Gadgetory

man versus machine is an argument that will become painfully increasingly relevant as humans desperately he grasped at their last fits of non obsolescence before the robots take over and take all of our jobs and so in part of this celebrating of man versus machine and the advent of automation we're going to be benchmarking automated overclocking versus manual overclocking the competitor is the EVGA 1070 TI SC which uses an auto SE to bypass the Nvidia rules of hey don't overclock this and they said okay we'll just make the sauce for do it so I'm gonna see if I can beat that within the time frame that this thing auto over clocks itself so I've got the same amount of time to do an OC here before that this video is brought to you by thermal Grizzly makers of the conductor not liquid metal that we recently use to drop 20 degrees off of our coffee leak temperatures thermal grizzly also makes traditional thermal compounds we use on top of the IHS like cryo not and hydro not pastes learn more at the link below so the way this thing works this is part of precision and basically it's a tool that does the Volt frequency curve scan so it's a VF scan and it looks at the what it thinks would be the ideal voltage and frequency now there's two different things here to look at one is the basic mode basic mode basically says that it has a voltage and frequency lookup table in the card so EVGA I think has three predefined frequencies and voltages for an overclock that the basic mode checks against and then whichever one appears stable is the one it will pick I don't know what the table is I just know there's at least three options the other mode steps through manually and increments the VF across the entire range it's a bit more complicated and it also seems a bit harder to work with so we're primarily gonna be testing on basic mode because it works better right now and this version of precision and the the takeaway here is that it's there's a bit of a challenge the better the silicon is the bigger the disparity between my overclock and the machine because I fully anticipate that I would be able to beat this but as the silicon is worse and quality and lower down in the tier of the ASIC quality so to speak the overclocks will be closer because the table is already predefined in a way that it won't overshoot it will never shoot beyond the capability as if any of the silicon really that these cards ship with so the closer it is to the maximum clock out of the box due to silicon quality the closer it will be to my potential overclock because I would hit the same wall as the automation so depending on how good that quality is will basically dictate how much I win by because I am assuming I will win here just like just like any other human who has their job threatened by automation maybe you think that you can drive cars better than Tesla can I'm going to boldly state that I can do it better all the way until I'm screaming down the highway flying into an accident because I didn't trust the automated car to do it for me so what we're gonna do is pretend basic now before this I did test the longer VF and it just it didn't work quite as well I this perversion of precision is still in development and it needs some work still so we're gonna start this we have three options here manual which is what I'll be doing later quick test which takes 15 to 20 minutes to do and full task takes damnit I clicked full and full test which takes 30 to 60 minutes to do but doesn't fully work so we need to not do that one right now okay there we go quick test so what this is doing now is it's running a rebranded version of fur Marc fur Marc is a bit tricky for overclocked testing because it's a power virus workload it's not really representative of games and sometimes for Marc will give you clocks in terms of offsets that are higher than what is realistically achievable in a game or a 3d mark but it's what they're using it works sometimes as a baseline and that I guess as long as you're conservative which I think this Auto OC will be it's fine so this is just scanning for artifacts right now it's with that graphic indicates and all its gonna do is scan a whole bunch of the different OSI range ISM it's at least three but there's more options in there for it to scan especially if you do the manual VF scan and then determine what the overclock will be so we're just gonna let this run for a little bit finish log how much time it takes to finish and then I will have that amount of time to apply a competitive overclock to this one okay so all this did was basically apply a 114 megahertz offset on the cores this is again a 1070 Ti so that's gonna matter so 114 on the core it did not increase the power target and it doesn't increase the memory offset either it looks like so it's basically just a core overclock of about the value that EVGA or any other company would do out of the box normally under scenarios where NVIDIA hasn't restricted them so 114 is what we have to beat now I'm going to run this in fire strike and we're not doing games or anything there's really no point because I mean it's just performance scaling and all we're doing this clock for his clock matchup here so we're gonna do each fire strike extreme ultra at normal and run all of them collect