Gadgetory

this video topic had a terrifying start its conception was build Zoid who came to me and said I want to do something and that is an immediate sign that you should be afraid so builds I'd wanted to provide a custom BIOS for our X 570 that had some timings tweaks to it this is something that you might have done a year two years ago if your remaining or something like that but you can also do some timings tweaks that might affect performance in things like games and so we started the idea with a custom BIOS and with some memory timing tuning on the video card so for the rx 570 and expanded from there to use a and these built-in timing tuning through their drivers today we're looking at how that affects performance if at all between an rx 570 and Vega 56 before that this video is brought to you by drop and their HD 6 xx headphones the 6 xx headphones offer high-quality audio output with wide device compatibility mouth drops 6 XX also includes a 6 foot cable with 3.5 millimeter plug alongside a quarter inch adapter for those who might want to plug into an app the 6 XX focuses its energy into balancing the sound leaning toward warmer and bass your audio they're also easy to disassemble and replace individual parts making these headphones trivial to maintain for long-term use learn more at the link below this content ended up being really straightforward so unfortunately the BIOS that we applied from builds hood we did have issues getting it to really work properly it worked it applied builds I did provide a BIOS that just changed the the memory strap and the end result was using the different variants that builds it supplied us the card which is not stable it was freezing it was crashing and it just couldn't complete benchmarks so there is more tuning we can do and for this specific card it was just struggling so we might need a different video card with different memory modules or something but there's still stuff we can do through software through third-party software especially from AMD there's a lot of memory tools out there for various and the devices CPUs and GPUs alike so there's salt or we can use but what we ended up doing for this one is just sticking with the drivers because ages ago when Scott Watson visited us we said we wanted to benchmark the AMD timing tuning utility in the new driver set that tuning utility is really just a drop-down so it's got automatic which is stocks that's the stock timings so when you're thinking of different memory timings for GPU is you have gddr5 and HBM these are different from DDR but in some sense the actual timing options are similar it's just the numbers will be different and there are a few different options mixed in there as well so it's it's very simple for the drivers it's automatic which is stock level one which we don't know exactly what that does it's all hidden in the software or level two don't know what that does either but we tested those three and then hopefully our plan is to either work with Bill Boyd and get a BIOS for a different card that'll accept it and run without crashing or use a software utility to do manual / timing tuning on GPUs which we're not particularly skilled at because we've never had to do it but it's something we could hack around with and see if there's any difference in the results when changing the memory timings for the video card all we're doing for these tests then is running the RX 570 and Vega 56 through some benchmarks with the AMD driver changes to level 1 level 2 timings we did a lot of test passes so for fire striking four times by it ran dozens of test passes per per line in the chart and that allows us to established a really tight tolerance for errors so we can see exactly how much variance there is and then for games we ran the usual games and test passes we did validation passes for these old stock results so there's have been updated for this test and then the rest of it is well it's stuff like we did testing with and without rebooting and that didn't change anything either so let's get through the results and talk about the the data if Andy's driver utility for memory timings does anything in games and gaming workloads which fire striking time so I are a quick note as well our understanding is that these were initially introduced to the AMD drivers because crypto miners were doing timing tuning and it actually did Matt it actually encrypt through cryptocurrency and mining our understanding from from speaking with fields right and AMD is that it really did influence the performance games are different a lot different they're so different that it was cryptocurrency mining you can plug into an x1 PCIe slot and you're fine with games if you do that you're completely ruined for your performance so they behave very differently and this is why the performance in gaming is going to be a lot different than what you'd see in crypto mining in terms of the improvement from using these drivers so let's get into the numbers after running dozens of tests we found firestrike to have a standard deviation of about 0.2 to F ps4 graphics 1 and 0.06 FPS for graphics 2 with the RX 570 and for Vega 56 for the gigabyte RX 570 with stock timings we measured graphics 1 at 65 point 5 6 FPS average across more than a dozen test passes with graphics 2 at 50 2.2 FPS average graphics 2 will be more accurate of the two numbers and should also be more sensitive to memory changes memory timings level 1 put the graphics 1 score at 65 point 8 3 FPS average a difference of plus a zero point 2 7 FPS considering the standard deviation of 0.2 to FPS for this test this is not a meaningful difference the result of 66 point zero 3 follows the upward trend begun by timings level 1 and does technically exit deviation against stock even at the extremes but only barely we still can't confidently state if the difference is statistically significant but it does appear to have an uptick in performance even if it does perform better than stock though the gains are irrelevant for this workload graphics to shows more meaningful gains if looked at from the perspective of error but they are still meaningless to the user these differences that would be undetectable to a human even overclocked and with memory over clocks the difference between memory timings at stock and level 2 whatever that means is effectively zero they give 56 showed a range of 0.17 from top to bottom of the average results in graphics 1 or 0.15 FPS for graphics 2 technically we see a decline in these results but they are mo within error we cannot state any meaningful difference once again times pi has much closer final results than firestrike with a standard deviation of about 0.03 FPS run to run for both graphics tests and by the way that's another couple dozen tests per device on the chart for this one the RX 570 stock moves from 25 point five nine fps graphics one to twenty five point nine one FPS for memory timings level two although this is statistically significant as a change based on our error from times by itself we're still too close to feel comfortable declaring a meaningful result remember that although we