mitch_de Posted June 28, 2010 Author Share Posted June 28, 2010 ARRRG, 1.7.0 was build with an older slg engine 1.5B4 insted of 1.6Beta1 I uploaded an 1.7.0 V2 to macupdate.com, which fixes that. Now the new sppm method can be used in interactive mode. Also maybe some little speed diffs to older slg engine. Link to comment Share on other sites More sharing options...
mitch_de Posted July 4, 2010 Author Share Posted July 4, 2010 New working ATI 58xx Users wantend Link to comment Share on other sites More sharing options...
MaLd0n Posted July 4, 2010 Share Posted July 4, 2010 line 58 .. is already working on OS X? mitch_de I can control the gpu cooler (bios mod) on Osx but overclocking does not work You know what? :S Link to comment Share on other sites More sharing options...
JwKuo87 Posted July 4, 2010 Share Posted July 4, 2010 Sapphire Radeon HD4870 Toxic Link to comment Share on other sites More sharing options...
mitch_de Posted July 6, 2010 Author Share Posted July 6, 2010 line 58 ..is already working on OS X? mitch_de I can control the gpu cooler (bios mod) on Osx but overclocking does not work You know what? :S YEP, they got 58xx and 57xx line to WORK. At least OpenGL works. OpenCL is untested (smallluxGPU will show - also good that since 1.6.4 i added the console tab which will show errors (OpenCL Info about found devices). Sorry, i didnt now anything beside the normal powermanagement and/or the chameleon video bios injection Link to comment Share on other sites More sharing options...
MaLd0n Posted July 6, 2010 Share Posted July 6, 2010 Thanks Link to comment Share on other sites More sharing options...
jasjeet Posted July 7, 2010 Share Posted July 7, 2010 Hi got the Asus 5850 running today Mid, Type 1, 5.9 seconds High, Type 1, 11.2 seconds Ultra High, Type 1, 28.4 seconds Link to comment Share on other sites More sharing options...
mitch_de Posted July 8, 2010 Author Share Posted July 8, 2010 THANKS for first 5870 results ! Could you try also the OpenCL Pixelbench (similar to CoreImage, but its OpenCL code) ? Link to comment Share on other sites More sharing options...
jasjeet Posted July 8, 2010 Share Posted July 8, 2010 Those were 5850 results. PixelBench doesnt work, it just stays on the same screen, and after 30 secs nothing happens. Link to comment Share on other sites More sharing options...
mitch_de Posted July 8, 2010 Author Share Posted July 8, 2010 OK, than OpenCL (pixeldevice functions) arent ready for 5xxx gpu - not so big problem. You will see some error messages in the Console Tab of the smallluxGPU after trying the Pixelbench. Fermi 4xx (already on start) cant even run normal OpenCL benchmarks with latest drivers. Link to comment Share on other sites More sharing options...
Spanky Deluxe Posted July 8, 2010 Share Posted July 8, 2010 Radeon 5970 Mid GPU only (Type 0) - 10.9 seconds Mid GPU only (Type 1) - 5.9 seconds High GPU only (Type 0) - 20.3 seconds High GPU only (Type 1) - 11.2 seconds Ultrahigh GPU only (Type 0) - 54.5 seconds Ultrahigh GPU only (Type 1) - 28.4 seconds This is a 5970 with stock clock speeds. Obviously only one GPU works in OS X. 5870s should be a little faster due to their higher clock speeds although I'm not quite sure what's going on since jasjeet's 5850 got exactly the same results. The 5970 and 5850 do run at the same clock speeds but the GPUs in the 5970s have 160 more stream cores. At some point I'll probably get round to flashing the BIOS with 5870 speeds and voltages. Link to comment Share on other sites More sharing options...
