From cd6c55c1ab14fcab15329cde599cf79e8d555657 Mon Sep 17 00:00:00 2001 From: pepe10-gpu Date: Sun, 6 Nov 2022 17:05:51 -0800 Subject: 16xx card fix cudnn --- modules/devices.py | 3 +++ 1 file changed, 3 insertions(+) (limited to 'modules/devices.py') diff --git a/modules/devices.py b/modules/devices.py index 7511e1dc..858bf399 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -39,10 +39,13 @@ def torch_gc(): def enable_tf32(): if torch.cuda.is_available(): + torch.backends.cudnn.benchmark = True + torch.backends.cudnn.enabled = True torch.backends.cuda.matmul.allow_tf32 = True torch.backends.cudnn.allow_tf32 = True + errors.run(enable_tf32, "Enabling TF32") device = device_interrogate = device_gfpgan = device_swinir = device_esrgan = device_scunet = device_codeformer = None -- cgit v1.2.3 From 29eff4a194d22f0f0e7a7a976d746a71a4193cf5 Mon Sep 17 00:00:00 2001 From: pepe10-gpu Date: Mon, 7 Nov 2022 18:06:48 -0800 Subject: terrible hack --- modules/devices.py | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) (limited to 'modules/devices.py') diff --git a/modules/devices.py b/modules/devices.py index 858bf399..4c63f465 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -39,8 +39,15 @@ def torch_gc(): def enable_tf32(): if torch.cuda.is_available(): - torch.backends.cudnn.benchmark = True - torch.backends.cudnn.enabled = True + #TODO: make this better; find a way to check if it is a turing card + turing = ["1630","1650","1660","Quadro RTX 3000","Quadro RTX 4000","Quadro RTX 4000","Quadro RTX 5000","Quadro RTX 5000","Quadro RTX 6000","Quadro RTX 6000","Quadro RTX 8000","Quadro RTX T400","Quadro RTX T400","Quadro RTX T600","Quadro RTX T1000","Quadro RTX T1000","2060","2070","2080","Titan RTX","Tesla T4","MX450","MX550"] + for devid in range(0,torch.cuda.device_count()): + for i in turing: + if i in torch.cuda.get_device_name(devid): + shd = True + if shd: + torch.backends.cudnn.benchmark = True + torch.backends.cudnn.enabled = True torch.backends.cuda.matmul.allow_tf32 = True torch.backends.cudnn.allow_tf32 = True -- cgit v1.2.3 From 62e9fec3df8518da3a2c35fa090bb54946c856b2 Mon Sep 17 00:00:00 2001 From: pepe10-gpu Date: Tue, 8 Nov 2022 15:19:09 -0800 Subject: actual better fix thanks C43H66N12O12S2 --- modules/devices.py | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-) (limited to 'modules/devices.py') diff --git a/modules/devices.py b/modules/devices.py index 4c63f465..058a5e00 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -39,12 +39,9 @@ def torch_gc(): def enable_tf32(): if torch.cuda.is_available(): - #TODO: make this better; find a way to check if it is a turing card - turing = ["1630","1650","1660","Quadro RTX 3000","Quadro RTX 4000","Quadro RTX 4000","Quadro RTX 5000","Quadro RTX 5000","Quadro RTX 6000","Quadro RTX 6000","Quadro RTX 8000","Quadro RTX T400","Quadro RTX T400","Quadro RTX T600","Quadro RTX T1000","Quadro RTX T1000","2060","2070","2080","Titan RTX","Tesla T4","MX450","MX550"] for devid in range(0,torch.cuda.device_count()): - for i in turing: - if i in torch.cuda.get_device_name(devid): - shd = True + if torch.cuda.get_device_capability(devid) == (7, 5): + shd = True if shd: torch.backends.cudnn.benchmark = True torch.backends.cudnn.enabled = True -- cgit v1.2.3 From 0fddb4a1c06a6e2122add7eee3b001a6d473baee Mon Sep 17 00:00:00 2001 From: brkirch Date: Wed, 30 Nov 2022 08:02:39 -0500 Subject: Rework MPS randn fix, add randn_like fix torch.manual_seed() already sets a CPU generator, so there is no reason to create a CPU generator manually. torch.randn_like also needs a MPS fix for k-diffusion, but a torch hijack with randn_like already exists so it can also be used for that. --- modules/devices.py | 15 +++------------ 1 file changed, 3 insertions(+), 12 deletions(-) (limited to 'modules/devices.py') diff --git a/modules/devices.py b/modules/devices.py index f00079c6..046460fa 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -66,24 +66,15 @@ dtype_vae = torch.float16 def