in today's edition of unicorn dust we're going to review we're talking about the 5070 Fe still confused as to why they didn't make a 570 TI Fe but since this one is an actual Fe card maybe it'll help with MSRP pricing a little bit probably not for those looking for a high-end custom gaming experience look no further than Falcon Northwest Falcon Northwest has been building PCS made for gamers for over 30 years with a focus on a true high-end gaming experience custom cases available only through Falcon Northwest feature state-of-the-art testing and design to sure that every component is performing at their best through thermal imaging and rigorous lab testing designed and overseen by the Falcon Northwest founder himself with a complete lineup of systems ranging from small to large every Falcon Northwest system includes a three-year warranty policy and a year of two-way overnight shipping coverage providing the ultimate pce of mind to see all that Falcon Northwest has to offer follow the sponsored Link in the description below all right so this is the Nvidia Founders edition of the uh RTX 5070 uh embargo lift this today they go on shelves tomorrow the whatever the date is the 5th of March is when they're actually going to be on shelves theoretically uh we're only going to be reviewing the Fe we're not doing any partner cards this time around so Usual Suspects uh when it comes to the type of review we're going to do here we got average FPS 95th percentile 99th percentile in terms of FPS uh we did some power testing and a little bit of Investigation as to why we were actually getting some pretty poor 1% lows with this card so I got some interesting visualized data for you guys to see um and uh yeah so we tested a lot of cards for this and the included some AMD stuff and this may not be as exciting to talk about today because the Embargo lift is tomorrow for the 970 and 970 XT so I have a feeling a lot of people are going to probably be sitting on pins and needles really seeing how those cards compare considering the pricing has finally been announced uh speaking of pricing this is a $549 MSRP RP card which is exactly the MSRP of the9 70 non XT and the 970 XT comes in at $5. 99 so it is getting very crowded in that $550 to $600 price point um unfortunately since there is no Founders Frontier SL reference card for the 97970 XT that might be a free-for-all when it comes to pricing so anyway moving forward uh 549 um MSRP if you look at the styling on it it definitely looks like I'm holding a 90 like it's I'm a big dude like people see me in real life they're like oh my God you are way taller and way bigger than I expected uh so just I feel I feel like a giant holding the 590 this is a 90 I'm just that big no this is a 90 right here so it's just like little brother like literally uh notably it is not a pass through cooler um if you can imagine how small the piece B would have to be although I bet you anything they could have done it with this card if they wanted to um so it is not a true flowthrough design this side is fully open though the PCB does have a curve made uh built into it so that the air can pass through and then the air going in this fan does exhaust out the back of the case as you can see with the vent right there in a very traditional layout um which is pretty much exactly how the 4070 uh design is and then the 37 or the 370 is just a dual axial fan that exhausts out the rear of the card um it's right here actually no it is a blowr I forgot that is one of the first blowr they did whatever um that's old though no one wants to play with that anymore okay I kid so talking about key specifications though there's 6,140 Cuda cor on the DB 205 core found in the 570 both the 470 and the 3070 have 5,888 Cuda cores but they're on two different architectures RT core count though is like really hardly moved at all it's just obviously running the fourth gen RT cores in the 5070 a total 48 RT cores 46 found on both 470 and 370 the difference is we have second gen RT cores found on the 370 and third gen RT cores found on the 4070 so the performance uplift you're going to see when it comes to RT performance is a generational uplift rather than it being more of a Brute Force approach of adding more RT cores um when it comes to tensor cores we have 192 of the fifth gen tensor cores found in the 57 or 570 184 of them found in the 470 those are fourth gen cores and then 184 also in the 3070 but again third gen cores robs that's always been an interesting discussion well actually no it's been a quite boring discussion that only people started talking about or even thinking about because of the fact that they've been missing on the 50 90 and the 80 and the 5070 ti so I have zero reason to assume there won't be 507s that pop up missing robs as well ours is not though we are having all 80 uh accounted for so there's 80 on the 570 64 on the 470 and interestingly enough 96 of them on the 3070 So based on the date on the box of when this card was manufactured and when news started popping up of certain cards missing uh what appears to be an entire cluster of RS and when Nvidia says hey we figured it out I I don't it's weird that they had it already figured out and apparently already fixed prior to anyone's videos going live tells us