Chrome WebGL Blacklist

[Alecazam]

We have a customers that can't run our WebGL app due to recent blacklist changes.   There are quite a few hoops just to get WebGL to run.  First there is the "Enable hardware acceleration where possible" flag, and now the "Override software rendering list" flag in chrome://flags.  The blacklist affects Chrome and Chromium builds, and in Chromium there often isn't any user control of the blacklist or the acceleration flag.  The blacklist seems to have included Intel HD 3000 GPUs and prior (f.e. a 2011 Macbook Air).   

I realize that WebGL 2 is on the way, but these machines should be more than capable of running a WebGL 1 app (with ES2 level functionality).  I'm concerned about overly-aggressive disabling of vendor-specific GPUs like this in the future. 

[Andre Weissflog]

I remember that this was brought up by AlteredQualia a few days ago on Twitter:

https://twitter.com/alteredq/status/783234688731283457

Latest developments:

https://twitter.com/alteredq/status/783375344032243712

The original crbugs issue states that 30% of GPU crashes on Mac are caused by HD3000, also see discussion:

https://groups.google.com/a/chromium.org/forum/#!topic/chromium-reviews/I6dL_9IAwiE

I agree that it's real bad for WebGL in general if GPUs are taken fairly randomly on and off the blacklist. There needs to be at least a very clear option built into browsers to inform the user what's happening, and whether he wants to add an exception.

[Tarek Sherif]

The key issue is that the Chrome team needs to be transparent about these decisions. We just suddenly saw an uptick in "WebGL not supported" errors from our users and had to hunt down an obscure commit message to discover the root cause. At the very least, if the change had been announced ahead have time, we would have been better prepared to support our customers. We're just starting to see clients trust WebGL as a viable platform, and surprises like this only serve to damage its reputation.

[Kenneth Russell]

Everyone,

I apologize for the lack of public communication about this decision. As discussed at length on the CL above, the Intel HD 3000 was responsible for a disproportionately large fraction of Chrome's GPU process crashes. In an effort to stem the tide of crash reports, the Chrome team stopped utilizing that GPU.

As mentioned on http://crbug.com/592130 , as a result of the user response, we don't have any plans to blacklist the Intel HD 4000 on the same platform -- though it is responsible once again for a disproportionately large fraction of GPU process hangs. We will try to gather information that will allow the underlying graphics driver bugs to be fixed at their source, instead of worked around.

-Ken

--
You received this message because you are subscribed to the Google Groups "WebGL Dev List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to webgl-dev-list+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Tarek Sherif]

Thanks, Ken. A couple of notes.

First, those bug tickets you've been linking to (e.g. http://crbug.com/592130) don't seem to be open to the public. I would be interested to follow these conversations, so is there anywhere they would be publicly available?

Second, is there any documentation available about Chrome's (or Firefox's for that matter) blacklists? Would the best approach be to just keep an eye on this file or is there something more human-readable we could follow? It would be nice to have something I could give our support staff to refer to when customers have issues.

Tarek Sherif
http://tareksherif.net/

https://www.biodigital.com/

--
You received this message because you are subscribed to a topic in the Google Groups "WebGL Dev List" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/webgl-dev-list/03i2dNQSwqM/unsubscribe.
To unsubscribe from this group and all its topics, send an email to webgl-dev-list+unsubscribe@googlegroups.com.

[Florian Bösch]

I'd like to add a few points to Intel HD.

• Roughly half of GPUs on Web visitors machines from the desktop segment are Intel HDs
• It's hard to say how many of them are HD 3000's and HD 4000's because most of the time the reported GPU string is just "Intel(R) HD Graphics". However it could easily be as much as 10-20% each.

A decision to blacklist Intel HD can therefore bring WebGL support levels that're hovering around 95% support back down to as low as 55%.

[Kenneth Russell]

On Thu, Oct 20, 2016 at 6:28 AM, Tarek Sherif <tsh...@gmail.com> wrote:

Thanks, Ken. A couple of notes.

First, those bug tickets you've been linking to (e.g. http://crbug.com/592130) don't seem to be open to the public. I would be interested to follow these conversations, so is there anywhere they would be publicly available?

Apologies for that -- the Chromium project leaves most bugs open -- but that bug was filed by a Google employee and there might be non-public information in the report, so I'm not comfortable opening it up. It basically has analyses of some crash reports and which GPUs they were coming from. I'll try to make sure future bugs in this area are made public early on.

 

Second, is there any documentation available about Chrome's (or Firefox's for that matter) blacklists? Would the best approach be to just keep an eye on this file or is there something more human-readable we could follow? It would be nice to have something I could give our support staff to refer to when customers have issues.

