Gadgetory

after the post-apocalyptic hellscape that was the r-tx 20/80 launch and videos following it up with lessons learned for the r-tx 2070 launched by and large technical media took issue with the 2080s price hike without proper introduction to its namesake feature that be r-tx which is still unused on the 2070 this time however the RT x 2070 launches at a much more tenable price of 500 to $600 putting it at rough price parity with the GT X 1080 Hana on stock before that this video is brought to you by Thermaltake and the view 71 enclosure the view 71 is a full tower case that's capable of fitting three video cards and most configurations it's also one of the better cooling cases in our recent case testing bench lineup the view 71 has hinged a tempered glass doors on either side that make it easy to open and show off and it comes with at least one rain fan though you can get the RGB version if you prefer learn more at the link in the description below the RT x 2070 still has the same problem that's it's it's basically a dead horse on the ground that's been beaten for about a month now and that is the fact that our TX isn't really a thing you can use yet so we can't really test that feature there are things that use it but they're still all tech demos so waiting on shadow of the Tomb Raider to kind of get moving in that direction a set of course of metro Exodus things like that so today we're still focusing on what makes sense rasterization performance focusing heavily on thermals focusing on noise things to that extent and there are a couple things here that are also worth going over which is this is a good study of how resolution impacts performance so in this specific use case the 2070 we're looking at a scenario where memory bandwidth becomes a potential limitation as you approach those higher resolutions so the 2070 you'll see in the charts will deviate from the 1080 as the resolution decreases as you get towards 1080 in some games not on all of them you'll see that too but in some games which become memory bandwidth bound the point of deviation is about 1080p where the 27 he starts really pulling ahead of the 1080 otherwise that rough parity and performance as you'll see through the charts and there also a rough price parity so this card is the EVGA 2070 that's it's just EVGA 2070 there's there's actually not a name on it anywhere but it's the EVGA 2070 technically black edition I think is what they're calling it and it's a $500 model which is why they couldn't afford the extra ink to put black on it somewhere it's just EVGA r-tx 2070 so it's a very cheap model it's about the cheapest they can make and this this then puts it pretty close to the GTX 1080s that are left over not too far from the 1070 TI so we're gonna be focusing on those today it also means that AMD is now back on the table for consideration something that hasn't been true for quite some time because NVIDIA has been dominating that ultra-high underwear and he just has nothing and so it's interesting because because of Nvidia's dominance of that high-end it's all anyone's been talking about cuz there's no reason to compare it to stuff that is four hundred dollars cheaper typically and it almost takes retraining yourself to remember that you have to look at AMD and their Vegas 64 and 56 cards because they're actually priced pretty close to this these days so that is another thing we want you to pay attention to in the charts today quick notes here on architecture it's all the same as the other touring stuff but the block diagram for the t 106 400 GPU is on the screen now that's the 27 t it's the same structure as the T 102 that we discussed except it's hacked in half for a total of three G pcs the rest of the containerization remains the same and we'd recommend watching our touring the architecture discussion for more of that in short there's still one overarching command processor at top that deals with shoving things into the GPU memory from the PCIe bus and tasking the workload there are then three gbc's under it each of which contains 60 pcs of which contains two SMS and then there are 64 FP use within those streaming multi processors the end total is 2300 for FP 32 FPU is CUDA cores if you prefer to use on videos language here but they're floating-point units they were being honest so it's it's an interesting comparison to the once high-end Pascal cards so let's get into it we're gonna go through the review we've got gaming notes thermals noise and overall value F 1 2018 is an ego engine game by Codemasters offering a stand-in benchmark for comparison against most other Codemasters titles or racing games this game is highly GPU intensive even at 1080p work towards that resolution as we go at 4k we observed f1 2018 in place in the EVGA r-tx 2070 at approximately the same level of performance as a board partner GT X 1080 FTW from last generation also from EVGA we're at around 60 FPS on each with the difference within margin of error we managed to overclock