NVIDIA RTX 3090 Founders Edition Review: How to Nuke Your Launch

when nvidia first announced its rtx 3090 founders edition video card and the subsequent partner models for this time it was pretty clear actually in the marketing slides that this would be a titan replacement the titan video cards have never been really meant for gaming workloads but that hasn't stopped nvidia in the past from marketing them as such the titan rtx for example made claims about being the fastest gaming gpu ever but it was never really emphasized that hard especially not in media typically with these titan class cards really what you're looking for is workstation style performance but this time there's been heavy emphasis on gaming for the 30 90 and specifically 8k gaming for a 1500 video card uh for which the selection of 8k displays is probably lower in overall quantity than the quantity of 3090s that will ever exist so today we're reviewing the rtx 3090 and we are asking the primary question of why namely why nvidia why did you do it before that this video is brought to you by us and the gn store for our 1 million subscribers limited edition foil t-shirt the front has the gn logo and a gold play button like the ones youtube sends out and the back has our history on youtube dating back over a decade the shirt is custom made with a high quality 100 cotton and features a shiny gold foil one million crept up on us way faster than expected so these shirts are still in production but you can guarantee you'll get one by backordering on store.gamersnexus.net our previous limited shirts have sold out in just three to four weeks each time and like those previous shirts we'll be using the funding from our one million subs special shirt to purchase new testing equipment for our reviews back order the limited 1 million subscriber on store.gamersnexus.net so as a workstation card the 3090 does gen this is very heavy does genuinely have a a lot of potential upside and that's something that we have been looking into andrew for example on our team works in blender and game engines regularly and that's a use case that makes a lot of sense for the 3090 and the reasoning is that when you're working with especially early on in the art pipeline with assets that are not yet optimized not yet compressed or finalized you really just want something that can run the viewport relatively smoothly relatively high fps while you're sorting everything out the optimization can be done later so it's often the developer side that would benefit from a card like this there are also things like quadros in existence and at that point you're looking at things like certified drivers as well if you're for example boeing where you need some kind of certification almost like you want a nist traceable certificate for certain testing utilities to back you up legally but this is a different deployment so in a game engine or a game development environment it makes sense to have a better computer than the average player would be on because you do need to just build the thing first and worry about optimizing it later surely this is not news to anybody if you look at the level of optimization in the number of games over the past few years the rtx 3090 might cost 114 more than the rtx 3080 but don't worry this is a workstation card not a gaming not a we're being told it's a gaming card so we're going to be looking at the gaming performance today of the rtx 3090. apparently millions of people have now been told that this is in fact a gaming video card and they've been told that the 8k gaming experience is nothing short of excellent at least for the titles thus far shown and controlled by nvidia so now that we've received the message that this card is actually meant for gaming and it's not just a boring workstation pro viz card it makes sense to test it for 8k gaming without the mantle of nvidia's marketing requirements sitting on top of the chart uh so it's probably time to look at that we'll look through that today as an aside there are a lot of games that don't even run 8k properly some of the ones we tested for example were running at 6651 by 3741 natively separately two we should note that 8k 8k in giant air quotes there is in and of itself a [ __ ] moniker this is because 8k is 7640 by 4320 pixels and that's about 33 million pixels if you multiply the two numbers so if you look at that against so-called 4k which is at this point a universally accepted shorthand for another technically inaccurate uh naming convention that's 8.3 million pixels 8k is not two 4k displays you would need to run quad sli displays at 4k in order to get quote-unquote 8k so that stated that's obviously a hell of a lot more pixels to cram through the gpu pipeline and anyone expecting fully representative depictions out of doom eternal which could run fluidly on a tamagotchi is going to be sorely disappointed and obviously that was hyperbole but it's not far off so nvidia started its marketing with a simple throwaway reference of 8k gaming and it did technically have the caveat on the slide of being rt enabled and dlss enabled which means it's not actually 8k it's just sort of kind of 8k but that ran away from nvidia once the marketing team or whomever was in charge of it started pushing media down the pipes one day before launch of actual third-party reviews with a dictation over the type