These two files are the primary ones which control the blacklisting of features:

https://chromium.googlesource.com/chromium/src/+/master/gpu/config/gpu_driver_bug_list_json.cc

https://chromium.googlesource.com/chromium/src/+/master/gpu/config/software_rendering_list_json.cc

The best way to help the Chrome team specifically is to periodically test on either the Dev or Canary channels, as well as the Beta channel, and report bugs early. We don't have the resources to test all GPU types in-house so we rely on customer bug reports. In this case there were zero bug reports about the Intel HD 3000 blacklisting change before it went to Chrome Stable.

For Firefox, the same applies to their Nightly and Beta builds.

-Ken

[Florian Bösch]

On Thursday, October 20, 2016 at 11:58:56 PM UTC+2, Kenneth Russell wrote:

In this case there were zero bug reports about the Intel HD 3000 blacklisting change before it went to Chrome Stable.

There wouldn't be any bug reports until 20-40% of people start complaining their WebGL don't work no more. The Chrome team has the numbers of how many machines a given blacklist change will impact. If we're getting to the double digits of the entire WebGL population, I'd expect some very, very serious consideration going on. 

[Tarek Sherif]

Thanks for the info, Ken.

Florian's right. This is not a case of bugs to report; it's a conscious decision to deprecate functionality. As such, it's far more straightforward for the Chrome team to communicate the change than to hope other developers happen to come across the particular configuration that triggers it in their testing.

Tarek Sherif
http://tareksherif.net/

https://www.biodigital.com/

--

[Florian Bösch]

So far I am observing a retreat of WebGL on OSX off the height of 98% August 25th to 94% October 20th on a 30-day moving average, so roughly ~2%/month:

Below is the non filtered day by day view of the same data:

[Florian Bösch]

I should note that this is just Chrome, neither Safari nor Firefox seem to have picked up the new blacklist entry yet.

[Tarek Sherif]

That fits with what I saw on the Steam hardware survey, which puts Intel HD 3000 usage at 5-6% on OSX: http://store.steampowered.com/hwsurvey/?platform=mac

Is there any expectation that Firefox or Safari would follow suit on this? My understanding is that Firefox has it's own, much less aggressive, blacklisting policy, which is why we're likely going to start recommending our users prefer it over Chrome.

Tarek Sherif
http://tareksherif.net/

https://www.biodigital.com/

On Sun, Oct 23, 2016 at 4:32 PM, Florian Bösch <pya...@gmail.com> wrote:

I should note that this is just Chrome, neither Safari nor Firefox seem to have picked up the new blacklist entry yet.

--

[Florian Bösch]

On Sunday, October 23, 2016 at 10:46:08 PM UTC+2, Tarek Sherif wrote:

Is there any expectation that Firefox or Safari would follow suit on this?

I don't know. On a 30 day average the current support levels look like this:

• Firefox: 98%
• Chrome: 94%
• Safari: 93%

So presently Firefox is definitely the better choice on OSX.

[Tarek Sherif]

Thanks, Florian. Are your stats for Windows and Linux similar? 

Tarek Sherif
http://tareksherif.net/

https://www.biodigital.com/

--

[Andrew Varga]

My Chrome just upgraded itself to 54.0 and WebGL stopped working.

"Override software rendering list" flag is not helping,

I have just updated to the latest nvidia driver, which had no effect either. The GPU is a Geforce 8800GT, sure, not a very new model but it's been working perfectly for me for years so I wonder what happened out of the blue just now?

Is this related to my GPU?

On Monday, October 24, 2016 at 3:36:27 PM UTC+2, Tarek Sherif wrote:

Thanks, Florian. Are your stats for Windows and Linux similar? 

Tarek Sherif
http://tareksherif.net/

https://www.biodigital.com/

On Sun, Oct 23, 2016 at 5:26 PM, Florian Bösch <pya...@gmail.com> wrote:

On Sunday, October 23, 2016 at 10:46:08 PM UTC+2, Tarek Sherif wrote:

Is there any expectation that Firefox or Safari would follow suit on this?

I don't know. On a 30 day average the current support levels look like this:

• Firefox: 98%
• Chrome: 94%
• Safari: 93%

So presently Firefox is definitely the better choice on OSX.

--
You received this message because you are subscribed to a topic in the Google Groups "WebGL Dev List" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/webgl-dev-list/03i2dNQSwqM/unsubscribe.

To unsubscribe from this group and all its topics, send an email to webgl-dev-lis...@googlegroups.com.