Gadgetory

and he made a lot of claims with its rx 5700 series of GPUs like saying that the new blower cooler is quote quiet where Andy launched with tremendous power in the CPU category especially the r5 3600 that we reviewed earlier today and we recommend the company has had a much more difficult time pulling together a competitive GPU today we're benchmarking the RX 5700 XT for thermals noise normalized thermals acoustics power consumption and gaming versus relevance and video competition like the freshly released Super Series before that this video is brought to you by Thermaltake c 360 DD c hard tubing water cooling kit if you're ready to dip your toes into the water and build your first open-loop cooling system the Thermaltake c 360 DD c hard tubing kit comes with all the components you need the kit includes a three sixty millimeter radiator 3 128 a RGB fans a copper w-4 a RGB water block for the CPU a pump and res DDC combo and all the fittings needed to build a full CPU open-loop learn more at the link in the description below we have a lot of thoughts on na'vi that we need to get through most of them will be in the conclusion we're gonna get into the benchmarks pretty quickly today but the primer here if you missed the architecture discussion we have a separate video with david cantor talking about the architectural changes in na'vi our DNA is actually a genuinely new backbone versus GCN so this isn't some reason it has meaningful updates to it so Navi is a new GPU to look at and separately we do need to address some decisions that Andy made with Navi that will probably confuse some people the most important of them is boost clock so what everybody on earth understands us boost clock especially people a dandy because they've used it this way before is that boost is basically it's it's your expected clock in a reasonable workload like gaming and then you have base clock which is the expected floor and that's sort of it that's what you get for numbers there's a higher clock that would be the peak clock but what AMD has done is rebranded their clocks because if you don't like the numbers you invent new ones and the boost clock is now a basically a peak opportunistic clock is I believe the phrasing that AMD is so this is something that could be hit for milliseconds at a time for example so when you see the calculations on the spec sheets that have for example the amount of teraflops one can expect be careful to distinguish between the one at the peak o'clock which is not realistic and the one at a reasonable clock like the gaming clock so anyway that's all that means game clock in Andy's phrasing means basically the same thing as Nvidia's boost clock peak clock you should ignore and base clock you should treat the same way you treat Nvidia's base clock those two are mostly identical so that's the marketing side explained there are some other changes as well like to overclocking the 5709 XC can't overclock past a certain frequency it is artificially locked and that is done for product segmentation purposes so that AMD can sell more of the high-end cards and you don't just go out and buy the low-end card and then overclock it which is odd because the CPU division works completely the inverse to that you can buy whatever CPU you want and overclock it and it can be pretty damn good at the end of the day so those are the primary differences there are others too there were a lot of driver issues AMD push the second driver version to press prior to launch we did end up using that and the other one and this is now becoming a trend with AMD launches so we'll talk about that more than conclusion let's get into the testing we're going to start with thermals and frequency plots will go through games get through some power and noise figures and then talk about the card on the whole just to establish what sort of frequency to expect with na'vi we'll start with a frequency over time plot in a fixed 3dmark workload and then move on to game testing the RX 5700 xt doesn't have the same frequency fall-off curve as nvidia we're touring boost behavior is extremely thermal dependent but it does still have some thermal dependencies the RX 5700 xt averages in the range of 17 77 megahertz to 1890 megahertz with a lot of small fluctuations along the way you'll notice a few outward spikes and negative spikes on this plot we considered deleting these data points as they're inaccurate so it's not an accurate representation of reality so they shouldn't really be shown but we decided to leave them to illustrate an important point which is that the drivers will still occasionally misreport the frequency so the alleged spike to twenty to fifty megahertz and the dip to 700 megahertz both of these are fake performance during this benchmark this fixed workload did not reflect those changes AMD spent a lot of time talking out how its new cooler is quiet this time and so we'll look at that before we get into the