nodes/IamCreateAI-Ruyi-Models
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Update low memory table

Browse Source
This commit is contained in:
cellzero
2024-12-24 16:35:14 +08:00
parent 405a4ddc3a
commit 5c690e1ed9
5 changed files with 239 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+2
View File
@@ -165,6 +165,8 @@ We provide the options **`GPU_memory_mode` and `GPU_offload_steps` to reduce GPU
Generally speaking, using **less GPU memory requires more RAM and results in longer generation times**. Below is a reference table of expected GPU memory usage and generation times. Note that, the GPU memory reported below is the `max_memory_allocated()` value. The values read from nvidia-smi may be higher than the reported values because CUDA occupies some GPU memory (usually between 500 - 800 MiB), and PyTorch's caching mechanism also requests additional GPU memory.
Additionally, the community and we have created a detailed table featuring various resolutions and option combinations, which can be found in the [gpu_memory_appendix.md](assets/gpu_memory_appendix.md). We encourage community members to help us complete the table.
### A100 Results
- Resolution of 512
README_CN.md
+2
View File
@@ -154,6 +154,8 @@ python3 predict_i2v.py
通常来说，**使用更少的显存，需要更多的内存和更长的生成时间**。以下列出预期显存使用和生成时间的参考表格。请注意，下面报告的显存是 `max_memory_allocated()` 的返回值，而 nvidia-smi 的显存数值通常会高于报告的数值。因为 CUDA 会占用一些显存（通常在500 - 800 MiB之间），而 PyTorch 的缓存机制也会请求额外的显存。
此外，在社区的帮助下创建了一张详细的表格 [gpu_memory_appendix.md](assets/gpu_memory_appendix.md)，包含了更多分辨率和选项组合。欢迎社区成员帮助完善表格。
### A100 的显存占用与运行时间
- Resolution of 512
assets/gpu_memory_appendix.md
+233
View File
@@ -0,0 +1,233 @@
# GPU Memory Appendix
On this page, we will display the required memory and execution time under different combinations of memory optimization parameters. We would also like to thank the community for their support in helping us enrich the table data.
Please note that due to variations in the running environment, the content in the table may differ from actual conditions and is for reference only.
## RTX 3090 Results
### Results from [bmgjet](https://github.com/bmgjet)
- Environments
| CPU | Ram | Disk | GPU |
| :---: | :---: | :---: | :---: |
| 7950X | 64GB 6200mhz, 30,32,32 | NVME 990 Pro | 3090 (Water Cooled 500W PL, +1000mhz Vram) Sits about 2.1ghz on core |
Memory usage reported from GPUz logs.
Torch was not compiled with flash attention.
Windows 11 Using Python Environment from Forge (webui_forge_cu124_torch24.7z) Feb 5 2024.
- Resolution of 512
| Num frames | Parameters | GPU memory | Time |
| :---: | :---: | :---: | :---: |
| 24 frames | normal 0 | 18117mb | 1:32s - 3.80s/it |
| 48 frames | normal 0 | 19995mb | 4:06s - 9.98s/it |
| 72 frames | normal 0 | 22133mb | 9:57s - 23.88s/it |
| 96 frames | normal 10 | 22200mb | 12:51s - 30.84s/it |
| 120 frames | normal 7 | 23195mb | 15:57s - 38.26s/it |
- Resolution of 768
| Num frames | Parameters | GPU memory | Time |
| :---: | :---: | :---: | :---: |
| 24 frames | normal 0 | 22105mb | 4:07s - 9.94s/it |
| 48 frames | normal 7 | 22195mb | 14:34s - 34.98s/it |
| 72 frames | lowmem 0 | 17998mb | 38:29s - 91.99s/it |
| 96 frames | lowmem 0 | 23000mb | 1:04:16s - 153.74s/it |
| 120 frames | lowmem 0 | 28195mb | 5:29:54s - 797.28s/it <br> (froze system during vram overflows for a few sec each pass) |
## RTX 4090
### Results from CreateAI
> The GPU memory reported below is the `max_memory_allocated()` value. The values read from nvidia-smi may be higher than the reported values because CUDA occupies some GPU memory (usually between 500 - 800 MiB), and PyTorch's caching mechanism also requests additional GPU memory.