our our TX 2070 fairly high something we'll discuss momentarily to outperform the overclocked GTX 1080 by about 5.6 percent this is pretty close to Silicon Lottery territory granted and we previously found that the overclocked 1070 is mostly equate overclocked n 80s if you're curious about where those would land in comparison generationally the our TX 2070 operates approximately 28% faster than the GTX 970 SC from the last generation it's about 94% faster than one of the world's best-selling video cards the GTX 970 from a few years ago and note that the scaling here should diminish with the resolution and these vegam 56 operates at 49 FPS average permitting the 2070 about a 20% lead and 64 is a bit more comparable as we'll see throughout the rest of these results at 1440p the our TX 2070 scale is approximately 4% over the GTX 1080 that we tested operating at an average frame rate of about 102 FPS the GTX 780 Ti maintains a 91 FPS average and the GTX 970 SC maintains a 78 FPS average the our TX 2070 holds about a 30% lead over the gtx 1070 SC at this resolution Vega 56 maintains an 81 FPS average placing it between the 1070 TI on the 1070 for anyone to on a GTX 970 you're looking at an approximate 2x increase in performance in this test the RT X 2080 offers a gain of about 30% over the 2070 price hike of $300 so significantly more percentage-wise than what you're gaining for performance finally for 1080p the r-tx 2070 manage is a 136 FPS average which still isn't CPU bound at least on these cards overclocking pushes us another 12 FPS higher for a market gain of 9% here's where it actually gets interesting the RTS 2070 now posts a gain of about 8% over the GTX 1080 whereas 4k had the met near equivalence we believe this is because 4k pushes the memory bandwidth requirement harder where the 27t had relatively marginal gains over the GTX 1080 just to visualize it here's a chart showing performance deltas between the 2070 and 1080 as resolution scales notice that the performance Delta starts to grow up 1440p then widens at 1080p at 4k the 2070 becomes choked by its memory bandwidth and meets the gtx 1088 for is another good test scenario this game runs on the asura engine and is among the best built games from a performance standpoint we use this game sometimes partly synthetically as it's DirectX 12 and asynchronous compute integrations are the primary points of interest we restrict testing to 4k on this one at least until the lower end the cards come out this is because it already post such strong performance and all we really care about is the scaling differences at 4k the r-tx 2070 manages a 64 FPS average and Sniper Elite for marking parity with a GTX 980 66 FPS average this is technically outside of our margin of error but is overall insignificant to the player and is insignificant when considering silicon quality differences anyway for scaling we see the gtx 1070 at 48 FPS average and that allows the RT x 2070 a gain of about 32% the GTX 970 operates 834 FPS average again nearing a 2x gain in the 2070 this chart shows the frame times between taurine Pascal and Maxwell generations this is the best way to look at performance although frame rate is flawed in significant ways it remains standardized and easiest to read but we've been pushing for more frame time plots as performance measures instead just because they are ultimately the only real metric that matters with regard to frame x or frame anything lower is better this chart and even more important than being low is being consistent the left axis depicts the on present frames of frame interval in milliseconds the spike here the present interval the worse the experience and fluidity of frame throughput for reference 16 milliseconds would be about 60fps the GTX 970 is significantly more prone to choppy frame throughput and its frame times are inconsistent overall this inconsistency is more problematic than anything else the gtx 1070 and our TX 2070 both maintain consistent frame to frame intervals and that is where you get a more fluid experience regardless of the frame rate because you can have a high frame rate relatively when looking at an average that smooths over some of those spikes that we see although the 970 is quite bad far cry 5 is next this one uses the dunya engine and is a Ubisoft title the game is another GPU intensive option for testing thriving particularly on things like particle effects and geometric complexity and character models at 4k the r-tx 2070 operates a 46 FPS average with a low is relatively tightly timed and in the 30s the GM coined 1% in 0.1 percent low values are representative of similarly smooth frame time performance to what you'd see in the previous charge the GTX 1080 is again at rough parity with the gtx 1070 at 36 FPS average in this title and at 4k the 2070 performs about 29% faster than the gtx 1070 the GTX 970 is about 1/2 the 20 70s performance and the 20 80s 59 FPS average leads the 2070 by about 26% at 1440p we see the 2070 again at rough