Gadgetory

so despite having just talked about how the Final Fantasy benchmark is basically pointless right now we still wanted to put out the last bit of data we had on it before moving on and that's CPU data so the primary interesting point here is testing the thread count and the utilization of thread count which you can specify through command line with the Final Fantasy benchmark hence why it's actually it'd be a great tool if it worked properly but anyway we're gonna be testing SMT and hyper threading a couple of other things today for CPUs not doing a whole suite of CPUs because halfway through doing all this data collection is when we realize like oh this benchmarks actually not great so but I did want to publish what we got before we just move on completely and forget about it till launch before that this video is brought to you by EVGA and their GTX ten ATT is c2 video card with icx technology the 1080 is c2 has a nine thermal sensors spread across the board which allows you to easily check the cooling performance of the vram the view RM power components and the GPU this makes for better noise to performance tuning in software and you can learn more about the sc2 at the link in the description below a few notes here before we get started if you missed it we have a video about why the Final Fantasy benchmark we think is sort of flawed this comes back to an issue of being optimization on square enix side I think we talked about this in the article a lot more than the video there's like six paragraphs at the end of that article talking about how ultimately comes down to square enix not down a game works because although game works is implicated in this and then Vidya does have a responsibility to clear their name and make sure the developers are implementing their stuff properly it's still Square Enix who are calling basically every object in the game which we're working on confirming with more tools by the way they're calling it in and rendering it and just loading down the GP resources all the time so it doesn't matter if it's game works or not game works objects like hair works applied to the buffalo are particularly impactful and are what tipped us off to this problem because you can toggle game works through an i and i hack and run the test in an area with zero game works and you'll still see it performance Delta when you turn it on versus off in an a/b test hence leading us to the discovery that a whole bunch of stuff is being rendered all the time we're now using another tool called render dock still learning a lot about it and we've pulled some of the meshes out from frames that don't contain those meshes in the camera in the viewport at all so yeah some interesting stuff we're still working on just out of curiosity at this point but basically there's a lot of stuff being drawn and that means this is actually an awful CPU benchmark as well because at 1080p medium we're still bumping into a framerate limiter on a 1080i at 1080p 1920 by 1080 despite the name being a 1080 Ti it is not actually supposed to stop at 1080p resolution so it's not a great benchmark 1080 low we're still pretty much about mean up against the cap so it's it's really just not optimized right now and if it is then Wow huh yikes but yeah so let's just let's go through the numbers and you'll see what I mean we can start this piece by illustrating just how easily the game bottlenecks on the GPU even when we're trying to do a CPU bench this is at 1080p medium settings for the first chart and we're clearly hitting a bottleneck at around 137 FPS average on the GPU GPU is a gtx 1080i FTW 3 i'm on one of the best gaming cards you can get right now and it's at 1080p medium settings and that's still too much to be a viable cpu benchmark with these settings we only start seeing real divergence from high-end parts when we step down to $100 our 3 CPUs for example so yeah illustrating the point this comes back to what we found with Final Fantasy 15 s benchmarks silent rendering of nearly everything on the map it's not just game works it's basically all the 3d objects like the Buffalo itself which is a game works object it's the host of game works objects but here's another example this is a frame we analyzed from the game where it's the main character on a fishing dock and even during this frame the game is still rendering cars that aren't nearby rendering item chests that aren't in frame were nearby and rendering large portions of highway that are located miles away we have another shot where the characters are drive around and we're still rendering for example the birds and the iguanas or whatever they are and things that are 2 minutes further into the benchmark than they are at the current scene so we're still looking into this still need to clarify if this is actually stuff that's being drawn or if it's just this utility intercepting these things and we'll talk about this more later but anyway it looks like there's a lot of stuff going on that shouldn't be so after stepping down to 1080p low we can finally start to plot some actual CPU performance differences we're still bottle necking at the high end but not as flatly as before with these settings the Intel i7 8700 K demonstrates our point of Givi limitations overclocked to 5 gigahertz or stock we're still bumping up against a rough 174 fps checkpoint the GPU utilization is nearly 100% at this point further illustrating that limitations of usefulness for this benchmark are bound by even GPUs at 1080p low anyway a shiny note here is that the game does seem to like threads but only up to a point with am these r7 1700 we noticed that performance improves a thousand he disabled and we saw performance uplift at 5.1 percent from the stock r7 1700 to the rs7 1700 with SMT off for the r5 1600 X we observed a 4% vorontsov list by disabling SMT and note also that frame time consistency is not hugely impacted we are technically plotting a downtrend in low-end