@raymundoojeda1: How fast are two DGX Sparks? GLM 5.3 Flash: 48 tokens a second. DeepSeek V4.1 Flash: a whole snake game in 24 seconds at 115 tokens a second. Recorded in Ownlight today. Posting every day. #dgxspark #localai #glm #deepseek #llm

Raymundo Ojeda
Raymundo Ojeda
Open In TikTok:
Region: US
Monday 16 November 2026 01:03:04 GMT
10140
289
24
6

Music

Download

Comments

semperjason
Jason Bourgeois :
That’s surprisingly slow for what those are. I expected way better. I’m currently running dual 12GB 3080’s and I’m getting 43 tok/s with Qwen 3.8 27b (160K context) and 248 tok/s with Llama 3.2 3b. I’m assuming you must be running larger contexts?
2026-11-16 03:42:09
6
csibinorbert
Norbert 🐺 :
I have one and I'm very disappointed 😔
2026-11-17 15:14:59
0
zabwie
Zabi 🦇 :
Atleast I'm happy with my M6 😔
2026-11-16 01:48:34
3
a.b.b.c4
A.b B.c :
What quantization are you using for DeepSeek V4.1 Flash to run it on 2 NVIDIA Sparks?
2026-11-16 14:05:49
0
uzr12ms
uza :
now try to run 1b or 0.5b model just to measure the speed
2026-11-16 13:30:19
1
runauwooffical
RunaUwO :
Yo bro you should lowkey give me one 🥺
2026-11-16 23:03:38
0
an1mexhunter
AnimeHunter :
what are Specs of These Like vram and RAM etc. and what does one cost?
2026-11-16 01:49:36
0
moneyprintergobrr4
moneyprintergobrrrr :
dual dgx army
2026-11-16 19:51:22
0
duniaku003
Duniaku :
Nice this info that I'm looking for
2026-11-16 13:46:12
0
bebbi_1
Bebbi :
The boasting without any specs is a plague — it’s just annoying at this point.
2026-11-17 00:47:06
0
blauscha7
blauscha :
can u give those guys a task like, programm a Stocks Dashboard in python as one html file and test it yourself, or gibe them 10 min to make a game and show us how good those free offline models are?(i also want to buy one but in my experience those models where useless)😅
2026-11-16 01:51:03
0
To see more videos from user @raymundoojeda1, please go to the Tikwm homepage.

Other Videos


About