Gadgetory


All Cool Mind-blowing Gadgets You Love in One Place

Radeon VII Powerplay Overclocking Results & Liquid Mod, Pt 2/2

2019-03-02
we're finally back with a second and probably final part of our Radeon 7 the quad cooling mod and we did a bit more than just the liquid cooling so first of all you may have noticed a few things changed we get to that and secondly we install the power plate tables mod so we're able to overdo the power by a couple hundred watts beyond what it was doing stock so today we're going through the results of that thermal power gaming and anything else related to those mostly overclocking in general before that this video is brought to you by the power color red dragon rx 588 gigabyte card the RX 580 red dragon is a good fit for 1080p and 1440p gaming priced very competitively at $170 and including two free games from AMD like the division 2 or Resident Evil 2 the red dragon series got a big name on our channel when we pushed the Vega 56 red dragon to the limits but it's now available at lower prices with the RX 588 gigabyte and rx 574 gigabyte the latter of which is presently $130 while the promo lasts learn more at the links below before getting to the first round of charts then the the main change was all of it so the liquid cooler has been replaced originally it was an alpha cool GPX kit of some kind and unfortunately that kit improved the the GPU edge temperatures by a lot but the GPU Junction temperatures were worse they were throttling pretty much constantly and so the reason for that is because the edge temperature is literally the edge of the die and the contact there was good so no issues there the junction temperatures it's the hottest of the 64 sensors and somewhere in the middle there wasn't good contact and it was bad enough that even paste wasn't filling the gaps and so we took the cooler off looked at the cold plate a bit closer and if you run your hand across it you see that there are actually some very noticeable bumps in the finish of the cold plate on that particular cooler so unfortunately because of the imperfections on the surface on that cold plate it just was not usable maybe a like a Hitachi HMO 3 thermal pad might have worked but then you're kind of defeating half of the purpose of the mod so replaced it with just an ace attack see I'll see you might remember a clip from our Vega frontier edition mod where we had my neighbor drill some holes through the AC tech mounting bracket and we dug that back out stuck it on the thermaltake flow 360 closed-loop liquid cooler to Gen 4.5 pumps so it has a faster rpm which is useful here through three random fans on it that were lying around and we ended up with this so it works and the temperatures are actually quite good so there's a lot to go over here we're gonna go through thermals frequency and the difference in frequency stock for stock with the original cooler be talking about power consumption overclocking at the end I'm going to go through some of the very important peculiar behaviors with wot man still so the absolute biggest thing before we get into the charts that you need to know is that the number you type in for frequency with Radeon settings is not the number you get out at least not with this card and so if you type in 2030 megahertz and it says it's running you need to check the actual frequency because it's potty running more like 1952 maybe 2000 is what 2030 means and then separately everything must be validated with performance because as an example we were able to get the card up to let's call it 20 to 50 megahertz and big air quotes there because you can type in 20 to 50 in watt man and provided you have the other settings presumably in a buggy enough state where it'll accept the number the the tool will tell you that it's accepted 20 to 50 or whatever but in reality once you run the performance numbers you'll see that performance I actually decays this is a bug that was in Radian settings when Vega first came out so you line up with the score of 1,800 instead of 5,000 or whatever and so the most important thing is score must be used to validate the overclocking success you can't just go based on input in a number then does it crash or not there needs to be a score there which makes it take a lot longer but let's get through the numbers and then I'll talk you through some of the rest at the end of the video at quick sidenote here we might do some extra testing on this we really want to but we're going on a trip to Taiwan and then China and we'll be there for a while so if we do a part three no guarantees it won't be till we get back so anyway let's get into it we'll get to the gaming benchmarks posthaste but power consumption is too fun to put off any longer let's start with plotting total system power consumption of the Radeon seven stock card unmodified in any way for our ashes of the singularity 4k benchmark total system power consumption of the stock card is at about 420 to 430 watts peak with the average closer to 390 to 400 remember that's total system power consumption but we do control the system carefully to ensure only the GPU causes power fluctuations next is to plot the water cooling mod but still with stock settings this line would potentially reveal any power drop from reducing power leakage something we've seen in previous liquid cooling mods where every 10 degree Celsius drop will occasionally get you about a 4% reduction power consumption in this instance unfortunately there's no meaningful change in power consumption but there's a reason for that and we'll talk about that after the game benchmarks we don't get the drop we sometimes see from power leakage reduction our next line plot is the overclocks water cooled card which doesn't use any power play table mods yet and only overclocks using the normal watt Mann procedure our settings were technically set to 2030 megahertz core and