Speed team: who can help with the setup of this bot?

I can take this bug. Ken: how do you usually go about handling this kind of bug? Should we just loop lab folks in & let them check the driver?

Yes, that's the best course of action. You should also confirm that this machine is hooked up to an IP KVM in order to activate the graphics card.

Peter: can you help checking the GPU driver & whether the machine is correctly hooked up to an IP KVM in order to activate the graphics card.

THe machine of interest is build168-m1 (Win 7 Intel GPU Perf). If you found the problem, it may be worth checking other machine in this pool as well 

Pool:
cpu: x86-64
gpu: Intel Haswell HD Graphics (8086:041a)
os: Windows-2008ServerR2-SP1
pool: Chrome-perf

The swarming hosts are:  $ swarming.py bots -d pool "Chrome-perf" -d gpu "8086:041a-10.18.14.4414" -d os "Windows" -S https://chromium-swarm.appspot.com --bare
build164-m1
build166-m1
build167-m1
build168-m1

Of these build154-m1 had active graphics on the kvm.  The graphics on the other 3 came back up after a power cycle.

The driver is pretty old.  Want them updated?

I still see "GPU device 0: VENDOR = 0x0, DEVICE = 0x0" in build168-m1. Peter: can you update the drivers on all these machines? Thanks!
 

Can you give a pointer where you are seeing "GPU device 0: VENDOR = 0x0, DEVICE = 0x0" on build168-m1?

I installed chrome on build168-m1.  See attached 

I will go ahead and update the drivers though.

Deleted: build168-m1.jpg 107 KB

	build168-m1.jpg 107 KB View Download

Peter, I was seeing it in the swarming log

https://chromium-swarm.appspot.com/task?id=3af3b1d5a919b610&refresh=10&show_raw=1 (this is on 01-09)

Interestingly, now I am checking the swarming log again, it's showing:
(INFO) 2018-01-11 05:07:52,010 browser._LogBrowserInfo:117  GPU device 0: VENDOR = 0x8086 (Intel), DEVICE = 0x41a
(INFO) 2018-01-11 05:07:52,010 browser._LogBrowserInfo:117  GPU device 1: VENDOR = 0x102b, DEVICE = 0x522

(https://chromium-swarm.appspot.com/task?id=3afe0b2ee9cda810&refresh=10&show_raw=1)

So I am guessing the problem is solved. Any idea what might cause the flaky problem?

That example was from before I power cycled the host.   Since then it's been o.k.
btw GPU device1 is the onboard matrox card whic his disabled in the bios.

Side note:  kbr@ has migrated the GPU folks to using desktop Intel HD 630 based graphics systems:   See crbug.com/754875

If you want to do the same please open up a new bug for us.

The Intel HD 630 systems that Peter deployed for us in Issue 754875 are working very well.

https://ci.chromium.org/buildbot/chromium.gpu.fyi/Win10%20Release%20%28Intel%20HD%20630%29/
https://ci.chromium.org/buildbot/chromium.gpu.fyi/Linux%20Release%20%28Intel%20HD%20630%29/

We are going to add them to our optional GPU tryservers in  Issue 801323 .

Before driver upgrade:
[build164-m1] out: Intel(R) HD Graphics P4600/P4700        10.18.14.4414
[build165-m1] out: Intel(R) HD Graphics P4600/P4700        10.18.14.4414
[build166-m1] out: Intel(R) HD Graphics P4600/P4700        10.18.14.4414
[build167-m1] out: Intel(R) HD Graphics P4600/P4700        10.18.14.4414
[build168-m1] out: Intel(R) HD Graphics P4600/P4700        10.18.14.4414

After driver upgrade:

[build164-m1] out: Intel(R) HD Graphics P4600/P4700        10.18.14.4578
[build165-m1] out: Intel(R) HD Graphics P4600/P4700        10.18.14.4578
[build166-m1] out: Intel(R) HD Graphics P4600/P4700        10.18.14.4578
[build167-m1] out: Intel(R) HD Graphics P4600/P4700        10.18.14.4578
[build168-m1] out: Intel(R) HD Graphics P4600/P4700        10.18.14.4578

Re #8: should the benchmark (or something) have a check to read that GPU config from the browser and complain if it believes GPU's should be found but aren't?

#12: that's  issue 771597  :-)

I think this is fixed?

We still identify two GPUs, but it seems Chrome is picking up Intel GPU. So I think we can mark this as Fixed.

Thanks Peter!