- Resolution of 384
| Num <br> frames | normal_mode + 0 steps | normal_mode + 10 steps | normal_mode + 7 steps | normal_mode + 5 steps | normal_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 15387MiB <br> _00:42s_ | 15052MiB <br> _00:50s_ | 14941MiB <br> _00:50s_ | 14867MiB <br> _00:57s_ | 14718MiB <br> _00:58s_ |
| 48 <br> frames | 16686MiB <br> _01:15s_ | 16018MiB <br> _01:33s_ | 15795MiB <br> _01:35s_ | 15647MiB <br> _01:38s_ | 15350MiB <br> _01:40s_ |
| 72 <br> frames | 17996MiB <br> _01:51s_ | 16993MiB <br> _02:12s_ | 16659MiB <br> _02:21s_ | 16436MiB <br> _02:22s_ | 15990MiB <br> _02:31s_ |
| 96 <br> frames | 19297MiB <br> _02:31s_ | 17959MiB <br> _02:58s_ | 17514MiB <br> _03:09s_ | 17216MiB <br> _03:13s_ | 16621MiB <br> _03:27s_ |
| 120 <br> frames | 20599MiB <br> _03:17s_ | 18928MiB <br> _03:50s_ | 18371MiB <br> _04:09s_ | 17999MiB <br> _04:14s_ | 17256MiB <br> _04:29s_ |
| Num <br> frames | low_gpu_mode + 0 steps | low_gpu_mode + 10 steps | low_gpu_mode + 7 steps | low_gpu_mode + 5 steps | low_gpu_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 4096MiB <br> _02:51s_ | 4096MiB <br> _03:25s_ | 4096MiB <br> _03:35s_ | 4096MiB <br> _02:45s_ | 4095MiB <br> _03:13s_ |
| 48 <br> frames | 4148MiB <br> _03:37s_ | 4148MiB <br> _04:01s_ | 4148MiB <br> _04:19s_ | 4148MiB <br> _03:43s_ | 4147MiB <br> _04:52s_ |
| 72 <br> frames | 4200MiB <br> _05:01s_ | 4200MiB <br> _05:40s_ | 4200MiB <br> _07:51s_ | 4200MiB <br> _05:01s_ | 4200MiB <br> _06:18s_ |
| 96 <br> frames | 5248MiB <br> _05:59s_ | 4249MiB <br> _06:49s_ | 4249MiB <br> _08:43s_ | 4249MiB <br> _06:03s_ | 4249MiB <br> _07:09s_ |
| 120 <br> frames | 6549MiB <br> _07:24s_ | 4876MiB <br> _08:12s_ | 4321MiB <br> _10:45s_ | 4299MiB <br> _08:09s_ | 4299MiB <br> _08:32s_ |
- Resolution of 512
| Num <br> frames | normal_mode + 0 steps | normal_mode + 10 steps | normal_mode + 7 steps | normal_mode + 5 steps | normal_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 16366MiB <br> _01:18s_ | 15805MiB <br> _01:26s_ | 15607MiB <br> _01:37s_ | 15475MiB <br> _01:36s_ | 15211MiB <br> _01:39s_ |
| 48 <br> frames | 18720MiB <br> _02:21s_ | 17532MiB <br> _02:49s_ | 17136MiB <br> _02:55s_ | 16872MiB <br> _02:58s_ | 16344MiB <br> _03:01s_ |
| 72 <br> frames | 21036MiB <br> _03:41s_ | 19254MiB <br> _04:25s_ | 18660MiB <br> _04:34s_ | 18264MiB <br> _04:36s_ | 17472MiB <br> _04:51s_ |
| 96 <br> frames | -----MiB <br> _--:--s_ | 20972MiB <br> _06:18s_ | 20180MiB <br> _06:24s_ | 19652MiB <br> _06:36s_ | 18596MiB <br> _06:56s_ |
| 120 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | 21704MiB <br> _08:50s_ | 21044MiB <br> _08:53s_ | 19724MiB <br> _09:08s_ |
| Num <br> frames | low_gpu_mode + 0 steps | low_gpu_mode + 10 steps | low_gpu_mode + 7 steps | low_gpu_mode + 5 steps | low_gpu_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 4212MiB <br> _04:31s_ | 4212MiB <br> _04:44s_ | 4211MiB <br> _05:12s_ | 4211MiB <br> _04:51s_ | 4212MiB <br> _04:14s_ |
