Selur's Little Message Board

send you the links,..

Cu Selur

(01.11.2023, 06:42)Selur Wrote: [ -> ]The torch-addon only works with NVIDIA cards.
I could send you a link to the vs-mlrt addon (https://github.com/AmusementClub/vs-mlrt/).

Yep , i have read about it afterwards   damn..

the upscaling looked promising though from what ive seen at github..

But then again, my new GPU is an RDNA3 architecture with AI acceleration ... 

Isn't there any alternative that supports amd newest cards ? 


cheers,

Quote:Isn't there any alternative that supports amd newest cards ?

Aside from what vs-mlrt offers: not that I know of.

Cu Selur

(01.11.2023, 19:00)Selur Wrote: [ -> ]

Quote:Isn't there any alternative that supports amd newest cards ?

Aside from what vs-mlrt offers: not that I know of.

Cu Selur

Oh well, cuz i think the vulkan api could have an advantage in speed using these addons.

cheers,
TD,

Quote: i think the vulkan api could have an advantage in speed using these addons.

Okay.
From what I have seen so far in ml relates stuff, the direct cuda implementations are usually faster than their vulkan alternatives, especially if TensorRT is used.


Cu Selur

(01.11.2023, 19:11)Selur Wrote: [ -> ]

Quote: i think the vulkan api could have an advantage in speed using these addons.

Okay.
From what I have seen so far in ml relates stuff, the direct cuda implementations are usually faster than their vulkan alternatives, especially if TensorRT is used.


Cu Selur

Is there much difference speedwise ?

I mean, beside comparing an rtx4090 high tier with an rx480 low end gpu ofcourse 


cheers,

Quote:Is there much difference speedwise ?

No clue, I bought an NVIDIA card since I knew that:
a. they are widely used in universities for years (due to apis&co), so most new research projects&coder use them.
b. the video encoders of AMD hardware were always worse than from Intel and NVIDIA.
According to https://www.hpcwire.com/off-the-wire/amd...-rocm-5-7/ newer pytorch might support AMD cards, so there is hope, but since I don't own a RDNA3 card I can't really say how much trouble it would be for the developers to port their code for that.

Cu Selur

(01.11.2023, 19:44)Selur Wrote: [ -> ]

Quote:Is there much difference speedwise ?

No clue, I bought an NVIDIA card since I knew that:
a. they are widely used in universities for years (due to apis&co), so most new research projects&coder use them.
b. the video encoders of AMD hardware were always worse than from Intel and NVIDIA.
According to https://www.hpcwire.com/off-the-wire/amd...-rocm-5-7/ newer pytorch might support AMD cards, so there is hope, but since I don't own a RDNA3 card I can't really say how much trouble it would be for the developers to port their code for that.

Cu Selur

I gather you are in favor of team blue/green 

Granted Nvidia still is quite ahead speed wise, but general AMD is the best in terms of Value / pricewise IMO .
And as far as I know, Nvidia has never beaten AMD in that regard 

And with the current economics.. they will have a hard time doing so.. in the next decade..


cheers,

Quote:AMD is the best in terms of Value / pricewise IMO .

Only for gaming.
For everything that is video related, they are behind.

(01.11.2023, 20:23)Selur Wrote: [ -> ]

Quote:AMD is the best in terms of Value / pricewise IMO .

Only for gaming.
For everything that is video related, they are behind.

Gaming obsolutely, goes without saying ..

In terms of video performance i can't complain..

But then again, just recently purchased an 16GB Vram GPU that noticeable makes an huge difference compared to my old 8GB gpu !

Like using filters like KNLmeans   wich is  very gpu taxing, improved the performance by ALLOT..  With old gpu i gained only 15 - 25fps / SD > HD 1080p content.
And cpu utilized like 5% tops LoL with that gpu.

Now, i get like 140 - 150+ fps Cpu @max   

I say , thats ↑ not bad for an upgrade   

cheers,
td