This guy is such a fucking G, he straight up just answered everything in the start of the video. Knowing his quality content will keep viewers watching longer.
I haven't played CS in about 7 years and I still watch every single one of your videos. I cannot get enough of your insight, you got the funny and the analytical.
@@Ultimaximus pretty much what Jake said 🤣 I used to grind csgo back in 2014-2016, clocked in 2k hours, and then dropped it. But seeing the hype around CS2, I feel like itd be nice to jump back in at least for casual gamemodes, just to feel the movement and shooting again.
You're in good company, because a while back Philip himself mentioned that he doesn't play the game either -- he just loves to study its design and mess with the tech.
As for Vulkan, Valve has admitted that they're misusing the API. The API exposes the raw Pipeline State Objects, which are a combination of the shader code, pipeline primitive input, etc. These are slow to create, because the driver must take the shaders, match them to the state and compile it all for your GPU. These are the things that should be done during the loading screen, but CS2 right now creates them on the fly, and caches them, which results in heavy stutters the first time a particular PSO is used. It gets better over time but game engines are expected to know ahead of time what PSOs they'd need, which CS2 has no way of tracking. What's very interesting to me is how much CPU utilization there is while you're playing on an empty map. What is the game doing with all that CPU time? Performing physics simulations on a single player?
The fps is uncapped, and the game is multi threaded now. Of course it'll have high utilisation. It may not be calculating much, but its doing it over 800 times a second
@@Derpynewb But here's the thing: Multithreading a game is *not* a trivial task. It's not like you can just put random stuff to be done on other threads, if you do things wrong you will end up with *lower* framerate and CPU utilization because you will be waiting on thread synchronization. I just don't know what the game is doing there to occupy the CPU. The recording can be done from multiple threads but realistically how many draw calls can you have in an empty server? Calculating of bone displacements can only be done in one thread since there's only one player to animate, so there's at *most* 2-3 active threads doing some work client-side. The built-in server is running but that's only 64 ticks so it should barely show up if at all.
Considering the graphical improvement compared to fps loss, this is very much still well optimized. They are likely to be putting the most effort of optimization into the new maps/overhauled effects where it's needed most. So we can't expect much better performance and dust 2 is best case scenario
@@pandiem Making things brighter is not a graphical improvement lol. In terms of quality and detail it's mostly unchanged with visible downgrades in places
in my opinion, the data in the video suggests the contrary. It is better at using more of your processor, and can perform similarly or even better despite being years newer and having graphical upgrades, so it seems much more optimised
I love your videos so much you probably get this comment everytime but im serious. i found your channel when i was getting into the hammer editor you helped me out so much, i started watching more and more of your videos to the point where ive watched all of your videos and especially when you upload quite alot i really look foward in watching your videos. :)
@@3kliksphilip Wow i was not expecting you to reply i enjoy your videos alot and to see how good your video quality has gotten through the years is amazing
Vulkan and Dx12 statistically run slower when directly compared to DirectX 11. Hypothetically both should be faster, but for the programmers achieving that in practice is rather hard to do.
@@griffin1366 It also really depends by driver implementations, not just complexity of their renderer. It's all about priorities and how mature drivers are. DX11 driver on NVIDIA is already really well optimized, which is why DX12 sometimes look slower than DX11. Vulkan is pretty much king on Linux (for example RADV for AMD GPUs, basically fastest Vulkan driver in the existence, also used on Steam Deck).
The fact that 1% lows are so much worse is actually a bit alarming since the main reason cs doesn't feel anywhere as smooth as other games is in the very inconsistent framtimes. This might be a much bigger problem than average fps.
Performance tuning and QC is the last thing devs do before the stable release, so the chances of the performance increase is very high. Especially in the low 1% because DX11 is a more mature platform with easier api to customize different aspects of a game. CS2 final performance drivers from either Nvidia or AMD has also not been released yet, which in CS2 case will help immensely as compared to Vulkan or DX12 drivers play a more important role in DX11 due to system overhead. They'll most likely change render prioritization, then there will be new CPU profiles for P+E core systems by Valve as DX11 is primarily an old Windows feature. There will be networking upgrades as well, they might even push for 128 sub-tick system, they're probably more focused on fixing the movements, and some common bugs noted in the beta access, as well as the porting or remastering of other maps, fixing animations. Then they'll start profiling for performance issues.
@@jonan2199 Probably because Source 2 had already matured by the time DX12 became a widely accepted library. Also they might've avoided DX12 because older hardware might not support the complete featureset. For Vulkan it was kind of like a requirement for their linux system. They still might have DX12 in the pipeline in the longer run, or have DX12 option as well.
it's probably even lower than that. there's very few individual/specific cpu models that even have 1% or above marketshare. there's just so many computers, so many different cpus from over the years and so few who buy the highest end part every year.
The one thing that you didnt mention and is probably why this happend. Look at the task manager on 8:59. When you disable the e cores power doesnt go throught them and you can achieve faster Ghz on the main cores(p cores). So you can see in the comparisson that on the top the processor is at 5ghz but in the bottom one its 5.47... Its quite a significant margin to be honest... More ghz sometimes mean more fps because its actually going faster. Try testing same ghz and probably you should get same results. Hope this makes some sense and can help!
Never rely on task manager to show you the correct clockspeed. Each core can set its clock multiplier independently depending on its current activity, and E-cores max out more than a GHz below P-cores. What task manager shows you is some kind of an average, so of course it'll go up when you disable E-cores. HWInfo can show you clocks for each individual core, that'll give you a more accurate idea of how hard your cpu is actually boosting.
Yeah that's what I thought as well when I saw that. That power that would go into E cores can then be shifted into P cores. Of course this still has to take voltage and thermals into account, but considering he mentioned he power limited his CPU, this seems like a reasonable explanation. It could also just be a matter of scheduling tasks because it's not easy to balance things over so many threads and especially threads with different architectures. Using just P cores might make this scheduling easier and more reliable in this case.
