glmark2 has odd results

[Clive McCarthy]

I have found that glmark2 shows Intel Xe graphics to be inferior to Iris graphics 655 by a factor of three. This is most odd. I am evaluating a NUC12WSK comparing it to a NUC8i5BEK. The screen is 4K. On the old box I get 264 and the new box just 85.

Could this be an artifact of the "Efficiency" cores ? I'm using glmark2 straight from the repository and is a version from 2021.

[Eirikr1848]

Confirming you have the correct GPU drivers and are not using software rendering?

[Clive McCarthy]

I'm using the latest drivers from Oibaf. I have been using glmark2 for years on the various generations of Intel integrated graphics. My application is an artwork, not a game. Something odd has happened using glmark2 with this new generation of integrated graphics.

[Eirikr1848]

I'm using the latest drivers from Oibaf. I have been using glmark2 for years on the various generations of Intel integrated graphics. My application is an artwork, not a game. Something odd has happened using glmark2 with this new generation of integrated graphics.

What’s glxinfo - B give you? (Think it’s part of mesa-utils but don’t quote me on that)

Then, have you tried the stock drivers instead of the PPA drivers?

[Clive McCarthy]

This is a pure A-B comparison. Same drivers, same OS, same screen, same glmark2. The only difference is the platform. I have used Intel integrated graphics for my artwork ever since the first NUCs arrived with the i3-4010U and HD 4400 graphics (ten years ?). The big change now, with the Alder Lake CPUs, is the addition of E-cores. This, ostensibly, has nothing to do with the GPU but might mess up ancillary crunching done by the CPU within the benchmark.

This is not a critical item for me. I just like to keep track of how well the integrated graphics have progressed. Intel's naming of their GPUs is appalling -- it's all flash and dash. First Arctic Sound, then, Xe, then ARC, then... god knows what and when one looks at what the driver reports you get yet other stuff. It's even hard to find out exactly how many execution units a GPU has. Consequently I have tended to simply see what glmark2 reports. And now it doesn't.

For all that Gelsinger has said, I'm disappointed that he hasn't fired all the people that make up the "code" names. There is probably a whole department of worthless confusion makers. This from someone who rather liked 8080, 8085, 8086, 8087, 80286, 80386, 80486 -- until AMD/copyright forced the name change to Pentium. Then the rot set in.

[Clive McCarthy]

Here is a little more data:
i7-1165G7 Xe Graphics (TGL GT2)
================================================== =====
glmark2 2021.02
================================================== =====
OpenGL Information
GL_VENDOR: Intel
GL_RENDERER: Mesa Intel(R) Xe Graphics (TGL GT2)
GL_VERSION: 4.6 (Compatibility Profile) Mesa 22.2.0-devel (git-e8fc5cc 2022-06-22 focal-oibaf-ppa)
================================================== =====
glmark2 Score: 200
================================================== =====


i5-1240Px16 Graphics (ADL GT2) [the box says Xe graphics]
================================================== =====
glmark2 2021.02
================================================== =====
OpenGL Information
GL_VENDOR: Intel
GL_RENDERER: Mesa Intel(R) Graphics (ADL GT2)
GL_VERSION: 4.6 (Compatibility Profile) Mesa 22.2.0-devel (git-e8fc5cc 2022-06-22 focal-oibaf-ppa)
================================================== =====
glmark2 Score: 103
================================================== =====

[Clive McCarthy]

The Intel driver writers choose obscure abbreviations for the GL_RENDERER description part of which is, inconsistent, marketing nonsense. The glmark2 score for this new (16 core) device is much inferior to the much, much older Iris graphics 655 results I have and half the performance of Xe Graphics (TGL GT2).

There are two possibilities:

• Intel Xe graphics is inferior to the older Iris graphics 655 (unlikely).
• There is something glmark2 is getting wrong.