handle gpu without fan

.vscode/launch.json

@@ -11,6 +11,7 @@
             "program": "${file}",
             "cwd": "${workspaceFolder}",
             "console": "integratedTerminal",
+            "justMyCode": false,
             "args": [
                 "--verbose",
             ],
@@ -54,7 +55,8 @@
             "program": "${workspaceFolder}/gpuctl/gpu_main.py",
             "console": "integratedTerminal",
             "args": [
-                "--slots", "0000:01:00.0", 
+                // "--slots", "0000:01:00.0", 
+                // "--slots", "0000:81:00.0", 
                 "--set-speed", "50"
             ],
             "env": {

.vscode/settings.json

@@ -6,8 +6,11 @@
         "-p",
         "test_*.py"
     ],
-    "python.testing.pytestEnabled": false,
+    "python.testing.pytestEnabled": true,
     "python.testing.nosetestsEnabled": false,
-    "python.testing.unittestEnabled": true,
-    "python.pythonPath": "venv/bin/python"
+    "python.testing.unittestEnabled": false,
+    "python.pythonPath": "venv/bin/python",
+    "python.testing.pytestArgs": [
+        "tests"
+    ]
 }

README.md

@@ -16,7 +16,7 @@ Currently, the example of action script are based on the [miner's remote managem
 ## Environment and Installation
 
 Intel CPU/16G RAM
-Ubuntu 18.4/Python3
+HiveOS/Ubuntu 18.4/Python3
 
 * Environment setup
 
@@ -231,11 +231,9 @@ If a failure is detected, the gpuctl will invoke the given script with slot name
     ```
 
     ```shell
-    ID Slot Name    Vendor   PCI-ID      Temp. Fan  PWR     Working
-    -- ------------ -------- ----------- ----- ---- ------- -------
-    1 0000:01:00.0  NVIDIA   [10DE:1C03]   61c  50%  74.76w True
-    2 0000:0b:00.0  AMD      [1002:67DF]   78c  47%  81.00w True
-    3 0000:0d:00.0  NVIDIA   [10DE:1C03]   49c  50%  72.60w True
+    Set slot 0000:01:00.0 fan speed to 50%
+    Set slot 0000:0b:00.0 fan speed to 50%
+    Set slot 0000:0d:00.0 fan speed to 50%
     ```
 
 * Example 3) If the temperature of a GPU is over 65c, activate the fan speed control for the specific GPU.
@@ -522,3 +520,4 @@ Use PhoenixMiner 5.3b as example:
 * [GPUFan](https://github.com/milani/gpufan)
 * [PyOpenCL Samples](https://github.com/virus-warnning/pyopencl_samples)
 * [Associating OpenCL device ids with GPUs](https://anteru.net/blog/2014/associating-opencl-device-ids-with-gpus/)
+* [NVIDIA/Tips and tricks](https://wiki.archlinux.org/title/NVIDIA/Tips_and_tricks)

gpuctl/eth_ctl.py

@@ -198,6 +198,11 @@ class EthCtl:
     def __init__(self, **kwargs):
 
         valid_keys = ["base", "temp", "rate", "wait", "rmode", "delay", "script", "verbose"]
+
+        for k in kwargs.keys():
+            if k not in valid_keys:
+                return None
+
         for key in valid_keys:
             setattr(self, key, kwargs.get(key))
 

gpuctl/eth_main.py

@@ -8,7 +8,7 @@
 import syslog
 
 from gpuctl import __version__
-from gpuctl import DRYRUN, GpuCtl, logger
+from gpuctl import DRYRUN, logger
 from gpuctl import PciDev, GpuDev, GpuAMD, GpuNV
 
 from gpuctl import EthCtl, scan_miner

gpuctl/gpu_amd.py

@@ -79,6 +79,8 @@ def set_speed(self, speed):
         self.speed = speed
         logger.debug(f'[{self.pci_dev.slot_name}/{self.name}] set speed {speed}% pwm {pwm}')
 
+        return True
+
     def get_speed(self):
         pwn = fv.read_file_value(self.hwmon_dir, GpuAMD.PWM_VAL)
         speed = int(pwn/(self.max-self.min) * 100)

gpuctl/gpu_ctl.py

@@ -28,8 +28,13 @@ def __init__(self, **kwargs):
         self.vendors = None
 
         # overwrite default value with arguments
-        valid_keys = ["slots", "gpu_devices", 
+        valid_keys = ["gpu_devices", 
                       "fan", "curve", "delta", "temp", "tas", "verbose"]
+
+        for k in kwargs.keys():
+            if k not in valid_keys:
+                return None
+
         for key in valid_keys:
             setattr(self, key, kwargs.get(key))
 
@@ -170,7 +175,9 @@ def stop(self):
         self.thread.join()
 
     def set_interval(self, intvl=None, wait_period=None):
-
+        """
+        wait interval must greater than interval
+        """
         interval = intvl if intvl else self.interval
         wait_period = wait_period if wait_period else self.wait
 

gpuctl/gpu_main.py

@@ -161,6 +161,16 @@ def sigstop(a, b):
             if gpu_dev and gpu_dev.is_gpu() and pdev.slot_name not in slot_names:
                 gpu_devices.append(gpu_dev)
 
