Software Engineering Courses (SECourses)•10mo ago

full model uses 52.16 gb - 14b image to video

FFurkan Gözükara SECourses full model uses 52.16 gb - 14b image to video

1nfern0•2/25/25, 11:17 PM

Is that the FP32?

11nfern0 Is that the FP32?

Furkan Gözükara SECoursesOP•2/25/25, 11:17 PM

bf16

Furkan Gözükara SECoursesOP•2/25/25, 11:17 PM

Furkan Gözükara SECoursesOP•2/25/25, 11:18 PM

but this is almost as if fp32 vram

Furkan Gözükara SECoursesOP•2/25/25, 11:18 PM

we need 1.3b image to video model :d

Furkan Gözükara SECoursesOP•2/25/25, 11:33 PM

i will test on h200 lets see the speed

Zaulinator•2/25/25, 11:37 PM

all very exciting

Furkan Gözükara SECoursesOP•2/25/25, 11:38 PM

ye really cool

Furkan Gözükara SECoursesOP•2/25/25, 11:38 PM

i will publish on Patreon hopefully today before sleep

Furkan Gözükara SECoursesOP•2/25/25, 11:38 PM

final test

Furkan Gözükara SECoursesOP•2/25/25, 11:39 PM

14b text to video broken atm though i expect fix tomorrow

Furkan Gözükara SECoursesOP•2/25/25, 11:40 PM

download speed with special tech

Furkan Gözükara SECoursesOP•2/25/25, 11:41 PM

h200 is fast

Furkan Gözükara SECoursesOP•2/25/25, 11:42 PM

h200 is 1.84 second id on my rtx 3090 ti around 8.5 second / it - 1.3b model

Furkan Gözükara SECoursesOP•2/25/25, 11:51 PM

text to video 14b fits into 24gb vram - min 17 gb

Furkan Gözükara SECoursesOP•2/25/25, 11:56 PM

h200 is 36 second / it for 720x1280 image to video at 14b for 81 frames

Furkan Gözükara SECoursesOP•2/26/25, 12:12 AM

waiting first image to video

Furkan Gözükara SECoursesOP•2/26/25, 12:27 AM

lol i worked full day today at least 14 hours now

Furkan Gözükara SECoursesOP•2/26/25, 12:39 AM

just wow

Furkan Gözükara SECoursesOP•2/26/25, 12:39 AM

FFurkan Gözükara SECourses Click to see attachment

Yobin•2/26/25, 2:43 AM

Yobin•2/26/25, 2:44 AM

you have to tell me how to do this specific vid from prompt to approach

Yobin•2/26/25, 2:44 AM

so gooood

FFurkan Gözükara SECourses waiting first image to video

Yobin•2/26/25, 2:44 AM

okay imma try it lol

Yobin•2/26/25, 2:44 AM

pls tell me you already dropped a tutorial in your patreon

YYobin pls tell me you already dropped a tutorial in your patreon

Cyph3r•2/26/25, 2:49 AM

looks like he did

CCyph3r looks like he did

Yobin•2/26/25, 2:51 AM

https://tenor.com/view/my-man-denzel-my-man-denzel-washington-myyyy-man-gif-376647294737879812

Tenor

AiInfluence•2/26/25, 7:18 AM

just amazing

AiInfluence•2/26/25, 7:19 AM

also refreshed my Patreon subscription

AiInfluence•2/26/25, 7:19 AM

Doctor I think it is very near kling tbh

AiInfluence•2/26/25, 7:19 AM

I'm going to try today and experiment with it

FFurkan Gözükara SECourses waiting first image to video

Timson•2/26/25, 7:38 AM

so around 30 minutes on h200 per generation? That's nuts

Timson•2/26/25, 7:40 AM

how many iterations per frame does it do?

AiInfluence•2/26/25, 7:41 AM

@Dr. Furkan Gözükara what gpu should I rent for the large 14b model on massed ?

daroth•2/26/25, 9:33 AM

good grief, how big are these models - 60+gb to download

daroth•2/26/25, 9:45 AM

its a bit confusing how such big models can be run with consumer gpus, but I'm hopeful

Ddaroth its a bit confusing how such big models can be run with consumer gpus, but I'm h...

Timson•2/26/25, 10:14 AM

with extreme quantization and quality loss

Timson•2/26/25, 10:15 AM

Consumer GPU need to get better, not the other way around

Timson•2/26/25, 10:16 AM

I was hopeful 5090 release would drop GPU compute prices but they continue to rise. L40S time is ever more expensive.

Timson•2/26/25, 10:16 AM

maybe because libraries does not fully support 5090 as Dr. Furkan reporting.
Seems vary strange how long it takes

TrumpYouAll•2/26/25, 10:24 AM

@Dr. Furkan Gözükara Hello, I watched this video https://www.youtube.com/watch?v=jHlGzaDLkto
which is about how to use 5090 for Swarm UI
However, I got this error .

I got several questions:

I got this error (can you help me solve it?) Is it 5090-specific?
I forgot (I asked you before but I since my computer was fried so I don't have the record) where to put the trained Flux Dev models with Dreambooth ; I remember somewhere I need to put something like 'diffusion_models' (?). Can you give me a a specific answer where to create this folder (or if my folder name is corrrct to put my Dreambooth-trained models)
I forgot how you type the yoloface command in the input textbox (again sorry, I lost the record when my PC is fried)

Thanks in advance ! I always love ur videos!!!! U are the king in AI image !

@Dr. Furkan Gözükara

YouTubeSECourses

RTX 5090 Tested Against FLUX DEV, SD 3.5 Large, SD 3.5 Medium, SDXL...

In this video I have intensively compared RTX 5090 speed on FLUX DEV, FLUX Fill, SD 3.5 Large, SD 3.5 Medium, Stable Diffusion XL (SDXL) and Stable Diffusion 1.5 (SD 1.5) models. For each benchmark, I have compared RTX 5090 against RTX 3090 TI so we see the speed improvement. Moreover, I have tested FP8 vs 16-bit precision for FLUX and SD 3.5 La...