Gadgetory

Demitri from hardware kinetics got a whole lot of hate for switching back to Intel from risin in his video talking about the new Intel hardware acceleration in Adobe Premiere and to a very large extent it pretty much proves the point that the blind fanboy ISM is just completely deprived of any kind of logic its bereft of any any logical ability to look at data and and agree that yes for his use case it made sense to do what he was doing and today we're actually validating some of that testing not for any other reason then we've been looking at building a new rendering rig and at the end of the day Dmitri makes different types of videos than us so we might have different needs we thought it's worth running some of the similar similar tests with the IGP hardware acceleration to see if it makes sense to go with a thousand plus dollar CPU or a $300 one with an AI GP that would otherwise never be used so that's what we're doing today before that this video is brought to you by thermal Grizzly's high-end thermal paste and liquid metal thermal Grizzly's cryo knot is an affordable high quality thermal compound that doesn't face some of the aging limitations of other pastes on the market cryo not has a thermal conductivity of 12.5 watts per meter Kelvin focuses on endurance is easy to spread and isn't electrically conductive making it safe to use on GPU dies thermal grizzly also makes conductor not liquid metal which we've used to drop 20 degrees off some temperatures than our dee-lighted tests by a tube at the link in the description below so for these tests we're looking at Adobe Premiere encoding and rendering exclusively there are obviously use cases for high core count CPUs look at Blender we've done those tests it's not the point today it's not the point to try and make your favorite 1000 plus dollar CPU until there Andy look good the point is we're doing this for ourselves selfishly to figure out what we need and if you happen to be someone who does a lot of video editing and production maybe your videos are similar to ours in some ways this data should be comparable for you so you can figure out what's the best choice for your build if you're working with Adobe Premiere primarily now obviously if you're working with things like blender you have more stuff to consider but for Adobe Premiere they got an update recently it's it's one of the few updates from a doe that doesn't instantly cause crashing and freezes and all kinds of fun features and in fact adds IGP utilization on the CPU so if you have an integrated graphics processor on an Intel CPU ie and 8700 K then premier can now leverage that and rendering to accelerate the rendering just like you can leverage CUDA and it can actually do both at the same time it's actually very interesting because we've been looking at using a sixty nine hundred K for a new system build we want to do but a seven hundred K might actually be better and it's easier to get in terms of it's cheaper and it's worth less on our shelf than high-end CPU so these are 100% real videos we're testing with today these are videos that we actually made in the last week or so rendered in their entirety and uploaded and we're using 4k videos for the most part but we have one 1080 video and that's a builds or video that we have erm analysis we uploaded so there's an interesting main here and that having the significantly reduced pixel count with a 1080 video versus a 4k video has actually serious implications for rendering performance so if you're doing video editing for your own YouTube channel or whatever and you do 1080 maybe 1080 60 like we do instead of 4k then the numbers are not going to be the same as 4k numbers for rendering they leverage different parts of the hardware in different ways and also we noted a couple of interesting things where GPU usage is significantly higher in our videos where we have charts being drawn so actually in this video you'll see some charts when those are on the screen the CPU usage plummets during rendering and the GPU usage goes up a lot as an example with thread Ripper and we'll show this later 1950 X we saw CPU utilization and a dual encode meaning two videos simultaneously dropped from a hundred percent during a role which is this right now this this is a role where I'm on camera there was a hundred percent with that and that's with color correction and other small effects may be some transforms opposed the color correction and just a roll which is a lot of pixels so one hundred percent usage and then a chart pops up and now CPU utilization on the 1950 X was 43 percent but GPU usage went up from 40 to 88% and that's on the cuda accelerated 1080i so no IGP there so that's the kind of thing we're talking about where usage matters and dimitri certainly had a point to switch to intel for his use case i frankly don't fully understand all the hate they got for that specific video it's not like it was a vacuum video it was him switching platforms because it works better for him and honestly the intense love of the downvote button on anything where someone's like i'm switching from xcp of hawaii cpu is mind-blowing the fact that people can't look past their their sheer blind fanboy ISM which by itself is mind-boggling even have that mentality to see that yes this particular individual is better served by another component than the component that I personally like I don't I don't know how we get there but it's it's really just perplexing to see I really truly don't get it and honestly it's stupid if you have that mentality it's a very bad one you should get rid of it so let's move on G 5,600 review is the first one this is a review of a CPU we did recently it is the entire review so it gives us an idea of how long it takes to render an actual review video that we make which is again for our selfish purposes for figuring out what we're going to use in the future or entering machines so we'll start with a rendering of our Intel G 5600 review it's a 16 minute video that uses color correction to adaptive noise reduction the Intel i7 8700 K isn't impressive when looking at the older method of software rendering and proves to be the slowest of all the CPUs tested the stock 8700 K takes 46 minutes or about 12 percent longer than the r7 2,700 ex stock CPU and when