It could be 600 or 1500 fps and the changes would only be on input delay The amount of images that'll actually be shown is still limited by your own screen's refresh rate or in this video's case, TH-cam's 60FPS Limit
Keep in mind that the E core utilization could also very well mean that Windows is shifting background tasks to them more efficiently while playing CS2. Why that would be more pronounced running a Source 2 game rather than a Source 1 game I'm not quite sure - maybe it's DirectX 9 related? Also, it could mean that all the network stuff is being handled more favorable by E cores rather than P cores. Again, not sure why, but the engine's thread/core prioritization could play a significant role here. Or maybe all I said is wrong and the Intel/Windows combination is just smart and figures it all out on its own. Who knows.
The extra details even on low settings might just be during the Beta so that all the footage looks presentable. And then they might gradually add even lower settings.
@@mindrover777 valve gave beta access to a bunch of pros -- people who are the most likely to put graphics settings to a minimum -- so it would make sense for them to limit how low settings can go during beta so people don't see clips of pros playing on something that looks like Half-life 1 and think the game looks like shit
Throughout this video I thought that something was wrong but I couldn't place what. Then I suddenly realised that there wasn't any of the classic 3kliksphilip hallmark music. What a time to be alive!
@@lilholm9446 How is one brick on the floor interfering with gameplay? Its not even possible to have it in line of sight. No item is small enough to get hidden in it even grenades would roll around to some.other spot. It doesnt affect your movement as its just cosmetic.
two things, I'd like to see testing of hyperthreading off and e cores on, and also thank you for doing benchmarks of CS2 when not many others have bothered yet!
For the engine that looks that much better while working marginally slower (like single-digit percentage on highest) that's really impressive. Source 2 was built with the performance in mind in the first place. And it still looks like 2023 game (unlike games that a releasing today with x360 grade visuals and performance of Crysis on pentium 3). After all it was made for VR first, to run on stupidly high framerate and resolution while keeping clarity. And there's definitely field to improve it.
e cores don’t actually cripple performance. it’s just when e-cores are disabled, cpu packet power can be channeled only to p-cores, therefore higher clocks on p-cores. e-core, even if they don’t use power that much, they still use power. without them, p-cores clock higher, only by 300-400 mhz though. e-cores actually serve perfectly fine in scenarios when single-core performances are greatly improved, but you don’t wanna stack too much traditional cores in a single processor, otherwise improvements are much limited. (ex. ryzen 9 5900x vs 5950x, where 4 extra cores only managed to improve multi-core performance just by %5 margin) e-cores don’t use too much power, do not cripple the other p-cores single-core performance and and enormously improve multi-core performance. another example: i7-11800h (8 core config) vs i7-12700h (6-p 8-e cores) 12th literally improves both single and multi-core performances by %50, while still limited to 45 watts, only one generation apart.
You mention that Directx 9 was old when CSGO was released, which is true that when CSGO was released DirectX 9 was already 9 years old. What you didn't mention was that directX 11 will be nearly 14 years old by the time CS2 is released. DirectX 11 was released with Windows Vista in 2009. Directx 12 was released in 2015 with Windows 10. Whether DX12 would actually be better for CS2 is something valve only knows, though I still wonder why it isn't the standard or Vulkan is. Vulkan might be better for lower end hardware than your current high-end hardware though. Still I wonder why DX12 wasn't chosen instead of DX11.
E-core gaming has always been so interesting to me, its neat that these modern intel processors have a bunch of extra cores for background tasks that are roughly equivalent to the oc'd 4790k that I had in my first gaming rig, and we look at them and think "eww bad". Gives me an appreciation for how fast tech progresses.
@@3kliksphilip yeah I really wish they would give us better control over which programs use what, kind of like how we can select which gpu a program uses in the graphics preferences menu
Always love seeing these investigations, keep doing what you're doing! Just two quick questions: why do you focus on average fps rather than 1% lows? I generally find the latter more relevant, would love your thoughts on it. And second is of course "yet another thing to test"! Have you tried using core affinity to force the game to run on specific cores without having to tweak bios settings? Perhaps paired with Process Lasso, which I think also lets you keep other processes away from the cores that are dedicated to CS. I don't think it warrants redoing all these tests, but could be worth giving out as a tip if you find it works the same or better than the bios option, since it's far less hassle and less detrimental to overall system performance. :) Okay sorry, guess those questions weren't as quick as I imagined... Anyway, thanks again for all you do, you're always testing things that in curious about and can't be bothered to test, or better yet, things I hadn't thought of but that you make me curious of! For me, yours is hands down the best CS content on TH-cam!
Awesome complete review. Thank you. Just feel bad because I will be forced to change my CPU + RAM + MB to take 240 fps on CS2... 1% LOW sucks on both game versions. Ps.: my PC i7-6700k + 32GB 2666MHZ + 2070S.
Saying "20% worse" sounds way scarier than "800fps instead of 1000fps" lol. I remember playing CSS on my first real gaming rig and being blown away that I had 300 fps.
Just got a 13900k my self... got it at 6ghz on 2 cores and rest at 5.7ghz... the fps boost over my 12900k on cs go is insane.. also would recommend you getting the anti bending bracket :) stock clocks i was under 80c on cinebench r23 with a manual vcore of 1.2v... with the oc I get around 89-95c on cinebench r23 with a score of 42k. stock was 40960
Hello Philip, I adore you for the work you put into this video and gathering such useful information. But what I have been missing the whole video (if I have not literally missed it myself) is the mention of a early development build you are comparing CS:GO with. It's nice to see already acceptable performance on CS2, but usually performance and optimization are two of the last things developers really but effort into in their development stage. Might have to redo the whole video on CS2 release.
An input delay would be the most interesting thing to test, because HL:Alex as a VR game does a lot to reduce it and CS2 probably has that carried over, so with its lower FPS it still might effectively run better.
I could say that it runs better on lower-mid tier PCs. I've tested it on mine with Ryzen 5 2600 and 1660 Ti with both DX and Vulkan. Vulkan actually did better by 5-10% on average. If I compare it to CS:GO, the fps does drop a bit, but the game runs much smoother. CS:GO can be sometimes choppy even with framerates higher than refresh rate of the monitor. No such thing with CS2, even when FPS drops below the refresh rate, it still feels smooth (maybe it's a nvidia reflex thing)
I'm not a fan of those 1% drops which are like 5 times less than avarage. This suggest that if you run cs2 at 400 fps you will get drops to 80 fps!!. It would be really unplayable, hope that valve manage these drops ASAP.
