December 31, 201510 yr Author You're right. So, the cards do not "drop down to the P2 state" when water is set to ultra. Only the video RAM declocks to P2....The difference in the video RAM clock speed is generally less than 10% and often not even 5%. What evidence do you base these statements on? Nvidia inspector clearly shows the card is operating in P2 power state and on my 980 memory speed drops from 3500 to 3000mhz, approximately 16%. Also P2 power state brings with it a maximum attainable GPU clock speed which is lower than when operating in P0. Whether, under normal p3d GPU utilisation, this lower P2 maximum would be reached depends on user settings and the balance of a given system (i.e. can the CPU feed the GPU fast enough to stress it? and probably more likely, how high is AA set? or are other GPU intensive features enabled, e.g shadows?)
December 31, 201510 yr Jay ... true, the fact that CUDA seems to trigger the lower P state and has been doing this thru version 4.x, 5.x, and now 6.x -- I gotta believe this is intentional on nVidia's part. I'd be really surprised (given a google search shows this pState issue is certainly not specific to P3D) and has been discussed for several years. Has nVidia "officially" commented on the pState in relation to CUDA? Cheers, Rob.
December 31, 201510 yr Not too many games use CUDA, but nVidia's various video encoders do and if one that uses CUDA is built into a piece of software, the problem would show up. Here's a good example of an official response from nVidia staff regarding Open Broadcaster Software (which uses the nVidia video encoder nVEnc): "CUDA is used in NVFBC. Because many of the use cases where an application uses CUDA, precision has a higher priority over speed. This is part of the reason why workstation CPUs and GPUs typically run at a lower clock speed than their consumer equivalents. We are developing an app profile system to catch cases where CUDA is used for consumer level applications (such as OBS) so that the memory clock speed will not clock." The original response from nVidia was in March, 2015, so they don't seem to be in too much of hurry to fix it.
January 1, 201610 yr Like you said and I agree, maybe issue is the wrong word. It´s more like a characteristic of this technology. 9800X3D@H150i // Msi RTX 5090 Trio OC // 64GB DDR5 6000mhz CL30 // 2TB + 1TB Nvme Dell 27" 2127DGF - 1440p - Gsync - 165hz Thrustmaster TCA Sidestick Airbus // TCA Quadrant Airbus // TFRP T.Flight Rudder Pedals // Logitech Flight Multi Panel
January 1, 201610 yr Commercial Member Just checked and my cards change from PO to P2 with water on ultra. I run high anyway because some float planes work better with water on high. I will say that the new update really works my three 980's and in cloudy, misty heavy scenery I can see them peak at 90 percent gpu usage. In fact I set fps at 30 to settle them down a bit. Paul Grubich 2017 - Professional texture artist painting virtual aircraft I love. Be sure to check out my aged cockpits for the A2A B-377, B-17 and Connie at Flightsim.com and Avsim library
January 1, 201610 yr I think you can disable CUDA in the triton.cfg file? Interesting to see if the P0 stays on when water is set to Ultra when CUDA is disabled ... and the performance difference. Cheers, Rob.
January 1, 201610 yr Commercial Member I don't want to disable CUDA because it is nice to see this sim taking full advantage of all my cards. Before they were peaking at around 60 but now they are working up a sweat. With the AM tweak cranking most of my cpu cores and the new CUDA things are running fast and smooth finally. Water on high setting is fine with me as it still looks great and running at high rather than ultra lets me use all my float planes. Paul Grubich 2017 - Professional texture artist painting virtual aircraft I love. Be sure to check out my aged cockpits for the A2A B-377, B-17 and Connie at Flightsim.com and Avsim library
Create an account or sign in to comment