overclocked to 5 gigahertz it's still at 43 minutes an 8 percent reduction in time required but still slower than even a stock 2700 ex using Hardware encoding pushes work to the IGP which is typically left completely unused when leaning on a D GPU for CUDA acceleration and also even when you're not even just software accelerated before this update it didn't do anything integrated graphics processors in the ACE how hard K accelerate the rendering to 32 to 30 three minutes a reduction of 30% in time required to render this is also 20% shorter in render duration than the 2700 X stock CPU resulting in a complete shuffling of the leader board for high-end parts specifically for purposes of our rendering workstation this is basically invalidating both the threader for 1950 X and the Intel I 979 ATX II technically speaking both are faster than an 80 700 K but not proportionately to their price the 1950 X is functionally equivalent in render time with stocks to stock differences totally in three minutes against a 33 minute render so about 8 to 10 ish percent sure it's an op lift but it's not worth the extra money of only using premier on the system the 79ad XE is even worse in value overclocked in to 4.6 gigahertz which is our back supported overclock on a 280 cooler allows us to finish the render in twenty four point five minutes and we wanted to pull 700 watts from the wall and stick a 360 millimeter radiator with loud fans on it we can push 4.9 gigahertz but just would not be worth it and it's not really a good fit for 24/7 use at all either way stocks the 30-minute completion time isn't worth the $2,000 plus costs not counting the platform 87 hard K blows all of this away and is a clear performer when it has the IGP enabled for a 4k 60 DiDio review basically with some charts and a couple of other things that you can actually see that's the cool thing about these tests all of the rendering all of the test cases go watch her videos and you can see them and also we get more views so it's it's a nice feedback loop there definitely why we made it that way not because it was just easier to test that way so there's an argument to be made for just raw performance and in a production environment sure I've actually argued this in the past tune our reviews if you're a professional and you have a lot of money to spend on something then it might make sense to go with something that's fractionally faster single digit percentages sure if it's a difference of a thousand bucks maybe that means nothing to you but here's the thing this is our production environment we're talking about so I don't really care what anyone else is using the CPU for keep that in mind for our production environment we do have budgets to consider contrary to popular belief I don't just get any CPU I want from any vendor all the time the 79 80 XE we have the 1950 X we have they have to stay in the CPU track tray for further testing for benchmarking for content that's more important than putting them in a machine so if I wanted to use one I'd have to buy it and put it in our next rendering machine but we're talking about single digit percentages before over clocks in terms of advantages with those CPUs and their 800 or $2,000 depending so you go with an 8700 K in this instance with the IGP that's like 300 something bucks and that's it it's within single digit percentages of a two thousand dollar Intel CPU it's like premier has single-handedly for this specific use case of ours and validated the high-end CPUs that we actually wanted to use by enabling IGP rendering so it's awesome to see it's kind of sad that it took so long to get this and hopefully they can stay on this and fix things like you know the soft are not working in the future as well let's look at this really interesting chart next Adobe allows users to perform simultaneous renders about putting from the same project file like from two separate sequences in this use case the high-thread-count CPUs showed their potential benefits we're at 46 minutes to render two files simultaneously on the 79 80 X e whereas we would have taken 60 minutes if rendering them one by one on the 79 80 X E and that's I mean refer to the previous chart for those numbers it's 30 times to the 8700 K has the opposite response it takes 76 minutes on a software and hardware render 77 minutes for dual software rendering and so many two minutes for dual hardware or IGP rendering in all three of these cases it would be faster to render two files sequentially rather than two files synchronously or simultaneously on the 8700 K a single render would have taken 33 minutes two sequential renders would have taken 66 and dual rendering takes 72 minutes on the 8700 K easy math this tells us something critical and gives us some insight we didn't have a moment ago and that thing that it tells us is that a 79 ad XE or 19:50 ice for that matter can be faster when you're talking about rendering multiple things simultaneously and that's an interesting use case but not one that applies to us so for our purposes everything we do is sequential we get the videos done and it's more or less one at a time render it keep moving because we don't have a whole bunch of product projects to just queue up and let them rip all at once so there is a use case there if you do that first of all I'd be curious to know why like what what is your job that you render two things at the same time or more in the same Adobe Media encoder window because that's interesting but also that would be a potential use case for a high end CPU where you do actually get significant benefit from having the threads but what this teaches us is that Adobe Media encoder premier in general when rendering they care far more about frequency for the most part there's obviously there's a bit of a lien on the GPU potentially on the IEP but frequency matters the most for our type of rendering and having more threads there's diminishing returns there because you do sacrifice frequency for it unless you're doing dual or multi rendering at the same time you're encoding multiple things at once then yes it does actually help to have all those threads because the the frequency doesn't matter as much anymore but if you're doing synchronous versus sequential sequential does actually seem to care about