I don't play CSGO, but if i did, i'd be really upset with this update if it's something future players are forced into. A drop in performance sucks, sure. but what _really_ sucks is that it's for the sake of these new visuals, which for me personally, the lack of contrast and more, "pastel" colors, i guess you could say? it hurts my eyes. And i don't even find it aesthetically pleasing. I didn't play this game before because the audio mixing with the headshot noise was so loud it was physically painful- now the game's apparently become even more inaccessible to people like me. Bit of a shame.
In the task manager you can view how many threads are spawned for each individual process. Now, most of the threads won't actually be active, programs tend to be extremely wasteful with these and most of them will be spawned by external libraries, but when joining a game you can see the thread count go from 70 threads to 78, so it wouldn't be too wild of a guess to say that CS:GO can use at most 8 threads during gameplay.
You need to make hard choices about in what order you investigate things. Start from what people would be the most interested in or things that are the most relevant for most people and then consider things you really want to investigate. You can make more videos about the more nice stuff in the future, I love the nice stuff.
Makes sense why turning off HT would increase wattage. The CPU core is using more instructions and cycles to manage/wait/queue/share/swap workloads which hyperthreading would normally do. So to get the same amount of total work done, more overhead/wattage/speed is required when hyperthreading is turned off. It also allows identical computations to be computed at once. Ex: If you get asked what 2+2 is twice, you don't have to do the math twice ;) Understanding CPU architectures can be very informative :)
so, philip, I need to know: did you use the adobe AI voice enhance for this videos voiceover? I use it quite a lot myself and have started being able to detect it, and I'm getting some subtle subtle vibes of it here and there in this.
Something important to think about is that having your E cores active will allow common apps to have more breathing room, even if that means they use some P core while cs uses a little more E core - ex: discord, browser, steam itself (ex: overlay), windows. These apps can eat quite a bit, most notably a playing youtube video & discord calls
9:23 You can see that Windows report that the results with the e-cores active are 5ghz while the results with the p cores only have 5.5 GHz. I haven't played around Windows 11 and with Intel's newer CPU but what I'm assuming is that the e-cores is processing stuff and that heats up the chip and it underclock itself and or CS 2 is trying to balance the performance with all the cores so it when it utilizes the e-cores the performance dips due to the fact of trying to manage low powered cores. We can potentially test the e-core theory by setting the affinity of CS 2 to only use p-core and see what difference it makes. Also, it might be fun to see CS 2 running on only E-cores and see how it performs.
It would be interesting to get the devs' opinions on why 0.1% lows are disproportionately lower than the average decrease in performance. Asset streaming maybe?
Philip, this is not CS2 related but do you know anything about the bug in Counter Strike where everyone's FPS in a given server absolutely tanks at random? I've experienced this so often for years, dropping from like 600 to 100 in the middle of a game, with some friends dropping from 200 to about 15. Even people on the enemy team usually type something about "fps," so it seems server-wide. Do you know what causes this? It seems really odd that it's been a problem for so long.
I'm a bad PC gamer: I hardly know what most of this means and have no idea how to utilize any information you've given me. But I was entertained regardless
the difference between vulkan and dx9 or dx11 is that vulkan is a lower level api, which basically means it has more potential for optimization, as well as improving cpu performance. whether or not it actually is faster or not depends on how well the developers optimized their code and how cpu bound the game is. in extremely cpu bound things like emulation or some older games like fallout new vegas, you can see major improvements in performance by using vulkan even without well optimized code, but in something like cs i'd imagine it's much harder to get vulkan running better than dx11, even if it is possible
I wonder if the difference is the i7 hitting it's thermal/voltage limits and pulling back on the clock's boost. The Task Manager does show 5ghz when the E-cores are enabled, and 5.4Ghz when just the P-Cores are enabled. So we might be seeing the limitations of the 13700k's architecture.
In the test with P cores only vs P cores w/ HT and E cores, CPU freq is not the same and may be something to keep in mind to those who are watching the video.
True, the P cores alone are clearly able to boost much higher due to lower heat output and power usage which is a clear explanation for the higher 1% low.
Dual core gang, let's all gather and mourn the likely possibility of not even managing 60fps in CS2.. Goodbye my dear frames per second, you will be missed.
if you meant Intel dual-core (like 1st gen i3 or i5), then you can replace it with a used Xeon that have 4 core. It will work if it has same socket number (like for example: LGA 1156). The 4 core is faster than the dual-core in real usage even when Xeon is 2ghz vs 3ghz dual-core.
I wonder if the poor Vulkan performance you noticed is part of the reason why Linux was left out of the CS2 beta. CSGO doesn't let you use the Windows build with DX9 on Linux, you need to use the native build on opengl or the expiremental vulkan mode. So I'd imagine Valve is ironing out some kinks with Vulkan to make that the default on Linux. Can't really picture them shipping it with opengl in 2023 That being said I've played a good bit of CSGO on Vulkan in recrnt months and it has worked pretty well, I hope the Vulkan performance looks better by release for CS2
Interesting how in your CPU utilisation comparison, CS:GO had your CPU clocking ~400 MHz faster (5.4GHz vs. 5.0GHz). Would be interesting if you can see why this might be, and if you can get it to be the same speed for both (maybe lock to a lower clock speed) and then compare performance again. The same also happens in your E-cores/hyperthreading comparison, and with how CPU dependant CS is, this could be making a big difference.
Would love to see a framerate and frametime comparison when a smoke grenade is in your view, since that used to cause major fps drops on my old laptop back in the day. Would be interesting to see if the effect is still as taxing in CS2 as it is in CSGO.
When I read about people describing CS2's performance, a much smoother frametime was one of the things that I've seen brought up several times - where despite the framerate technically being lower, the game felt noticeably smoother despite of that because the frametime graph didn't represent a hacksaw. I think this is definitely something worth investigating further - especially on a more modest configuration that doesn't involve top-spec parts...
I suspect CS:GO processes everything on the P cores, while CS2 is able to more efficiently separate tasks on certain cores. So information that the GPU needs may be exclusively processed on the P cores, and all but GPU information may be dedicated to the E cores.