+    if args.set_speed != None:
+        for gpu in gpu_devices:
+            pdev = gpu.pci_dev
+            # set fan speed
+            print(f"Set slot {pdev.slot_name} fan speed to {args.set_speed}%")
+            rv = gpu.set_speed(args.set_speed)
+            if not rv:
+                print(f"Failed to set slot {pdev.slot_name} fan speed !!!")
+        sys.exit(0)
+
     # list monitored devices
     print("\n")
     print("ID Slot Name    Vendor   PCI-ID      Temp. Fan  PWR     Working")
@@ -169,12 +179,13 @@ def sigstop(a, b):
     cnt = 1
     for gpu in gpu_devices:
         pdev = gpu.pci_dev
-        # set fan speed
-        if args.set_speed != None:
-            gpu.set_speed(args.set_speed)
         working = gpu.is_working()
+        t_str = gpu.get_temperature() if gpu.get_temperature() else '--'
+        s_str = gpu.get_speed() if gpu.get_speed() else '--'
+        p_str = gpu.get_pwr() if gpu.get_pwr() else '--'
+
         msg  = f"{cnt:2} {pdev.slot_name} {pdev.vendor_name():8} [{pdev.vendor_id}:{pdev.device_id}] "
-        msg += f"{gpu.get_temperature():4}c {gpu.get_speed():3}% {gpu.get_pwr():6.2f}w {working}"
+        msg += f"{t_str:4}c {s_str:3}% {p_str:6.2f}w {working}"
         print(msg)
         cnt += 1
 
@@ -183,16 +194,14 @@ def sigstop(a, b):
     if args.list:
         sys.exit(0)
 
-    if args.set_speed != None:
-        sys.exit(0)
-
     if len(gpu_devices) == 0:
         print('No GPU found, abort !\n')
         sys.exit(0)
 
     # remove not working devices
     for gpu in list(gpu_devices):
         if gpu.is_working() == False:
+            print(f'slot {gpu.pci_dev.slot_name} is malfunction, removed !\n')
             gpu_devices.remove(gpu)
 
     gpu_ctl = GpuCtl(gpu_devices=gpu_devices, 

gpuctl/gpu_mock.py

@@ -25,20 +25,22 @@ def get_speed(self):
 
     def set_speed(self, percentage):
         self.speed = percentage
+        return True
 
 class GpuNak(GpuDev):
     def is_gpu(self):
         return True
 
     def is_working(self):
-        return True
+        return False
 
     def get_temperature(self):
-        return 70
+        return None
 
     def get_speed(self):
-        return 10
+        return None
 
     def set_speed(self, percentage):
         self.speed = percentage
+        return False
 

gpuctl/gpu_nv.py

@@ -65,55 +65,62 @@ def is_gpu(self):
     def set_speed(self, speed):
 
         if DRYRUN:
-            return
+            return True
 
         if self.nvh == None:
-            return
+            return False
 
         cmd = f"nvidia-settings -c {self.display} -a [gpu:{self.nv_id}]/GPUFanControlState=1 -a [fan:{self.nv_id}]/GPUTargetFanSpeed={speed}"
         # logger.debug(f'exec: {cmd}')
+        rv = True
         try:
             sc.exec_cmd(cmd)
             logger.debug(
                 f'[{self.pci_dev.slot_name}/{self.name}] set speed {speed}%')
             self.speed = speed
         except:
-            logger.error(f"{self.pci_dev.slot_name}/{self.name}] set fan speed failed !!")
+            # some GPU card does not install fan (ie. Tesla P4)
+            # logger.error(f"{self.pci_dev.slot_name}/{self.name}] set fan speed failed !!")
+            rv = False
+
+        return rv
 
     def get_speed(self):
         # NOTE: nvmlDeviceGetFanSpeed report wrong speed, use nvidia-settings instead
         # s = nv.nvmlDeviceGetFanSpeed(self.nvh)
         cmd = f"nvidia-settings -t -q  [fan:{self.nv_id}]/GPUTargetFanSpeed"
         # logger.debug(f'exec: {cmd}')
-        speed = 0
+        speed = None
         try:
             s = sc.exec_cmd(cmd)
             speed = int(s)
             # logger.debug(
             #     f'[{self.pci_dev.slot_name}/{self.name}] speed {speed}%')
         except:
-            logger.error(f"{self.pci_dev.slot_name}/{self.name}] get fan speed failed !!")
+            # some GPU card does not install fan (ie. Tesla P4)
+            # logger.error(f"{self.pci_dev.slot_name}/{self.name}] get fan speed failed !!")
+            pass
         return speed
 
     def get_temperature(self):
         if self.nvh == None:
             return None
+        t = None
         try:
             t = nv.nvmlDeviceGetTemperature(self.nvh, nv.NVML_TEMPERATURE_GPU)
             self.temperature = t
         except:
             logger.error(f"{self.pci_dev.slot_name}/{self.name}] get temperature failed !!")
-            t = None
         return t
 
