Intel Haswell Might Have Regressed Hard On Linux

[phoronix]

Phoronix: Intel Haswell Might Have Regressed Hard On Linux

For your viewing pleasure this weekend are some extra benchmarks of various Intel Sandy Bridge, Ivy Bridge, and Haswell HD Graphics when using an Ubuntu 14.04 Linux development snapshot with the Linux 3.13 kernel and Mesa 10.0.1. The processors tested included the Core i3 2120, Core i5 2500K, Core i5 3470, Core i7 3770K, Core i3 4130, and Core i7 4770K. These tests appear to represent a huge drawback in performance for Intel Haswell on Linux compared to earlier results.

http://www.phoronix.com/vr.php?view=19646

[halo9en]

Whoa, the block layer regressed, Haswell cpus regressed... 3.13 sounds less and less desirable each passing day.

[mikkl]

Shocking test results.

[Sn3ipen]

Awfull results. How come Haswell is so much slower than previous gen?

[tomtomme]

Awfull results. How come Haswell is so much slower than previous gen?

it wasn?t with older kernels...

[mendieta]

it wasn?t with older kernels...

Hold on, do we know it is the kernel? It could be something in mesa as well. I am running linux-3.13 git with mesa 9.2.1 from kubuntu saucy, and this has no significant regressions in my testing. Michael mentioned he'll investigate, let's see how it goes.

Regardless, I've been saying this for months now. Ever since mesa 9.2 + kernel 3.10, where haswell improved tremendously, nothing much happened for haswell, at least in my testing with qgears2, glmark2 and unigine. In fact, many times, kernels newer than 3.10 gave me regressions, though not as bad as this. so i've been mostly running the default kubuntu saucy stack.

Worth mentioning, too, is the fact that both xorg-edgers and oibaf on top of saucy have been really really unstable, to the point that i try every now and then, just to revert to the default saucy stack.

~$ cat /proc/cpuinfo |grep model
model           : 60
model name      : Intel(R) Core(TM) i5-4670K CPU @ 3.40GHz

[zanny]

I'm running 3.12.7 with Mesa 10.0.1 and I'm not noticing any big slowdowns for my hd 4600 either. What are your benchmark results mendieta? I'll run them myself quickly and post results.

[sarcaza]

Benchmarks and hardware details for 3.13 rc7 with haswell 4600 and Mesa 10.1.0-devel.

$ uname -r
3.13.0-031300rc7-generic

================================================== =====
glmark2 2012.08
================================================== =====
OpenGL Information
GL_VENDOR: Intel Open Source Technology Center
GL_RENDERER: Mesa DRI Intel(R) Haswell Mobile
GL_VERSION: 3.0 Mesa 10.1.0-devel (git-2dc35a6 saucy-oibaf-ppa+curaga)
================================================== =====
[build] use-vbo=false: FPS: 3047 FrameTime: 0.328 ms
[build] use-vbo=true: FPS: 3293 FrameTime: 0.304 ms
[texture] texture-filter=nearest: FPS: 3573 FrameTime: 0.280 ms
[texture] texture-filter=linear: FPS: 3603 FrameTime: 0.278 ms
[texture] texture-filter=mipmap: FPS: 3662 FrameTime: 0.273 ms
[shading] shading=gouraud: FPS: 2843 FrameTime: 0.352 ms
[shading] shading=blinn-phong-inf: FPS: 2873 FrameTime: 0.348 ms
[shading] shading=phong: FPS: 2815 FrameTime: 0.355 ms
[bump] bump-render=high-poly: FPS: 1439 FrameTime: 0.695 ms
[bump] bump-render=normals: FPS: 3855 FrameTime: 0.259 ms
[bump] bump-render=height: FPS: 3692 FrameTime: 0.271 ms
[effect2d] kernel=0,1,0;1,-4,1;0,1,0;: FPS: 1997 FrameTime: 0.501 ms
[effect2d] kernel=1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;: FPS: 848 FrameTime: 1.179 ms
[pulsar] light=false:quads=5:texture=false: FPS: 2656 FrameTime: 0.377 ms
[desktop] blur-radius=5:effect=blurasses=1:separable=true:windows=4: FPS: 811 FrameTime: 1.233 ms
[desktop] effect=shadow:windows=4: FPS: 1295 FrameTime: 0.772 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 721 FrameTime: 1.387 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=subdata: FPS: 1000 FrameTime: 1.000 ms
[buffer] columns=200:interleave=true:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 772 FrameTime: 1.295 ms
[ideas] speed=duration: FPS: 2208 FrameTime: 0.453 ms
[jellyfish] <default>: FPS: 1513 FrameTime: 0.661 ms
[terrain] <default>: FPS: 174 FrameTime: 5.747 ms
[conditionals] fragment-steps=0:vertex-steps=0: FPS: 3109 FrameTime: 0.322 ms
[conditionals] fragment-steps=5:vertex-steps=0: FPS: 3217 FrameTime: 0.311 ms
[conditionals] fragment-steps=0:vertex-steps=5: FPS: 3114 FrameTime: 0.321 ms
[function] fragment-complexity=low:fragment-steps=5: FPS: 3131 FrameTime: 0.319 ms
[function] fragment-complexity=medium:fragment-steps=5: FPS: 3219 FrameTime: 0.311 ms
[loop] fragment-loop=false:fragment-steps=5:vertex-steps=5: FPS: 3126 FrameTime: 0.320 ms
[loop] fragment-steps=5:fragment-uniform=false:vertex-steps=5: FPS: 3117 FrameTime: 0.321 ms
[loop] fragment-steps=5:fragment-uniform=true:vertex-steps=5: FPS: 3175 FrameTime: 0.315 ms
================================================== =====
glmark2 Score: 2463
================================================== =====
$ cat /proc/cpuinfo |grep model
model name : Intel(R) Core(TM) i7-4700MQ CPU @ 2.40GHz

