Don't Buy an RTX Spark Laptop Until You See This
Ash
0:00 / 0:00
Don't Buy an RTX Spark Laptop Until You See This
36 752 просмотра · 2 дн. назад
Ash
503 подписчика
36 752 просмотра · 2 дн. назад
Nvidia RTX Spark laptops for local AI: I did the math on the Surface Laptop Ultra and the new RTX Spark lineup announced at Microsoft's October 7 event — memory, real speed, price, and who should actually buy one.
RTX Spark (N1X) pairs a 20-core Grace CPU with a Blackwell GPU and up to 128GB of unified memory. Surface Laptop Ultra starts at $2,599, the Surface RTX Spark Dev Box is $5,999, and laptops from Dell, HP, Lenovo, ASUS and MSI ship October 16. HP lists 32GB models from $2,999 and 64GB models at $4,499–$4,999; 128GB pricing isn't public yet.
The catch: the "600 GB/s" in the headlines is the NVLink-C2C link between CPU and GPU, not memory bandwidth. Memory reads at roughly 300 GB/s (DGX Spark, same chip family: 273 GB/s) vs 614 GB/s on an M5 Max. On the same chip, Qwen 3.8 27B measures ~12 tokens/s raw — but 65–71 tok/s with NVFP4 + speculative decoding (DFlash2). 70B dense models crawl at ~5–6 tok/s; mixture-of-experts models are where it shines: Qwen3.8-Flash-Next 125B runs at 30–48 tok/s, and DeepSeek V4 Flash (284B, 13B active) fits in ~60GB at 1.6-bit. Also covered: why "faster than M5 Pro" measures time-to-first-token, why Jensen's "one petaflop" is FP4, RTX Spark vs MacBook Pro M5 Max 128GB ($6,699) vs DGX Spark ($4,699) vs Mac Studio, which memory config to buy, and honest catches (Arm, battery claims).
Subscribe for daily breakdowns of local AI with real numbers — and tell me in the comments: would you buy one, and what would you run on it?
Chapters:
0:00 Intro
0:49 The desk and the book
1:08 What was announced
1:24 128 GB: the biggest desk on Windows
2:03 Trap #1: the 600 GB/s headline
2:41 What 300 GB/s means: 27B and 70B
3:25 Trap #2: faster than M5 Pro?
4:06 What the keynote didn't say
4:23 Trap #3: not the same petaflop
4:45 Twist #1: Nvidia's software
5:37 Twist #2: expert models
6:32 The 1.6-bit warning
6:56 Which version: 32, 64 or 128 GB
7:31 Where it really shines
8:05 RTX Spark vs MacBook Pro
8:38 The money
9:38 Honest catches
10:04 Who should buy it
10:41 The receipt
#RTXSpark #LocalAI #SurfaceLaptopUltra