of discussion that would happen in the content and that is something for which we don't necessarily fault the content creators but rather nvidia's own teams which has decided that now that it has been given a pistol it will repeatedly shoot itself in both feet not just one so following up what could have been an easy slam dunk of a launch with the impressive performance gains of the 3080 a card which we could actually recommend at the end of the review process unlike a lot of the 20 series cards it it ran out of stock and then it followed it up with this continual push on the 8k aspect of a card which has 24 gigabytes of memory which is really meant for the artist someone who needs or maybe a cad user an engineer someone who actually needs the memory to have a high enough speed viewport or a highly responsive work application it's being shown as 8k beneficial so uh here's the chart for 8k gaming it's well it's a console experience that's the best spin that we have for it gta 5 a game from 2015 from 2013 gets 32 fps average the lows are not great and call attention to the need for a frame time plot at least with other games which we'll get to shortly but that's okay we'll look at something else shadow of the tomb raider does a really poor really fluid console like experience at at 30.4 at 30.4 fps average that 0.4 that makes it better than consoles really a lot better than consoles those suck and with shadow of the tomb raider which has that fluid console like experience you might think that 11 fps for the lows is bad but actually moving on strange brigade a game that many people play and some enjoy ran at 50 fps average this was the best score that we plotted in a game that actually rendered that so-called eight so called k because dlss doesn't count as actually 8k and this isn't running it it's also one of the best optimized games out there so this is not common nor is it fair to say that it can be extrapolated market wide to even hit this 50 number horizon zero dawn as an example of this had us at 32 fps average with lows at 24 and 16. red dead 2 on only high settings put us at 34 fps average with lows as far down as 15.7 0.1 percent hitman 2 which is old had us at 37 fps average which is more dumpster tier performance here's a frame time plot from shadow of the tomb raider as a reminder lower is better but more consistent his best it's sometimes difficult to weigh low frame times versus consistent frame time excursions of 4 to 8 milliseconds fortunately this chart makes it easy to call this an objectively bad experience we're running between 4 milliseconds and 90 frame to frame so just about every single frame produced in this game feels like the google stadia launch this is one of the worst frame time plots we've ever set eyes upon so our congratulations to nvidia for making 8k truly possible there are more numbers that we included too and ran but they're technically not valid because the games didn't actually support 8k and when we checked in later by taking simple screenshots we realized that they were running below 8k for example at 66.51 by 37.41 so even if you like 30 fps average you shouldn't expect all games to be 8k gaming ready and that's not to mention the cost of actual displays or availability of them where you have one primary desktop like computer display available and then you've got the marketed ones that were shown in media online previously where you're talking thirty thousand sixty thousand dollars per display and it's all very unfortunate that nvidia decided to nuke its launch with this 8k marketing because otherwise if it had stayed free and clear of it our review basically would have been here's a couple of production benchmarks here's some gaming benchmarks but by the way the gaming stuff doesn't matter that much it's not marketed to those people it's a workstation card and we'll revisit with more workstation stuff later and here's a whole bunch of thermals and acoustics too instead what's happened is here's this [ __ ] marketing claim that nvidia has driven into the ground to the effect that it has now moved away via a game of telephone from all of the original caveats that nvidia had on its slides during jensen juan's presentation and is now just 8k gaming is possible not 8k gaming asterisk is possible but just that 8k is finally here and that's where the message got lost in the game of telephone so ultimately this should be a workstation card but we've got a lot more stuff to do and the benchmarks will still be helpful for those of you who are trying to buy one of these especially for a workstation application even though we're not looking at those numbers today because we were busy validating the 8k claims by the way invalid doom doesn't count and neither does dlss in certain titles it's got to be more than that so we're looking at thermals contacts quality we'll have a separate teardown video check back for the teardown video of course later in the day and then we are also planning to do a whole bunch of overclocking with this in live streams which will hopefully go up on day of launch or maybe the day after there are places for the card uh unfortunately it was mis-marketed but we can still help out with the thermals the acoustics the gaming performance if you actually care about that outside of 8k and then workstation will have to come later with viewport validation