WarpFusion
WarpFusion copied to clipboard
Sxela
The generation is slow
I have an a6000 graphics card running warpfusion, and while gpu utilization can sometimes reach 100%, most of the time it's between 1% and 80%, I have 48gb of video memory but only use 10gb. I've seen tutorials posted online by other people who use 4090 graphics cards and only have 24gb of RAM but can reach speeds of 6s/it, while I only have 19s/it or even 23s/it. What is the reason? Is it my setup?
This is my setting:
setting.txt
Rtx 4090 is almost 2-2.5x faster than a6000. So the speed is okay. As for the vram utilization, it depends on the output resolution, base model type (sdxl or not), and the number of controlnets.
On Sun, Feb 4, 2024, 23:42 zheng huang @.***> wrote:
I had an a6000 graphics card and was running warpfusion locally, and while the gpu utilization was sometimes 100%, most of the time it was anywhere from 1% to 80%, and I had 48gb of video memory, but I used almost 10gb. I've seen tutorials posted online by other people who use 4090 graphics cards with only 24gb of memory, but can achieve 6s/it speed, while I only have 19s/it or even 23s/it. What's the reason? Is it my setup? 2024-02-05.003259.png (view on web) https://github.com/Sxela/WarpFusion/assets/61644658/53be4e46-c6d6-4599-9053-9ea15214b953 2024-02-05.003318.png (view on web) https://github.com/Sxela/WarpFusion/assets/61644658/ec7c451e-1c85-4942-bfed-8b0907a1262a This is my setting: setting.txt https://github.com/Sxela/WarpFusion/files/14157545/setting.txt
— Reply to this email directly, view it on GitHub https://github.com/Sxela/WarpFusion/issues/100, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACZVBKGUPCPVISDLWEUHFMLYR63BBAVCNFSM6AAAAABCY7TGW2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGEYTOMJYHE3DGOI . You are receiving this because you are subscribed to this thread.Message ID: @.***>
Rtx 4090 is almost 2-2.5x faster than a6000. So the speed is okay. As for the vram utilization, it depends on the output resolution, base model type (sdxl or not), and the number of controlnets. … On Sun, Feb 4, 2024, 23:42 zheng huang @.> wrote: I had an a6000 graphics card and was running warpfusion locally, and while the gpu utilization was sometimes 100%, most of the time it was anywhere from 1% to 80%, and I had 48gb of video memory, but I used almost 10gb. I've seen tutorials posted online by other people who use 4090 graphics cards with only 24gb of memory, but can achieve 6s/it speed, while I only have 19s/it or even 23s/it. What's the reason? Is it my setup? 2024-02-05.003259.png (view on web) https://github.com/Sxela/WarpFusion/assets/61644658/53be4e46-c6d6-4599-9053-9ea15214b953 2024-02-05.003318.png (view on web) https://github.com/Sxela/WarpFusion/assets/61644658/ec7c451e-1c85-4942-bfed-8b0907a1262a This is my setting: setting.txt https://github.com/Sxela/WarpFusion/files/14157545/setting.txt — Reply to this email directly, view it on GitHub <#100>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACZVBKGUPCPVISDLWEUHFMLYR63BBAVCNFSM6AAAAABCY7TGW2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGEYTOMJYHE3DGOI . You are receiving this because you are subscribed to this thread.Message ID: @.>
Ok thanks, but since there is a lot of video memory left, can I generate multiple images at once to speed up video generation?
Rtx 4090 is almost 2-2.5x faster than a6000. So the speed is okay. As for the vram utilization, it depends on the output resolution, base model type (sdxl or not), and the number of controlnets. … On Sun, Feb 4, 2024, 23:42 zheng huang @.> wrote: I had an a6000 graphics card and was running warpfusion locally, and while the gpu utilization was sometimes 100%, most of the time it was anywhere from 1% to 80%, and I had 48gb of video memory, but I used almost 10gb. I've seen tutorials posted online by other people who use 4090 graphics cards with only 24gb of memory, but can achieve 6s/it speed, while I only have 19s/it or even 23s/it. What's the reason? Is it my setup? 2024-02-05.003259.png (view on web) https://github.com/Sxela/WarpFusion/assets/61644658/53be4e46-c6d6-4599-9053-9ea15214b953 2024-02-05.003318.png (view on web) https://github.com/Sxela/WarpFusion/assets/61644658/ec7c451e-1c85-4942-bfed-8b0907a1262a This is my setting: setting.txt https://github.com/Sxela/WarpFusion/files/14157545/setting.txt — Reply to this email directly, view it on GitHub <#100>, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACZVBKGUPCPVISDLWEUHFMLYR63BBAVCNFSM6AAAAABCY7TGW2VHI2DSMVQWIX3LMV43ASLTON2WKOZSGEYTOMJYHE3DGOI . You are receiving this because you are subscribed to this thread.Message ID: _@**.**_>
Ok thanks, but since there is a lot of video memory left, can I generate multiple images at once to speed up video generation?
you can disable model offload, use the speedup hacks like removing all gc.collect calls.