet上牆用?
大棍巴 2025-07-09 02:00:55
都想問下嗰部咩來,tablet?
如果Amazon Fire Tablet可以改到咁用就幾好。
Hagayyy 2025-07-09 02:34:58
唔知係咪樓主自己寫既? 幾正
應該係咩tablet 都得, 開個browser 顯示dashboard 同 allow microphone access就得
但仲要多部機去host whisper.cpp 去 speech-to-text
大棍巴 2025-07-09 02:59:49




密謀緊
Llsudhejsidbejd 2025-07-09 07:06:15
香港多選擇,唔一定揀佢,等下先

Amazon AU prime day,執定咗套sodimm ddr5 corsair 64gb kit

大家好似都好多嘢玩緊,我自己都攪緊一個info portal放出去public俾人用,目標為本嘅話perplexity寫code真係幾好,特別係load API,俾足example佢就有高質response
debugger; 2025-07-09 09:33:27
係呀
debugger; 2025-07-09 09:34:30
Amazon fire tablet絕對可以,邊部tablet都得,有browser就得

我用緊二手galaxy tab s7+貪佢有12.4 oled
debugger; 2025-07-09 09:42:03
有兩個問題

廣東話voice input既話tiny未必搞得掂,我依家係large v3 bf16都20-30%機會錯,遠距離就更嚴重,未有解法

llm果邊,0.6b既細模型連理解指令我都驚有困難。之前試過4090行gemma3 27b qat都唔太滿意,tool call成日扮做左,先死死氣轉用gemini 2.5 flash

但2.5flash都有時好智障,搞緊個tool係上網讀html抽文字,佢死都reject user話自己上唔到網,個tool只係用黎分析文字,幫唔到我開網頁,改左好多次prompt都唔掂搞到我都嬲左
大棍巴 2025-07-09 10:11:38
大棍巴 2025-07-09 10:17:39
我應該會用返英文,希望夠做。本身想行Gemma 3n直接做埋transcript ,但好似好多inference backend都未support audio input (諗住用SGLang+FlashInfer)

Model方面的確係問題,但Jetson Nano Super得8GB,要行哂成個pipeline,容量有限之下選擇唔多。呢刻我想先試左個基本flow先,之後再玩下fetch weather infomation再俾佢report/做下tool calling計下數,Google下嘢咁。開關哩啲就暫時用返Alexa先
叔叔受不了 2025-07-09 10:45:34
個博主話短期內都唔support 50系
Llsudhejsidbejd 2025-07-09 16:25:33
CL9528 2025-07-09 16:37:57
Llsudhejsidbejd 2025-07-09 17:54:55
CL9528 2025-07-09 18:12:51
吹水台自選台熱 門最 新手機台時事台政事台World體育台娛樂台動漫台Apps台遊戲台影視台講故台健康台感情台家庭台潮流台美容台上班台財經台房屋台飲食台旅遊台學術台校園台汽車台音樂台創意台硬件台電器台攝影台玩具台寵物台軟件台活動台電訊台直播台站務台黑 洞