and performance power testing is next we're putting this forward in the review because it's becoming interesting with these cards mostly because of how high it is the rtx 3090 is driving towards big impact on power supply selection as we've been doing for over a year now we're testing out the pcie slot and pcie cables not at the wall with a custom interposer that gamersnexus built but we've also added pcat recently for more test accessibility in the process furmark is first here's the chart this is a power virus workload to fully load the cards the rtx 39d fe stock card draws 356 watts alone not total system draw but card drop and that has it 10 percent higher in power consumption than the 3080 for some of the most impressively inefficient performance gains since amd's rx 590. nvidia talks huge game in its press day marketing where it claims that you can overclock the cards and throw increasingly high power at them and they'll continue to scale well this hasn't proven correct in even one remote aspect of performance increases thus far just overclocking the 3080 as we showed in our live streams gets you about three percent in gains without liquid nitrogen and only 15 that but this card shows even worse scaling than that thus far the cards aren't as power hungry as leaks may have suggested but the efficiency for frame rate gains hits diminishing returns and did so long ago this is not the scaling we were led to expect for something that is being marketed as a gaming card despite being probably more adequately marketed as a production or workstation cart overclocking runs 408 watts or 15 increase whereas the 3080 oc ran 373 watts a 50 watt increase against stock for its two and a half to three percent gains if you're wondering about gaming power consumption it's not much different since we're ultimately bound by power targets in the ev bios total war three kingdoms at 4k ultra which is our most intensive gaming workload we test runs the 3090 stock card at 352 watts to 358 watts depending on resolution and even still it's 409 watts with an overclock at 4k the 3080 fe required 314 to 321 watts for the same test when stock if you're a professional and can use the 3090 for those applications the power is potentially worth it you might not care for efficiency and you'll need to budget for additional cooling and if building a smaller render box with multiple of these you'll seriously need to consider room temperature during renders server farms don't need to worry about this but they're also technically not allowed to use geforce cards but for a render farm maybe in your house or small business you will need to think about the accessibility of ac to a room with multiple these in one machine our next test was for pressure of the cold plate as it contacts the die this was done after all other testing as it required disassembly of the cooler the tear down process will be in a video going up later today the pseudo color pressure map shows the points of highest contact and pressure similar to the 3080 fe cooler we see a triangulated pattern of three key holes in the coverage which are filled with thermal paste at full assembly and then we also see some additional holes in coverage that were not present on the 3080 fe overall we'd rate this contact as acceptable but it could still be a lot flatter we'll show the comparison chart in our cold plate flatness test for a better idea of what's going on here's the cold plate flatness chart measured in microns of depth from a known zero point the rtx 3090 cooler is overall more consistent in measurement than the rtx 3080 fe cooler but has a much greater peak and a wider range from the best to the worst measurement by comparison the 30 90 fe cooler is overall flatter than our 3080 fe cooler with a few exceptions in those peaks but still not quite as good as a strix cooler or other heat pipe and cold plate solution here's a look at the thermal frequency response of the rtx 3090 fe for this chart we're running fully auto controlled to determine the v bios temperature target and the frequency that's sustained at said target the frequency ends up peaking at about 2 000 megahertz at the start then falls to around 1890 to 1905 megahertz once the temperature rises to target the target in this v bios appears to be about 68 degrees celsius as a reminder the temperature of the card is controlled almost more by v-bios than even the cooler itself because the cooler simply follows the v-bios target to get true comparisons between coolers you need to noise normalize and ideally power normalize for perspective the 3080 fe was maintained in about 1920 to 1935 megahertz when fully auto controlled so as is usual the lower skew card is running a bit higher frequency due to the cut down die here's the thermal rpm response showing how the automatic fan curve responds to temperature rise in this one the rtx 390 fe temperature quickly settled at around 68 degrees celsius with fans ramping on about a 100 second delay to the load start that's a long soak time which is good and speaks to the size of the heatsink this means that if you're doing bursty workloads that are short-lived the fans might not even spin up the fans ultimately climb to about 300 rpm to maintain 68 degrees celsius over 21c ambient not delta t for this mind you and the rtx 380 fe for comparison maintained a target of 75 