When I was playing the CS2 beta I thought it felt kind of stuttery, so I decided to measure frame times and frame time variance (the variance of the distribution of frame times, basically the frequency of microstutters) and found that, at least on my machine (5800x + 3070), the frame time variance was measurably worse in CS2. You can use CapFrameX to measure this. These were my results just running around Dust2 with bots: CSGO: < 2 ms : 97.55% < 4 ms : 2.42% < 8 ms : 0.03% < 12 ms : 0% > 12 ms : 0% CS2: < 2 ms : 53.11% < 4 ms : 23.2% < 8 ms : 23.63% < 12 ms : 0.06% > 12 ms : 0% So I don't think I'm imagining it at least, there is a difference in the frametime distribution (for my machine). You can see the precise thread count for CSGO/CS2 (or any single process) with Process Explorer, which might interest you. If there's more threads, then the game should benefit more from higher core count systems. Even on multicore systems the OS scheduler will move processes around if there's too much load on a single core.
Could it be that CS2's server is more demanding than CS:GO's and causing more of a performance hit, assuming this testing was done on a local server, if so could testing be done with the server running on a different pc, might show a more accurate difference of just the game client?
Seems like this beta only has a DX11 renderer, and uses DXVK to translate it into Vulkan (just like L4D2 and Portal 2 does with -vulkan variable). Some performance loss is expected because it isn't running natively. I seriously hope they develop/switch to native Vulkan renderer like they did on DotA 2. DX11 and Vulkan won't make a big difference on Nvidia hardware as their DX11 driver is optimized to a black magic degree. But it'll make lots of difference on AMD hardware.., at least on Windows.
I had to modernise my setup for the future smooth 144hz CS2 gameplay, since coupled with RTX 2070 super, Ryzen 5 1600x wasnt enough to provide it during alpha testing. There were a lot of drops, 1% low was something around 90fps at 1290x920 resolution. Now with the same GPU and Ryzen 5 5600x - the most cost-effective upgrade its something like 2x the fps at the same resolution and all visuals on the highest settings. I hope valve is going to suck everything from S2 to optimise it for as old equipment as possible, for the sake of all the people that are going to play it despite low performance setup (just like I did, and still do with my shitty internet lol). They live from those people money.
Why Valve just not do some kind of benchmark in game?! So after change of new version, gpu drivers or even cpu/gpu we could test if smth changed about fps. In CSGO we can only use „timedemo” for that with our 1 min demo with bots but in CS2 it’s not working. Greets, micronn
I'm entirely self taught so I could be way off base but, My guess on why the e cores lower framerates is due to the p cores having to still wait for the e cores to finish the lower priority operations before being able to process the information for the next frame, and by trying to more efficiently spread the load it's slightly slower for the e cores to complete the work than if the p cores just did it in the middle of everything else. Basically it's not optimised for the speed of your p cores and just assumes offloading some lower priority work will be faster, but it's not. The hyper threading thing is interesting, I have no idea how hyperthreading actually works but, I'm guessing hyperthreading leaves a tiny bit of performance for more concurrent operations, so with it disabled the individual cores can put out slightly more iops than with hyperthreading enabled, but less than if cs2 fully utilised 16 threads (likely due to not having enough operations to fill out the threads so some of them end up waiting for other threads to finish what they're doing before they can do their next thing, or perhaps before the engine will hand out the next operation) Alternatively it could simply be an optimisation problem and there is no tangible benefit to disabling hyperthreading and source 2 just doesn't know how to properly use this specific cpu properly. question to anyone who would be able to correct me, is there any potential benefit to disabling hyperthreading (stability, a minor single core boost, etc.) or is it purely a compatibility thing?
The thing with Vulkan is that it's so different from DX11 or OpenGL To put it simple, in DX9/10/11 or OpenGL you are expected to talk to the driver, in Vulkan/DX12 you are expected to write your driver. Its low level capabilities will give you an edge in performance, only if you invest the time in knowing how a GPU works
I believe Vulkan is slower as it's not fully optimized yet. In Dota 2, which is also Source 2 it can go head to head with DirectX any day. That may also be the reason why there's no CS2 beta build for Linux yet.
you want all your cores being used actually. the e-cores being used means the p-cores are busy with stuff. the more cores you have the more likely you'll find situations like this where the load is spread across all the available cores and this results in many peaks and valleys on the graphs. also, the higher the graph the longer the waiting time. cpu usage is determined by how much time it took to complete the work. i can't remember how or where i learned this or if i'm correct at all though, but for digital systems it is vital to measure time as accurately as possible so there might be some truth there.
You give yourself too little credit for how well you chose what to test and what not to test. You anticipated and hit exactly all the questions I would have had around performance in relation to power limits and p/e core allocation. Could consider making a separate video investigating the same questions with AMD Ryzen, and see whether the scaling goes beyond 8 physical cores. Even better if you do a counterpart to the p/e comparison with the X3D cores - assuming you have one of those lying around.
8 P-Core, No E-Core with HT Off surprised me... I would like to see a how scalable this is at this point, somehow I imagine 16 P-Core would only be a 20% uplift
For implementing Vulkan they should hire ID Software engineers, cause they nailed it with ID Tech 7 (DOOM Eternal engine); Minimum CPU bottleneck with all cores been used + GPU maximum usage ! I played it low settings on my laptop i7-8750H (6 cores; 12 threads) with GTX 1050 (Potato compared to current Gen); Constant 140 FPS in Final Sin arenas without signle freeze or stuttering !! Or maybe ID makes ID Tech 7 open source and it's late 90s early 20s all over again !!!
This guy is such a fucking G, he straight up just answered everything in the start of the video. Knowing his quality content will keep viewers watching longer.
Wtf is a "fucking G" Andrew Tate stan ?
@@Cr4ZzZy 🤓🤓🤓
@@Cr4ZzZy this comment is so retarded i physically flinched upon reading it, might have snapped a neuron or something
@@Cr4ZzZythe term "a G" has been widely used by all types of people long before Andrew Tate's dumbass
@@Cr4ZzZy calling someone a G has been around for decades, it didnt start with tate
It would be amazing if you took the average PC in Steams hardware survey and tested that
Brokey
@@xanmancan ???
@@xanmancan orphey
❤
@@xanmancan tweaker
To quote the half life 2 E3 show, "Will this run on my 486?"