| 48 <br> frames | 4666MiB <br> _05:17s_ | 4401MiB <br> _06:11s_ | 4400MiB <br> _06:33s_ | 4399MiB <br> _08:29s_ | 4401MiB <br> _06:18s_ |
| 72 <br> frames | 6981MiB <br> _07:27s_ | 5199MiB <br> _08:00s_ | 4605MiB <br> _08:57s_ | 4598MiB <br> _11:02s_ | 4598MiB <br> _08:46s_ |
| 96 <br> frames | 9298MiB <br> _09:33s_ | 6922MiB <br> _10:13s_ | 6130MiB <br> _11:31s_ | 5602MiB <br> _13:46s_ | 4794MiB <br> _11:29s_ |
| 120 <br> frames | 11613MiB <br> _11:33s_ | 8643MiB <br> _13:11s_ | 7653MiB <br> _14:18s_ | 6993MiB <br> _14:23s_ | 5673MiB <br> _14:16s_ |
- Resolution of 640
| Num <br> frames | normal_mode + 0 steps | normal_mode + 10 steps | normal_mode + 7 steps | normal_mode + 5 steps | normal_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 17671MiB <br> _01:47s_ | 16739MiB <br> _02:09s_ | 16429MiB <br> _02:17s_ | 16234MiB <br> _02:23s_ | 15850MiB <br> _02:28s_ |
| 48 <br> frames | 21324MiB <br> _03:49s_ | 19468MiB <br> _04:23s_ | 18849MiB <br> _04:55s_ | 18437MiB <br> _04:59s_ | 17612MiB <br> _05:14s_ |
| 72 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | 21230MiB <br> _07:57s_ | 20611MiB <br> _08:13s_ | 19373MiB <br> _08:25s_ |
| 96 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | 21134MiB <br> _11:57s_ |
| 120 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ |
| Num <br> frames | low_gpu_mode + 0 steps | low_gpu_mode + 10 steps | low_gpu_mode + 7 steps | low_gpu_mode + 5 steps | low_gpu_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 4301MiB <br> _03:49s_ | 4301MiB <br> _04:31s_ | 4301MiB <br> _04:55s_ | 4301MiB <br> _06:33s_ | 4301MiB <br> _04:52s_ |
| 48 <br> frames | 7271MiB <br> _07:05s_ | 5416MiB <br> _08:04s_ | 4798MiB <br> _08:07s_ | 4589MiB <br> _09:42s_ | 4589MiB <br> _09:18s_ |
| 72 <br> frames | 10889MiB <br> _10:38s_ | 8106MiB <br> _11:54s_ | 7179MiB <br> _11:58s_ | 6560MiB <br> _12:52s_ | 5322MiB <br> _13:13s_ |
| 96 <br> frames | 14509MiB <br> _14:13s_ | 10795MiB <br> _15:50s_ | 9557MiB <br> _16:10s_ | 8732MiB <br> _17:08s_ | 7082MiB <br> _17:39s_ |
| 120 <br> frames | 18128MiB <br> _18:26s_ | 13487MiB <br> _19:57s_ | 11942MiB <br> _21:12s_ | 10910MiB <br> _21:55s_ | 8847MiB <br> _26:20s_ |
- Resolution of 768
| Num <br> frames | normal_mode + 0 steps | normal_mode + 10 steps | normal_mode + 7 steps | normal_mode + 5 steps | normal_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 19223MiB <br> _02:38s_ | 17900MiB <br> _03:06s_ | 17448MiB <br> _03:18s_ | 17153MiB <br> _03:23s_ | 16624MiB <br> _03:34s_ |
| 48 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | 20946MiB <br> _07:28s_ | 20352MiB <br> _07:35s_ | 19164MiB <br> _08:04s_ |
| 72 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ |
| 96 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ |
| 120 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ |
| Num <br> frames | low_gpu_mode + 0 steps | low_gpu_mode + 10 steps | low_gpu_mode + 7 steps | low_gpu_mode + 5 steps | low_gpu_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 5251MiB <br> _05:19s_ | 4398MiB <br> _05:55s_ | 4398MiB <br> _06:52s_ | 4398MiB <br> _07:24s_ | 4399MiB <br> _07:35s_ |