to 76 degrees with its own fan ramp to around 1830-1900 rpm the 3080 fe cooler ends up running 2 to 3 dba higher than the 3090 fe cooler in this scenario anyway this next chart is for the automatic fan adjustment and noise levels versus a 100 workload we run this workload until the rpm stabilizes and it's useful for showing us hysteresis and noise levels during use change in noise is more noticeable than anything else with noise so we're looking for as few swings in the noise as possible in this line plot starting at a noise floor of about 26 27 db we can see the cooler ramps until about 38 dba at 20 inches in our testing environment this puts it around where most of the 20 series partner models were targeting when they were actually good ones it's not as loud as past reference coolers from either amd or nvidia so that seems to be the main positive and the longer soak time of just the heatsink when passive is also a major benefit of the cooler the density of the fins and the surface area of the cooler allows it to get away with this heat load at a lower noise level than typically we didn't run the 3080fe as long in our review but we'll plot this line anyway it reaches maximum fan ramp much quicker which is a result of the relative reduction in the ability of the heatsink to soak change the 3080fe ramps much faster and more aggressively and reaches a higher noise level than the 3090. that's because the 3090s fans are spinning slower which is benefited by the larger size of them and the increased heatsink surface area for this one we just manually changed the noise level and normalized for rpm the rtx 3090 fe runs louder than the 3080 fe which makes sense since the outer edge of the fan is spinning faster there's also a little bit more coil line on this card although that'll vary unit to unit the end noise level is just under 60 dba with our measurement approach when at 100 fan speed but in testing the card never reached 100 fan speed that would be if you installed multiple in a case and built a small render farm or similar or if you're trying to play 8k games and you need every megahertz possible to get anywhere close to the playable frame rates that nvidia led you to expect back to games at resolutions suitable for people other than jeff bezos here's red dead 2. at 4k the rtx 3090 fe manages a world-changing uplift of 2fps against the overclocked 3080fe a gain of 2 percent while pulling 8 percent ahead of the stock 3080 fe in exchange for 114 percent more dollars this wouldn't matter if it weren't a gaming card except nvidia has decided that it's a gaming cart so actually it does matter at 1440p the rtx 3090 fe is six percent higher than the rtx 3080 fe stock to stock not particularly exciting for the expense don't come anywhere near this card for 1440p or 1080p gaming and if you do you should be putting frame rate limits in place because the fumbling that will happen on the cpu frame pacing side will make the overall experience worse in some games as for 4k it wasn't really impressive enough there either this should be marketed only or primarily as a workstation card at 1080p the results are identical between the 3080feoc and the 3090 fe and basically the same stock to stock that's because we're cpu bound unsurprisingly here's a look at total war 3 kingdoms at 4k the rtx 3090 fe stock card ran at 54 fps average which is 16 higher than the rtx 3080 fe considering most of the cost really is the memory and the workstation audience the gaming language attached to this again doesn't make any sense at 1440p the difference remains 15 that's better than the rumored 10 gains but against the pricing it's still bad traditionally with a workstation targeted card we'd hand wave this and say that you're buying the card for the memory and nothing else but that's not how the marketing worked out finally gains dropped to 10 percent at 1080p due to cpu bottlenecking moving on rainbow six siege at 4k has the 390 fe stock at 203 fps average or 13 ahead of the rtx 3080 fe that may be useful if you're trying to play rainbow six competitively and also at 4k for some reason and also want to approach 240 hertz but still that's a big price jump for the performance the 3080 fe already gets most of the way there compared to 1440p the 3090 runs at nine percent faster than the 380 fe at 358 fps average rather than 329 fps average lows are about the same in shadow of the tomb raider the 4k testing positions the rtx 3090 fe at 104 fps average which has about 13 ahead of the 3080 fe stock card one percent and 0.1 lows remain roughly proportionate to the average fps and so there is consistent scaling here just not enough to justify the cost again that's expected and that's also what we said with the titan rtx you should be deploying this in workstations don't buy this card for gaming 8k in scare quotes or otherwise because you'd be better off putting that money towards a better cpu cooling system or almost literally anything else in the computer performance at 1440p is 9.8 percent higher than the 3080fe stock with the 3090 fe stock for whatever that's worth anyway it's about seven percent faster than the 3080 oc which is similar to some stock partner models horizon zero dawn is our last rasterized only title because we've already established performance ranges and we don't need to keep proving it with other games the gains here