twelve hundred megahertz memory but the actual operating frequency is much lower than this due to miss reporting or inaccuracies in watt man realistically we're more in the range of highly variable 19 50 megahertz the 2009 Hertz depending the result of this configuration is a total system power consumption though peaking at 520 watts and averaging about 455 to 460 watts the last line is the most impressive for this one we're running a 100 percent power target offset and pushing the card draw towards the 500 watt marker total system power consumption maxes out at about 620 watts an increase of around 200 watts over the stock Radeon 70s that we first plotted performance doesn't scale linearly with this naturally but that doesn't matter for what we're doing today we're just trying to figure out how far we can reasonably get Radeon 7 and it starts with this 620 watt peak total system power consumption for performance benchmarks we'll start with x by extreme just because it's a synthetic workload that heavily loads the GPU and memory independently so we get a fuller understanding of the maximum theoretical performance differences these differences don't necessarily scale the actual gameplay but they're typically good indicators just of whether the overclock is even working remember with Vega the biggest challenge is that wot man might look like it's accepting frequency over clocks but the actual stability is worse and performance will be worse as a result so you can't trust the number that you set the frequency to you have to validate with performance otherwise it doesn't count with time spy extreme we placed a baseline score of 40 to 78 points under complete auto with a full stock cooler for the coolant solution our overclocking test with the first driver revision also failed often causing performance regressions even with small overclock so you can see in some of the results that we have here we placed at about 1,800 points to 1900 which is a massive regression in performance this issue has been pretty much resolved with the newest driver which is why revisiting today there's still performance regressions with unstable overclocked settings and there's no way to validate that other than performance testing for the first overclocked attempt with the new drivers we scored 45 62 points by operating with a 200 megahertz offset on our liquid cooled mod again 200 Hertz offset doesn't mean it's just straight 200 or it's higher because it's variable using that a 120 percent power target 1.125 GPU voltage we ended up at that 45 62 score with the clock set to 20 megahertz offset and a one-point 162 volt GPU core voltage still using a 1,000 megahertz stock memory frequency and 120 power target 120 percent we scored 46 17 points this is an increase over stock of about 8% which gives you an idea for the upper limits of our liquid cooled Radeon 7 before using any other mods like Power Plate able registry mods to increase the power target beyond stock the final result was 48 97 points sparing everyone the slow increases in between where we set the frequency to a 250 mega Hertz offset voltage to 1.2 37 volts which is over the stock spec and uses power plate table mods and the power target was set to 100 percent or technically 99 percent offset for a total system power of well two hundred percent if you added up that way the memory was twelve hundred megahertz this result is at fifteen percent over the stock performance number moving to another times by extreme chart next we can look at the individualized GT wanted you to do scores presented as FPS which helps us better visualize these specific areas of performance uplift gt2 traditionally gets the just gains from memory overclocking whereas GT Wan gains the most from core over clocks as it better supports the specific workloads the final overclock allowed for GT scores of 25.8 5 for gt2 and 35 point three eight for GT one better than FPS and showing individual gains of 13.5 percent over the stock 31 point one eight FPS for GT one and fifteen point two percent improvement over the stock GT to score from the original Radeon seven tests although the Delta isn't massive between GT one and two it is common that we see and these gt2 performance drag more of the weight upward as memory frequencies increase and he needs that memory bandwidth on a tire and GPUs and it does benefit from memory overclocking sometimes it more than from core overclocking depending on the application on the workload before plotting thermals and talking about our overclock stepping and challenges in overclocking it'd be good to get some gaming results presented for those most curious about performance gains and linearity of the overclock first up is apex Legends which is already demonstrated in an uncounted performance advantage to the Radeon seven rich the RDX 20/80 competitor we're starting here because it's sort of a best-case scenario for a DM 7 so keep in mind that these results won't extrapolate to all their games equally at 4k with all settings configured a high using our river village benchmark the Radeon seven card placed at 56 FPS average with lows at 43 and 44 fps performance was overall good functionally tie in with the gtx 980ti there was no meaningful difference between the 1080i and Radeon 7 cards so our TX 2080 tree i/o stretched its compute targeted legs with a 65 FPS average leading the stock Radeon 7 by about 16% with the base overclocked and a water cooling mod noting again that the 2030 megahertz in quotes there is just the setting not the actual output frequency we see a frame rate of 62 FPS average that's not polluted 10.5% over the stock Radeon 7 performance and encroaches on stock RDS 2080 territory granted you could overclock the 2080 as well and power would be lower but this is closing in and an impressive way at least on the stock card 3d on 7 power play tables mod puts it at 65 FPS average