[10016478] Firefox leaks memory running GUIMark3 vector test with Intel GPU

[Stefan3D]

Note: DXDIAG.TXT without GeForce driver (nvwgf2um.dll) which crashes DXDIAG

Category

Questions

Answers (N/A if not applicable)

Description

Provide a detailed description of the issue

Firefox leaks memory running GUIMark3 vector test with Intel GPU.

Edge + Intel works, Firefox + NVIDA works, so i'm not sure if it's a bug in Firefox or Intel graphics driver

Does it fail every single time, or only sometimes?

If you can offer a % rate please do.

100%

Hardware (HW)

Brand and Model of the system.

Acer E5-571G

Hybrid or switchable graphics system?

ie Does it have AMD or NV graphics too?

NVIDIA Optimus

Make and model of any Displays that are used to see the issue (see note2 below). 

LFP = Local Flat Panel (Laptop panel) EFP = External Flat Panel (Monitor you plug in)

LFP

How much memory [RAM] in the system (see note2 below).

8 GB

Provide any other hardware needed to replicate the issue.

ie: Cables&brand, cable type [vga, hdmi, DP, etc], dock, dongles/adapters, etc

N/A

Hardware Stepping (see note1 below).

<td style="border: 0px solid black; padding: 0 5.4pt 0 5.4pt; border-right: solid windowtext 1.0pt; border-bottom...

[gbhj]

this also happens on Intel iris 5100 with the latest firefox nightly 44.0a1

the same also happens on all browsers that use hardware accelerated vector(edge does not)

the memory leak also affects other applications GPU memory requiring a driver restart to fix the memory leak

[Allan_J_Intel1]

If this issue were a graphics driver, I would expect to have same problem when running test on Chrome and Internet Explorer but the test ran for more than 10 minutes until I closed the window.

I did experience same problem when I tested with Firefox.

Allan.

[Stefan3D]

Re-checked with Firefox 42.0 final version and Graphics driver 20.19.15.4312

Memory leak has been fixed, in deed by the new graphics driver

EDIT: after a reboot the leak came back

[Bryce__Intel]

Stefan3D

https://www/ https://www.youtube.com/watch?v=GNRuWJLIc8c

[Bryce__Intel]

Bug# 10016478

Issue is under investigation, it has cleared Debug level and has been progressed to Development queue for further analysis.

[Bryce__Intel · Accepted solution]

The conclusion from the Developer came back (below). He indicates this should be resolved by the app/benchmark owner.

This is regarding issue 10016478 - Memory leaks running GUIMark3 vector test with Intel GPU

While debugging this issue, we found out we are running out of Graphics Virtual address space and that's because we get a lot of CreateResource calls from the app though DestroyResource is not being called from the app properly, leading to memory hogging.

Below is the debug description:

Found that only 0-1% of the CreateResource calls were getting Destroyed.

When run with RenderPerPrimitive key (where issue is still reproducible), 35-40% of the CreateResource calls were Destroyed.

I have attached a screenshot VMAP to show that we are actually out of Gfx virtual address space.

With this above information, it is concluded that iGPU is not responsible for this behavior. This should be reported to the ISV to fix on their side, we should not resolve app bugs in our driver.