#3203 additional info in cuda detect by rawwar · Pull Request #6809 · numba/numba

rawwar · 2021-03-09T15:06:01Z

PR related to Issue #3203.

numba.cuda.detect() now prints Watchdog Enabled Status, Compute Mode and Single precision to Double-precision performance ratio

…info_in_cuda_detect

gmarkall · 2021-03-09T15:27:18Z

On my Windows machine:

(numba) C:\Users\Graham Markall> python -c "from numba import cuda; cuda.detect()"
c:\work\numbadev\numba\numba\__init__.py:106: UserWarning: llvmlite version format not recognized!
  warnings.warn("llvmlite version format not recognized!")
Found 2 CUDA devices
id 0      b'Quadro RTX 8000'                              [SUPPORTED]
                      compute capability: 7.5
                           pci device id: 0
                              pci bus id: 21
                        Watchdog Enabled: 0
                            Compute Mode: TCC
             FP32/FP64 Performance Ratio: 32
id 1      b'Quadro RTX 8000'                              [SUPPORTED]
                      compute capability: 7.5
                           pci device id: 0
                              pci bus id: 45
                        Watchdog Enabled: 1
                            Compute Mode: WDDM
             FP32/FP64 Performance Ratio: 32
Summary:
        2/2 devices are supported

The watchdog status and driver mode aligns with nvidia-smi output:

(numba) C:\Users\Graham Markall> nvidia-smi
Tue Mar  9 15:25:56 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 461.40       Driver Version: 461.40       CUDA Version: 11.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name            TCC/WDDM | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  Quadro RTX 8000     TCC  | 00000000:15:00.0 Off |                  Off |
| 33%   33C    P8    31W / 260W |      1MiB / 48767MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
|   1  Quadro RTX 8000    WDDM  | 00000000:2D:00.0  On |                  Off |
| 33%   44C    P8    35W / 260W |    918MiB / 49152MiB |     12%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

gmarkall · 2021-03-09T15:30:20Z

On my Linux machine:

$ python -c "from numba import cuda; cuda.detect()"
Found 2 CUDA devices
id 0    b'NVIDIA Quadro RTX 8000'                              [SUPPORTED]
                      compute capability: 7.5
                           pci device id: 0
                              pci bus id: 21
                        Watchdog Enabled: 1
                            Compute Mode: WDDM
             FP32/FP64 Performance Ratio: 32
id 1    b'NVIDIA Quadro RTX 8000'                              [SUPPORTED]
                      compute capability: 7.5
                           pci device id: 0
                              pci bus id: 45
                        Watchdog Enabled: 0
                            Compute Mode: WDDM
             FP32/FP64 Performance Ratio: 32
Summary:
	2/2 devices are supported

The "Watchdog Enabled" status matches the "Interactive" setting in xorg.conf (as expected). However, showing a "Compute Mode" doesn't make sense here.

rawwar · 2021-03-09T15:45:53Z

The "Watchdog Enabled" status matches the "Interactive" setting in xorg.conf (as expected). However, showing a "Compute Mode" doesn't make sense here.

Should I show this information only on NT platform?

EDIT: Please correct me if I am wrong. Once can understand that the compute mode is WDDM if Watchdog is Enabled right? Do you think this information is redundant?

EDIT2: Wondering if the output is consistent with nvidia-smi's output on linux machine?

EDIT3: I was going through this link and understand that TCC is a driver for windows. So, I am assuming that you meant, printing Compute Mode on linux machines is not relevant. Am i correct?

gmarkall · 2021-03-09T17:42:39Z

The "Watchdog Enabled" status matches the "Interactive" setting in xorg.conf (as expected). However, showing a "Compute Mode" doesn't make sense here.

Should I show this information only on NT platform?

Yes, WDDM / TCC should be Windows-only.

EDIT: Please correct me if I am wrong. Once can understand that the compute mode is WDDM if Watchdog is Enabled right? Do you think this information is redundant?

I'm not sure whether you can have a watchdog in TCC mode - maybe you can. You can have a watchdog in Linux even though there is no WDDM / TCC. I would leave both the Compute Mode and Watchdog Enabled fields in.

EDIT2: Wondering if the output is consistent with nvidia-smi's output on linux machine?

It was, yes. Apologies for the omission.

EDIT3: I was going through this link and understand that TCC is a driver for windows. So, I am assuming that you meant, printing Compute Mode on linux machines is not relevant. Am i correct?

That is correct.

A couple of other small suggestions:

As part of this PR, could you fix the capitalisation of the other fields to match your additions (e.g. "compute capability" -> "Compute Capability", etc.)
Displaying "Watchdog: Enabled" or "Watchdog: Disabled" would be a bit nicer than "Watchdog Enabled: 0" (or "1").

rawwar · 2021-03-09T17:49:31Z

I updated the code with capitalisation, watchdog output and the following is the output now

In [3]: cuda.detect()
Found 1 CUDA devices
id 0    b'GeForce GTX 1660 Ti'                              [SUPPORTED]
                      Compute Capability: 7.5
                           PCI Device ID: 0
                              PCI Bus ID: 1
                                Watchdog: Enabled
                            Compute Mode: WDDM
             FP32/FP64 Performance Ratio: 32
Summary:
        1/1 devices are supported

rawwar · 2021-03-09T17:54:32Z

My bad. Did not realize commit to my repo's branch would automatically trigger build

gmarkall · 2021-03-09T23:19:19Z

Thanks for the updates! Will re-review / re-test this tomorrow.