"But my spreadsheet ran so well!"
Blast from the past
*laughter in the background*
weezer.
Ah the good old times, it was on the coast section too I remember it well!
I haven't played CS in about 7 years and I still watch every single one of your videos. I cannot get enough of your insight, you got the funny and the analytical.
Same, but i am thinking of returning when CS2 drops 💀
@@vator_rs What does the skull emoji mean here?
@@Ultimaximus pretty much what Jake said 🤣
I used to grind csgo back in 2014-2016, clocked in 2k hours, and then dropped it.
But seeing the hype around CS2, I feel like itd be nice to jump back in at least for casual gamemodes, just to feel the movement and shooting again.
You're in good company, because a while back Philip himself mentioned that he doesn't play the game either -- he just loves to study its design and mess with the tech.
As for Vulkan, Valve has admitted that they're misusing the API. The API exposes the raw Pipeline State Objects, which are a combination of the shader code, pipeline primitive input, etc. These are slow to create, because the driver must take the shaders, match them to the state and compile it all for your GPU. These are the things that should be done during the loading screen, but CS2 right now creates them on the fly, and caches them, which results in heavy stutters the first time a particular PSO is used. It gets better over time but game engines are expected to know ahead of time what PSOs they'd need, which CS2 has no way of tracking.
What's very interesting to me is how much CPU utilization there is while you're playing on an empty map. What is the game doing with all that CPU time? Performing physics simulations on a single player?
The fps is uncapped, and the game is multi threaded now. Of course it'll have high utilisation. It may not be calculating much, but its doing it over 800 times a second
@@Derpynewb But here's the thing: Multithreading a game is *not* a trivial task. It's not like you can just put random stuff to be done on other threads, if you do things wrong you will end up with *lower* framerate and CPU utilization because you will be waiting on thread synchronization. I just don't know what the game is doing there to occupy the CPU. The recording can be done from multiple threads but realistically how many draw calls can you have in an empty server? Calculating of bone displacements can only be done in one thread since there's only one player to animate, so there's at *most* 2-3 active threads doing some work client-side. The built-in server is running but that's only 64 ticks so it should barely show up if at all.
Hey, can I get a source for the claim that Valve has admitted to using Vulkan incorrectly?
@@V1etnow Source: My delusions
But really I think I've heard it in a GDC talk once, can't remember which one, sorry.
@@kiroma0 All good :) I’ll have to check out their talks sometime
Great stuff! In my opinion, CS2 is not optimized yet. You should test it again after the beta testing is done and CS2 is fully launched.
I was expecting this to be the major thing mentioned at the end of the vid
Considering the graphical improvement compared to fps loss, this is very much still well optimized. They are likely to be putting the most effort of optimization into the new maps/overhauled effects where it's needed most. So we can't expect much better performance and dust 2 is best case scenario
@@pandiem Making things brighter is not a graphical improvement lol. In terms of quality and detail it's mostly unchanged with visible downgrades in places
@@maskettaman1488 how did u get onto the internet monkey
in my opinion, the data in the video suggests the contrary. It is better at using more of your processor, and can perform similarly or even better despite being years newer and having graphical upgrades, so it seems much more optimised
I love your videos so much you probably get this comment everytime but im serious. i found your channel when i was getting into the hammer editor you helped me out so much, i started watching more and more of your videos to the point where ive watched all of your videos and especially when you upload quite alot i really look foward in watching your videos. :)
@@3kliksphilip Wow i was not expecting you to reply i enjoy your videos alot and to see how good your video quality has gotten through the years is amazing
@@txripzy4694x hxh mт тим лдс м н шоошоометр. о . хүний санааг шүү орших л вт
Vulkan and Dx12 statistically run slower when directly compared to DirectX 11. Hypothetically both should be faster, but for the programmers achieving that in practice is rather hard to do.
DX12 is faster but is far more complex to code for AFAIK.
they need john carmack lol
@@griffin1366 that's literally what he said
DX12 is faster but is far more complex to code for AFAIK.
@@griffin1366 It also really depends by driver implementations, not just complexity of their renderer. It's all about priorities and how mature drivers are. DX11 driver on NVIDIA is already really well optimized, which is why DX12 sometimes look slower than DX11. Vulkan is pretty much king on Linux (for example RADV for AMD GPUs, basically fastest Vulkan driver in the existence, also used on Steam Deck).
The fact that 1% lows are so much worse is actually a bit alarming since the main reason cs doesn't feel anywhere as smooth as other games is in the very inconsistent framtimes. This might be a much bigger problem than average fps.
Performance tuning and QC is the last thing devs do before the stable release, so the chances of the performance increase is very high. Especially in the low 1% because DX11 is a more mature platform with easier api to customize different aspects of a game. CS2 final performance drivers from either Nvidia or AMD has also not been released yet, which in CS2 case will help immensely as compared to Vulkan or DX12 drivers play a more important role in DX11 due to system overhead. They'll most likely change render prioritization, then there will be new CPU profiles for P+E core systems by Valve as DX11 is primarily an old Windows feature. There will be networking upgrades as well, they might even push for 128 sub-tick system, they're probably more focused on fixing the movements, and some common bugs noted in the beta access, as well as the porting or remastering of other maps, fixing animations. Then they'll start profiling for performance issues.
Indeed...
@@therealgamingmaniacwhy did they not go for dx12?
@@jonan2199 Probably because Source 2 had already matured by the time DX12 became a widely accepted library. Also they might've avoided DX12 because older hardware might not support the complete featureset.
For Vulkan it was kind of like a requirement for their linux system.
They still might have DX12 in the pipeline in the longer run, or have DX12 option as well.
Now what I'd love to see is once CS2 gets Danger Zone maps how they'll compare, since those maps pushed the limits of what's reasonable for Source 1.
thanks. no clickbait, no unnecessary dragging of content till the end of the video and straight to the point. liked and subbed
Just want to say: Thanks! What you're doing has value to us.
Thanks! Really helps those 0.01% people with 13900k
it's probably even lower than that. there's very few individual/specific cpu models that even have 1% or above marketshare. there's just so many computers, so many different cpus from over the years and so few who buy the highest end part every year.