giving us a disappointing improvement over the overclocked average performance of 4.2% in plain terms and more straightforward and non stat mathy terms it's 2 to 3 FPS which is an invisible improvement in other words so the percentage sounds a bit better than the reality it's an improvement on the last the end result is that we tie with the RT x 2080 trio which isn't a bad result just not as big of an improvement as you'd expect given a power increase let's show a frame time performance plot to better illustrate the frames of frame interval differences in this test the lower frame times are better but consistence is better than lower testing is repeated in the same area and test variance is under one FPS average per run so this is very consistent and accurate as a test pattern 3d im7 stock card ends up averaging closer to 19 to 20 milliseconds per frame with frame to frame an interval deviation never greater than two milliseconds on average plus or minus this is excellent consistency despite slower than 60 FPS average frame rates and the power mod in liquid cooling get our frame times down to 14 to 17 milliseconds on average depending frames of frame interval variance does not meaningfully widen so our overclock is considered stable for this testing at 1440p for Apex legends the gradient 7 stock card ran a baseline frame rate of 106 FPS average which was significantly outdone by the overclocked variants 120 FPS average just like last time we see about a 13% increase in performance over baseline with the water-cooled overclocked tests and also like last time we see very little difference with a power-play mod actually it's worse this time unfortunately with this test the difference was within test variants it's really no different at all there was zero benefit from increased power consumption to the core we believe this to be a limitation of the core frequency as opposed to the 4k results where more of the memory bandwidth is utilized so the differences don't come out as strongly with a more core limited scenario at least that's our hypothesis for this one time spy extremes earlier GT 2 results reinforced this belief the end result is that the Radeon 7 / clock ends up leading the r-tx 2080 by about 14% and is nearly tie in the RT x 28 ETI again you cannot extrapolate this across all games but it's still good information for this game unfortunately some people will see this and tell everyone that Radeon 7 is almost as good as a stock 28 ETI but that's not always true that said let's try another best-case scenario let's look at Sniper Elite 4 and see how close the Radeon 7 can get to the 2080 Ti and that one although we do need to keep everyone's expectations in check by looking at some DX 11 games after that so I'm currently at 4 with high settings and DX 12 gave us stock Radeon 7 performance of about 85 FPS average with low as well spaced behind this positions Radeon seven is about tied with the RDX 20 ATF II and just behind the gtx 980ti overclocking and water cooling gave a significant uplift to 97 FPS average or 14 4.4 percent increase in performance allowing the radium 7 to outpace the stock 1080i and 20 ATF II though overclocked results obviously reshuffle things a bit again the power mod only gives us another couple percent increase in performance disappointingly so the real story is in the uplift from cooling and a more basic overclock either way the power mod does start to approach 20 atti stock performance but doesn't quite make it there and is still led by the stock 28 ET i by 9 percent we've looked at more compute intensive games the apex although using dx11 and sniper is in the x12 so now it's time to balance with a more traditionally devolved game GTA 5 at 4k and very high ultra settings produces the Radeon sevens FPS at 51 FPS average with lows at 41 and 39 FPS this ranked it as just ahead of the RT X 2070 only about 7% while still being $200 more expensive than our tested 2070 the water-cooling mod and overclocked improves the Radeon 7 performance 256 FPS average an increase of about 9% and allowing it to get closer to the 61 FPS average of these stock RT X 2080 on our charts overclocked in the 2080 didn't get it much in this game and it only created a few percent gap versus the stock 2080 the powerplay tables mod allowed the Radeon seven card to get 257 FPS average showing an improvement consistent with the previous results of about two to three percent uplift then that change is 11% versus the stock Radeon 7 with performance just under the stock RT X 28 EF II when the 7 is pushed to our power flameout limits it's a very far distance from the 28 ET I stock card which leads the Radeon 7 power play mod by 55% that's a pretty hard counter to our previous two charts and gives us another perspective of how a game might perform at 1440p GTA 5 shows the stock Radeon 7 at 99 FPS average leading the 1080 FTW by about 6.6 percent the overclocked and water-cooled Radeon 7 places at 107 point 6 FPS average improving over stock performance by 9 percent again with a power-play mod then showing some similarly poor gains as seen in the apex legend 1440p benchmark we think that most of the performance gains may be once again more resultant of the memory increase than the flimsy and unpredictable core increases for f1 2018 at 4k the Radeon seven card performs at 73 FPS average when stocked which is the 20 eighties 81 FPS average although we previously Illustrated that the Radeon 7 does post stronger frames I'm performance in this particular title overclocking and water cooling the Radeon 7 gets it up to 70 9.6 FPS average posting an increase of 9% once again with a very plain mod putting it 284 FPS average and improving at an additional 5 percent over the previous overclock this is one of the bigger jumps we've seen from just the power play tables the net gain versus stock is 14.5% Landing it close to an overclocked 