the scores and once we have those scores I'm going to run my own overclock within the same about 18 minute window that it took to finish and then we'll have the data to see who did it better so here's where the actual competition starts I have about 18 minutes to beat the auto OC and what we're gonna do is first off immediately set this power target 220 percent the Auto C does not do this and I did set it to 120 percent after its Auto C just as he helped out at all but it did because we weren't hitting the power limit so we were hitting other limits first so reset that's 120 secret here for anyone who doesn't know the minute that Pascal goes over 60 degrees Celsius you start dropping clocks it's by design it's part of boost 3.0 it'll go down 25 50 megahertz as you increment up and towards 80 to to 84 so 82 to 84 although the more obvious throttle point is not the only one it is an incremented stepping and yes I know this counts against my time but because I'm so confident that I'm going to win this battle I don't really care and you get to learn a bit about Pascal if you didn't already know it so let's start with the fan speed at 65 I don't want to cheat by setting it to 100 and making it too easy so you know what let's let's take it down to 60 let's make it even easier for the program to win start 120 I'm gonna set up a stress test extreme 60 loops windowed and we need gpu-z and make sure OBS is going yes so gpu-z we're just gonna look at the sensors tab while this is running and I already know of course that 114 is a minimum stable overclock because that's what it did so there's definitely Headroom there above it now it did not do any overclocking on the memory but I know from benchmarking dozens of Pascal cards at this point that they've never really failed to do less than 500 it's pretty rare for them to be incapable of hitting that we used 400 megahertz offset on the memory so are 150 offset on the core 300 on the memory right now and you know what I think we've got some time here still we're doing pretty well on time we're already better than it in the overclock Department let me just take a minute here resize my windows and make sure the environments all get hit okay so we're at 20 37 megahertz core now this is a bit misleading because boost 3.0 as you all know fluctuates depending on what's going on so it's gonna bounce down to 2025 sometimes things like that that will stabilize more as you approach 40 degrees Celsius but we're not gonna push for that on this air cooler so let's go forward we're at 150 right now let's do 175 and just set apply on that see if we remain stable I'm gonna give this a minute or so to burn in and while it's burning in we're still at 55 on the GPU we're good there I want to stay under 60 ideally if I really want the best chance of hitting a higher overclock or push that up to 70% now if so if if we can't stay under 60 it's not the end of the world it just means that we might be down anywhere from like 13 to 25 megahertz on the core not a big deal because we're gonna beat the thane anyway but if you're doing this for your own competitive scoring reasons the fun thing about fire strike is that it's a short test so you can blast the fan speed keep it under 60 easily for the entire test and you'll score higher because the frequency is gonna be more stable and higher so we're looking good here let's do 200 and then all this will have to be burned in separately and we'll do the benchmarks and things like that it's not quite as easy as like it was stable for one minute so it's therefore stable that's not how it works but that's not the point of this video either so typically what we do for reviews is you let it burn in separately for a while and then you play a bunch of different games cuz some games like for honor will enumerate the clock differently than others and then synthetic applications and then you can use all that experience to figure out where the clock should actually be right now we're at two 25 megahertz offset experience tells me that this should fail at some point soon because we're pushing into the 20 100 megahertz range on the core which is absurd to be able to hold on to for most cards that we've worked with in this class so we will see if it holds on but typically does not for now that what we can do is start pushing up the memory a bit too because that will benefit us a little bit oh I didn't even hit apply so I didn't hit apply yet and it's crashed which means it was not the memory that caused that crash although it would seem that way if you thought I hit apply so that means we're not stable at 225 let's go around 200 I think we still have time here to play around with the memory so start that test back up we got 200 now let's go up to 500 megahertz on memory make sure this is 200 see how that goes waiting on fire strike costing me valuable time okay so that's running 200 offset core 500 offset memory we might be able to push it a little more but let's just see if this is stable and we're gonna have to call the clock in a minute so if it is we'll keep these frequencies for the GPU so this is looking stable looking pretty good we're gonna run the tests at these frequencies because they're fine they seem to be holding you can push it a little further if you were really