😂😂😂😂 Fr
indian guy salty
the video is literally called "CS:GO VS CS2 Performance on a 13900K processor", were you expecting an aftermarket pentium?
Exactly what u said ¡black god!
The one thing that you didnt mention and is probably why this happend.
Look at the task manager on 8:59.
When you disable the e cores power doesnt go throught them and you can achieve faster Ghz on the main cores(p cores).
So you can see in the comparisson that on the top the processor is at 5ghz but in the bottom one its 5.47...
Its quite a significant margin to be honest...
More ghz sometimes mean more fps because its actually going faster.
Try testing same ghz and probably you should get same results.
Hope this makes some sense and can help!
Never rely on task manager to show you the correct clockspeed. Each core can set its clock multiplier independently depending on its current activity, and E-cores max out more than a GHz below P-cores. What task manager shows you is some kind of an average, so of course it'll go up when you disable E-cores.
HWInfo can show you clocks for each individual core, that'll give you a more accurate idea of how hard your cpu is actually boosting.
@@tiarkrezar i said that because he used task manager in the video.
Of course you should use differenet tools like you said above!
Yeah that's what I thought as well when I saw that. That power that would go into E cores can then be shifted into P cores. Of course this still has to take voltage and thermals into account, but considering he mentioned he power limited his CPU, this seems like a reasonable explanation.
It could also just be a matter of scheduling tasks because it's not easy to balance things over so many threads and especially threads with different architectures. Using just P cores might make this scheduling easier and more reliable in this case.
Still impressed that fps can still be in the 600+ range while looking so good.
It could be 600 or 1500 fps and the changes would only be on input delay
The amount of images that'll actually be shown is still limited by your own screen's refresh rate or in this video's case, TH-cam's 60FPS Limit
@@marcosm1223 That isn't even his point...
@@marcosm1223 The input delay is so minimal. Best to just cap at 400 FPS for stability and a predictable game.
@@marcosm1223 🤣🤣🤣🤣 ZAZA smoker spotted
there was a time when your balls fell off when someone told you that you can run Quake 3 with over 50 fps
Keep in mind that the E core utilization could also very well mean that Windows is shifting background tasks to them more efficiently while playing CS2. Why that would be more pronounced running a Source 2 game rather than a Source 1 game I'm not quite sure - maybe it's DirectX 9 related? Also, it could mean that all the network stuff is being handled more favorable by E cores rather than P cores. Again, not sure why, but the engine's thread/core prioritization could play a significant role here. Or maybe all I said is wrong and the Intel/Windows combination is just smart and figures it all out on its own. Who knows.
Personally I love the level of detail in this video and I really appreciate all the work you put into testing all of these scenarios.
Such an interesting video I love these little adventures into disabling and changing the cpus config to see how it affects game performance.
You are the only TH-camr that can pull these view numbers on this sort of content.
The extra details even on low settings might just be during the Beta so that all the footage looks presentable. And then they might gradually add even lower settings.
What does that mean? They will dumb down the graphics?
@@mindrover777 they might optimize it, better culling and less unnesesary models.
@@mindrover777 valve gave beta access to a bunch of pros -- people who are the most likely to put graphics settings to a minimum -- so it would make sense for them to limit how low settings can go during beta so people don't see clips of pros playing on something that looks like Half-life 1 and think the game looks like shit
@@RafsterMC let's hope it's better optimized than csgo that had crazy frame pacing issues unless u could push very high fps.
@@mindrover777 The same was done to Team Fortress 2 over time, it's entirely possible for the same to be done to CS2.
Throughout this video I thought that something was wrong but I couldn't place what. Then I suddenly realised that there wasn't any of the classic 3kliksphilip hallmark music. What a time to be alive!
Can we just appreciate the ammount of effort was put in this
Yes, testing and manipulating k series Intel CPUs is a very tedious process. But knowledge is power
I love when Philip starts clicking that tech stuff. He said he failed to keep this simple. And I'm just really pleased by this amount of information
CS2's dust 2 is both less and more detailed. Its missing those ground details like elevated bricks and some things look far too clean.
it's actually pretty unimpressingly the same, only the higher overall brightsness in CS2 makes everything way too smooth and less textury
it really reminds me of tf2
Its not supposed to be a beautiful game but a competitive one, valve has done a great job future proofing it
@@lilholm9446 How is one brick on the floor interfering with gameplay? Its not even possible to have it in line of sight. No item is small enough to get hidden in it even grenades would roll around to some.other spot. It doesnt affect your movement as its just cosmetic.
@@Kacpa2 visuals lower fps
Love your videos man. Its nice to see others keep going for the answers as well as more questions to be answered. Cheers man have a great week!
two things, I'd like to see testing of hyperthreading off and e cores on, and also thank you for doing benchmarks of CS2 when not many others have bothered yet!
wow, so much effort for quality content, appreciate it! liked & suscribbed
As a Statistician, I love your graphs Philip :)
For the engine that looks that much better while working marginally slower (like single-digit percentage on highest) that's really impressive. Source 2 was built with the performance in mind in the first place. And it still looks like 2023 game (unlike games that a releasing today with x360 grade visuals and performance of Crysis on pentium 3). After all it was made for VR first, to run on stupidly high framerate and resolution while keeping clarity. And there's definitely field to improve it.
True, Source 2 is a marvel of optimization. Judging by how culling works and other trickery stuff. But it was first used on Dota 2.
Did Phillip's sharp inhales before talking give anyone else ASMR chills??? Am I Weird???
no, but LoathsomeLarryGamingYT did
Thanks for putting the results in the intro!
can we get a 10 minute e-core video please, your style of video just gets across all the info and data so perfectly
e cores don’t actually cripple performance. it’s just when e-cores are disabled, cpu packet power can be channeled only to p-cores, therefore higher clocks on p-cores. e-core, even if they don’t use power that much, they still use power. without them, p-cores clock higher, only by 300-400 mhz though. e-cores actually serve perfectly fine in scenarios when single-core performances are greatly improved, but you don’t wanna stack too much traditional cores in a single processor, otherwise improvements are much limited. (ex. ryzen 9 5900x vs 5950x, where 4 extra cores only managed to improve multi-core performance just by %5 margin) e-cores don’t use too much power, do not cripple the other p-cores single-core performance and and enormously improve multi-core performance. another example: i7-11800h (8 core config) vs i7-12700h (6-p 8-e cores) 12th literally improves both single and multi-core performances by %50, while still limited to 45 watts, only one generation apart.
really good video, Philip. Well done!!