randn(seed, shape): - # Pytorch currently doesn't handle setting randomness correctly when the metal backend is used. - if device.type == 'mps': - generator = torch.Generator(device=cpu) - generator.manual_seed(seed) - noise = torch.randn(shape, generator=generator, device=cpu).to(device) - return noise - torch.manual_seed(seed) + if device.type == 'mps': + return torch.randn(shape, device=cpu).to(device) return torch.randn(shape, device=device) def randn_without_seed(shape): - # Pytorch currently doesn't handle setting randomness correctly when the metal backend is used. if device.type == 'mps': - generator = torch.Generator(device=cpu) - noise = torch.randn(shape, generator=generator, device=cpu).to(device) - return noise - + return torch.randn(shape, device=cpu).to(device) return torch.randn(shape, device=device) -- cgit v1.2.3 From 2651267e3af5886b8b6b1dc3023f2507f7079118 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Sat, 3 Dec 2022 15:57:52 +0300 Subject: fix #4407 breaking UI entirely for card other than ones related to the PR --- modules/devices.py | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) (limited to 'modules/devices.py') diff --git a/modules/devices.py b/modules/devices.py index 1325569c..547ea46c 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -53,12 +53,10 @@ def torch_gc(): def enable_tf32(): if torch.cuda.is_available(): - for devid in range(0,torch.cuda.device_count()): - if torch.cuda.get_device_capability(devid) == (7, 5): - shd = True - if shd: + if any([torch.cuda.get_device_capability(devid) == (7, 5) for devid in range(0, torch.cuda.device_count())]): torch.backends.cudnn.benchmark = True torch.backends.cudnn.enabled = True + torch.backends.cuda.matmul.allow_tf32 = True torch.backends.cudnn.allow_tf32 = True -- cgit v1.2.3 From 46b0d230e7c13e247eabb22e1103ce512e7ed6b1 Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Sat, 3 Dec 2022 16:01:23 +0300 Subject: add comment for #4407 and remove seemingly unnecessary cudnn.enabled --- modules/devices.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) (limited to 'modules/devices.py') diff --git a/modules/devices.py b/modules/devices.py index 547ea46c..d6a76844 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -53,9 +53,11 @@ def torch_gc(): def enable_tf32(): if torch.cuda.is_available(): + + # enabling benchmark option seems to enable a range of cards to do fp16 when they otherwise can't + # see https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/4407 if any([torch.cuda.get_device_capability(devid) == (7, 5) for devid in range(0, torch.cuda.device_count())]): torch.backends.cudnn.benchmark = True - torch.backends.cudnn.enabled = True torch.backends.cuda.matmul.allow_tf32 = True torch.backends.cudnn.allow_tf32 = True -- cgit v1.2.3 From b6e5edd74657e3fd1fbd04f341b7a84625d4aa7a Mon Sep 17 00:00:00 2001 From: AUTOMATIC <16777216c@gmail.com> Date: Sat, 3 Dec 2022 18:06:33 +0300 Subject: add built-in extension system add support for adding upscalers in extensions move LDSR, ScuNET and SwinIR to built-in extensions --- modules/devices.py | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) (limited to 'modules/devices.py') diff --git a/modules/devices.py b/modules/devices.py index d6a76844..f8cffae1 100644 --- a/modules/devices.py +++ b/modules/devices.py @@ -44,6 +44,15 @@ def get_optimal_device(): return cpu +def get_device_for(task): + from modules import shared + + if task in shared.cmd_opts.use_cpu: + return cpu + + return get_optimal_device() + + def torch_gc(): if torch.cuda.is_available(): with torch.cuda.device(get_cuda_device_string()): @@ -67,7 +76,7 @@ def enable_tf32(): errors.run(enable_tf32, "Enabling TF32") cpu = torch.device("cpu") -device = device_interrogate = device_gfpgan = device_swinir = device_esrgan = device_scunet = device_codeformer = None +device = device_interrogate = device_gfpgan = device_esrgan = device_codeformer = None dtype = torch.float16 dtype_vae = torch.float16 -- cgit v1.2.3