they obviously already knew about it and they hoped nobody would talk about it or notice but unfortunately we did but again based on the whole timeline I have no reason to believe 507s won't pop up with some missing clusters either um it's just all to bunched together and they're all coming from the same mass manufactured wafer and whatever is causing uh some of those RS to get fused off has not been totally discovered yet so just know if you buy a 5070 you might end up missing some RS just the timing is all too close together so the 5070 has 12 GB of gddr 7 an 192bit bus it's a PCI and pcie Gen 5 device which as we know in the 5090 barely means anything so obviously means nothing on the 55070 for a uh total uh memory throughput of 672 GB per second bandwidth now it's important to note the 470 also is on 12 gbt of gddr6 but the 370 is actually running 8 GB of gddr6 so you'll notice actually on stalker 2 in our benchmarks when we get there and you might have noticed this also with our 370 TI which is also 8 GB the game wouldn't run it actually would run for a few seconds and then completely the bed uh giving us about 05 FPS once that frame buffer became fully utilized so that's why there are big fat zeros on our chart because it's the first test we've actually got with our setting and the way it's set up on the Epic settings which is we test everything at almost Max settings to kind of see what the worst case scenario would be it's really the first time we have a hardware that is too far behind the the requirements and although they're not 4K cards we continue to show the data because if you ever wanted a 4K panel we think it's important to know what the FPS impact might be going from something like 1440p to 4K it's important note the 5070 though is a very strong Contender at 1440p many games is capable 4K if you lower the settings obviously it's a 1080p card uh no problem whatsoever if you wanted to run like Esports titles or whatever and get high frame rates uh speaking of frame rates and such the base clock is a 2325 with a boost clock of 2512 on the 5070 obviously GPU boost uh logic is going to take over where if you have plenty of power Headroom and temperature Headroom it will push those clocks much higher and our instances it was pushing them over 2,800 MHz which is an additional 300 MHz overclock giving us about 550 MHz of total automatic overclocking that takes place over the base clock uh it is a 250 W thermal design though that's where you start to realize most of the generational gains that we're seeing here although it's got a bump in cacor and such really comes from 50 Series uh mentality of let's just turn the power knob up some more I mean we saw that across the board every card pulls more power than its previous uh or versus its predecessor so this is a 250 watt card the 4070 right here is a 200 watt card versus the 220 watt card found on the 370 so it's funny the 470 was actually a power drop from 220 to 2 and then now it is 25% more power on the 5070 at 250 Watts so let's talk about gaming performance here we go and put the charts up typical fashion uh we're just going to kind of go through them we're not going to talk about them just pause them on the screen you can see where they land versus uh previous gen and obviously the higher end card so if you were considering maybe going with like a 5070 TI it's on there obviously if you want to go like 80 or compared to a 480 or 470 TI or 4070 or 470 super 470 TI super and all the stupid 70s that exist the one thing to consider here is the price and performance Gap I really want you to pay attention to on the gaming charts as you scroll through through them is the gap between the 5070 and the 5070 TI not only is there a huge performance Gap there is a $200 price gap between the two remember the 5070 TI uh MSRP price is $7. 49 we know that is a launch number uh there are a handful of cards out there that exist at that price point but realistically it is closer to $850 to $900 which is absolutely ridiculous and still selling out at that price which is ridiculous with this being a $549 card tells me you can obviously expect a 5070 super at some point and probably a 5070 TI super which I don't know how you could do that theoretically when the 580 is supposed to be $1,000 unless all this pricing was always by Design moving on from pricing it's the depressing subject we all know and expect um performance uplift generationally this is the first time I've actually done a performance generation uplift uh graph which is kind of interesting so if we compare the 570 uplift versus the 470 across the our gaming Suite this time including our synthetics I want to include the synthetics to just kind of show you guys how those impact the overall numbers we have a 22.
1% uh uplift across the board all of them from 4070 to 40 to 5070 uh 4070 compared to 3070 was actually 32. 3 so you can see we're about 10% lower when compared to the generational uplift versus 470 now remember it is just a small tweak on a couple more things a few more Cuda Cor uh but just pushing the power limit and the clock as far as you possibly can now the the thing that's really pulling that number up though is cyberpunk coming in at 40. 6% uplift from 570 versus 470 but it's a cyberpunk version 2.
