唔使天價GPU!六千蚊搞掂DeepSeek R1 AI,單人試玩都得!

Ai




以$6,000硬件運行DeepSeek R1聊天機器人

你是否以為要運行一個前沿的聊天機器人必須花費巨額的顯示卡?其實不然。來自AI工具公司HuggingFace的工程師Matthew Carrigan表示,僅需$6,000的電腦硬件,就能運行最新的DeepSeek R1大型語言模型(LLM),而且根本不需要高端的顯示卡。

Carrigan建議的配置包括一塊雙插槽的AMD EPYC主板和幾個兼容的AMD處理器。他指出,CPU的具體規格其實並不是最關鍵的,重點在於內存的配置。

他解釋道:“我們需要768GB的內存(以容納模型),這需要24個內存通道(以獲得足夠的帶寬來快速運行)。這意味著需要24個32GB的DDR5-RDIMM模塊。”僅內存的花費就約為$3,400。然後還需要一個機殼、電源、一個1TB的SSD,以及一些散熱器和風扇。

Carrigan表示,這套配置能夠讓你獲得完整的DeepSeek R1體驗,沒有任何妥協。“這是真正的模型,沒有任何蒸餾,並且使用Q8量化以達到最佳質量,”他說。

接下來,只需在系統上安裝Linux,下載700GB的權重,並輸入Carrigan提供的命令行字符串,這樣就能在本地運行大型語言模型了。

在這一切中,有一個值得注意的地方,就是完全沒有提到昂貴的Nvidia顯示卡。Carrigan提供了一段視頻,展示了這個設定下LLM的本地運行效果以及大致的性能指標。

他指出:“這個配置的生成速度為每秒6到8個標記,具體取決於CPU和內存的速度,或者如果聊天歷史較長,速度會稍微降低。”視頻中顯示模型以合理的速度生成文本,但這僅針對單一用戶。如果將這個設置擴展到多個用戶,每位用戶的性能顯然會變得無法使用。

簡而言之,這意味著$6,000的硬件實際上僅能支持一位用戶。因此,這種方法顯然不適合用於建立一個能夠服務數百、數千甚至數百萬用戶的AI業務。對於這樣的應用場景,顯示卡可能會更具成本效益,即使它們的單位價格非常高。

Carrigan還提到,依賴顯示卡的配置可能很快就會達到三位數的成本,儘管性能會更好。

AI的真實性與運行成本

這一發現令人著迷,因為它告訴我們其實不需要花費巨額金額就能在本地運行全功能的LLM。這也讓我們更深入地理解了最新LLM所暗示的智能規模。

作為終端用戶,當我們體驗這些機器人所呈現的似乎是意識的輸出時,往往會認為生成LLM的輸出需要巨大的計算能力。但這套配置卻是依靠幾顆AMD處理器來實現。

因此,除非你認為幾顆AMD處理器能夠擁有意識,否則這一硬件解決方案展示了即使是最新、最先進的LLM的平凡現實。或許,AI末日並不會如我們想象中那麼快來臨。

在這個快速發展的技術世界中,我們需要保持警覺,認識到即使是最先進的技術也可能並不需要我們想象中的巨額投資。這不僅是對硬件需求的重新評估,也是對AI技術本身的一次反思。是否我們過於依賴高端硬件來證明AI的能力,而忽視了其背後的算法和模型的真正價值?

以上文章由特價GPT API KEY所翻譯及撰寫。而圖片則由FLUX根據內容自動生成。

🎨 Nano Banana Pro 圖像生成器|打幾句說話就出圖

想畫人像、產品圖、插畫?SSFuture 圖像生成器支援 Flux Gemini Nano Banana Pro 改圖 / 合成, 打廣東話都得,仲可以沿用上一張圖繼續微調。

🆓 Flux 模型即玩,不用登入
🤖 登入後解鎖 Gemini 改圖
📷 支援上載參考圖再生成
⚡ 每天免費額度任你玩
✨ 即刻玩 AI 畫圖
Google Gemini / Meta AI 🎯💯 

Promte ✅ 

A hyper-realistic night portrait of a handsome young man leaning casually against a black luxury car, laughing heartily with head tilted slightly back. He has a groomed beard, short styled dark hair, and is wearing a plain black long-sleeve crewneck sweatshirt. The lighting is low-key and cinematic, with the subject's face warmly illuminated against the darkness. The background features a dark street with beautiful, creamy golden bokeh orbs from distant streetlights and car headlights. Shot on a Sony A7R IV with an 85mm f/1.4 lens, shallow depth of field, sharp focus on the smile, high contrast, moody urban night atmosphere, 8k resolution. Size 4:5 ratio 100% use my upload reference image Create a very realistic, cinematic image of a woman, with 100% of her original facial features maintained from the provided image. Do not alter the face, focusing on realistic details of the facial structure, eyes, eyebrows, nose, mouth, and expression. The lower left corner of the image features the Facebook logo and the text Nomahtan Kung.
Makeup: sharp Asian-style eyebrows, sharp Cat Eye eyeliner, and gradient red lipstick
​Her lips are parted slightly, revealing a hint of teeth, and are coated in a glossy, natural-toned red lipstick that catches the light, giving them a luscious appearance. The subtle lines and contours of her philtrum and cupid's bow are visible. with the subtle texture of skin visible. Wears a necklace with the text Kunika and other small matching accessories, such as earrings or a wrist bracelet. Base Setup
keep 100 percent facial information adherence of the attached image and turn her into a lone traveler posed on a sunlit desert dune ridge, captured as a live action photograph or movie still, not an illustration or CGI render, with a sexy, confident, heat soaked editorial mood.

Shot and Camera
Full body shot from slightly low height, framing her on the right third, with sweeping dunes rolling into the distance on the left, using a wide cinematic lens feel that keeps the landscape vast and minimal.

Identity and Pose
She has a slim, toned build, medium height impression, sun kissed skin, and long loose hair blown by the wind. She stands barefoot on the crest, one leg forward so the hip shifts naturally, one hand resting on her upper thigh and the other lightly gripping a sheer wrap at her side, wearing a high cut desert bikini with a gauzy open sarong that shows plenty of leg and midriff without nudity, head turned over her shoulder with a subtle, knowing smile, 8k Photorealistic and hyper realistic.

Lighting and Environment
Harsh midday sun from high right casts crisp shadows along her body and the dune ripples, with soft sky bounce filling the shadows enough to keep detail. The sand textures, wind carved ridges, and pale blue sky remain exactly like the reference, with faint footprints and slight imperfections grounding her on the slope.

Masking and Constraints
Change only by adding the subject and her wardrobe, keep dune shapes, lighting direction, perspective, horizon line, and white balance identical, preserving realistic body proportions and clear contact between her feet and the sand. Absolutely no added text, no painterly or toon look, no CGI plastic skin, no see through fabric on intimate areas, no warped limbs or floating feet, strictly require consistent perspective, natural film like grain, fine skin texture and sand detail, and physically correct contact shadows in the sand depressions around her feet.