parity with the GTX 1080 including a rough price parity Vega 64 is also nearby operating an 81 FPS average and again rough parity with the 2070 and the 1080 1080 P is less valuable at the high end as we begin hitting CPU bottlenecks with the ti class cards the 2070 manages a 118 FPS average about equal to the 117 FPS average of the 1080 we are within silicon quality variance between these two chips where within test variance between the jibs so given how an videos boost algorithm behaves given how a silicon quality is given how game tests behave it's fair to say these are functionally equal here let's move on to another chart next up shadow of the Tomb Raider the eventual r-tx title that is still awaiting proper r-tx support and tuning this is an AI dose and Square Enix game that runs on a modified crystal engine at 4k the r-tx 2070 manages a frame rate of about 41 fps not bad really but not good enough to hold these settings dropping to lower settings would permit playability at 4k and compared to the GTX 10 70s 30fps though it's still a significant performance increase and one which allows parody with the GTX 1080 14:14 heat positions the 2017 more playable spot when running these settings holding an 80 FPS average when overclocked but we did have to drop our OC and it's about 73 fps when stock the 2070 is again at parity with the 1080 again posts rough 33% gains over the 1070 and again what's the difference with the 1070 T I will leave 1080p for the article link to the description below if you'd like to see those results Grand Theft Auto 5 remains one of the most played games today and runs on the rage engine by Rockstar Games the benefit of this game is that it is a fairly mature game at this point and so future updates that impact performance are extremely unlikely this is valuable at 4k the 2070 operates at functional equivalents to the 1080 and is within error margins overclocking provides the 2070 a slight lead but not one which exits silicon quality variants the RT X 2080 manages a 27% boost over the 2070 and the 2070 manages a 25% boost over the 1070 the GTX 970 is again about half the speed of the brand-new 2070 here's another interesting spin in this game going to 1440p allows the GTX 1080 to take the lead contrary to what we saw in f1 this is the opposite problem memory bandwidth isn't the issue in this one as the performance of scaling worse than the 1080 as resolution scales down it's more likely that this behavior is resultant of better driver tuning and Pascal for GTA or game tuning or something else 1080p also shows scaling opposite of expectations set by f1 as did 1440p and we're presently uncertain as to the exact reason why this game performs differently from f1 and resolution scaling but we're still looking into it finally ashes of the benchmark runs on the nitrous engine and as more parts benchmarks than game at this point to start ox credit it is very well optimized as a DirectX 12 benchmark this one is almost entirely synthetic if we're honest because no one actually really plays it for the most part it can become CPU bound pretty quickly so we'll stick to 4k and crazy with a GPU focused test at these settings the RT X 2070 is comparable again to the GTX 1080 but as outside of error margins in its differences the 1080 maintains a 2.2 percent lead meanwhile the RT X 2070 is about 6 percent faster than the similarly priced Vega 64 card in this regard Andy is starting to reenter the conversation as an option either we're no longer only talking about the ultra high-end where Nvidia is left alone we never particularly liked the 64 and instead favored 56 as a great deal over 64 but the point remains the same Vega is a reasonable competitor and is a bit cheaper depending on when you check the prices and therefore should be considered in addition to the 1080 alternative to the 2070 those are three cards you should be considering at this price point overclocking follows the same process is what we extensively detailed in our RT X 2080 review but we now enter the fray with more knowledge of memory straps and Delta clocks this is largely thanks to guidance from professional overclocker kingpin who provided oversight on GP memory overclocking during our rip series streams the EVGA r-tx 20 70 base model operates a peak frequency of about 1800 megahertz when full stock averaging roughly 17 30 megahertz in our times by extreme workload note that this will vary based upon workload as always we immediately noticed that power is the only perfect cap here increasing the power target to a maximum of only a hundred and fourteen percent and changing nothing else we were able to push the peak clock up another thirty megahertz with the average frequency increasing by about 50 megahertz from there we manually tuned until finding stability at about 2000 megahertz peak with an offset of 980 megahertz on the memory in our rift J testing we found that 980 megahertz is one of the memory straps or so we think anyway as is 930 and we also think 