JahStories Posted November 10, 2015 Share Posted November 10, 2015 60fps with this new version vs 8fps on the older one But btw visually it's seems that there are a lot more particles showed on the older one... :/ take a look: 1 Link to comment Share on other sites More sharing options...
Ramalama Posted November 10, 2015 Share Posted November 10, 2015 60fps with this new version vs 8fps on the older one But btw visually it's seems that there are a lot more particles showed on the older one... :/ take a look: Indeed, the 33Mill version have bugs like it seems... Compare this 3 pictures: JahStories @60 FPS (Nvidia GTX 780) mich_de @10 FPS (Nvidia GT 740): Ramalama @5 FPS (Intel Iris 5100): It looks like only the intel card have 33m pixels... Cheers :-) 1 Link to comment Share on other sites More sharing options...
Carstiman Posted November 10, 2015 Share Posted November 10, 2015 Hello i5 3570 K / HD 4000 Gainward Nvidia GTX 650 2 GB vram Link to comment Share on other sites More sharing options...
carbo178 Posted November 10, 2015 Share Posted November 10, 2015 Gigabyte GTX650 10fps Link to comment Share on other sites More sharing options...
mitch_de Posted November 10, 2015 Author Share Posted November 10, 2015 Thanks reporting results with 33 Mill version. The shown, computed particles were cut off / limited by changing discrete VRAM cache option. So i build an 2 Mill final version. 12 FPS GT 740, 99% GPU load. If very fast GPUs run in the 59/60 FPS Metal limit OR! gpu load goes below 80% please resport - in this case i must use more particles count. final version on first page. Link to comment Share on other sites More sharing options...
Ramalama Posted November 10, 2015 Share Posted November 10, 2015 Thanks reporting results with 33 Mill version. The shown, computed particles were cut off / limited by changing discrete VRAM cache option. So i build an 2 Mill final version. 12 FPS GT 740, 99% GPU load. If very fast GPUs run in the 59/60 FPS Metal limit OR! gpu load goes below 80% please resport - in this case i must use more particles count. final version on first page. Bildschirmfoto 2015-11-10 um 12.25.00.jpg Hi mitch, It running now at 60 FPS on the Iris 5100... the 2m version i mean... But now it seems, this is the first version I've not getting all pixels... so its running now @60 fps... EDIT: However, its working as a demo, it means we can check if meal is supported or not... so thank you very much for this :-) Everything else doesn't matter :-) Cheers :-) Link to comment Share on other sites More sharing options...
mitch_de Posted November 10, 2015 Author Share Posted November 10, 2015 You got around 40 FPS with first version, but there were more (4 Mill particles) . All 2 Mill should be shown with the 2 Mill version. Only if other gpus also run in the 60 fps limit i will use more particles. 1 Link to comment Share on other sites More sharing options...
Carstiman Posted November 10, 2015 Share Posted November 10, 2015 OSXMetalParticles_2Mill_final GTX 650 2 GB = 20 FPS Link to comment Share on other sites More sharing options...
mitch_de Posted November 10, 2015 Author Share Posted November 10, 2015 Thanks!. If you no tool to check the GPU load (%) you can use atMonitor (attached). In settings enable GPU in menue oder floating window. If gpu load < 85% (on very fast gpus) i may add more particles to compute. atMonitor.zip Link to comment Share on other sites More sharing options...
droples Posted November 10, 2015 Share Posted November 10, 2015 GT-740,DDR-5,2048MB (GV-N740D5OC-2GI) Link to comment Share on other sites More sharing options...
carbo178 Posted November 10, 2015 Share Posted November 10, 2015 Gigabyte GTX650 2Gb DDR5 Link to comment Share on other sites More sharing options...
mitch_de Posted November 10, 2015 Author Share Posted November 10, 2015 96% / 98% gpu load = GPU has enough work Maybe only he GTX 970/80 or AMD 79xx are getting boring Oh dear: AMD 7950 hits the 60 FPS limit - must build new 3-4 Million version. Link to comment Share on other sites More sharing options...
mitch_de Posted November 10, 2015 Author Share Posted November 10, 2015 HD 7950 final 60 fps "GPU load small , 60 fps is max ?" Yep, METAL seems to have fixed vsync with 60 HZ(FPS) max. Cant bei changed. Even not by dev tools (disable vsync) in the past. That is bad, because we can setup more particles and run slower to avoid that 60 FPS limit. But then, the lowend gpus are stalling near with 2-5 fps. Please try the 3Mill test build if you also hitting the 60 FPS limit. I now get only 8 FPS GT 740. Perhaps wie need 4 Mill if your fast GPU is to close to 60 FPS or GPU load less than 80%. EDIT: If you have an AMD cpu, i dont know the speed compared to Intel i5/i7 , the cpu maybe also an bottleneck firing the fast gpu with work. How is your CPU load running that stuff? Normaly should be low (at least <= 25%). OSXMetalParticles_3Mill_test.zip Link to comment Share on other sites More sharing options...
mitch_de Posted November 10, 2015 Author Share Posted November 10, 2015 Looks the window normal, someone in the past posted an AMD gpu result with wried coulors = failure redering= fps not correct. Looks it as the others posted here? Link to comment Share on other sites More sharing options...
mitch_de Posted November 10, 2015 Author Share Posted November 10, 2015 Yep, then some driver bugs in the AMD Metal part. Link to comment Share on other sites More sharing options...