| 48 <br> frames | 10457MiB <br> _09:38s_ | 7786MiB <br> _11:09s_ | 6896MiB <br> _12:43s_ | 6304MiB <br> _12:52s_ | 5114MiB <br> _13:20s_ |
| 72 <br> frames | 15671MiB <br> _15:14s_ | 11661MiB <br> _17:50s_ | 10325MiB <br> _18:22s_ | 9433MiB <br> _19:32s_ | 7652MiB <br> _19:19s_ |
| 96 <br> frames | -----MiB <br> _--:--s_ | 15534MiB <br> _24:30s_ | 13752MiB <br> _24:28s_ | 12564MiB <br> _25:54s_ | 10188MiB <br> _29:19s_ |
| 120 <br> frames | -----MiB <br> _--:--s_ | 19406MiB <br> _33:04s_ | 17179MiB <br> _33:14s_ | 15694MiB <br> _35:01s_ | 12724MiB <br> _39:11s_ |
- Resolution of 896
| Num <br> frames | normal_mode + 0 steps | normal_mode + 10 steps | normal_mode + 7 steps | normal_mode + 5 steps | normal_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 21085MiB <br> _03:46s_ | 19265MiB <br> _04:25s_ | 18659MiB <br> _04:39s_ | 18248MiB <br> _04:43s_ | 17543MiB <br> _04:59s_ |
| 48 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | 20994MiB <br> _11:34s_ |
| 72 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ |
| 96 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ |
| 120 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ |
| Num <br> frames | low_gpu_mode + 0 steps | low_gpu_mode + 10 steps | low_gpu_mode + 7 steps | low_gpu_mode + 5 steps | low_gpu_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 7128MiB <br> _07:26s_ | 5309MiB <br> _07:45s_ | 4703MiB <br> _07:50s_ | 4517MiB <br> _08:29s_ | 4517MiB <br> _09:31s_ |
| 48 <br> frames | 14220MiB <br> _14:05s_ | 10582MiB <br> _15:01s_ | 9370MiB <br> _15:43s_ | 8562MiB <br> _16:34s_ | 6946MiB <br> _17:36s_ |
| 72 <br> frames | -----MiB <br> _--:--s_ | 15857MiB <br> _24:53s_ | 14038MiB <br> _25:06s_ | 12826MiB <br> _26:49s_ | 10401MiB <br> _27:30s_ |
| 96 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | 17087MiB <br> _38:52s_ | 13853MiB <br> _45:19s_ |
| 120 <br> frames | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | -----MiB <br> _--:--s_ | 17310MiB <br> _55:50s_ |
## A100
### Results from CreateAI
> The GPU memory reported below is the `max_memory_allocated()` value. The values read from nvidia-smi may be higher than the reported values because CUDA occupies some GPU memory (usually between 500 - 800 MiB), and PyTorch's caching mechanism also requests additional GPU memory.
- Resolution of 384
| Num <br> frames | normal_mode + 0 steps | normal_mode + 10 steps | normal_mode + 7 steps | normal_mode + 5 steps | normal_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 15387MiB <br> _00:36s_ | 15052MiB <br> _00:43s_ | 14941MiB <br> _00:50s_ | 14867MiB <br> _00:50s_ | 14717MiB <br> _00:56s_ |
| 48 <br> frames | 16686MiB <br> _00:57s_ | 16018MiB <br> _01:10s_ | 15795MiB <br> _01:28s_ | 15647MiB <br> _01:16s_ | 15350MiB <br> _01:21s_ |
| 72 <br> frames | 17996MiB <br> _01:29s_ | 16993MiB <br> _01:43s_ | 16659MiB <br> _01:53s_ | 16436MiB <br> _01:53s_ | 15990MiB <br> _01:57s_ |