frame time consistency but we can't confidently state whether this is statistically significant or accurate as the benchmark is simply too inconsistent to establish confidence in that 0.1% lowest wane this is further illustrated by the opposite behavior on the r7 1700 where we still saw average FPS performance uplift but we also saw 0.1% low performance uplift relating this back to our previous research with the num threads commands we believe that the game encounters a point of diminishing returns at around 8 threat up until that point more threads is better and after that point though we either lose performance from an inefficient load balancing across the threads or we stagnate in performance this leads to a greater discussion on CPU utilization Asian for which we also have charts from previous research because lower utilization is not in fact a good thing there's a misconception that a game utilizing minimal amounts of the CPU means that the CV has more Headroom for background processing in reality what this means is that we're load balancing across all the threads inefficiently and losing performance as a result with any component you want to be fully engaged or close to it in any task and a closer to 100% the better because that means if you're able to leverage the component to its fullest potential not wasting any performance the background operations that exist should have some native load balancing and the OS should work with them to distribute resources as needed otherwise you can manually do it and back to the Final Fantasy 15 1080p low chart the 7700 K and the r5 1400 both demonstrate we're disabling hyper-threading or as some tea results and a net negative in these instances the r5 1400 CPU sees a deficit in average FPS and frame time consistency the 7700 K achieves an appreciably different average FPS but has halved 0.1% lows we attributed this to a four thread limitation on both devices which the game really seems to not like at the low end highlighting the r3 CPU the r5 1400 and the 7700 K with us on T off the game does not like working with four threads at all at the high end with the r7 1700 and r5 1600 X the extra threads should actually be toggled down to a count of eight for peak performance we are uncertain about the 87 100 KS performance behaviors without hyper threading because we can't reduce GPU load enough to limit the CPU eliminating the CPU would require 480p or some other really low resolution which enters a realm of becoming a strict academic study and exits any usefulness whatsoever looking back at our CPU utilization chart from a few days ago we can show again that using num threads commands to limit thread utilization does have a noteworthy impact the result is improved or equal performance with half of the r7 seventeen hundred's threads interestingly despite this game seemingly hitting a point of diminishing returns at eight threads it will still attempt to use every thread you give it just in a less efficient way also interestingly this type of limitation would indicate an IPC bias or a frequency bias in the very least so when we consider that but you look at the numbers Final Fantasy 15 as a benchmark is actually doing pretty well on rising despite having performance behaviors that would typically suggest a frequency bias so in this instance we have for example an overclocked r7 1700 at 4 gigahertz versus an i7 7700 K at 5 gigahertz and the r7 is still favored by the Final Fantasy 15 benchmark which is certainly noteworthy and potentially impressive we'll see what happens as the square-enix developers continue to refine their game so that it's more than just a 4 gigabyte benchmark I mean this is not representative of anything however from a cpu performance standpoint I shouldn't suspect that would change very much it seems things would change more on the GPU side with all the rendering issues were encountering so from what we're seeing now especially given one month left there's no time to refactor the engine or anything like that it seems likely that the r7 will in fact retain a pretty good performance advantage in this particular title you will want to look into the option of running the numbers command will have to revisit at launch but running num threads equals eight would limit the thread count down to a point where you actually get some more performance than if you let it use all 16 and it will use all 16 if you allow the game to however the load balancing is a bit more favorable on eight threads rather than 16 and I guess depend on where you benchmark it seems to outperform a 7700 K when both are overclocked which is absolutely noteworthy and the 8700 K does outperform both of those devices as it does have both frequency and threads however we can't actually pinpoint how high it would go without a GPU bottleneck without just without doing 480p or something which is pointless so that's kind of where we're we're stuck now the 8700 K may be doing better than we're seeing on the charts because we just we don't have a way to know how many more frames it would be drawing if it were unencumbered by the GPU a very interesting game benchmark anyway from a an optimization standpoint and he is fullness standpoint I do genuinely like the Final Fantasy 15 benchmark as a tool for testing I just needed to be better so we'll see what happens when the game launches but I think that's more or less the last bit of data we might talk about it and it asked Gian or some of the short video at some point but for now that'll wrap up most the Final Fantasy 15 benchmark stuff so as always subscribe for more check the other ffxv videos on the channel and go to patreon.com/scishow can next it's helps that directly and go to store gamers and access net slash mod matt to pick up a mat like this one they're on backorder because we blew through our entire first production run so thank you to all of you who ordered that's all for this one I'll see you all next time