mitch_de Posted July 9, 2010 Author Share Posted July 9, 2010 Thanks ! Type 1 OpenCL Code runs mostly on all newer GPUs (GTX 2xxx, ATI 48xx) better than Type 0, which runs faster on older/low end GPUs like 8800 GT, 9400M. Even new 5xxx GPUs runns well with OpenCL (fastest GPU now), offical+next OpenCL drivers will perform even better. About smallluxGPU benches: They are GPU only, but that doesnt mean that is is independend fromcpu+bus speed. The faster the GPU the more the cpu+bus speed have also an effect on the time used in the gpu only benches. CPU+bus must feed(and read) GPU fast enough. So you may measure some time diff running 5870 between an C2D + Highend i7 CPU. Diff between both systems running 8800GT will be much smaller/near zero. You may also check the GPU LOAD (=high % Load means GPU must not wait for CPU) in the interactive scene running GPU only. If LOAD of GPU gets less than 80% (some scenes with my 8800GTX/C2D) CPU cant feed gpu fast enough and GPU must wait. You see that problem running hybrid (CPU+GPU) there my 8800GTX mostly has some wait time (only 30-70% LOAD, depends on scene). Would be fine if you check some scenes , select GPU only and look after the LOAD % (marked red in the sceenshoot)- all above 90% is perfect. Link to comment Share on other sites More sharing options...
atlee Posted July 9, 2010 Share Posted July 9, 2010 ATI RADEON 5770 1GB 825Mhz Edition But program only shows as 725mhz. Type 1 Midrange GPU 10.3 Sec Type 1 Highend GPU 18.3 Sec Type 1 Ultrahigh GPU 50 Sec Type 0 Midrange GPU 20.3 Sec Type 0 Highend GPU 38 Sec Type 0 Ultrahigh GPU 103.3 Sec All my LOADS are 98% and goto 100% Link to comment Share on other sites More sharing options...
mitch_de Posted July 9, 2010 Author Share Posted July 9, 2010 Perfect if GPU LOAD (shown in Interactive Mode, GPU only) is such high(>= 95%)! Final drivers will of course get higher OpenCL speed. The OpenCL Framework now will for sure not be optimized for 5xxx series. And OpenCL speed is much dependend from the IQ /AI of the on the fly OpenCL compiler - beside the overallspeed of the gpu (units). So lay back and be happy with new 5xxx ATIs - Fermi cant do any OpenCL this time Link to comment Share on other sites More sharing options...
mitch_de Posted July 9, 2010 Author Share Posted July 9, 2010 "Radeon 5970 Mid GPU only (Type 0) - 10.9 seconds Mid GPU only (Type 1) - 5.9 seconds High GPU only (Type 0) - 20.3 seconds High GPU only (Type 1) - 11.2 seconds Ultrahigh GPU only (Type 0) - 54.5 seconds Ultrahigh GPU only (Type 1) - 28.4 seconds This is a 5970 with stock clock speeds. Obviously only one GPU works in OS X" Q: Can you check that ? Use interactive Mode, gpu only and look in the shown info lines at the bottom. If there is only one line (Cy..., Load, ...) listed , only one GPU is used. If there are two (sure little diffenet/changing) info lines then both GPUs are used. Normally smallluygpu uses automatic all gpus which it can find - that eqaual in interactive mode and benchmark mode. Link to comment Share on other sites More sharing options...
jeanlain Posted July 11, 2010 Share Posted July 11, 2010 I assume that testers of radeonHD 5XXX still have the stock openCL framework (10.6.4) that hasn't been updated for the cards. Can this affect performance ? Link to comment Share on other sites More sharing options...
mitch_de Posted July 11, 2010 Author Share Posted July 11, 2010 Sure ! ATI 5xxx design is different to older gpus - so OpemCL framework will be optimized/updated in the future. OpenCL on the fly compiler (ATI Part) has very much effect by how much optimized for the OpenCL speed. Its like normal compiler (gcc) compiling code on newest CPU with no/less optimizing parameters for that new cpu type (like Pentium I compiled for i7 CPU= runs but much slower). Beside that, OpenCL is in the early stage - all drivers/frameworks, even "final" ones will get major optimisations within the next 2 years. So there is no hurry to get OpenCL working until OpenCL isnt used / rare used in the apps. Link to comment Share on other sites More sharing options...