12 which actually really benefits from RT improvements because we run RT on in our testing so that one test is actually pulling it up quite a bit if we just omit the synthetics and get rid of cyberpunk entirely we actually get 19% uplift so you can see that one title pulled it up 3. 1% just that one title um if we talk about per title uplift stalker 2 was 18. 4 Avatar was 17 Horizon forbidden West is 14.
6 Borderlands was 22. 1 black myth 20. 4 and f124 21.
6 so it's really landing around the generational uplift we've been seeing sort of across the board with the exception of 590 versus 490 um nothing really to write home about if you could get this card at 549 it actually would be a pretty solid Contender for fewer future facing um gaming when it comes to AAA titles I didn't even test multi gen uh or frame gen at all in this particular generation I'm kind of over it to be honest or this particular card let us not forget this is the card that Jens Sun probably got up on stage and said RTX 570 4090 performance at $549 now get the fine print that you've got to use the magnifying glass from the Santa Claus movie to be able to read says when multiframe frame gen uh IND D lss4 is enabled obviously we didn't do any of that and if you look at our gaming charts as you can see it's not exactly the way it really truly pans out this in my opinion borderline Border Lines uh or probably just straight up crosses the line on false advertising let's talk about some of the uh more specific charts here when it comes to power draw and such so when it comes to power draw I do two different tests here I do furmark which is just a straight up heavy load it's got the fuzzy donut just rotating around um we use that to test both temperatures and to test power uh so if we go and take a look at the uh furmark temperature test it's kind of interesting the 5070 it ramps up pretty quickly actually ramps up faster than the other two cards uh but then it levels off and actually starts to dip a little bit uh so it did surpass about 76. 5 C and then level out right at 75c as the fan curve ramped up it's almost like the temp overshot the fan curve and then the fan curve caught up and brought the temp back down the other two cards behaved uh exactly as they the the behavior normally is the fan speed adjusts more quickly to the load which means it can sort of keep it from climbing quite as high as the uh the 5070 did Under full load though even obviously you would expect more temperature out of a more power hungry card so at 250 Watts obviously it leveling out at 75 degrees the uh 370 which is a 220 watt card leveling out at approximately 72° and then obviously the 20000 watt uh RTX 470 leveling out about 66° so the temperature lands exactly where their power spacing lands but real world scenario here we can take a look at uh cyberpunk 2077 uh when it comes to actual temperatures the 3070 is actually the highest or hottest card at just over 70c the 470 uh capping out at just about 66 and a half degrees and ironically the 5070 coming in as the coolest card Landing at about 7 or 63. 5 cels and the reason for that is the fact that the card doesn't actually pull 250 Watts when you're doing actual gaming it actually runs out of voltage and clock speed before it runs out of power in a real world scenario if you take a look at the actual power draw in furmark you can see it's a fairly solid yellow line for the 5 70 at 200 it says 250 Watts but remember I measure with the PMD what the physical power draw is plus the PCI Express um so it's actually pulling over 260 Watts out of uh total power so it does overshoot that all the cards do there's a difference between what software reports and what actual physical limitations or physical measurements are after heat losses and and other power losses are taken into account as you can see the 370 Fe kind of hovering right around 225 to 227 and then the 4070 has some weird Behavior I think because it's such a low wat TDP card and how difficult the fuzzy donut is to run you see this weird kind of a wave happening there and as the doughnut turns it does different things in the test so you can see which parts of the test are harder to run based on the power going up and which parts of the test are easier to run as the power starts to go down so it's not a solid line it's a looks like an ocean wave which is kind of interesting on that one but again a 200 watt card uh you can run out of power before you'll run out of other limitations uh on that card but again going to real world power draw testing and games all of them pull under their uh well except for the 4070 they all pretty much pull under their rating the most efficient card coming in again is the 5070 Fe in this test at just over 177 watts in real world gaming um one thing I want to note both the 470 and the 3070 pull additional power from the PCI Express 3070 was actually pulling 60 wats from the PCI Express slot with the rest of the power coming from the uh 8 Pin PCI Express to the uh pre1 12volt high power adapter it's not quite 12volt high power but it looks a lot like it uh it is pulling all of that power UH 60 watts of it from the socket so that or the slot that's the highest of the three the uh 4070 is pulling approximately 11 to 12 Watts from the PCI Express slot and then the 570 is pulling nothing from the PCI Express slot it pulls one watt consistently through the PCI