gaming results here's an overtime thermal chart in our torture workload like with Radeon seven have a has to GPU temperatures to know about there's the GPU temperature as normally presented it's an edge temperature from the edge of the die that offers a lower number and then there's the GPU junction temperature which is the single hottest temperature facing clock boosts off of Junction temperature allows for higher maximum clocks as the card doesn't have to run conservatively against an edge temperature t.j.maxx is 110 degrees before it starts throttling with the non overclocked at lower power policy with the cooler left fully stock and with no overclock supplied we're climbing up to 95 degrees Celsius toward the end of the test and still slowly climbing and this is with the drivers freshly installed so no weird driver bugs with thermal policy issues here at all actually this was with the newer drivers to the memory temperature measured at about 84 degrees toward the end of the test which is approaching the 95 degree temperature range of GTR six where we stopped being comfortable and video to be fair also has issues with memory thermals shown in our super reviews either card in a standard case with a case ambience of about 30 degrees which is easily doable on a 20 degree room will be running hotter than we're comfortable with for memory temperature on the other hand be REM thermals look pretty good the hotspot V RMS plot in the range of 65 to 74 degrees Celsius which is completely acceptable and well under the recommended range for MOSFETs for the record Auto fan speeds sat at around 2100 rpm during this test which would put you in the range of about 52 DBA at a 20 inch measurement that's very loud and doesn't at all match a Andy's marketing of it being quiet it's not it's the opposite of that so we'll have noise levels after the gaming later the next test will fix the fan speed to 40 DBA in our measurements normalizing the fan rpm to look at how the thermals respond this is a cooler efficiency test so we're looking at the efficiency between a 2070 super cool in a moment and the 5700 xt cooler both at 40 BA the goal is to see if a user could lock this cooler down to a reasonable noise level like 40 DBA this is actually one of the worst stock coolers we've ever worked with following our standardized testing to normalize for for DBA the GPU climbs to 110 degrees Junction temperature or about a hundred degrees for edge temperature which is still unacceptable and begins to hit a thermal runaway scenario only prevented because of frequency throttling on the core we're dropping clocks here this isn't an open-air bench with an ambient temperature of just 20 to 22 degrees Celsius so it's only going to get worse in basically any case that's out you can see our case reviews for more of that memory temperatures at 40 DBA are hitting about 97 degrees Celsius which is uncomfortably warm for longevity especially once you do put it in a case and the MOSFETs end up in the 90s well add the 2070 super GP thermals now a competing card with a higher power consumption of the 2060 super and so it should be hotter if anything this would put things in a Handy's favor because the 2060 super in the 2070 super use the same fans but the 2070 super is a higher power consumption part even still it doesn't matter the 2070 super founders edition card plots at 67 degrees Celsius under full load for its GP temperature and that's at 40 DBA despite how much we complain about the screws and the super glue of the FE cards will give Nvidia full credit for making something that doesn't blow quick note on overclocking before we get into the charts you'll see a plus 50% power target offset and some of the benchmarks for games especially and note that overclocking is presently mostly broken it was very broken on the first press driver and that's because the fan speed was not set incorrectly so we manually overrode that and set the a better fan speed like 75% and then with the updated press driver we still had issues with overclocking so we ended up it was basically impossible to get it stable anywhere and we ended up just setting it to plus 50% power target we got 25 to 50 megahertz out of the memory and we couldn't get anything manually out of the core because the software doesn't it's just something's not working so anyway we set the power target and then we let the card boost and it did gain about 100 megahertz by reporting if we can trust that so it's still kind of overclocked itself but we need to revisit overclocking separately once the drivers are fixed so just wanted to note that so that when you see the OC results in the chart hopefully those will improve once it's functional so I freely it for is our first test as always we like to use this one for reliability consistency and well optimized DirectX 12 implementation slightly it was actually originally one of AMD's best case scenarios in our testing equina end video was