The Mac failures are  Issue 550016 , which has been mitigated.

However, you're right, ScreenshotSync.WithCanvas has somehow become flaky on the Windows ATI bots. These flakes:

https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/203877
https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/203775
https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/203662
https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/203442
https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/202929
https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/202622
https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/202057
https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng/builds/201979

are all the same timeout on the same test.

I'd initially suspect something wrong with the specific machines since it seems to be happening on a small subset:

build60-m4
build63-m4
build70-m4
build71-m4

I filed Issue 603338 about build71-m4 earlier today, but need to check on the others.

By the way, since this is affecting the CQ, P0 is the right priority until that's resolved.

https://codereview.chromium.org/1885963004/ is in the CQ to mark this test flaky.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/a3ea1e39abb968f3ab111751c079bebc6601dde4

commit a3ea1e39abb968f3ab111751c079bebc6601dde4
Author: kbr <kbr@chromium.org>
Date: Thu Apr 14 05:32:45 2016

Suppress flaky ScreenshotSync.WithCanvas failures on Win AMD.

BUG= 602313 
CQ_INCLUDE_TRYBOTS=tryserver.chromium.linux:linux_optional_gpu_tests_rel;tryserver.chromium.mac:mac_optional_gpu_tests_rel;tryserver.chromium.win:win_optional_gpu_tests_rel
TBR=zmo@chromium.org

Review URL: https://codereview.chromium.org/1885963004

Cr-Commit-Position: refs/heads/master@{#387242}

[modify] https://crrev.com/a3ea1e39abb968f3ab111751c079bebc6601dde4/content/test/gpu/gpu_tests/screenshot_sync_expectations.py

Downgrading to P2 now that the failures have been suppressed. Please indicate here and re-raise the priority if it looks like this test is still failing on win_chromium_rel_ng.

Note: I spot-checked https://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_rel_ng?numbuilds=200 and don't see any failures of screenshot_sync_tests.

There seem to be more problems with the screenshot_sync_tests. Duplicating this into a previously reported bug.