Gugiwuz Posted July 11, 2010 Share Posted July 11, 2010 Sapphire Radeon 5770 1GB with SmallLux 1.7.0 (10.6.4) Mid GPU only (Type 0) - 15.4 seconds Mid GPU only (Type 1) - 8.4 seconds High GPU only (Type 0) - 28.9 seconds High GPU only (Type 1) - 15.8 seconds Ultrahigh GPU only (Type 0) - 78.2 seconds Ultrahigh GPU only (Type 1) - 40.1 seconds Link to comment Share on other sites More sharing options...
mitch_de Posted July 17, 2010 Author Share Posted July 17, 2010 Soon there will be an update to V171, which has slg V1.62Dev insie and also some fixes. I removed High gpu only bench and added UltraHigh Sponza gpu only. Would be fine if you check new Sponza Benchmark. I uploaded the V171 to RS for getting some reference benchmark times. No major changes in known Luxball UltraHigh benchmark . Link to comment Share on other sites More sharing options...
danielkza Posted July 17, 2010 Share Posted July 17, 2010 Weird, I can't get it running because my HD4830 seems to be detect as an HD4870... Link to comment Share on other sites More sharing options...
mitch_de Posted July 17, 2010 Author Share Posted July 17, 2010 Weird, I can't get it running because my HD4830 seems to be detect as an HD4870... The OpenCL driver from Apple detects all ATI 48xx as 4870 - doesnt matter if 4850, 4870, 4890 or 4830. So its not the main problem if slg didnt run (error message in the result window / not run and error message in the console tab of the app) What happens if you run the bench (Midrange,....) ? Some error messages in the console of the App (TAB console) ? It could be that Apple didnt support 4830 gpus with their OpenCl drivers. Link to comment Share on other sites More sharing options...
mitch_de Posted July 27, 2010 Author Share Posted July 27, 2010 Now 1.7.1 PreFinal: DL Link in later posting ! - removed OpenCL Code Type selector (0/1) from Benchmark. Always use now Type 1. - added SPPM Method Type - special raytracing method in OpenCL. Glass / lights look mostly better, but more cpu usage - added intances 130 Mill Scene (fixed res / special OpenCL Code = no res+no Type selector for that scene) Instances 130 Million (triangles) may NOT work on ATI OS X GPUs. At least on Windows the ATI OpenCL drivers fail, Nvidia drivers can run that scene on Win/OS X Also huge RAM usage, therefore 3GB RAM needed to perform fast. I would NOT run that GPU only on lowend GPUs. - added again CPU cores count selector (2 or 4 cores) for interactive scenes Would be nice if someone with ATI 5xxx could test the new smalluxGPU(OpenCL) instances 130 Mill triangles Scene. That huge scene may not run already (Nvdia does, ATI Mac unknown, ATI Win not). Please post interactive mode instances scene gpu only - if it worked. Speed (in the interactive mode) is shown as xyKile samples/Sec - my 8800GTX gets around 27k samples/sec. in gpu only. EDIT: Nvidia GTX 285 gets 148k samples/sec (gpu only). Will befine to see how fast ATI will raytraycing that huge scene (instances). It can be that the new OpenCL Framework (comes with the iMac Mid 2010 update) already fixed that - but untested. Also benchmark times for sponza and luxball uktrahigh are welcome. Link to comment Share on other sites More sharing options...
mitch_de Posted July 31, 2010 Author Share Posted July 31, 2010 DL 55 MB: 171V3 with benchmark instances scene (8800GTX = 150 sec) http://rapidshare.com/files/410151278/smallluxGPU171_V3.zip PS: You get an ATI Info message box which inform about perhaps not working instances scene with actual ATI OpenCL drivers (iMac Mid 2010 or 10.6.5 Update may solve this). Link to comment Share on other sites More sharing options...
mitch_de Posted July 31, 2010 Author Share Posted July 31, 2010 Rob from barefeats tested slg on new iMac Core i7 with Radeon HD 5750. On the Benchmarks,I get 43.8 sec on Luxball Ultra GPU and 46.0 sec on Sponza Ultra GPU. Really fast for an iMac !!! Even with iMac Mid 2010 Update the instances scene fails in OpenCL - only Nvidia OpenCL drivers work. Link to comment Share on other sites More sharing options...
mitch_de Posted August 11, 2010 Author Share Posted August 11, 2010 You could also check/Bench FPS with my new smallOpenGL FPS Bench - (reference results already available) http://www.insanelymac.com/forum/index.php?showtopic=228175 Be free to post your results in that thread for compare. Link to comment Share on other sites More sharing options...
Recommended Posts