still on Pascal and just beginning to play around with asynchronous compute implementations in retail products unfortunately for the 5700 XT today's landscape is different the 5700 XT manages 71 FPS average in our Sniper Elite benchmark which places it as 4.2 percent ahead of the RT X 2060 super before an overclocked is applied with a trivial overclock on the 2060 super and ours was one of the worst overclockers you could ask for it manages to tie the 5700 XT low end frame times are the same here now it's becoming clear why Andy dropped the price this wasn't some 5 d chess mental gymnastics this was a requirement and something that we confirmed speaking with people inside of the companies that work with Andy and also Andy the RT X 2070 that AMD originally marketed the 5700 XT against is also tied with the XT ranking again at 71 FPS average on some retailers like Amazon with an MSI RT x 2070 the non-super 27 TS are now available for about 450 to 460 dollars as manufacturers try to clear out the stock the RTR 27 is super now $100 over the 5,700 XD after the price reduction manages a lead of eleven point seven percent over the 5700 XT one both are stock overclocking the 5700 XT resulted in an increase of about 6% over the 5700 XT stock the clock averaged in the range of 1970 to 2017 megahertz as reported by the Psalter with this plus 50% power target but it's hard to tell since the frequency is misreporting through software anyway it's sometimes show for instance 65,000 megahertz which would be and probably the best overclock ever in history if it were accurate Radeon 7 actually before reasonably in this game which isn't true in every scenario the extra see you count seems to help in a significant way for this title strange Brigade is up next this one is built by the same company as Sniper Elite 4 and it's built equally well but instead gives us coverage of the Vulcan API additionally we can use strange Brigade to demonstrate the scaling Delta from Vulcan to DirectX 12 with each GPU well get some 1080p tests shortly but starting with 1440p and the Vulkan API and these are our X 60 700 XT runs at 122 FPS average with lows of 105 fps or 1% 104 for 0.1% the 5700 XT ends up about 1.3 percent ahead of the non-ace qrt x 2070 and roughly within our error margins this marks the two as functionally identical and a skew 2070 would plays higher here approaching the 2070 overclock that we have on this chart but it would still be between the 20s OD black and 2070 black OC which our performs the 5700 XT the RT X 2060 supercard is led by the 5700 XT by about 4% overclocking suber puts it beyond the 5700 xt stock and overclocked results the latter of which does bugged as a result of poor launch drivers something for which aim the Radeon has become a repeat offender in terms of price the 2060 super compares most directly to the 5700 XT and that's before getting into things like noise thermals and power something we'll do later for gaming there within a few percentage points of each other in this benchmark the RT X 2070 super is on another level here placing 12% ahead of in performance when stock but after those price changes it is also now another price category as well sometimes we see the performance stack change as resolution increases potentially illustrating Rob's or memory limitations things like that bandwidth limitations at 4k the RX 5700 XT sits at Sony one FPS average ranking as equal to NRT X 2060 supercard when it's overclocked 1% lows are within error margins here performance over a non a skew 2070 the lower bin one the one that launched at MSRP in October that's why it's on the chart because it was actually MSRP it shows a performance lead about 5.6 percent for the xt rel it's to 1440p this is better performance when scaled against the old 20 70 the new r-tx 20 70 now $100 more expensive again so 250 ends up at 77 FPS average and leads the 5700 XT by 8% one stock overclocking increases that lead to 18% overall at 4k the 5700 XT recovers much of its losses from the 1440p results this next chart looks at performance scaling between the two API is available in strange brigade the chart is ordered by best average FPS in Vulcan and 100% number here bar would mean equivalence with Vulcan whereas numbers below 100% indicates that performance is worse on DX 12 and numbers over 100% indicate better performance on DX 12 and this is all DX 12 relative performance scaling versus Vulcan baseline the RX 5700 xt performs about 1.5 percent better with DirectX 12 nearing error margins and the overclocked puts a 3% over Vulcan performance and thee is doing better in DX 12 in Vulcan for this title although its marginal in this instance f1 2018 is next and give us a look at an older more standard DirectX 11 implementation for this one the rx 5700 XT runs at 150 FPS average with lows at 102 and 73 this positions the stock 5700 XT is about nine point nine percent ahead of the RT X 2017 on a GPU for everyone