1080i but not quite passing at an average frame rate although the low frame time performance is superior as we discussed in our previous review far cry 5 is the last one at 4k the Radeon 7 stock card ran at 60fps average with the water-cooled overclocked at 67 point 5 FPS average which is functionally tied with a 20 ATT I tested back on for 17.3 5 we've not looked at far cry again on the 28 TI since 417 YT 5 but so far the two are about equal with the overclock and water cooling on the Radeon 7 card adding the power mod only gives us a couple of percentage points of improvement hitting 69 FPS average and improving 15% over baseline stock and adding water and an overclock to the 28 ET I gives it 282 FPS average leading the more power-hungry Radeon 7 by 18.6% and this is without any BIOS flashes so it can't go to higher power targets like the Radeon 7 is doing in our original baseline testing ignoring vrm thermals for a moment we placed the Radeon 7 and its Itachi graphite thermal pad at a junction temperature of 108 degrees Celsius under load as a reminder Junction temperature is the hottest of the 64 thermal sensors across the package whereas GPU temperature is the edge temperature literally the temperature at a cooler edge of the GPU die it is therefore less useful but it's the more traditional measurement that you're used to seeing the junction temperature has a t.j.maxx of 110 degrees when stock so 108 is technically under t.j.maxx GPU edge temperature was about 80 degrees when tested stock and the clocks will boost based on the junction temperature so that's the really the more relevant one here our water-cooled mod under the same stock clock and voltage conditions ran a junction temperature approximately 39 degrees cooler than the stock temperatures at about 68 to 70 degrees Celsius Junction the GPU edge temperature also ended up about 40 degrees below a stock as you would expect and that ended up at a firm 40 degrees Celsius note that the ambient temperature was controlled at 22 degrees Celsius for both tests and was logged every second of the test of the thermocouple reader we are not doing a delta T over ambient here so this is just the straight temperature readout and our Delta would be well subtract 22 or so and that's your Delta moving on to the frequency overtime test for the same test case we see that critically performance is actually up versus the stock air cooled card and this is without any overclock settings applied that means that boosting is utilizing the extra thermal Headroom which may also explain why we didn't see reduce power consumption from leakage earlier if it's boosting clocks and voltages meet the thermal headroom until a hitting a power target power consumption would remain the same our average frequency for the water-cooled mod is about 1720 to 1750 megahertz with some spikes although rare up to 1800 megahertz the stock card averages closer to 16 50 megahertz allowing us a 70 megahertz to 100 megahertz average increase just from liquid cooling the card there are no changes in watt man or any other overclocking software and there are no mods beyond the water cool that's a very good result and is essentially a pre overclock achieved just by using a better cooling solution which illustrates that the boost parameters on this card are functioning as they should it is sort of comparable to the NVIDIA GPU boost routine where it'll boost based on thermals voltage things like that still the water mod is more variable and frequency likely due to the power limits but it a verge is a higher result and is overall better finally the newest gpu-z rendition has given us softer monitoring of the sensors inside of the power components Rhian Radeon 7 bill joins video on our channel talks about these parts in more depth but we can show the thermals of our mod here in a more practical fashion just to prove that you don't need a base plate on the vrm and be validated this with hardware k-type thermocouple as well the hottest component was a GPU vrm MOSFET which plotted at about 50 degrees Celsius that's not over ambient or anything that's just straight 50 C in a 22 °c ambient environment so our Delta is about 30 degrees that's damn good particularly considering we have no base plate no direct contact cooling and we only positioned a knock to a fan to blow over the VR at air is enough here we can even run this passively without the knock to a fan although that wouldn't be advisable once you actually start overclocking with power mods you probably want the cooling that either way Radeon 7vr M is efficient enough that it doesn't run too hot the GPU memory temperature plotted at about 47 to 48 degrees Celsius will just complicate the chart now and throw the rest of the measurements up there all at once the memory vrm and SRC v RM temperatures ended up in the range of 35 to 44 degrees Celsius ensuring that this card ran well within spec and better than stock even without the direct contact heatsink on the VXR so closing out then as we prepare to get on the plane to go to taiwan i would i keep this a cut a few charts out that i did want to include but i'll just present the numbers verbally instead so the junction temperature with the card using its power mod and the liquid cooling mods that's the one thermal chart we did not show the junction temperature there is about 99 degrees celsius it goes up quite a bit but it's still not hitting t.j.maxx which is 110 stock or 120 degrees by Andy's own configuration when you start playing out with overclocking setting so we were still 20 degrees off of t.j.maxx which means we have a bit more Headroom but we're starting to hit some limits here where the power offset just seems like it stops working at some point so you keep increasing increasing increasing and eventually you look at the power draw it like a current clamp look at a wall meter look at gpu-z and they're all