Asgorath Posted November 10, 2015 Share Posted November 10, 2015 You should try rendering to an offscreen surface only to avoid the 60 FPS vsync limit that all on-screen Metal apps are stuck with (due to CoreAnimation controlling the display pipeline). That would allow a much fairer comparison between GPUs and drivers. Link to comment Share on other sites More sharing options...
mendietinha Posted November 10, 2015 Share Posted November 10, 2015 thanks for this. i have a problem. i cannot see my fps lol: Link to comment Share on other sites More sharing options...
npwski Posted November 13, 2015 Share Posted November 13, 2015 3Mill test, 53 fps GTX Titan (Kepler GK110), Nvidia driver official 310.42.05f01 CPU load ~5% Link to comment Share on other sites More sharing options...
joe75 Posted November 13, 2015 Share Posted November 13, 2015 46fps for 770, beta3 C40a drivers Link to comment Share on other sites More sharing options...
mitch_de Posted November 13, 2015 Author Share Posted November 13, 2015 some new METAL playground N-body Metal demo , well known as cuda, opencl bench - now computes with metal - but i see that more as an demo,not as an benchmark threadwidth + maxthreadspergroup may very between drivers/gpu models and are shown as info Key s switches between METAL devices - if you have more than one. Info: Metal is limited to 60 FPS - if you run in that limit you can use key + to get more bodies (up to 96K) = more gpu work, less FPS 16k bodies (= min) GT 740 45 FPS 32K bodies GT 740 11 FPS 64K bodies GT 740 1,8 - 2,8 FPS ( stalls, sometimes crashes) DL: NBody-Metal_test1.zip Link to comment Share on other sites More sharing options...
Hangten Posted November 14, 2015 Share Posted November 14, 2015 EVGA GTX 670 GPU core speed 1.12GHz GPU memory speed 6.1GHz As read on HWMonitor.app Tried it again but have different results don't know why? Link to comment Share on other sites More sharing options...
maromi Posted November 15, 2015 Share Posted November 15, 2015 definitely optimized for IGPUIntel 4200U 1.6 GHZ GPU 4400(half the cores of 5000/5100) max 1 GHzusing Intel Power Gadgetbenchmark starts at 14 Watts CPU 1.6 GHz/ GPU 1 GHz then changes to 9 Watts CPU 800 MHz(idle speed)/GPU 1 GHz and the framerate actually improves metal particles2M- 35FPS Link to comment Share on other sites More sharing options...
mitch_de Posted November 15, 2015 Author Share Posted November 15, 2015 Yep, main goal of metal is iGPU/embedded usage. Anyone has already tried new Nbody-Metal? (look post #70) Would be interesting if the switch (key s) function works and you can test any METAL device which is available. Link to comment Share on other sites More sharing options...
Ramalama Posted November 15, 2015 Share Posted November 15, 2015 Yep, main goal of metal is iGPU/embedded usage. Anyone has already tried new Nbody-Metal? (look post #70) Bildschirmfoto 2015-11-15 um 11.42.55.jpg Would be interesting if the switch (key s) function works and you can test any METAL device which is available. I haven't second Graphics... only Iris 5100 Cheers :-) Link to comment Share on other sites More sharing options...
Asgorath Posted November 15, 2015 Share Posted November 15, 2015 Yep, main goal of metal is iGPU/embedded usage. Anyone has already tried new Nbody-Metal? (look post #70) Bildschirmfoto 2015-11-15 um 11.42.55.jpg Would be interesting if the switch (key s) function works and you can test any METAL device which is available. Well, you should probably enable an offscreen rendering mode to let the discrete GPUs actually run at full speed. Also, you should move all your resources to the framebuffer memory by using MTLStorageModeManaged and [MTLBuffer didModifyRange:], as leaving the particle data in system memory (i.e. MTLStorageModeShared) will be a huge penalty for discrete GPUs. The goal of Metal is to dramatically reduce the CPU overhead typically associated with OpenGL, but you can't just take code that runs on iOS or iGPU and expect it to work well on a big discrete GPU. Link to comment Share on other sites More sharing options...
Recommended Posts