frequency first next one builds a video for us we have a 1080p file with no effects and it behaves a lot differently from our usual 4k correction having final outputs we used one of our builds with videos for this which is rendered at 1080p 60 a lower bitrate and has no special effects in this test we observed no measurable performance improvement with hardware acceleration there's just nothing for it to accelerate you won't see the gains of GPU acceleration if there's nothing to use a GPU on you'd want transforms or similar effects to get some value out of the IGP for a straight 1080p file even running a 798 exe or 1950 X doesn't really change things meaningfully anyone rendering files similar to this one would not need high-end hardware or even IGP acceleration that said CUDA acceleration still helps and is enabled on all of these tests speaking of CUDA acceleration we decided to torture ourselves just to give you an idea of how much that still matters with CUDA acceleration disabled in our NZXT m22 review render it took the 8700 ka 350 minutes that's five point eight hours to render the same clip that was finished in 34 minutes with software rendering and CUDA using Hardware rendering on the IGP and also CUDA by the way we brought this down further to 21 minutes a reduction from the 34 minute prior time of 38 percent and a reduction from the 350 minute time of a lot of percent that's hands-down impressive again and shows where we should move for our next render machine the 2700 X days closer to 30 minutes and the 1950 X at 23 minutes with the 780 XE at 20 minutes overclocked in the center 980 XE does get it a few minutes faster than the 8700 Kate stock CPU with the IGP assistance but at 2000 dollars versus 350 it's simply not justifiable for us remember we have to keep this CPU the i9 for benchmarking so we'd have to buy another one for rendering it's not worth the cost finally for an ask GN video one of our most popular series we measured the 8700 K at 90 minutes completion for both the stock and overclocked variants when rendered without IGP assistance enabling the IGP brought us down to 53 and 58 minutes not distant from the 1950 X and 79 80 X II this type of video is intensive on the CPU because there are very few transforms and GPU accelerated effects it's almost entirely straight a roll would just specifically CPU intensive so we're going to go over to the test bench just for a second to get some hands-on with what premiere does and what its resource utilization looks like just to give everyone a better idea of how it behaves these days then we'll come back here and do the wrap-up in conclusion with how the effects are impacted as well so here's the 8700 K now playing back one of our clips and we can actually set this to higher quality as well so in certain playback right now and if we look at task manager while it's playing back what you'll see is when it's playback a chart and it's struggling a little bit here it sees all that stuttered because I'm at full quality right now we're actually pushing the load primarily to the when it actually does load up but it's just kind of spiking all over the place and this is indicative of some Adobe premier optimization issues that we're all where there's really no there's there's no clear single resource right now let's take it on this processing task so let's just drop this down to more processable speed like half so you have that changes things so now we're getting is some more GPU load and this is the Intel GPU the IGP so you see this is actually a hundred percent load right now the CPU is bounced between fifteen and eighty percent depending on how much the IGP is taking up there's some latency between these two should not see a perfect response between them and then to the accelerations during some of the rest it just depends on what kind of image we're rendering presently in the playback monitor so if we kind of jump ahead maybe it's some arrow lure b-roll which has this one's actually slowed down so the speeds reduced on this to smooth out the clip and the result is the IGP is taking all the playback and this is in real-time playback mind you and if this stutters it gets really hard to do editing because I mean it's stuttering so you have to constantly pause and start or you can drop the quality to like an eighth and then you end up with a really bad picture and play backwards super grainy so right now we can sustain a half play back the full playback on the 8700 Kate stock as CPU is taking a good bit of the load the GPU on the scene is taking the rest and CUDA acceleration is taking some of it just depends on what's going on so that gives you an idea of how each of these works now we're gonna open up a test file with some different effects that we can toggle on and off so here we have a clip we do some really simple color correction applied and disabled and if we just do that while it's playing back there might be a bit of impact on the components but there's almost no impact in rendering which happens after this this is again previewing so overall luma tree doesn't really seem limit recolor just doesn't seem to impact much of anything and my playback or in rendering is a bit more in live playback than the rendering though and then we can also do a clip with some warp stabilization just for another test case and for this one what we'll do is take a clip that has warp stabiliser plied and see if there's any impact playing back versus not applied so instantly we're getting a lot of IGP load right now still like 14% cuda load so not a whole lot going on with cue the acceleration premieres never really properly leverage that during preview it uses it on and off during rendering but yeah lots of IGP load not so much on the cpu and then if we look at another example of applying warp stabiliser rather than enabling it we can see what that looks like because now it's gonna analyze and this will take a long time so it's analyzing frame by frame this entire clip to smooth it out and over here we can see what it's doing to analyze which appears to be not a whole lot of load on any particular resource right now bit on the CPU and a little bit on my GP but not a whole lot really on anything so this is going to take a while and we're not really sure what exactly is going on