showing the same sort of maximum power at like 470 to 500 watts I think there's a way to get around this I have seen successful users and forums who are able to get beyond 500 watts of power pushed into the GPU which is what we'd like to do as well unfortunately like it said leaving town so didn't dig into it any further than that I will leave that for later if you know of someone who is successfully act fully validated that their card is accepting more than 500 watts please let us know and with a link to maybe their thread form thread below and I'll look into it as soon as we're back and see if we can get a little further because I feel like there's still more in this card but the scaling obviously falls off a cliff even if there's more power we can push into it once we got to well once we did the power play mod the gains start really dropping fast and there might be I think there's more we can do in memory too we stopped at 1200 megahertz it kind of started artifacting in some applications and stopping there but I think in some gaming or benchmark applications we might be able to push a bit higher so that's something to revisit now overall the power offsets a bit buggy sometimes you do occasionally have to DD you reinstall remodel Phi the registry depending on we use the power play table so if you have one power play table modification installed and you try to do another one sometimes it doesn't take and you have to wipe everything and do it again we did some basic under voltage I have data on that we'll talk about that later again it's gonna be a while because we're gonna be in Asia but we got it successfully to about like point 9 something 0.94 maybe and then below that it just wasn't really holding and also quick note on under volting doesn't count as under volt in if you under volt it and then you increase the power offset to achieve this is the same power consumption with the same clocks so there are something there's some really important caveats to discuss there we talked about someone with builds right as well but we'll do that later so is it worth it the answer is well if you're buying radeon 7ne way it might be worth trying to liquid cool it whether that's with a proper block and those are probably coming out or something else but it might be worth it because the performance uplift just from liquid cooling it and doing a basic overclock with the I should have mentioned this earlier with AMD driver version nineteen point two point three it's it's the newest absolute newest version I think it's called an optional version as of today which is March 1st so a 90 point two point three I think is the newest one eyes of today so with that driver version overclocking does sort of work now which is good actually it does work I mean if you're if you're not playing around with power play tables it pretty much just works properly out of the box took Jensen's quote there by accident but it is it is more or less functional at this point so you can overclock now good news there and the kind of told us that you it wasn't a software issue earlier and they thought it was just the limit of the card but the card actually can do more so that's good news and can overclock you do a water cooling mod you get a pretty far up the charts first as the stock configuration it's outpacing our RT X xx 80s in some cases where it wasn't before it's approaching RT X xx ATT is in some cases that are really compute heavy so specific application workloads like compute intensive stuff very bandwidth intensive stuff you can see some pretty noteworthy gains with the basic overclock by which I mean no power mods even registry mods and a liquid cooler and so we think that's worth doing if you get one of these of course you know you start exiting spec so if that's not your thing then don't do it I guess but for anyone who's into modding we're just getting more out of a kind of fun card to hack around with and it's worth doing the power ploy tables kind of work just be careful once you start doing that stuff you can push more power into it then as it's meant for and you know if we ran it with our power play mod for a year I have no idea if the card would even still work it's just that that's probably not something we would recommend for daily use so keep that in mind but anyway nineteen point two point three is the important bit and then power play mods be careful with them is important as well don't use them for 24/7 if you're pushing not 500 watts into the GPU I know it might be ok not an expert there but it's just going past spec significantly enough that we're uncomfortable recommending it to anyone who spends this much money on a card and then intends to use it 24/7 as opposed to just for fun for overclocking you can of course step it down do like a 50% off set you're probably fine but no guarantees so that's it for this one thank you for watching please leave any any questions if you think we sort of skipped over something because we're in a rush to get the door please leave your questions in the comment section below and what we'll do is revisit the topic as soon as we are back in the studio and can do some additional testing it a really fun project we liked working on it a lot and I look forward to working on it some more so big is always pretty fun to overclock and play around with and and there's more room yet for us to do that so give us some time we'll do more hope you like the content though it's still got plenty of depth so subscribe for more go to store documents nexus net to become a mod mat like this medium one on the table with GPU themes to help us out with this type of content production it significantly helps when you pick up shirts anything mugs from our store because that supports artwork directly and you go to patreon.com/scishow sexist otherwise i'll see you all next time
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.