gmarkall · 2021-03-10T15:06:39Z

From my Linux machine:

$ python -c "from numba import cuda; cuda.detect()"
Found 2 CUDA devices
id 0    b'NVIDIA Quadro RTX 8000'                              [SUPPORTED]
                      Compute Capability: 7.5
                           PCI Device ID: 0
                              PCI Bus ID: 21
                                Watchdog: Enabled
             FP32/FP64 Performance Ratio: 32
id 1    b'NVIDIA Quadro RTX 8000'                              [SUPPORTED]
                      Compute Capability: 7.5
                           PCI Device ID: 0
                              PCI Bus ID: 45
                                Watchdog: Disabled
             FP32/FP64 Performance Ratio: 32
Summary:
	2/2 devices are supported

Looks good, and agrees with nvidia-smi. Will check on Windows shortly.

gmarkall · 2021-03-10T15:11:28Z

On Windows:

id 0      b'Quadro RTX 8000'                              [SUPPORTED]
                      Compute Capability: 7.5
                           PCI Device ID: 0
                              PCI Bus ID: 21
                                Watchdog: Disabled
                            Compute Mode: TCC
             FP32/FP64 Performance Ratio: 32
id 1      b'Quadro RTX 8000'                              [SUPPORTED]
                      Compute Capability: 7.5
                           PCI Device ID: 0
                              PCI Bus ID: 45
                                Watchdog: Enabled
                            Compute Mode: WDDM
             FP32/FP64 Performance Ratio: 32

Also looking good and agreeing with nvidia-smi.

gmarkall

Thanks for your efforts and the updates here - this looks good to me!

@esc Could this have a buildfarm run please?

esc · 2021-03-10T15:21:44Z

BFID: numba_smoketest_cuda_yaml_23

esc · 2021-03-10T16:11:34Z

BF passed

gmarkall · 2021-03-10T16:54:21Z

@esc Many thanks!

rawwar · 2021-03-10T18:16:05Z

Accidentally pressed close.

sklam

re-approving to allow merge

rawwar · 2021-03-26T17:27:33Z

closes #3203

stuartarchibald · 2021-03-26T17:32:06Z

closes #3203

Thanks, will close.

rawwar added 5 commits March 8, 2021 22:51

included watchdog,Compute mode and perf ratio

a10b732

changed compute mode output

b0376c0

updated performance ratio label

62e9aee

fixed lint error

5701ae2

Merge branch 'master' of github.com:numba/numba into 3203_additional_…

8cb992d

…info_in_cuda_detect

gmarkall added 4 - Waiting on author Waiting for author to respond to review CUDA CUDA related issue/PR labels Mar 9, 2021

updated capitalisations and Watchdog output

f7f5ae4

rawwar added 2 commits March 9, 2021 23:28

fixed lint errors

03aa76f

printing compute mode only for windows

cde5af3

gmarkall added 4 - Waiting on reviewer Waiting for reviewer to respond to author and removed 4 - Waiting on author Waiting for author to respond to review labels Mar 9, 2021

gmarkall approved these changes Mar 10, 2021

View reviewed changes

gmarkall added 4 - Waiting on CI Review etc done, waiting for CI to finish Pending BuildFarm For PRs that have been reviewed but pending a push through our buildfarm and removed 4 - Waiting on reviewer Waiting for reviewer to respond to author labels Mar 10, 2021

esc added BuildFarm Passed For PRs that have been through the buildfarm and passed and removed 4 - Waiting on CI Review etc done, waiting for CI to finish Pending BuildFarm For PRs that have been reviewed but pending a push through our buildfarm labels Mar 10, 2021

gmarkall added the 5 - Ready to merge Review and testing done, is ready to merge label Mar 10, 2021

rawwar closed this Mar 10, 2021

rawwar reopened this Mar 10, 2021

sklam approved these changes Mar 10, 2021

View reviewed changes

sklam merged commit 55e6831 into numba:master Mar 10, 2021

rawwar deleted the 3203_additional_info_in_cuda_detect branch March 10, 2021 22:41

stuartarchibald mentioned this pull request Mar 26, 2021

numba.cuda.detect() feature request #3203

Closed

2 tasks

Conversation

rawwar commented Mar 9, 2021

Uh oh!

gmarkall commented Mar 9, 2021

Uh oh!

gmarkall commented Mar 9, 2021

Uh oh!

rawwar commented Mar 9, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gmarkall commented Mar 9, 2021

Uh oh!

rawwar commented Mar 9, 2021

Uh oh!

rawwar commented Mar 9, 2021

Uh oh!

gmarkall commented Mar 9, 2021

Uh oh!

gmarkall commented Mar 10, 2021

Uh oh!

gmarkall commented Mar 10, 2021

Uh oh!

gmarkall left a comment

Choose a reason for hiding this comment

Uh oh!

esc commented Mar 10, 2021

Uh oh!

esc commented Mar 10, 2021

Uh oh!

gmarkall commented Mar 10, 2021

Uh oh!

rawwar commented Mar 10, 2021

Uh oh!

sklam left a comment

Choose a reason for hiding this comment

Uh oh!

rawwar commented Mar 26, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

stuartarchibald commented Mar 26, 2021

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

rawwar commented Mar 9, 2021 •

edited

Loading

rawwar commented Mar 26, 2021 •

edited

Loading