who felt buyer's remorse over there RT x 20 series purchases when super came out this week you're not alone now the AMD Radeon 7 buyers can feel the same remorse because an AMD card that's nearly three hundred dollars cheaper achieves the same performance in this game as the Radeon seven did when it launched five months ago if you're using Radeon seven and professional applications and you can get use of that memory you dodged some of this bullet by way of the extra VM having an actual use case for you but it just doesn't really matter in these gaming workloads more notably as has always been the case with this game AMD maintains significantly improved training times over Nvidia cards this is something we've observed with every other reason to Andy card in this specific title although that flips a bit as you look at other games this frame time plot will illustrate the peculiar behavior of the cards in this benchmark 5700 XT holds steady at about eight to nine millisecond frame to frame intervals with almost no excursions from that average we typically start seeing the excursions once they exceed eight to twelve millisecond spikes the 2070 bounces between five milliseconds at the low end and 16 milliseconds at the high end marking the 5,700 XC has a better experience in this title 1440p 4f 1 2018 positions that are x2 the 700 xt at 115 FPS average which is roughly tied with an overclocked 2070 black and superior to the stock RT x 2070 black the RT x 2060 super with an overclock reaches 110 FPS average placing the 5700 xt stock GPU 4% ahead of the 2016 Subaru that overclocked the rx 5700 XT with 50% more power roughly matches the stock Radeon 7 with the 2070 super leading the stock rx 57 or XT by 7% at 4k the 700 xt muscles ahead of the overclocked 2070 black in f1 2018 and lands between the 2016 Subaru C and 2070 super stock performance overclocking doesn't do anything meaningful for us here but we think that's primarily because the drivers are broken although we don't expect much Headroom here anyway because Andy has severely locked down overclocking to the point of it being a tremendous disappointment hip an - is up next first with DirectX 12 at 1080p hitman 2 positions the RX 5700 xt about equal with an overclocked RDX 2060 super or about 3.8 percent ahead of the RT x 2060 super 1 stock the 2070 super leads the 5700 XT by 9% in this title versus the RT X 2070 the 5700 XT leads by about 4 to 5 percent and average FPS and that's 2070 black edition to be clear with DirectX 12 and hitman all devices have a hard time with 0.1% low values but we'll note fairly that Nvidia is RT X 26 super is substantially improved over the 57 XD this is the inverse scenario as with f1 2018 so Nvidia is pulling ahead in this one specifically for player experience at 1440p that rx 5700 X he ends up with in error margins of the AMD Radeon 7gp who and tied an average FPS with the overclocked r-tx 2060 super the 2060 Super Stock GPU ends up at 76 FPS average to the 5700 XT is 81 FPS average but the 2060 super maintains better 0.1% lowest mark and that's a better experience again the inverse of F 1 2018 results DirectX 11 gives everyone in improvements in 0.1% low performance seen here at 1440p the RX with you 700 XT is now functionally tied with the RT X 2020 black and RT X 2060 founders Edition 1 overclocked that's not super mind you that's the 2060 the RTR 2060 super it leads marginally roughly with an error and a 2060 super with an overclocked leads the pack of $400 cards at 4k there's these 700 XT runs a 43 FPS average about tied of the 2070 black although the 2070 has significantly better lows the RT or 2060 super is also function tied with both of these cards and this is back on the x12 mind you shadow of the Tomb Raider at 1080p is up next for this one the 5700 XT place is just below the r-tx 20 somebody super so it does well in this benchmark performance has it at 124 FPS average compared to the 126 FPS average of the 2070 super this also positions the rx100 xt ahead of the Radeon 7 card one stock and about 15% ahead of the RT x 27 t Black Card 1440p places the 5700 XT marginally ahead of the overclocked to 27 t black card and about 15% the head of the RT X 2060 super the RT x 27 t super maintains a lead of 5.3% Radeon 7 can unfortunately once again join the RT X 2060 and 2070 owners in the morning in the corner at 4k and these are rx 5700 XT and 2070 super keep the same distance as before with the 2070 super about 6% ahead technically the super is gaining distance as resolution increases but it's not too much each time it is it is notable though the 5700 XT outperforms the stock RT X 20 so d black and the 2060 super by about 16% this is the last game before we get into power thermals and noise which are very important for this blower card far cry 5 at 1080 P positions the RT X 2060 super 2070 black and if it's 200 XT all in good company Susy's have 100 XT it leads the 2060 super by about 4.7 percent