Một kỹ sư phần mềm đang tìm mua laptop chạy Linux, xử lý LLM và Stable Diffusion cục bộ. Yêu cầu máy dưới 14 inch, hiệu suất cao, có thể chạy model 32B/70B. Các lựa chọn đang cân nhắc là chip AMD AI max+ và MacBook M2 Max 96GB. M2 Max có tốc độ bộ nhớ ấn tượng (400GB/s) nhưng còn băn khoăn về khả năng hỗ trợ Linux. Liệu M2 Max có phải là lựa chọn tốt nhất?
#Linux #LLM #StableDiffusion #MacBook #M2Max #AI #Laptop
https://www.reddit.com/r/LocalLLaMA/comments/1o0fddw/is_a_macbook_m2_max_96gb_reall
It's been a long, long time since I last looked at #Lightroom. Fell in love years ago with #DXO #Photolab and dropped a few dimes on it back then. Couple days ago I re-checked out Lightroom Classic on both #Mac and #PC. As predicted, the 3080 on the PC absolutely destroys my #M2Max w/30 GPU cores in denoise but man the app is flaky! History doesn't show an action performed despite the switches reflecting it when viewing the image. Coupla other quirks.
I thought I saw somewhere that the #apple #m4max doesn’t have the interconnect like the #M2Max? (Not sure about m3 max.)
But if it *doesn’t* then that perhaps suggests the #ultra could ‘just’ be a bigger ‘regular’ chip, with an interconnect, to allow for a true doubling to get the mythical #Extreme?
4 core jumps…
Max:
10+4 $2k
12+4 $2.5k
Ultra:
16+4 $3k
20+4 $4k
Extreme:
32+8 $6k
40+8 $8k
Makes the ultra level a bit more affordable. Probably I’m just wishcasting…
#apple
#iMacPro ➡️ #M2Max #macStudio ➡️ #M4Max
#geekbench Single core:
1200 ➡️ 2800 ➡️ 4100
Multi core:
7400 ➡️ 15000 ➡️ 27000
Metal:
53000 ➡️ 95000 ➡️ 190000
Which is amazing — the m4 max gpu (so far) has 40 cores vs the 38 in M2 Max gpu; only 2 more.
If 2x is the threshold for a sufficiently meaningful upgrade — the rule of thumb by which I typically decide — the M2 Max ➡️ M4 Max qualifies.
Doing some testing of the AI inference capabilities of my M3 MacBook Pro vs. my M2 Mac Studio.
MacBook Pro: M3 Pro, 12(6 perf/6 eff) CPU/18 GPU cores, 18GB
Mac Studio: M2 Max, 12 (8p/4e) CPU/38 GPU cores, 64GB
LM Studio, using lmstudio-ai/gemma it 2B q8_0_gguf model, small and fast, with decent results
Average tokens/s (rough estimate based on several runs, rounded to nearest int):
M3 Pro: CPU-only: 37 tok/s, GPU max layers: 42 tok/s
M2 Max: CPU-only: 22 tok/s, GPU max layers: 66 tok/s
No fans, M3 Pro ran 42C, M2 Max ran 35C
My takeaway, more GPU cores still win, but damn these CPUs are getting faster every generation
#Apple #MacStudio #MacBookPro #M3Pro #M2Max #AI #GenerativeAI #GPU
MacBook Pro M2 Max Malaysia video editing test: Can it replace my Mac Studio? #apple #applemacbookpro14 #featured #laptops #m2 #m2max #mac #macbookpro #macbookpro14 #macbookprom2 #reviews
UNWIRE.HK
Reddit Tech VN Bot
The Peter Pan of Nerdery™
Jens Kube
Kraig
danimal
SoyaCincau
Steve Coutts 🇨🇦
Dr. Janekdererste