| 96 <br> frames | 19297MiB <br> _02:01s_ | 17959MiB <br> _02:21s_ | 17514MiB <br> _02:29s_ | 17216MiB <br> _02:33s_ | 16621MiB <br> _02:54s_ |
| 120 <br> frames | 20599MiB <br> _02:33s_ | 18928MiB <br> _03:04s_ | 18371MiB <br> _03:12s_ | 17999MiB <br> _03:16s_ | 17256MiB <br> _03:25s_ |
| Num <br> frames | low_gpu_mode + 0 steps | low_gpu_mode + 10 steps | low_gpu_mode + 7 steps | low_gpu_mode + 5 steps | low_gpu_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 4096MiB <br> _03:56s_ | 4096MiB <br> _04:24s_ | 4096MiB <br> _04:16s_ | 4096MiB <br> _04:14s_ | 4095MiB <br> _04:41s_ |
| 48 <br> frames | 4148MiB <br> _04:45s_ | 4148MiB <br> _05:27s_ | 4148MiB <br> _05:29s_ | 4148MiB <br> _05:32s_ | 4147MiB <br> _06:06s_ |
| 72 <br> frames | 4200MiB <br> _06:10s_ | 4200MiB <br> _07:10s_ | 4200MiB <br> _07:13s_ | 4200MiB <br> _07:19s_ | 4200MiB <br> _07:24s_ |
| 96 <br> frames | 5248MiB <br> _07:53s_ | 4249MiB <br> _08:16s_ | 4249MiB <br> _08:30s_ | 4249MiB <br> _08:28s_ | 4249MiB <br> _09:05s_ |
| 120 <br> frames | 6547MiB <br> _08:43s_ | 4876MiB <br> _09:54s_ | 4321MiB <br> _10:16s_ | 4299MiB <br> _10:20s_ | 4299MiB <br> _11:26s_ |
- Resolution of 512
| Num <br> frames | normal_mode + 0 steps | normal_mode + 10 steps | normal_mode + 7 steps | normal_mode + 5 steps | normal_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 16399MiB <br> _00:59s_ | 15805MiB <br> _01:07s_ | 15607MiB <br> _01:13s_ | 15475MiB <br> _01:18s_ | 15211MiB <br> _01:16s_ |
| 48 <br> frames | 18720MiB <br> _01:47s_ | 17532MiB <br> _02:10s_ | 17136MiB <br> _02:25s_ | 16872MiB <br> _02:22s_ | 16344MiB <br> _02:24s_ |
| 72 <br> frames | 21036MiB <br> _02:55s_ | 19254MiB <br> _03:27s_ | 18660MiB <br> _03:36s_ | 18264MiB <br> _03:41s_ | 17472MiB <br> _03:51s_ |
| 96 <br> frames | 23348MiB <br> _04:08s_ | 20972MiB <br> _04:46s_ | 20180MiB <br> _05:05s_ | 19652MiB <br> _05:12s_ | 18596MiB <br> _05:38s_ |
| 120 <br> frames | 25664MiB <br> _05:42s_ | 22694MiB <br> _06:28s_ | 21704MiB <br> _06:47s_ | 21044MiB <br> _06:50s_ | 19724MiB <br> _07:06s_ |
| Num <br> frames | low_gpu_mode + 0 steps | low_gpu_mode + 10 steps | low_gpu_mode + 7 steps | low_gpu_mode + 5 steps | low_gpu_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 4212MiB <br> _05:03s_ | 4213MiB <br> _05:31s_ | 4212MiB <br> _05:23s_ | 4212MiB <br> _05:50s_ | 4212MiB <br> _05:51s_ |
| 48 <br> frames | 4666MiB <br> _06:47s_ | 4401MiB <br> _07:13s_ | 4401MiB <br> _07:34s_ | 4400MiB <br> _08:21s_ | 4400MiB <br> _08:19s_ |
| 72 <br> frames | 6981MiB <br> _08:41s_ | 5199MiB <br> _09:44s_ | 4605MiB <br> _10:10s_ | 4598MiB <br> _11:13s_ | 4598MiB <br> _12:00s_ |
| 96 <br> frames | 9298MiB <br> _10:59s_ | 6922MiB <br> _12:10s_ | 6130MiB <br> _12:16s_ | 5602MiB <br> _13:27s_ | 4794MiB <br> _15:17s_ |
| 120 <br> frames | 11613MiB <br> _13:27s_ | 8643MiB <br> _14:55s_ | 7653MiB <br> _15:22s_ | 6993MiB <br> _16:04s_ | 5673MiB <br> _18:04s_ |
- Resolution of 640