although the 2060 super maintains better frame pacing than the 5700 XT this is an instance where it's measurable but probably not perceptible overclocking the 2060 super ties it with the presently - rx 57 100 XT overclocked both of which reach upwards toward a stock RT acts 20:7 a super 3dmark time spy is last before we get into those power numbers this is useful for looking at performance more individually of the GPU and memory GT Wan is a GPU intensive workload whereas that's core clock whereas gt2 is more memory intensive this chart is sorted by total combined score not shown but measured in the thousands as gt1 and gt2 each have a weight contributing to that 4 times pi we measured the 5700 xt at 60 FPS average GT one across multiple test passes which positions Navi ahead of the 2062 / by 7% in GT 1 but allow is that the supercar deleted in GT 2 by 2.2 percent super has more efficient memory in GT 2 which could come down to things like Delta color compression but we need to dig into it more after release in a separate content piece potentially 50-100 XT ends up about where Radeon 7 is in this test but the 2060 super with a light overclock blasts it ahead in total score GT 1 ends up at 60 point 8 f PS not far from the 5700 XT but GT 2 polls far ahead note that every single tenth of a point in FPS is important in 3dmark scoring 2070 super leads in total score at 10000 181 points whereas the 5700 XT is eight thousand eight hundred ten points netting a 16% lead although we didn't show those numbers because they're too big and they blow out the scale but that's all you need to know for those power consumption is next our new power consumption testing for GPUs measures at the 12 volt rails although he took some 3.3 volt rail measurements for the new super and Navi cart this testing is done with a custom-built interposer to measure PCIe slot draw as well as the 12 volt PCIe cable power consumption down the power supply cables these charts will list the TDP for AMD and NVIDIA parts but note that Nvidia defines its power consumption based upon the GPU and AMD defines its power consumption based upon the total board power which is everything on the board we're looking at synthetics for the first test the power virus per mark puts AMD's RX 5700 xt at 217 watts with just the 12 volt cables and PCIe slot measured we separately measured the 3.3 volt slot power consumption plotted here as polling 0.9 amps or about 3 watts extra we're right at about the TDP figure plus or minus some measurement error the RT X 2060 super polls 164 watts or about 175 max if you include measurement error and maybe a couple of watts down 3.3 volt and overclocked in the 2060 super puts it at 209 watts power consumption is finally getting closer it took AMD an entire node shrink to almost match and videos power consumption at the same performance level but they got there eventually the our TX 2070 pulls at 214 watts for reference and amusingly applying a 50% power offset to the RX 5700 XT actually does what it suggests and increases power consumption by about 54% from the 217 watt figure previously with the over power setting applied we measure about 335 watts to average in the range of 1990 to 2017 megahertz assuming watt man and GPZ aren't totally inaccurate it's not worth the three to four FPS that we gained on average to generate that much more heat since the power into the PC does effectively just make it a space heater also considering what a twenty ATT I can get done for the same amount of power this highlights that Navi has difficulty scaling performance with higher power consumption than the spec it's just it unfortunately it's not linear it doesn't work that way f1 2018 at 1440p places the 5700 XT at 206 Watts from 12 volt power which is about 30 watts more than the r-tx 2070 supercard as a reminder in this game the 2070 super outperformed the rx with you 700 XT by 7.2 percent meaning that it operated at 15% less power while doing so the rx 2060 super runs at 164 point 4 watts in this workload for reference with an overclock pushing it to 200 watts for 4k resolution the 5700 XT runs at 214 watts for 4k with the overpowered 5700 XT at 281 watts for 4k back to noise levels despite and these marketing the rx 5700 XT runs loud and relatively hot 40 DBA seems to happen at about 1650 rpm or so about 20 inches away and noise levels are made worse by some sort of resonant frequency issue within the cooler maybe Turbie err making noise inside of the shroud or maybe a bearing for the fat the noise levels for the card end up looking like this the curve plots above everything else here for the 5700 XT with all the others being dual fan coolers are better like the Radeon seven triple fan axial cooler the 5700 XT is a hard regression in design MD needs to concede that it can't make what any reasonable person considers to be a quote actually pretty good and quiet blower cooler it's not happening here and this is certainly not