12 40 megahertz offset is another strap keeping these temperatures for the GPU I below 50 C will allow clocks to average about 19 70 megahertz of these settings and on this card but we found steady state restricting us closer to fifty three to sixty degrees Celsius depending on how we controlled the fans more on thermals next as previously and as indicated on this chart core temperature heavily impacts overclocking capabilities this card is a relatively strong overclocker but it's severely limited by his power consumption it's possible that higher-end cards with custom v bios builds will have more meaningful overclocking gains but with only a 114 percent offset maximum we really can't get that much more out of this card this next chart shows frequency overtime plotted against thermals the value in this chart is illustrating if or when the GPU clock throttles to control for thermal scenarios in this instance with EVGA is cooler we aren't observing temperature really climbed past about 61 to 63 degrees Celsius when left to Auto controls you BGA's v-- BIOS modulates fan speed to regulate for a rough 61 to 63 degree temperature on the GPU typically maintaining an RPM of about 1800 we'll talk about this as it impacts noise later the peak frequency starts at about 1845 megahertz in this fire strike extreme test and then drops down to arrange about 1772 1815 megahertz once for north of about 53 degrees this is a normal boost behavior for NVIDIA cards this next chart is a bit of a unique GN test that we do where we stick thermocouples to the hottest MOSFET and the hottest GDD are six memory module on the board we use a power virus workload for this test we're in the GPU remains close to about 60 degrees Celsius once again when following EVGA stock V BIOS for thermal noise response the hot spot NC P MOSFET measures at about 69 degrees Celsius and the hottest memory module measures 66 degrees Celsius these are actually very good results thermally memory modules typically want to stay under 90 degrees at least for gddr5 and MOSFETs can take up to and over 125 degrees Celsius in this instance EVGA Schouler is more than adequate even for a model which is clearly cheap in every other department killing LEDs and frills and investing all the spend into the cooler was a good move here aside from the obvious power constraints this MSRP unit is actually a decent thermal performer the power constraints are the biggest limitation here but if you're not overclocking it's also not as big of a concern and also to be fair power constraint is why thermals are tripped a complete non-issue for this one it's not really drawing that much power anyway so it's rarity for low-end models like this $500 units to actually be these in thermal performers as for noise results the EVGA r-tx 2070 black tries to stay around 60 degrees Celsius so it's acoustic performance will hinge upon your room and case ambient temperatures for our testing in an ambient environment of about 21 C this had us typically around 1800 to 1900 rpm where noise levels were about 40 DBA the EVGA card is overall quieter than the FE designs at each rpm level maxing out at 50 6.8 DBA when I 100% speed or 3800 rpm we typically found that the card sat around 40 DBA in a noise floor of about 26 DB and as for the FE cards we don't know yet if the 2070 Fe has a different fan speed curve than the 28 es typically they're very similar if not the same but the thermal profile is different enough between these chips that they could be different so we'll have to revisit that if we ever do get an Fe 2070 and that's the RT x 2070 so what we're looking at the trouble here we're looking at roughly a 30 percent performance increase for roughly a 30 percent price increase and while this might sound like a fair trade on paper it's one to one the reality is that we already had this option it's called the gtx 1080 so it's functionally a rerelease of the gtx 1080 in so far as the value proposition today buying this thing it's not any different than buying a 1080 it's the same price increase it's the same performance increase over something similar last generation so that part isn't all that exciting the only way you could really look at it and make it exciting is if you say well but instead of buying a 1080 from last gen you're buying the newer car to 2070 for roughly the same price for roughly the same performance increase over whatever other model you're looking at because they I mean they're about the same performance so it's gonna be the same linear increase for roughly the same price and performance jump you're getting something that will get more driver support going forward you're getting something that has r-tx for free included if it ever becomes a theme that's use that's kind of how you have to look at it to justify mentally the card is significantly better positioned than the r-tx 2080 that is absolutely certain the 2080 TI is the top of the class you can't really beat it period so hard to argue that 128 e is bad