     def get_pwr(self):
         if self.nvh == None:
-            return 0
+            return None
+        pwr = None
         try:
             pwr = nv.nvmlDeviceGetPowerUsage(self.nvh)/1000
             self.pwr = pwr
         except:
             logger.error(f"{self.pci_dev.slot_name}/{self.name}] get pwr failed !!")
-            pwr = 0
 
         return pwr

gpuctl/version.py

@@ -2,5 +2,5 @@
 
 __all__ = ['__version__']
 
-__version__ = '0.3.6'
+__version__ = '0.3.7'
 

scripts/nv-temp.sh

@@ -0,0 +1,5 @@
+#!/bin/bash
+
+# get temperature of NV cards
+
+nvidia-settings -q gpucoretemp -t

tests/test_gpu.py

@@ -3,22 +3,43 @@
 from re import U
 import unittest
 
-from gpuctl import PciDev, GpuDev, GpuCtl
+from gpuctl import PciDev, GpuDev, GpuAMD, GpuNV
 
-class TestSlot(unittest.TestCase):
-
-    def test_valid_slot(self):
-        pass
-
-    def test_invalid_slot(self):
-        gpu_ctl = None
-        try:
-            gpu_ctl = GpuCtl(slot_name='aaaa')
-            self.assertFalse(True)
-        except:
-            pass
 
 class TestGpu(unittest.TestCase):
+    def test_get_temp(self):
+        vendors = ['AMD', 'NVIDIA']
+        pci_devices = PciDev.discovery(vendor_filter=vendors)
+        gpu_devices = []
+        for pdev in pci_devices:
+            gpu = None
+            if pdev.is_amd():
+                gpu = GpuAMD(pdev)
+            if pdev.is_nvidia():
+                gpu = GpuNV(pdev)
+            if gpu and gpu.is_gpu():
+                gpu_devices.append(gpu)
+
+        for gpu in gpu_devices:
+            gpu.get_temperature()
+
+    def test_get_speed(self):
+        vendors = ['AMD', 'NVIDIA']
+        pci_devices = PciDev.discovery(vendor_filter=vendors)
+        gpu_devices = []
+        for pdev in pci_devices:
+            gpu = None
+            if pdev.is_amd():
+                gpu = GpuAMD(pdev)
+            if pdev.is_nvidia():
+                gpu = GpuNV(pdev)
+            if gpu and gpu.is_gpu():
+                gpu_devices.append(gpu)
+
+        for gpu in gpu_devices:
+            gpu.get_speed()
+
+class TestCurve(unittest.TestCase):
     def test_valid_curve(self):
         curve = [[0,0], [10,10], [50,50], [100,100]]
         rv = GpuDev.check_curve(curve)

tests/test_gpuctl.py

@@ -9,6 +9,15 @@
 from gpuctl import GpuOk, GpuNak
 
 class TestGpuCtl(unittest.TestCase):
+
+    def test_invalid_key(self):
+        gpu_ctl = None
+        try:
+            gpu_ctl = GpuCtl(slots='aaaa')
+            self.assertEqual(gpu_ctl, None)
+        except:
+            pass
+
     def test_discover_all(self):
 
         pci_devices = PciDev.discovery()
@@ -68,9 +77,12 @@ def test_over_temp(self):
         gpu_dev = GpuNak(pdev)
         self.assertIsNotNone(gpu_dev)
 
-        gpu_ctl = GpuCtl(gpu_devices=[gpu_dev], fam=10, temp=20, tas='./tests/ok.sh')
+        gpu_ctl = GpuCtl(gpu_devices=[gpu_dev], fan=10, temp=20, tas='./tests/ok.sh')
+        self.assertNotEqual(gpu_ctl, None)
+
+        rv = gpu_ctl.set_interval(wait_period=10)
+        self.assertTrue(rv)
 
-        gpu_ctl.set_interval(wait_period=3)
         gpu_ctl.start()
         time.sleep(5)
         gpu_ctl.stop()
@@ -96,5 +108,24 @@ def test_interval(self):
         rv = gpu_ctl.set_interval(intvl=2, wait_period=1)
         self.assertFalse(rv)
 
+    def test_nak_gpu(self):
+        slot_name = '1111:11:11.1'
+        pci_id = '1111:1111'
+        pdev = PciDev(slot_name, pci_id, 'mock pci')
+        self.assertIsNotNone(pdev)
+        self.assertEqual(pdev.vendor_name(), 'Other')
+
+        gpu_dev = GpuNak(pdev)
+        self.assertIsNotNone(gpu_dev)
+
+        gpu_ctl = GpuCtl(gpu_devices=[gpu_dev], verbose=True)
+
+        rv = gpu_ctl.set_interval(wait_period=10)
+        self.assertTrue(rv)
+
+        gpu_ctl.start()
+        time.sleep(5)
+        gpu_ctl.stop()
+
 if __name__ == '__main__':
     unittest.main()