trying but we're trying to just beat the auto overclock in the amount of time that it gave us to overclock so that's our competitive bench with the auto AC it's 200 and 500 power target being a big part of it and then we're gonna run some benchmarks and see what the numbers look like for each now I fully expect obviously now that we've seen the clocks are higher that these scores will be higher than the auto OC the question is is the autumn OC good enough that for a user who for whatever reason maybe is timid about overclocking or doesn't just that frankly doesn't have the time to learn the basics to get started I had a head start here I already knew how to do all that stuff so there's reasons to not manually OC the question is is the Otto's OC actually worth it or should you just use the card stock at that point so that's what we're gonna find out okay so the results are in let's start with the stock card the stock card with fire strike 1080p was scoring 20,000 229 points and the roughly 18 minute overclock done automatically by EVGA was scoring 20,000 885 points and when we applied a 20% power offset to trying to help it that didn't really change because we were already hitting other limits so the difference there Auto versus stock you gain 3.2 percent by running the auto overclocking utility which again it's a lookup table it tests each one and then if it crashes during that test or has artifacts it goes down to the next one that's how it works so that got us three percent ish and this is fire strike this isn't a video game your mileage may vary games behave differently some of them love clocks some of them really don't care and just need more shaders so it really depends on what you're doing but for purposes of scalability in comparison three percent versus mine was eight point one percent versus the OTO OC not versus stock but versus Auto so at plus 200 and plus 500 core in memory I was at 22,000 576 points versus the auto OC of 20,000 885 so an extra 8% their fire strike extreme the Ottawa OC got 3.7 percent higher than stock which is a comparison of 98 59 points versus 95 83 points and mine got eight point seven percent higher than the auto OC a difference of ten thousand eight hundred six points versus 98 fifty nine points and then if you wanted to look at the FPS numbers that's a difference of 57 for mine versus 52.4 for the auto OC whereas the fifty one point three for the stock configuration fire strike ultra which is a 4k test has the difference is slightly bigger still because we start benefiting from other things potentially that memory clock that the ottawa c did not have the auto OC is two point nine percent ahead of stock with a score of 48 74 versus forty seven thirty six and my OC is 53 29 versus the others so that's a nine point four percent increase over the auto OC for the overclock that I did so the takeaway here isn't to say anything like look at me I'm a great overclocker because that's not really the case it's just if you want to do this yourself it takes just as much time to do it as with the auto C if you have a foundation of knowing what you're doing and you already know where all the tools are and what the download things like that the auto overclock is still useful it's useful as a feature for people who want basically a one-click option to recover the clocks that were more or less taken from them by Nvidia with a 1070 TI because the board partners are not allowed to release with a pre overclock on a hardware level like with the other cards so one click button in software lets you get that back but if you're doing that anyway it's not that hard to do what I just did take same amount of time and again it's like another eight to ten percent Headroom in the clock over auto so not bad the fact that this can do the lookup table and everything get it done in 15 to 20 minutes is pretty good it's certainly a step up from what used to be the case the takeaway though if you do it manually it's always gonna be better for now until Tesla Motors gets into video card overclocking and wipes out the content creator jobs but until that time you might as well just do it yourself and the tools you need everyone here already knows this stuff but afterburner GPZ is helpful a synthetic software like fire strike or heaven are good the reason I like those is because they're really easy to repeat and the data is fairly consistent and it's easy for anyone to run the benchmark without having to understand a complex game that might may have special behaviors so that's what I would recommend but you know if you're just trying to buy a card and click a button then Auto C gets you another 3% if you really want that and then from there I would recommend use that as your foundation starting point of saying this is the minimum it can do safely for sure and then bump it up until it crashes and that's how you use the auto OC if you aren't fully sure where to start so thank you for watching as always you can get a patreon.com slash gamers and excellence to helps out directly or you can just subscribe for more because watching really is the biggest support we have right now and leave a comment below or go to store that gamers and access net to pick up a shirt like this one or one of our stickers I will see you all next time