during this process it might just kind of try to be a low intensive background process and maybe that's why it takes a while but either way not a lot of immediate impact playback though a while we work stabilized you'll see a lot of IDP load because that flip is warm stabilized already so there are some basic examples for you just a another firm reminder here this testing could be done a lot of different ways we're not doing it for your benefit really it was for our own to see what kind of components we should upgrade to and if your use cases happen to align with ours hopefully some of this teaches you something along with the charts we have otherwise it's mostly just for our own updates for the future and finally now back in the main set moving on to some studies of effects we noticed that rendering is largely unimpaired when toggling things like warp stabiliser luma tree color and scales and rotates a lot of the impact is seen in real-time playback when scrubbing the video or when analyzing the video with orb stabilizes but doesn't seem to much challenge the render times for our benchmarks orb stabilization during rendering for instance was 4.3 minutes with software each time or 2.2 minutes with hardware acceleration for each render that said the frame analysis pre-render would be more impacted than the render itself limit three colors showed the same thing 5.6 versus 5.7 minute surrender time with hardware acceleration which is completely within margin of error and that's when toggling the effect on versus off it's about 12 minutes with software acceleration when toggled on versus off so again no change at all from the toggle of Lumet recolor scale and rotate showed mostly the same on more bound configurations specifically GPU bound this one starts a challenge lower and CUDA devices but we mean very low-end here we use a 1080i for our production machine so this is a complete non-issue for us and for anyone wondering there's no difference at 1080p we won't bother with a chart for that we tested the same things there and it's all the same lack of difference and that is how adobe from your behaves at least for our use cases again a couple things here before all the especially and the fans out there freaked out and start typing in mean comments here's the thing they're perfect to use cases for the 1950 x4 the 79 80 XE for the 8700 k and for the 2700 X these renders not really the best use case for a lot of those CPUs for the high-end ones especially you'd really - for the most part sadly be better off with an 8700 K and IGP and that's again for our specific type of rendering so to give you an example of where the high on 1950 X which is great value for the dollar we've again repeatedly recommended it and have given it a lot of awards the 1950 X or the Saudi 90 XE at - 2 X the cost are great choices for rendering multiple things at once or if you do other stuff than premier on that machine like blender where you can definitely make use of all those threads and blenders advantage over CUDA in some instances when you just run out of cuda memory so if you have a cuda memory limitation with your project files then having a threader per cpu with all the 32 threads working at once and more or less limitless ram 32 64 gigabytes that's a lot of power and a perfect use case for it but for what we're looking at the 87 hard k is a better fit for us what we use right now just so you know is an X 79 platform it's several years old and we were looking at upgrading to x99 with a 6900 k because we had retired at 6,900 k recently and we have x99 boards really was looking forward to that build sadly we're not gonna do that builds because going with a high court count intel cpu is sixty nine hundred k with no IG p in it is gonna be worse than an eighty seven hundred with with an IG p and we can just overclock that when tribute as well so we're gonna go with eighty seven okay for a next build i think is what it's looking like currently our x79 platform is actually I think it's a twenty four thread system it's a Xeon CPU and we bought it specifically for the thread count which has been nice for blender but the frequencies low and up at 3 gigahertz that for premiere those threads are one basically never utilized and to when they are utilized the frequencies low enough that we don't get I mean it takes two or three times as long as the 8700 caters to render so it'll be a huge upgrade for us to go from X 79 and a twenty-something thread Xeon CPU which is just not the right fit for Adobe from ear to you know to a new one so that's what we're looking at there again if you have an AMD CPU don't feel bad about it there's no reason to feel buyer's remorse or anything else that you might be feeling because they're still good CPUs rise and CPUs are doing just fine they don't have to be good at every single thing all the time they're still good at premiere though it's just that this new update makes Intel's IGP and actually advantageous thing to have and it's not something really that I think anyone was expecting IGP has kind of been a stupid pointless component for the most part but suddenly it has it use and that's really interesting so we'll be using an 8700 K I think for a render machine and we'll stick with Rison options for things like blender rendering or something like that where we have just a lot of tiles to do and we don't have the frequency concern that premiere seems to have or the IGP advantage and one final thing as a reminder stock the 2700 X was doing better than the 8700 K stock in Adobe from your without hardware acceleration so I'll punt ill this point yeah the 2700 X was a pretty good choice and there's no reason for anyone to feel bad about that now because if you're happy with it or if you were happy with it then nothing's changed it's just the 87 hard K is suddenly using something it wasn't before so that's it for this one thank you for watching subscribe for more we go to patreon.com/scishow and exit stops out directly go to store dock cameras nexus dotnet let's pick up one of our mod mats like this one the anti-static building service that's four feet by two feet if you understand American metric system by which I mean not the metric system the check it out on the store that gamers exit on that I'll see you all next time