Express slot it is pulling every bit of it through the 12vt 2x6 or 12vt high power plug uh but as you can see in this efficiency chart here it is the lowest power draw of our 1440p cyberpunk with rate tracing going uh but 3070 pulling the most power at about 2117 watts and then kind of straddling the 200 watt line is the 4070 cuz again it's super low watt TDP car therefore it hits the power limit early now when it comes to cooling though all three of these cards do take a slightly different approach to how the card is cooled all three of them have a half flow through half uh 90° turn design but all three of them have different fans and overall different shapes so what we're going to do right now is we'll listen to some sound profiles of how all three cards sound at both 50% fan speed which is about the average we saw the auto fan curve go-to sometimes 60 sometimes 44 45 so 50% and 100% pay close attention to the the the sound of the air and its its turbulent properties because of the fact that all three of these cards have to make a 90° turn in Air Flow at least for part of the cooler it's never going to sound as smooth as the 590 uh FL double flow through design does It's actually an incredible design when it comes to the Acoustics um but all three of these have a very different sound profile so take a listen [Music] [Applause] [Applause] now let's talk about frame times frame time all FPS is is a calculation of the frame times how much in milliseconds it takes to draw the frame and how many of those do you get on average per second this is why we do 95th percentile basically on the advice of Gamers Nexus when he came here and gave us some of the uh ideologies he believed fifth percentile was actually more important than 0.
1 percentile so we do average we do 99th percentile and we do 95th percentile so as you can see the light blue line is the 370 uh which has the highest frame times on average if you take a look at the 4070 obviously it comes comes down and then the 5070 is the lowest coming in at about 11. 5 milliseconds on average but I want you to look through the lines here you might have noticed on the FPS charts that the 5070 had the worst 1% lows for us it was like in the 3070 1% low range and I went and just started double-checking all of our runs the amazing thing about the fact that every single run is recorded into a CSV now we can go back and verify the data make sure we didn't trans or transcribe something wrong and the data is what it is and I noticed something here if you look at the yellow lines so if we remove the other uh frame times here and look at just the 5070 you can see very consistent frame times uh that have us this High Spike and they're very consistent on uh very close in range on how high they're spiking so I thought maybe something was potentially off with the our capture or something on with the card but started looking at all of our data for that card on all the runs and every single game has this ex exact same weird interval Spike on the frame times that is why we are getting the really low 1% lows I can't tell you why it's spiking like that I can tell you that this is a pre-release driver which we have to use cuz it's the only driver that will recognize the 5070 and work um I can tell you that behavior did not happen on any of our other 50 Series cards 5070 TI 58 80 or 5090 nor any of the 40 or 307 30 series cards that we retested to prepare for this launch so I can't tell you why it's doing that but I can at least point to this and say this is why the 1% lows look like complete trash on the chart so this is something I'm going to have to go back once we get like full me uh non-media drivers or public facing drivers we're going to have to go through and retest and see if this continues to happen if it doesn't then we'll have to rerun our test and entirely for the 5070 which should give us some uh more accurate 1% lows but I don't feel like the overall the data the data affects the total like opinion on the card because the average FPS and the fifth percentile FPS are landing where we expect them to land versus their neighboring cards their price point uh peers as well as their generational peers that we compare them to it's just the 1% lows now none of this really was noticeable when gaming it's not like it's not like that was a noticeable stutter in the game it's not like it was a noticeable hiccup because it's not like a crazy spike it had been like a 100 milliseconds 200 milliseconds or something like that that would be very noticeable but we're talking going from 11 milliseconds up to about 22 milliseconds so it's double the time which does sound like it would be a hiccup or a stutter but it actually was not noticeable so I don't know if it's a potential reporting thing when it comes to present Monon we know present Monon is uh does need some tweaking when it comes to 50 Series cards especially when it comes to uh m multi frame gen and really being able to recognize what those numbers look like outside of average FPS so we've got some more digging to do here but this is the only card that does that so I don't think it's necessarily a present Monon thing I think it's more than likely a driver thing which shouldn't be all that surprising giving everything that has uh been happening with 50 Series launches at the end of the day I feel that just about the only card that will probably be$ 5.