value 2070 is significantly more competitive with even NVIDIA zone lineup and at $500 you're looking at primarily comparing versus the 1080 and overclocked 1070 TI and maybe Vegas 64 or if you feel like it heavily modded Vega 56 which becomes a Vega 64 with enough effort and without any power limitations which can be done for historical perspective the GTX 1070 offered upwards of 50 to 60 percent performance improvement over the GTX 970 generationally for a price jump that was about 25% Japan when you're looking at it so that it moved the bar from $330 on the bottom line 970 s to 380 dollars on the bottom 10 70s upwards of 450 most cards fell around 410 something like that so it was a much less significant price increase generationally than this one is and the performance improvement was greater leaving us feeling like this isn't quite as impressive yeah and video set a high bar for themselves and it seems like they're having trouble meeting that bar from last generation now this one we think is much more tenable than the previous the 2080 launch the 2070 is easier to argue it is not an exciting improvement but it is an improvement and if you want it comes with r-tx but that's probably something more suited for developers today because no games have it so don't buy based on a promise do not buy based on promises you don't know if they'll come to fruition and we'll stay on top of it as soon as this stuff gets implemented in games we'll be talking about it but overall the is one of those things where it's not a wishy-washy review it's just either you want the 2070 at 500 bucks or you get a 1080 I guess and then if you like the idea of AMD thank you 64 is kind of up and down in price so when it's down it's absolutely worth the consideration it's just it's still somewhat variable but stock for Vega should be a bit easier to get going forward than stock for Pascal so you have that to keep in mind as well demand how long you're waiting to buy your card 2070 though at 500 bucks is reasonable if slightly off-putting if let's let's bump that up reasonable if a decent amount off-putting but over 500 bucks 550 600 is gonna be harder to justify the purchase of this thing because then you might as well buy like 1080i or something like that and one more note here and video launch tactic so this is something that you might hear about today but there aren't too many of these 2070 cards going out for review actually they are pretty limited and there are a few reasons for that so Nvidia got the memo they heard that everyone was upset about the pricing of the 20 80 in the 20 80 TI and to address that and Vidya has to our knowledge not sampled any or if they have it's very few of the founders Edition cars which are a hundred dollars over the cost of this one they're about six hundred bucks so not many of those went out if any at all it's the US media and instead Nvidia asks the board partners to fulfill sampling and NVIDIA required we got confirmation from Nvidia on this that the board partners send a $500 model and if board partners want to send for example a $550 or a $600 model they also must send a $500 model and it's there's not necessarily anything too wrong with that depending on how you look at it but from from the view Nvidia presented it's basically they know that they have a price a perception problem with price and board partners typically do send only their best cards out for review which then really makes that look a lot worse in addition to the founders decision price that was already kind of bad on the 2080 so it's an attempt to make pricing less of a concern I suppose for the 2070 and that's why we end up with a $500 card for the launch day reviews so it's pretty rare that reviewers get the lowest end parts without buying them we have typically bought them ourselves and some of the manufacturers don't even want to send $500 cards because they know they're garbage and a lot of cases so very few and we'll give props to EVGA for being one of them vendors feel confident enough to send out the cheapest model because at $500 you have to realise just how how little room this leaves for EVGA or gigabyte or msi to make a good cooler it's a really slim low margin part here and the GPU cost a lot of money PCB is referenced that cost money and the cooler you're left with like 20 bucks left to spend on the cooler at the end of the day which is not a lot of money for a cooler something good like maybe Strix or the high-end gigabyte or high-end EVGA coolers there was typically cost about $50 plus or minus 10 50 to 60 for the really good ones and there's no room for that here without going to $550 so you're gonna see a lot of $500 cards that have very bad coolers for this generation but this is not one of them as you'll see in our thermal notes today so that's it for this one thanks for watching subscribe for more as always go to store that game is exits net to help us that directly or patreon.com slash gamers Nexus helps out there I'll see you all next time