3kliksphilip the counterstrike youtuber we need! Good stuff
You mention that Directx 9 was old when CSGO was released, which is true that when CSGO was released DirectX 9 was already 9 years old. What you didn't mention was that directX 11 will be nearly 14 years old by the time CS2 is released.
DirectX 11 was released with Windows Vista in 2009. Directx 12 was released in 2015 with Windows 10. Whether DX12 would actually be better for CS2 is something valve only knows, though I still wonder why it isn't the standard or Vulkan is. Vulkan might be better for lower end hardware than your current high-end hardware though. Still I wonder why DX12 wasn't chosen instead of DX11.
E-core gaming has always been so interesting to me, its neat that these modern intel processors have a bunch of extra cores for background tasks that are roughly equivalent to the oc'd 4790k that I had in my first gaming rig, and we look at them and think "eww bad". Gives me an appreciation for how fast tech progresses.
Intel is king now. AMD who ?
@@3kliksphilip yeah I really wish they would give us better control over which programs use what, kind of like how we can select which gpu a program uses in the graphics preferences menu
Intel i9-13900k and 4090? Philip living his best life. Literally the best rig you can buy.
as long as you have an AC inside your computer to tame the 13900k
Thanks for all the testing. It’s super interesting. :)
Always love seeing these investigations, keep doing what you're doing! Just two quick questions: why do you focus on average fps rather than 1% lows? I generally find the latter more relevant, would love your thoughts on it.
And second is of course "yet another thing to test"! Have you tried using core affinity to force the game to run on specific cores without having to tweak bios settings? Perhaps paired with Process Lasso, which I think also lets you keep other processes away from the cores that are dedicated to CS. I don't think it warrants redoing all these tests, but could be worth giving out as a tip if you find it works the same or better than the bios option, since it's far less hassle and less detrimental to overall system performance. :)
Okay sorry, guess those questions weren't as quick as I imagined... Anyway, thanks again for all you do, you're always testing things that in curious about and can't be bothered to test, or better yet, things I hadn't thought of but that you make me curious of! For me, yours is hands down the best CS content on TH-cam!
High value and high quality. Impressive and loved by the community. 10/10
and then pros with computer strong enough to power a Maglev plays on 1024 x 768 4:3 stretched and low-med settings.
maybe the 4:3 resolution they play gives them more frags on average than the native one. ever think of that?
@@tomasiskooo yes, that's why he said that...
My laptop is fried. Farewell Counter Strike, you have been the absolute best to me o7
Awesome complete review. Thank you. Just feel bad because I will be forced to change my CPU + RAM + MB to take 240 fps on CS2... 1% LOW sucks on both game versions. Ps.: my PC i7-6700k + 32GB 2666MHZ + 2070S.
Saying "20% worse" sounds way scarier than "800fps instead of 1000fps" lol. I remember playing CSS on my first real gaming rig and being blown away that I had 300 fps.
these kinda videos are definition of information overload
You must be new to his videos then, that's only the case if you aren't expecting a detailed analysis
Just got a 13900k my self... got it at 6ghz on 2 cores and rest at 5.7ghz... the fps boost over my 12900k on cs go is insane.. also would recommend you getting the anti bending bracket :) stock clocks i was under 80c on cinebench r23 with a manual vcore of 1.2v... with the oc I get around 89-95c on cinebench r23 with a score of 42k. stock was 40960
Hello Philip, I adore you for the work you put into this video and gathering such useful information.
But what I have been missing the whole video (if I have not literally missed it myself) is the mention of a early development build you are comparing CS:GO with. It's nice to see already acceptable performance on CS2, but usually performance and optimization are two of the last things developers really but effort into in their development stage.
Might have to redo the whole video on CS2 release.
What a delicious data analysis! thank you!
LoathsomeLarryGamingYT loves eating data!!!
An input delay would be the most interesting thing to test, because HL:Alex as a VR game does a lot to reduce it and CS2 probably has that carried over, so with its lower FPS it still might effectively run better.
very interested in seeing the slower pc tests!
I could say that it runs better on lower-mid tier PCs. I've tested it on mine with Ryzen 5 2600 and 1660 Ti with both DX and Vulkan. Vulkan actually did better by 5-10% on average. If I compare it to CS:GO, the fps does drop a bit, but the game runs much smoother. CS:GO can be sometimes choppy even with framerates higher than refresh rate of the monitor. No such thing with CS2, even when FPS drops below the refresh rate, it still feels smooth (maybe it's a nvidia reflex thing)
I'm not a fan of those 1% drops which are like 5 times less than avarage. This suggest that if you run cs2 at 400 fps you will get drops to 80 fps!!. It would be really unplayable, hope that valve manage these drops ASAP.
Very interesting. So far, what I've seen from other tests is that CS2 has more stable framerates, but definitely not true for this test
That's a lot of work. Thank you for sharing
I don't play CSGO, but if i did, i'd be really upset with this update if it's something future players are forced into. A drop in performance sucks, sure. but what _really_ sucks is that it's for the sake of these new visuals, which for me personally, the lack of contrast and more, "pastel" colors, i guess you could say? it hurts my eyes. And i don't even find it aesthetically pleasing.
I didn't play this game before because the audio mixing with the headshot noise was so loud it was physically painful- now the game's apparently become even more inaccessible to people like me. Bit of a shame.
In the task manager you can view how many threads are spawned for each individual process. Now, most of the threads won't actually be active, programs tend to be extremely wasteful with these and most of them will be spawned by external libraries, but when joining a game you can see the thread count go from 70 threads to 78, so it wouldn't be too wild of a guess to say that CS:GO can use at most 8 threads during gameplay.
I love the "Test" graphic at the end
You need to make hard choices about in what order you investigate things. Start from what people would be the most interested in or things that are the most relevant for most people and then consider things you really want to investigate. You can make more videos about the more nice stuff in the future, I love the nice stuff.