| Num <br> frames | normal_mode + 0 steps | normal_mode + 10 steps | normal_mode + 7 steps | normal_mode + 5 steps | normal_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 17707MiB <br> _01:26s_ | 16778MiB <br> _01:45s_ | 16470MiB <br> _01:48s_ | 16263MiB <br> _01:51s_ | 15850MiB <br> _01:53s_ |
| 48 <br> frames | 21324MiB <br> _03:09s_ | 19468MiB <br> _03:36s_ | 18849MiB <br> _03:53s_ | 18437MiB <br> _03:53s_ | 17612MiB <br> _04:03s_ |
| 72 <br> frames | 24943MiB <br> _05:06s_ | 22158MiB <br> _05:54s_ | 21230MiB <br> _06:12s_ | 20612MiB <br> _06:18s_ | 19374MiB <br> _06:35s_ |
| 96 <br> frames | 28562MiB <br> _07:35s_ | 24848MiB <br> _08:36s_ | 23611MiB <br> _09:04s_ | 22785MiB <br> _09:07s_ | 21134MiB <br> _09:35s_ |
| 120 <br> frames | 32182MiB <br> _10:32s_ | 27541MiB <br> _11:54s_ | 25994MiB <br> _12:20s_ | 24962MiB <br> _12:37s_ | 22899MiB <br> _13:05s_ |
| Num <br> frames | low_gpu_mode + 0 steps | low_gpu_mode + 10 steps | low_gpu_mode + 7 steps | low_gpu_mode + 5 steps | low_gpu_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 4301MiB <br> _05:40s_ | 4301MiB <br> _06:07s_ | 4301MiB <br> _06:27s_ | 4301MiB <br> _06:33s_ | 4301MiB <br> _06:57s_ |
| 48 <br> frames | 7271MiB <br> _09:18s_ | 5416MiB <br> _09:51s_ | 4798MiB <br> _10:26s_ | 4589MiB <br> _10:36s_ | 4589MiB <br> _11:52s_ |
| 72 <br> frames | 10889MiB <br> _13:03s_ | 8106MiB <br> _14:07s_ | 7179MiB <br> _14:35s_ | 6561MiB <br> _17:30s_ | 5322MiB <br> _16:33s_ |
| 96 <br> frames | 14509MiB <br> _16:57s_ | 10795MiB <br> _18:12s_ | 9557MiB <br> _18:29s_ | 8732MiB <br> _18:55s_ | 7082MiB <br> _21:10s_ |
| 120 <br> frames | 18128MiB <br> _22:05s_ | 13487MiB <br> _22:49s_ | 11942MiB <br> _23:47s_ | 10910MiB <br> _23:50s_ | 8847MiB <br> _32:46s_ |
- Resolution of 768
| Num <br> frames | normal_mode + 0 steps | normal_mode + 10 steps | normal_mode + 7 steps | normal_mode + 5 steps | normal_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 19300MiB <br> _02:09s_ | 17962MiB <br> _02:32s_ | 17517MiB <br> _02:37s_ | 17219MiB <br> _02:39s_ | 16624MiB <br> _02:46s_ |
| 48 <br> frames | 24510MiB <br> _04:46s_ | 21837MiB <br> _05:32s_ | 20946MiB <br> _05:51s_ | 20352MiB <br> _05:59s_ | 19164MiB <br> _06:10s_ |
| 72 <br> frames | 29725MiB <br> _08:23s_ | 25715MiB <br> _09:35s_ | 24378MiB <br> _09:56s_ | 23486MiB <br> _10:05s_ | 21703MiB <br> _10:40s_ |
| 96 <br> frames | 34925MiB <br> _12:52s_ | 29579MiB <br> _14:33s_ | 27797MiB <br> _15:01s_ | 26609MiB <br> _15:13s_ | 24233MiB <br> _15:53s_ |
| 120 <br> frames | 40134MiB <br> _18:29s_ | 33451MiB <br> _20:39s_ | 31223MiB <br> _21:31s_ | 29737MiB <br> _21:42s_ | 26766MiB <br> _22:51s_ |
| Num <br> frames | low_gpu_mode + 0 steps | low_gpu_mode + 10 steps | low_gpu_mode + 7 steps | low_gpu_mode + 5 steps | low_gpu_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 5251MiB <br> _08:04s_ | 4398MiB <br> _08:15s_ | 4398MiB <br> _08:21s_ | 4398MiB <br> _08:48s_ | 4399MiB <br> _09:14s_ |
| 48 <br> frames | 10457MiB <br> _12:42s_ | 7786MiB <br> _13:54s_ | 6896MiB <br> _14:17s_ | 6304MiB <br> _14:37s_ | 5114MiB <br> _16:21s_ |
