Oolite 1.84 performance issue on Nvidia Pascal (10xx series GTX) cards

[Cmdr Wyvern]

Greets, Commanders.

It may have been reported, but I couldn't find any reference... Anyway!

I recently upgraded to a GTX 1070, and whoa, this thing is a beast; able to handle anything I want to throw at it at 1080P @ 60fps with room to spare...

...Except, for some reason, Oolite.

I get 60 FPS when docked, yet in flight the framerate drops to half that, with a noticeable microstutter. Same thing in windowed mode or full screen. I played with vsync and triple buffer settings on the card, with no improvements.

In hardware monitors, the card is barely loaded: Doesn't go above it's idle clocks, shows at most an 18% load. In contrast, the CPU (FX 8350) gets a fairly heavy load and goes into turbo boost mode.

I should easily be getting 60FPS or greater consistently. It's almost like the game is afraid to let the 1070 off the leash.

Code: Select all

Opening log for Oolite version 1.84 (x86-64) under Windows 6.2.9200  64-bit at 2016-11-10 21:17:37 -0600.
8 processors detected.
Build options: OpenAL, new planets.

Note that the contents of the log file can be adjusted by editing logcontrol.plist.

21:17:37.815 [display.mode.list.native]: Windows native resolution detected: 1920 x 1080
21:17:37.995 [joystick.init]: Number of joysticks detected: 1
21:17:37.996 [rendering.opengl.version]: OpenGL renderer version: 4.5.0 ("4.5.0 NVIDIA 372.70"). Vendor: "NVIDIA Corporation". Renderer: "GeForce GTX 1070/PCIe/SSE2".
21:17:37.996 [rendering.opengl.extensions]: OpenGL extensions (362):
GL_NV_blend_square, GL_ARB_map_buffer_range, GL_ARB_instanced_arrays, GL_ARB_shader_viewport_layer_array, GL_ARB_direct_state_access, GL_ARB_clear_texture, GL_EXT_texture_compression_rgtc, GL_EXT_post_depth_coverage, GL_NV_texture_compression_vtc, GL_EXT_texture_storage, GL_NV_fragment_program, GL_ARB_query_buffer_object, GL_EXT_texture_shared_exponent, GL_EXT_sparse_texture2, GL_ARB_uniform_buffer_object, GL_EXT_texture_array, GL_ARB_shader_draw_parameters, GL_ARB_shading_language_include, GL_ARB_separate_shader_objects, GL_NV_fragment_shader_interlock, GL_EXT_packed_pixels, GL_ARB_texture_query_levels, GL_ARB_fragment_layer_viewport, GL_NV_fence, GL_ARB_explicit_uniform_location, GL_ARB_stencil_texturing, GL_NV_conservative_raster, GL_EXT_gpu_program_parameters, GL_EXT_vertex_array, GL_EXT_provoking_vertex, GL_NV_draw_vulkan_image, GL_ARB_texture_view, GL_ARB_ES3_1_compatibility, GL_NV_texgen_reflection, GL_EXT_multi_draw_arrays, GL_ARB_clear_buffer_object, GL_EXT_texture_object, GL_EXT_packed_float, GL_EXT_texture_filter_anisotropic, GL_ARB_get_program_binary, GL_ARB_half_float_vertex, GL_ARB_program_interface_query, GL_EXT_bindable_uniform, GL_ARB_multi_draw_indirect, GL_NV_texture_multisample, GL_ARB_texture_buffer_object, GL_ARB_gpu_shader_fp64, GL_EXT_blend_minmax, GL_ARB_vertex_attrib_64bit, GL_ARB_texture_query_lod, GL_ARB_compressed_texture_pixel_storage, GL_EXT_texture_integer, GL_ARB_texture_non_power_of_two, GL_EXT_texture_compression_s3tc, GL_ARB_texture_stencil8, GL_KHR_no_error, GL_ARB_texture_mirror_clamp_to_edge, GL_ARB_framebuffer_no_attachments, GL_NV_parameter_buffer_object2, GL_ARB_shading_language_420pack, GL_ARB_sample_locations, GL_NV_texture_rectangle, GL_KHR_robustness, GL_SGIX_shadow, GL_EXT_vertex_array_bgra, GL_EXT_depth_bounds_test, GL_NV_vertex_array_range2, GL_ARB_shader_texture_image_samples, GL_NV_vertex_program2, GL_ARB_texture_compression_bptc, GL_NV_vertex_program3, GL_EXT_stencil_wrap, GL_EXT_separate_shader_objects, GL_NV_shader_atomic_float, GL_NV_packed_depth_stencil, GL_ARB_texture_gather, GL_ARB_shader_image_size, GL_EXT_texture_cube_map, GL_EXT_shader_image_load_formatted, GL_NV_register_combiners2, GL_NV_half_float, GL_ARB_map_buffer_alignment, GL_SUN_slice_accum, GL_ARB_provoking_vertex, GL_NV_shader_thread_shuffle, GL_ARB_tessellation_shader, GL_NV_stereo_view_rendering, GL_EXT_direct_state_access, GL_NVX_multigpu_info, GL_ARB_texture_rg, GL_NV_texture_shader2, GL_NV_texture_shader3, GL_ARB_bindless_texture, GL_SGIX_depth_texture, GL_NV_texture_env_combine4, GL_EXT_polygon_offset_clamp, GL_EXT_draw_range_elements, GL_EXT_draw_buffers2, GL_EXT_texture_filter_minmax, GL_ARB_transform_feedback2, GL_ARB_geometry_shader4, GL_ARB_fragment_coord_conventions, GL_ARB_transform_feedback3, GL_ARB_shader_precision, GL_NV_gpu_program_fp64, GL_NV_blend_equation_advanced_coherent, GL_NV_shader_atomic_counters, GL_ARB_get_texture_sub_image, GL_EXT_blend_color, GL_ARB_vertex_array_object, GL_ATI_draw_buffers, GL_ARB_texture_mirrored_repeat, GL_EXT_stencil_two_side, GL_ARB_timer_query, WGL_EXT_swap_control, GL_NV_parameter_buffer_object, GL_EXT_texture_sRGB_decode, GL_NV_transform_feedback, GL_EXT_timer_query, GL_EXT_texture_edge_clamp, GL_EXT_shader_image_load_store, GL_NV_gpu_program4_1, GL_IBM_rasterpos_clip, GL_NV_framebuffer_multisample_coverage, GL_ARB_base_instance, GL_ARB_texture_float, GL_ARB_imaging, GL_ARB_draw_elements_base_vertex, GL_EXT_texture3D, GL_ARB_cull_distance, GL_EXT_packed_depth_stencil, GL_EXT_compiled_vertex_array, GL_ARB_texture_cube_map, GL_NV_uniform_buffer_unified_memory, GL_NV_primitive_restart, GL_EXT_blend_subtract, GL_EXT_separate_specular_color, GL_S3_s3tc, GL_NV_fragment_program_option, GL_EXTX_framebuffer_mixed_formats, GL_ARB_texture_compression, GL_ARB_gl_spirv, GL_ARB_texture_compression_rgtc, GL_ARB_ES2_compatibility, GL_ARB_shader_ballot, GL_NV_shader_atomic_float64, GL_NVX_blend_equation_advanced_multi_draw_buffers, GL_ARB_depth_buffer_float, GL_NV_light_max_exponent, GL_ARB_vertex_buffer_object, GL_KHR_blend_equation_advanced, GL_NV_shader_atomic_int64, GL_ARB_shading_language_100, GL_NV_shader_storage_buffer_object, GL_ARB_texture_env_add, GL_ARB_fragment_program, GL_NV_copy_depth_to_color, GL_ARB_sparse_buffer, GL_EXT_import_sync_object, GL_EXT_texture_swizzle, GL_ARB_shadow, GL_NV_conservative_raster_pre_snap_triangles, GL_ARB_texture_filter_minmax, GL_NV_copy_image, GL_NV_clip_space_w_scaling, GL_ARB_shader_texture_lod, GL_ARB_clip_control, GL_NV_conservative_raster_dilate, GL_ARB_texture_multisample, GL_EXT_texture_env_combine, GL_ARB_sync, GL_ATI_texture_mirror_once, GL_ARB_robustness, GL_ARB_compatibility, GL_EXT_texture_compression_dxt1, GL_NV_gpu_program4, GL_NV_gpu_program5, GL_ARB_transform_feedback_overflow_query, GL_NV_vertex_array_range, GL_ARB_multitexture, GL_EXT_transform_feedback2, GL_NV_vertex_program2_option, GL_ARB_buffer_storage, GL_NV_compute_program5, GL_NV_vertex_program1_1, GL_NV_gpu_program5_mem_extended, GL_EXT_shadow_funcs, GL_ARB_occlusion_query, GL_ARB_fragment_program_shadow, GL_NVX_nvenc_interop, GL_NV_vertex_buffer_unified_memory, GL_ARB_texture_barrier, GL_EXT_geometry_shader4, GL_ARB_texture_env_dot3, GL_ARB_seamless_cubemap_per_texture, GL_ARB_multisample, GL_ARB_copy_image, GL_ARB_gpu_shader_int64, GL_ARB_color_buffer_float, GL_EXT_texture_env_add, GL_ARB_ES3_compatibility, GL_ARB_point_parameters, GL_ARB_sampler_objects, GL_ARB_transform_feedback_instanced, GL_ARB_invalidate_subdata, GL_KTX_buffer_region, GL_NV_vertex_attrib_integer_64bit, GL_NV_bindless_multi_draw_indirect, GL_NV_sample_mask_override_coverage, GL_NVX_gpu_memory_info, GL_EXT_framebuffer_object, GL_NV_gpu_shader5, GL_ARB_texture_rgb10_a2ui, GL_NV_shader_atomic_fp16_vector, GL_NV_ES1_1_compatibility, GL_ARB_texture_border_clamp, GL_NV_alpha_to_coverage_dither_control, GL_ARB_shading_language_packing, GL_EXT_bgra, GL_ATI_texture_float, GL_ARB_occlusion_query2, GL_ARB_shader_clock, GL_ARB_texture_env_crossbar, GL_NV_shader_buffer_load, GL_NV_depth_clamp, GL_ARB_derivative_control, GL_ARB_conservative_depth, GL_AMD_vertex_shader_layer, GL_AMD_multi_draw_indirect, GL_ARB_shader_atomic_counters, GL_ARB_texture_env_combine, GL_EXT_framebuffer_multisample, GL_NV_occlusion_query, GL_EXT_Cg_shader, GL_NV_texture_barrier, GL_NV_path_rendering, GL_ARB_ES3_2_compatibility, GL_ARB_draw_buffers, GL_SGIS_generate_mipmap, GL_NV_path_rendering_shared_edge, GL_ARB_texture_buffer_object_rgb32, GL_EXT_window_rectangles, GL_NVX_conditional_render, GL_NV_register_combiners, GL_EXT_framebuffer_blit, GL_ARB_vertex_array_bgra, GL_NV_multisample_coverage, GL_EXT_fog_coord, GL_ARB_window_pos, GL_NV_fill_rectangle, GL_ARB_half_float_pixel, GL_SGIS_texture_lod, GL_IBM_texture_mirrored_repeat, GL_EXT_texture_lod, GL_EXT_blend_equation_separate, GL_ARB_point_sprite, GL_ARB_parallel_shader_compile, GL_NV_fragment_program2, GL_ARB_shader_group_vote, GL_ARB_fragment_shader, GL_ARB_framebuffer_sRGB, GL_ARB_shader_objects, GL_NV_internalformat_sample_query, GL_WIN_swap_hint, GL_EXT_framebuffer_multisample_blit_scaled, GL_ARB_sparse_texture_clamp, GL_ARB_blend_func_extended, GL_ARB_shader_subroutine, GL_NV_pixel_data_range, GL_NV_viewport_swizzle, GL_ARB_sparse_texture, GL_ARB_multi_bind, GL_KHR_debug, GL_NV_framebuffer_mixed_samples, GL_ARB_transpose_matrix, GL_EXT_texture_compression_latc, GL_ARB_texture_cube_map_array, GL_NV_ES3_1_compatibility, GL_ARB_texture_buffer_range, GL_ARB_robust_buffer_access_behavior, GL_ARB_texture_storage_multisample, GL_EXT_texture_env_dot3, GL_ARB_vertex_type_10f_11f_11f_rev, GL_EXT_texture_sRGB, GL_EXT_secondary_color, GL_EXT_point_parameters, GL_ARB_shader_image_load_store, GL_EXT_pixel_buffer_object, GL_ARB_explicit_attrib_location, GL_ARB_internalformat_query2, GL_ARB_seamless_cube_map, GL_ARB_draw_instanced, GL_ARB_draw_buffers_blend, GL_NV_sample_locations, GL_ARB_post_depth_coverage, GL_NV_fragment_coverage_to_color, GL_EXT_draw_instanced, GL_ARB_pipeline_statistics_query, GL_ARB_framebuffer_object, GL_ARB_internalformat_query, GL_EXT_texture_mirror_clamp, GL_ARB_texture_rectangle, GL_ARB_gpu_shader5, GL_EXT_gpu_shader4, GL_ARB_compute_shader, GL_AMD_seamless_cubemap_per_texture, GL_NV_fog_distance, GL_ARB_depth_clamp, GL_KHR_robust_buffer_access_behavior, GL_NV_texture_shader, GL_NV_shader_thread_group, GL_NV_float_buffer, GL_ARB_vertex_type_2_10_10_10_rev, GL_KHR_blend_equation_advanced_coherent, GL_AMD_vertex_shader_viewport_index, GL_ARB_shader_storage_buffer_object, GL_NV_conditional_render, GL_ARB_vertex_program, GL_ARB_enhanced_layouts, GL_ARB_compute_variable_group_size, GL_NV_bindless_texture, GL_ARB_draw_indirect, GL_ARB_arrays_of_arrays, GL_NV_multisample_filter_hint, GL_ARB_indirect_parameters, GL_ARB_viewport_array, GL_NV_point_sprite, GL_ARB_texture_storage, GL_NV_geometry_shader4, GL_ARB_texture_swizzle, GL_ARB_sparse_texture2, GL_EXT_texture_lod_bias, GL_EXT_abgr, GL_NV_viewport_array2, GL_ARB_vertex_shader, GL_ARB_vertex_attrib_binding, GL_NV_transform_feedback2, GL_ARB_depth_texture, GL_NV_vertex_program, GL_NV_blend_equation_advanced, GL_EXT_framebuffer_sRGB, GL_ARB_shader_atomic_counter_ops, GL_ARB_copy_buffer, GL_NV_depth_buffer_float, GL_EXT_raster_multisample, GL_ARB_pixel_buffer_object, GL_NV_explicit_multisample, GL_EXT_texture_buffer_object, GL_EXT_blend_func_separate, GL_ARB_fragment_shader_interlock, GL_NV_command_list, GL_NV_geometry_shader_passthrough, GL_NV_draw_texture, GL_EXT_vertex_attrib_64bit, GL_EXT_rescale_normal, GL_ARB_sample_shading, GL_ARB_shader_bit_encoding, GL_KHR_context_flush_control, GL_EXT_shader_integer_mix, GL_NV_bindless_multi_draw_indirect_count, GL_ARB_debug_output, GL_ARB_conditional_render_inverted
21:17:38.001 [rendering.opengl.shader.support]: Shaders are supported.

