Gadgetory

not long ago we open discussion about Andy's Oh cat tool which was a presentment updates i am be made and added an interface to present mom made it a lot easier to used for new api testing and the sort of replacement frats and we were sitting on that in a beta stage of the software and we've also been working with nvidia for the last few months on their new software fcat vr before getting to fcat vr this coverage is brought to you by thermaltake and the new contact silent 12 cooler which is a $25 cooler and a competitor to the hyper 212 so be our benchmarking is actually really hard to do properly we don't have the full numbers can't release them yet but very shortly we'll have the full benchmarks of VR on different video cards the reason we are benchmarking is hard is because it's not the normal type of testing we do first of all the headset has a force to be sync of 90 hurt and that is for motion sickness reasons among other technical reasons this means it's difficult to actually measure the true FPS get to do some calculations and softer side to guess at what the FPS is beyond 90 hurts and then if you're below 90 hurts there's another problem and that problem is when the NVR if below the if you're not meeting the runtime within the designated amount of time normally 11 milliseconds to produce a frame the runtime then has to decide what do I do for this next frame because it's got a few options one it can reap reject an old frame with no updates that's what we're calling a warp miss and in that scenario that's sort of a worst case that is when the user experience is absolutely no updates to their frames at all that means you're seeing the previous frame which would be comparable to a stutter in traditional FPS benchmarking or gaming and if you see a previous frame in VR a couple times anyway it will eventually start to induce sickness or uneasiness or just not good feelings and you're dead because you're wearing a thin on your face and it's the only thing you can see so that's a wart miss the other thing that we have to measure is a drop frame a drop frame is when the runtime decides to take a previous frame and then synthesize a new frame came by updating the head tracking but animation within the frame remains the same so if you have character movement if there's if you're looking at a character and a sort of firing position that animation will not change but if you move your head around the head tracking will update and that helps that's important to reduce the chance of user sickness it's not an ideal experience just like tearing in a game is not an ideal experience but it's probably not going to make you vomit so that's a good thing these things are not easy it turns out to measure with traditional tools like fraps or present mom we actually have no way to measure them and you can use things like steams 5 FPS monitoring and frame time tracking tools but it's not the best solution and it's got some limitations so the fcat be our approach is twofold one there is a hardware solution we set up a secondary hardware capture machine and that's one for validation but two there's a software solution that measures a dozens of different data points and variables creates a massive CSV file we have some b-roll of that I think and and then we can take that data and start crunching it into graphs and how many work mrs. are there drop frames what's the average latency frame to frame what's the worst plate and see what the best latency all that stuff we can get from there along with our traditional one percent and 0.1 percent lows so I'm a hardware capture side that we have that one the really important thing is epcat is an nvidia tool of course they say that it's fair and they are planning to release it as open source which will allow you to look into the code but it's still in that video tool so we have to validate it and make sure that's being fair and one of the best ways to do that is with hardware captured so we set up a secondary pc that has a vision i think it's an SD HD for something it's a pretty expensive capture card and that has very high bandwidth that card out of it has a splitter and into the splitter goes the sort of the the hdmi from a splitter box we actually get through the complicated the cables alone are the heart part of the our benchmarking the way the cables are set up and the connection order does matter by the way you have the benchmarking system with the video card under test out of that comes displayport to your monitor and then hdmi to a splitter box and then out of the splitter box comes to other cables one it goes to the link box for the headset so we've got two boxes in there and that goes to the headset the head mounted display so I'd be the vibe and then the other cable coming out of the splitter box not the link box with a splitter box goes to the capture system into that splitter and then that goes into the capture card and from there you're able to record using virtualdub which is a third party tool the actual output to the head-mounted display and then that output on it has an overlay a VR benchmarking overlay which cycles through something like eight colors so we can step through each of those colors one frame at a time and see if there are any that are missing or see if there are any other problems within the frame now traditionally you would use this for looking at tarian within frame but with fcat VR we're mostly looking at are there any dropped frames there can also be dropped frames because of improper testing so the big problem with with the our testing is that it's such high bandwidth that we have 21 use a high bandwidth capture card and to use either raid SSDs or something like an intel 715 we have an Intel 750 one terabyte drive from vs mods thank you Bob and rod for that and that is capable of keeping up with the recording and the capture it's also because one point two terabytes capable of storing everything this is a big problem too because each one minute file or so is about 50 gigabytes and then from the hair we analyze the file we output a CSV that says if anything is missing we can see if it's mind you hurt if there any drops and use that to validate against the software capture which was running on the capture system which has recorded all of the frames synthetic and otherwise from the video card so we can use that for validation the next step is to compress the 50 gigabyte file which we can do you're probably watching one of them at some point this video and that is done with the script that we wrote so we use it for our own media production the script compresses it to eighty percent or smaller of the original size or i should say it's it's eighty percent smaller or greater than that of the original size we go down sometimes to a couple hundred megabytes depending and we don't really lose any quality so that snakes to script that we wrote in house I can't release the full members yet you've probably seen some examples while going through all this stuff in the video but we'll have numbers shortly if you want to look into this the real main news item here there is now a good way the benchmark VR it has taken several months of back and forth with NVIDIA to get it functional it was originally rather than this nice interface they have now we were using a actually code so we had to modify Perl code to output the chart we have used code to align the benchmark data and we also has used things like regular expression editing to do anything more complex that was really not not easy to work with but it's a lot better now it's a UI and they're going to be releasing it to the public I think it's going to reviewers first and we'll have numbers soon if you're curious about it we'll try and put a tutorial out there as well as that you can use it yourself in terms of fairness will validate that but the hardware capture should pretty much resolve any concerns that because we're going to be able to see the actual raw output from the headset to the capture systems that will validate that but that's all for now we'll have a lot more on this so it's more depth really the house planning to get but go to patreon.com/scishow help that directly and help fund our testing efforts we looking at fcat vr shortly hopefully oak at once it's got a more final state subscribe tomorrow I'll see you all next time