| 72 <br> frames | 15671MiB <br> _18:18s_ | 11661MiB <br> _19:44s_ | 10325MiB <br> _20:53s_ | 9435MiB <br> _20:47s_ | 7652MiB <br> _23:42s_ |
| 96 <br> frames | 20880MiB <br> _24:55s_ | 15534MiB <br> _27:07s_ | 13752MiB <br> _28:13s_ | 12564MiB <br> _28:44s_ | 10188MiB <br> _30:30s_ |
| 120 <br> frames | 26092MiB <br> _31:21s_ | 19406MiB <br> _34:09s_ | 17179MiB <br> _35:13s_ | 15694MiB <br> _35:15s_ | 12724MiB <br> _32:55s_ |
- Resolution of 896
| Num <br> frames | normal_mode + 0 steps | normal_mode + 10 steps | normal_mode + 7 steps | normal_mode + 5 steps | normal_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 21181MiB <br> _02:58s_ | 19362MiB <br> _03:32s_ | 18755MiB <br> _03:48s_ | 18351MiB <br> _03:47s_ | 17543MiB <br> _03:54s_ |
| 48 <br> frames | 28271MiB <br> _07:18s_ | 24633MiB <br> _08:24s_ | 23420MiB <br> _08:49s_ | 22611MiB <br> _08:53s_ | 20994MiB <br> _09:10s_ |
| 72 <br> frames | 35360MiB <br> _13:22s_ | 29902MiB <br> _14:57s_ | 28083MiB <br> _15:33s_ | 26870MiB <br> _15:46s_ | 24445MiB <br> _16:13s_ |
| 96 <br> frames | 42447MiB <br> _21:08s_ | 35169MiB <br> _23:48s_ | 32744MiB <br> _24:41s_ | 31126MiB <br> _25:08s_ | 27891MiB <br> _26:33s_ |
| 120 <br> frames | 49538MiB <br> _30:55s_ | 40442MiB <br> _34:03s_ | 37410MiB <br> _35:11s_ | 35388MiB <br> _35:31s_ | 31345MiB <br> _36:40s_ |
| Num <br> frames | low_gpu_mode + 0 steps | low_gpu_mode + 10 steps | low_gpu_mode + 7 steps | low_gpu_mode + 5 steps | low_gpu_mode + 1 steps |
| :---: | :---: | :---: | :---: | :---: | :---: |
| 24 <br> frames | 7128MiB <br> _08:52s_ | 5309MiB <br> _09:30s_ | 4702MiB <br> _10:37s_ | 4517MiB <br> _10:55s_ | 4517MiB <br> _12:09s_ |
| 48 <br> frames | 14220MiB <br> _15:56s_ | 10582MiB <br> _17:04s_ | 9370MiB <br> _18:49s_ | 8562MiB <br> _19:08s_ | 6946MiB <br> _21:50s_ |
| 72 <br> frames | 21315MiB <br> _24:58s_ | 15857MiB <br> _26:54s_ | 14038MiB <br> _29:10s_ | 12826MiB <br> _29:42s_ | 10401MiB <br> _30:58s_ |
| 96 <br> frames | 28407MiB <br> _31:16s_ | 21130MiB <br> _34:22s_ | 18704MiB <br> _36:45s_ | 17087MiB <br> _36:34s_ | 13853MiB <br> _43:48s_ |
| 120 <br> frames | 35504MiB <br> _45:00s_ | 26405MiB <br> _48:05s_ | 23373MiB <br> _49:27s_ | 21352MiB <br> _50:34s_ | 17310MiB <br> _59:38s_ |
predict_i2v.py
+1 -4
View File
@@ -201,10 +201,7 @@ else:
height, width = video_size
# Set hidden states offload steps
if pipeline.model_cpu_offload_flag:
pipeline.transformer.hidden_cache_size = gpu_offload_steps
else:
pipeline.transformer.hidden_cache_size = 0
pipeline.transformer.hidden_cache_size = gpu_offload_steps
# Load Sampler
if scheduler_name == "DPM++":
predict_i2v_80g.py
+1 -4
View File
@@ -201,10 +201,7 @@ else:
height, width = video_size
# Set hidden states offload steps
if pipeline.model_cpu_offload_flag:
pipeline.transformer.hidden_cache_size = gpu_offload_steps
else:
pipeline.transformer.hidden_cache_size = 0
pipeline.transformer.hidden_cache_size = gpu_offload_steps
# Load Sampler
if scheduler_name == "DPM++":