[Cody]

Hello Wyvern - good to see you! Try disabling Threaded Optimisation in the nVidia CP - that might help.
Also, you could try setting V-sync to Adaptive - that might help with the CPU usage.



Damn the forum formatting - code seems to upset things.

[Cmdr Wyvern]

Heya Cody
Here's how I have my CP profile for Oolite set:

Code: Select all

ansiotropic  16X
FXAA off
Gamma correction on
AA mode override
AA setting 8X
AA transparency 8X supersample
MFAA on
max pre-rendered 1
VR pre-rendered 1
power optimal
shader cache on
Ansio sample opti  on
Negative LOD clamp
Texture quality  high quality
Trilinear opti on
Threaded opti off
Triple buffering on
Vsync adaptive

That gets me two cores of the CPU loaded about 50 to 60 % each, with turbo core kicking in.
And the 1070 operating about 30 to 50 %.
Framerates, 60FPS docked, around 40 to 45 FPS flying. Enabling triple buffer and setting vsync adaptive reduced the microstutter to a barely noticeable level.
Still not the butter smooth 60+ framerate I should be getting, yet tolerable.

FYI, I can get over 100FPS in Quake 4 with vsync set to fast or off. Oolite is nowhere as demanding as Q4, so Oolite, why be lagging?

[Cody]

@AC or any Dev: Oolite's FPS is limited to 75, yes? (Or have I got that wrong?) I've forgotten how to uncap it ('twas 199 before).

[Norby]

I've forgotten how to uncap it

Set animation_timer_interval in GNUstep\Defaults\.GNUstepDefaults .

[Cody]

Ah... so it's not capped at 75? For some reason, I thought the default (in 1.85) had been changed. Oh well... wrong again! Thanks, Norby!

[Cmdr Wyvern]

I've forgotten how to uncap it

Set animation_timer_interval in GNUstep\Defaults\.GNUstepDefaults .

This setting can be used for testing purposes to emulate low-end systems, or alternatively, to determine the maximum performance of a given system while running Oolite. A value of 0.001 will practically achieve this, as it will set the desired FPS to 1000. The default value is 0.05, which corresponds to a maximum of 200 frames per second.

This setting failed; animation_timer_interval totally wiped out my joystick settings, and I'm not seeing anywhere near the ballpark of 200FPS, much less 1000. Seems Oolite 1.84 is deliberately limiting itself to half my monitor's refresh, no matter what I try.

I do know that earlier versions timed themselves to your monitor's refresh, if your graphics was capable. And there was a secret setting to turn that behavior off. I can't recall what it was though.

[pagroove]

Also interested in this topic.. As my eye is on a new pc with this card.

[Cody]

This setting failed; animation_timer_interval totally wiped out my joystick settings, and I'm not seeing anywhere near the ballpark of 200FPS

You probably missed a semi-colon, amigo. With V-sync set to Off (and nothing added to .GNUstepDefaults), I'm getting 199fps (GTX 770).

[Cmdr Wyvern]

A little web walking got me a fix, and you'll be pleased to know it's not the game.

Here's the deal.
Two brands of memory are fitted to 10xx cards, Samsung and Micron. The cards equipped with Micron memory have been showing strange issues.

Wanna take a guess which one my 1070 has?
Right. Fortunately, EVGA issued a firmware patch. I grabbed it, installed it, and presto! Oolite is a happy camper, delivering consistent 60 FPS like it should.

Poke a fork in it, call it solved. And let it stand as a little warning to y'all with an eye on a Pascal powered Geforce.

[Cody]

Cool! I like EVGA - good kit and good support.

[another_commander]

Looks like I've arrived late to the party. Good to know that the problem was solved Cmdr Wyvern.

It would be interesting to try the unlimited FPS setting using the animation_timer_interval just to see where we can get to with a very powerful card. The reason it wiped out your joystick settings earlier is that there was a mistake made when inserting the line in .GNUstepDefaults (and I think it is very likely what Cody said about the semicolon missing, but any other syntax error would be the same). A syntax error in .GNUstepDefaults invalidates it when Oolite tries to load it and causes it to be re-created from zero, thus wiping out most settings it had at the time. Best to make a backup copy of .GNUstepDefaults first if you want to give it another try.

And to close the loop, Oolite goes for V-sync by default and, assuming no intervention using the animation_timer_interval parameter is attempted, it has an FPS cap of 200.

[Cody]

Best to make a backup copy of .GNUstepDefaults first...

<chortles> I learnt that lesson the hard way!

Insert animation_timer_interval = 0.001; for a theoretical max of 1000fps, eh? Hmm... tempting, I must admit.

[Norby]

If you want a test then check FPS in [wiki]Fireworks[/wiki] with [wiki]CustomShields[/wiki] and [wiki]Trails[/wiki] installed - both make the explosions more nice.

When I made screenshot for the Fireworks wiki page, my internal HD4000 (which usually do 30FPS in game) updated once in a few seconds only.

[Cody]

There was a really good explosion in an early version of Smivs' Xeptatl's Sword, as I recall. I like big explosions!

If I try it, it'll be with virgin Oolite. Not tonight though - brain's foggy as I had the pneumonia jab this morning.