are again amounting to about 13.6 percent versus the 3080 fe stock although overclocking the fe closes the range in that's still far from worth it but since nvidia decided to repeatedly brand this as a gaming card it's appropriate to test it as such quake is a good rtx game to look at developed on the vulcan api and it's entirely path traced and as such will benefit the 3000 series more than the 2000 series at 4k with rtx on the 3090 is running 39 fps average or 13.6 percent ahead once again the rt scaling is actually not any different than the rasterized scaling so far at 1440p the performance jaunt is again 13 with the 3090 at 85 fps average and 3080 at 75 fps average the 2080 ti strix for reference is led by 59 percent against the 30 90 but even that wasn't as expensive as the 30 90. at 1080p the gap shrinks to 12 percent against the 3080 fe from the 30 90 fe which is now running at 141 fps average if you're willing to run dlss control benchmarks at 120 fps average when at 1440p that's just 7.2 percent ahead of the rtx 3080 fe and the scaling has reduced in this title so that's it for the review then of the rtx 3090 video card we will be doing a lot with these in overclocking streams for world records things like that it'll be a lot of fun it is a top performing card in an objective sense but that doesn't mean you should be paying over 100 more for marketing claims which are totally irrelevant to the vast majority of people on planet earth so uh nvidia unfortunately took a very strong launch and it fumbled having a lack of inventory was something that's a bit more excusable because of the current situation because of the insane demand if you look at google search trends just as an example alone the rtx 30 series was far more popular than the 20 series and that was something where especially after talking to partners they had real inventory and there were enough excuses where it was like okay well so you screwed that up a little bit maybe i should have waited a little bit longer but it's not the complete end of the world now though they've compounded that which you can maybe hand wave a little bit because nothing's in stock right now ps5 is gone too phones are regularly gone on launch day all consoles are gone on launch they so that was a little easier to hand wave but now combining with the additional marketing that was put forth by nvidia's teams one day before the reviews properly went up for the 30 90. we'd have to say this is uh a botched launch this is nvidia was passed to the ball and then it looked around for a teammate and it saw amd and it said oh uh here you go so amd ball is firmly in your court don't screw it up this is this is like your chance to get some kind of goodwill but the product still has to be good and obviously we'll have to test it to see if they're there for now though nvidia is still isolated on the charts at the top end amd has nothing anywhere nearby and the rtx 3080 actually in an objective sense is a pretty good performer it left the 1080 ti behind finally if you still have one you should keep using it unless you're truly unhappy with it because it's still a really good car even same with the 980 ti's but if you're looking for 8k gaming you're not going to find it here that was we're gonna we're gonna go with uh we're gonna go with lie for that it was basically a lie because technically you can kind of play 8k games but good luck getting a display and if you can afford one good luck actually enjoying things other than the select titles that nvidia has promoted unless you're looking to revive say your counter-strike source or condition zero career from the early 2000s so then the cooler actually not bad and we've got testing for that in this video if you want to see what acoustics also not bad nvidia's done a much better job at those two things this generation the disassembly of this as you'll see in our teardown video is pretty similar to the 3080 disassembly so overall nvidia has improved massively in the mechanical design and thermal engineering department versus previous launches marketing not so much and performance well it's not worth spending for just gaming you should still be buying this basically only if you're a workstation user don't get tricked into it from the 8kbs so workstation users we're going to try and come back with some additional specialized tests that we're developing nvidia only gave us one week between the launches to work on this cart that's not enough time to actually do new tests it was enough time to do the ones that we had already developed so that's it for this one thank you for watching subscribe for more check back for the live stream with the overclocking of this card because it is still objectively good it's just that the value is objectively bad and then you can of course go to patreon.comgamersnexus or store.gamersnexus.net to support us directly like by grabbing one of these shirts we'll see you all next time [Music] the

Channel: Gamers Nexus

Views: 780,465

Keywords: gamersnexus, gamers nexus, computer hardware, nvidia rtx 3090 founders edition review, nvidia rtx 3090 review, nvidia rtx 3090 benchmarks, nvidia rtx 3090 vs rtx 3080, nvidia rtx 3090 8k gaming, nvidia rtx 3090 8k benchmarks, nvidia rtx 3090 overclocking, nvidia rtx 3090 worth it

Id: Xgs-VbqsuKo

Length: 27min 3sec (1623 seconds)

Published: Thu Sep 24 2020