is the same conclusion as we come to with all the other blower reference coolers which is don't buy it just buy a partner model if you're going to buy the card the partner models will be far far better for thermal and acoustic performance finally reaching the end of all of this data there are a couple of things first of all gaming performance is its up and down it's like a tennis match and some of the games we've tested like some of the older dx11 titles for example surprisingly Navi has done pretty well so it's it's actually genuinely fairly competitive in some of those titles after the price reduction is taken into account so this is not 5d chess this is they had to drop the price and that's fine it doesn't matter why they drop the price so all that matters is that it's lowered how and and it was before anyone ordered which oh that's a dummy sample it's it's actually not functional so don't worry about it that'll be in our teardown later it should've mentioned that earlier where this card doesn't do well is in thermals and we've given this part of the review to Nvidia in the past as well when they used to do blower cards and we still kind of ribbed them over there a lot for their since their new founders edition designs but the review we always gave to blow our cards was that they're not worth it and there are instances where you could justify it for example if you wanted to buy the reference board and stick a water cooler on it sure the V rhyme is good enough you should skip the blower design it's just not good so the card itself if you want to buy it that's fine it's okay it's competitive int to good and some of the scenarios like GTA 5 for example I don't know if we put that one in here but GTA 5 it did pretty well f1 2018 the frame times are fantastic and f1 2018 for any time to take a hit in the x12 with hitman 2 everything kind of does but Andy a bit more than others so it's a bit give-and-take but the card itself without the cooler is is a reasonable purchase get it with a different cooler though that's the recommendation we make so where do we go from here well drivers those are an issue the we ran into a lot of the driver bugs we've seen in the past and we'll put some on the screen here where it does the multicolored thing when you first install this isn't even just press drivers a lot of these photos are from the public drivers the new public drivers on RX 590 in the Radeon seven so it's not even unreleased stuff that this is happening on it's been happening for years this is on multiple different systems we've had it happen over multiple different years now and just in the last week three different systems three different monitors two different motherboards same thing happens we different operating systems and/or installs anyway 19:03 problem so drivers are a problem it's not a great user experience you can work around that one by rebooting but it'll happen every now and then if you leave the system item I don't let the screen go to sleep other issues with the drivers include overclocking being broken so this is the biggest point to bring it back to is that currently with the drivers we have overclocking does not function in a way that it seems like it makes sense even with the updated version with the updated thermal policy which really just means the fan is faster it's still not functioning well you should basically set the power target to plus 50% and then stop touching the dials because the frequency at that point will auto jump to something that's probably kind of stable and any tuning beyond that it was just we had all kinds of issues of the drivers so we'll revisit that once they fix them up and this is something any of these aware of it is something that AMD is working on very brief recap then the Navi card in comparison to the rise and launch is a letdown the r5 3600 was much more compelling and for example as is the 3900 X the Navi card has significant issues with thermal with noise it is it required a node process shrink to become competitive with NVIDIA and power consumption is still a higher on average but it's not significant anymore so that is will give credit there it's not as bad it's actually powers high earth and NVIDIA but it's not high so it's not to a point where it's actually going to matter for basically anyone so performance wise the GPU is defensible as a purchase it's fine don't buy this one buy one buy a partner with a good cooler on it and you'll be much happier I think that pretty much sums up our review it's fine with a different cooler and that'll cap it for us here so thank you for watching subscribe for more as always it's been a busy day one of them and you can go to store that Cameron's Nexus dotnet to support a strikethrough by picking up a tool kit or a mod mad and patreon.com slash gamers Nexus and definitely check back on the channel for the teardown of this card because we do some cool stuff with it so I'll see you all next time