[sarcaza]

Some benchmarks:

$ uname -r
3.13.0-031300rc7-generic
$ cat /proc/cpuinfo |grep model
model name : Intel(R) Core(TM) i7-4700MQ CPU @ 2.40GHz

================================================== =====
glmark2 2012.08
================================================== =====
OpenGL Information
GL_VENDOR: Intel Open Source Technology Center
GL_RENDERER: Mesa DRI Intel(R) Haswell Mobile
GL_VERSION: 3.0 Mesa 10.1.0-devel (git-2dc35a6 saucy-oibaf-ppa+curaga)
================================================== =====
[build] use-vbo=false: FPS: 3047 FrameTime: 0.328 ms
[build] use-vbo=true: FPS: 3293 FrameTime: 0.304 ms
[texture] texture-filter=nearest: FPS: 3573 FrameTime: 0.280 ms
[texture] texture-filter=linear: FPS: 3603 FrameTime: 0.278 ms
[texture] texture-filter=mipmap: FPS: 3662 FrameTime: 0.273 ms
[shading] shading=gouraud: FPS: 2843 FrameTime: 0.352 ms
[shading] shading=blinn-phong-inf: FPS: 2873 FrameTime: 0.348 ms
[shading] shading=phong: FPS: 2815 FrameTime: 0.355 ms
[bump] bump-render=high-poly: FPS: 1439 FrameTime: 0.695 ms
[bump] bump-render=normals: FPS: 3855 FrameTime: 0.259 ms
[bump] bump-render=height: FPS: 3692 FrameTime: 0.271 ms
[effect2d] kernel=0,1,0;1,-4,1;0,1,0;: FPS: 1997 FrameTime: 0.501 ms
[effect2d] kernel=1,1,1,1,1;1,1,1,1,1;1,1,1,1,1;: FPS: 848 FrameTime: 1.179 ms
[pulsar] light=false:quads=5:texture=false: FPS: 2656 FrameTime: 0.377 ms
[desktop] blur-radius=5:effect=blurasses=1:separable=true:windows=4: FPS: 811 FrameTime: 1.233 ms
[desktop] effect=shadow:windows=4: FPS: 1295 FrameTime: 0.772 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 721 FrameTime: 1.387 ms
[buffer] columns=200:interleave=false:update-dispersion=0.9:update-fraction=0.5:update-method=subdata: FPS: 1000 FrameTime: 1.000 ms
[buffer] columns=200:interleave=true:update-dispersion=0.9:update-fraction=0.5:update-method=map: FPS: 772 FrameTime: 1.295 ms
[ideas] speed=duration: FPS: 2208 FrameTime: 0.453 ms
[jellyfish] <default>: FPS: 1513 FrameTime: 0.661 ms
[terrain] <default>: FPS: 174 FrameTime: 5.747 ms
[conditionals] fragment-steps=0:vertex-steps=0: FPS: 3109 FrameTime: 0.322 ms
[conditionals] fragment-steps=5:vertex-steps=0: FPS: 3217 FrameTime: 0.311 ms
[conditionals] fragment-steps=0:vertex-steps=5: FPS: 3114 FrameTime: 0.321 ms
[function] fragment-complexity=low:fragment-steps=5: FPS: 3131 FrameTime: 0.319 ms
[function] fragment-complexity=medium:fragment-steps=5: FPS: 3219 FrameTime: 0.311 ms
[loop] fragment-loop=false:fragment-steps=5:vertex-steps=5: FPS: 3126 FrameTime: 0.320 ms
[loop] fragment-steps=5:fragment-uniform=false:vertex-steps=5: FPS: 3117 FrameTime: 0.321 ms
[loop] fragment-steps=5:fragment-uniform=true:vertex-steps=5: FPS: 3175 FrameTime: 0.315 ms
================================================== =====
glmark2 Score: 2463

[mendieta]

I'm running 3.12.7 with Mesa 10.0.1 and I'm not noticing any big slowdowns for my hd 4600 either. What are your benchmark results mendieta? I'll run them myself quickly and post results.

Thanks, Zanny!

This is an example, GLMark2 would be consistently lower with kernel 3.12, than 3.11. However, early 3.13 during the merger window was better.

http://openbenchmarking.org/result/1311169-MEND-131101985

In this other test at higher resolution, i have later kernels in the comparison, and it seems glmark2 is mostly ok with kernel 3.13

http://openbenchmarking.org/result/1401119-MEND-140103115