know time spies variants computers are complicated and could have unknown background tasks that may cause interference that could affect the numbers beyond what we see as our error for time spy in this benchmark either way under the best case assuming that we are perfectly 100% accurate we could declare an increase maximally of 1.25% graphics to shows an increase of 0.2 for FPS for the RX 570 from stock to timings level to a gain of 1.1 percent assuming again perfectly accurate results timing differences with an overclock are also insignificant although looser timings may allow for higher overclocks assuming we had the level of control required to actually manually push down the timings and they're just solved for that kind of does that and we've begun playing around with it but it's not necessarily so trivial Vega 56 again plots a downtrend from stock but like with the rx 570 we don't have full confidence to declare changes without making some assumptions assuming again that the numbers are 100% perfectly accurate which is unlikely when talking about such small changes the Vega 56 slight decays and performance with the other timing levels would be mismatched from what we saw with the RX 570 that said they use different types of memory and memory subsystems more realistically all these numbers are too close to call far cry 5 at 1080 P has the RX 574 gigabyte style card at 62 FPS average with lows at 54 for 1% and 50 for 0.1% the memory timing level 2 configuration puts us at 60 point 6 FPS average for the 2 FPS 1% and 50 FPS 0.1% these results were consistent and within 0.5 FPS average run Ron we can't declare an improvement for memory timing level two although we also can't necessarily confidently state that it diminishes performance either it does look that way here but we're also close enough to error that we have to settle for calling it unnoticeable if different at all they can 56 ran at 98 FPS average stock in ninety seven point eight FPS average with level two timings so we again observed no change 1440p further demonstrates a lack of difference our stock results were at forty three point nine FPS average with memory timing level two at forty three point four FPS average and this is well within error margins we did see more variable 0.1% low performance here but it's hard to know the precise cause of this as the data set is more limited the zero point one percent low range was wider for the changed memory timings potentially indicating some instability or memory errors assuming this change in software is working to begin with this is also something we commonly see with memory overclocking when pushing too hard at 1440p vega these six reference reported sixty eight point six FPS average with timing level two version at sixty eight point four FPS average so no difference here once again GTA v at 1080p put the RX 570 at sixty two point six FPS average with Louis at forty seven point six and forty five the memory timing level two test was almost exactly identical almost seems like nothing even happened we're at sixty two point two FPS average forty seven point six four one percent and forty five point one four zero point one percent that's not functionally the same but in some instances literally the same there's no difference here Fagen 56 had us at ninety four point six FPS average which is the same as level two timings changing and it's ninety four point eight FPS average lows are a little more variable at the zero point one percent mark with the timings change but there's naturally more variance in 0.1 percent anyway so it's difficult to determine the exact cause 1440p has the RX 570 of forty two point seven FPS average and forty two point five and PS average for memory timings level two lows are at thirty two point eight for each and about 31 to 32 for 0.1% lows which is with an error there's no difference between these big 56 ends up at sixty six point four FPS average in both tests so no difference again and finally last one we're showing is f1 2018 for this one we observed the rh5 seventy eight six six point three fps average 1-run stock 65.9 when run with memory timings level 2 and again we can't really see a difference here 1440p testing had the two cards once again identical 51.6 versus 50.9 which is within error margins and those were unaffected we did have more games but you get the idea they're all the same the other games we tested like shadow of the Tomb Raider and Sniper Elite 4 were also the same but there's just there's no point showing more charts not super exciting then no real differences in the tests we ran there were some that were sort of bordering on maybe a difference and that was typically more along the lines of timing level to potentially introducing some lower 0.1% loads than we would expect however that 0.1% low value is already a really narrow data set so we don't have as much data to average against we don't have as much data to check for things like margin of error and so the margin is going to be wider and this is it's difficult to actually confidently state if there's a meaningful difference there but overall it looks like no not really so we're not a hundred percent sure if the driver option is doing anything and if it is it doesn't appear to be working well it doesn't appear to be influencing the game and results even if it's working it might influence minor results but we're not going to run those tests now there's not much interested in anyway and we're not experts in figuring out how to even check for the how to benchmark that performance so what we're left with is the next step if this topic is of interest to you is for us to attempt to manually tune the timings through software or to work with build Zoid on building a bios for a card that will more readily accept it because this one just like I said it froze it crashed it didn't like any of the things that pilchard was trying to do so unfortunately the custom BIOS was a no-go for for this one from the start one final note here is that one thing we we don't really have the sample size to analyze is that changing the timings if you make them looser would theoretically increase your Headroom for frequency overclock and this is something that roman their bauer found with I believe it's the 780ti where when they loosened the timings they were able to get a higher frequency and that makes sense you see the same thing with the in system memory so that's a potential item of of note where you to actually see some realistic outcome or or meaningful outcome where there's a result difference but it's not one that's really worth testing with with the AMD drivers because we don't even know what they're doing to the timings precisely so that's it for this one subscribe for more we'll probably do a follow-up on this if you're interested you go to store documents access net to pick up the mod mats the medium and large are both back in stock especially the medium has been sold out immediately when we do the first run so that's back it should be now you can go to patreon.com/scishow sexes to get access to the behind-the-scenes videos I'll see you all next time