Makes sense why turning off HT would increase wattage. The CPU core is using more instructions and cycles to manage/wait/queue/share/swap workloads which hyperthreading would normally do. So to get the same amount of total work done, more overhead/wattage/speed is required when hyperthreading is turned off.
It also allows identical computations to be computed at once. Ex: If you get asked what 2+2 is twice, you don't have to do the math twice ;)
Understanding CPU architectures can be very informative :)
can't wait to test the full release on my i7-3770k
so, philip, I need to know: did you use the adobe AI voice enhance for this videos voiceover? I use it quite a lot myself and have started being able to detect it, and I'm getting some subtle subtle vibes of it here and there in this.
Something important to think about is that having your E cores active will allow common apps to have more breathing room, even if that means they use some P core while cs uses a little more E core - ex: discord, browser, steam itself (ex: overlay), windows. These apps can eat quite a bit, most notably a playing youtube video & discord calls
9:23 You can see that Windows report that the results with the e-cores active are 5ghz while the results with the p cores only have 5.5 GHz. I haven't played around Windows 11 and with Intel's newer CPU but what I'm assuming is that the e-cores is processing stuff and that heats up the chip and it underclock itself and or CS 2 is trying to balance the performance with all the cores so it when it utilizes the e-cores the performance dips due to the fact of trying to manage low powered cores. We can potentially test the e-core theory by setting the affinity of CS 2 to only use p-core and see what difference it makes. Also, it might be fun to see CS 2 running on only E-cores and see how it performs.
It would be interesting to get the devs' opinions on why 0.1% lows are disproportionately lower than the average decrease in performance. Asset streaming maybe?
Philip, this is not CS2 related but do you know anything about the bug in Counter Strike where everyone's FPS in a given server absolutely tanks at random? I've experienced this so often for years, dropping from like 600 to 100 in the middle of a game, with some friends dropping from 200 to about 15. Even people on the enemy team usually type something about "fps," so it seems server-wide. Do you know what causes this? It seems really odd that it's been a problem for so long.
i bet its esea mining crypto on our pc's xd
I always thought it was VAC analyzing everyone´s processes. I never really knew.
its prolly bc server got bugged and ur pc is getting too much information that cpu can handle, idk
thumbnail: overpass
overpass in the video: 12:03 - 12:08
I'm a bad PC gamer: I hardly know what most of this means and have no idea how to utilize any information you've given me. But I was entertained regardless
the difference between vulkan and dx9 or dx11 is that vulkan is a lower level api, which basically means it has more potential for optimization, as well as improving cpu performance. whether or not it actually is faster or not depends on how well the developers optimized their code and how cpu bound the game is. in extremely cpu bound things like emulation or some older games like fallout new vegas, you can see major improvements in performance by using vulkan even without well optimized code, but in something like cs i'd imagine it's much harder to get vulkan running better than dx11, even if it is possible
Interesting that CS2 performs better without both hyper-threading and E-cores enabled. Thanks for the performance review!
I wonder if the difference is the i7 hitting it's thermal/voltage limits and pulling back on the clock's boost. The Task Manager does show 5ghz when the E-cores are enabled, and 5.4Ghz when just the P-Cores are enabled. So we might be seeing the limitations of the 13700k's architecture.
Good frame rate testing, Phillip.
Valve: Yes, we leave the 21 year old DX9, for the 14 year old DX11.
I REALLY wish you tested CSGO with E cores disabled because it might've had higher fps and thrown off all the data.
You should check the perfomance on a low-end laptop with Intel graphics so the audience can see if they can play CS2
Nice video, however lets remember that the game hasnt launched yet, they didnt updated it in a while and it could launch performing a lot better.
I like the birds in the background (blackbird) 11:14 and up.
In the test with P cores only vs P cores w/ HT and E cores, CPU freq is not the same and may be something to keep in mind to those who are watching the video.
True, the P cores alone are clearly able to boost much higher due to lower heat output and power usage which is a clear explanation for the higher 1% low.
Dual core gang, let's all gather and mourn the likely possibility of not even managing 60fps in CS2.. Goodbye my dear frames per second, you will be missed.
At that point you gotta get a new pc lol. It must be older than CSGO itself 💀
if you meant Intel dual-core (like 1st gen i3 or i5), then you can replace it with a used Xeon that have 4 core. It will work if it has same socket number (like for example: LGA 1156). The 4 core is faster than the dual-core in real usage even when Xeon is 2ghz vs 3ghz dual-core.
Cool video man!
Are you looking forward to repeating all these tests when CS2 actually releases to see if they improved the performance over the Beta version?
proceeds from this video will hopefully pay his electricity bill
I wonder if the poor Vulkan performance you noticed is part of the reason why Linux was left out of the CS2 beta.
CSGO doesn't let you use the Windows build with DX9 on Linux, you need to use the native build on opengl or the expiremental vulkan mode. So I'd imagine Valve is ironing out some kinks with Vulkan to make that the default on Linux. Can't really picture them shipping it with opengl in 2023
That being said I've played a good bit of CSGO on Vulkan in recrnt months and it has worked pretty well, I hope the Vulkan performance looks better by release for CS2
Interesting how in your CPU utilisation comparison, CS:GO had your CPU clocking ~400 MHz faster (5.4GHz vs. 5.0GHz). Would be interesting if you can see why this might be, and if you can get it to be the same speed for both (maybe lock to a lower clock speed) and then compare performance again. The same also happens in your E-cores/hyperthreading comparison, and with how CPU dependant CS is, this could be making a big difference.
I gotta admit I thought the thumbnail was memeing really hard with an OMEGALUL down the middle, just me?
Would love to see a framerate and frametime comparison when a smoke grenade is in your view, since that used to cause major fps drops on my old laptop back in the day. Would be interesting to see if the effect is still as taxing in CS2 as it is in CSGO.
When I read about people describing CS2's performance, a much smoother frametime was one of the things that I've seen brought up several times - where despite the framerate technically being lower, the game felt noticeably smoother despite of that because the frametime graph didn't represent a hacksaw.
I think this is definitely something worth investigating further - especially on a more modest configuration that doesn't involve top-spec parts...
"I promised myself I would keep this video simple and I failed" task failed successfully. Thanks for the video.
I suspect CS:GO processes everything on the P cores, while CS2 is able to more efficiently separate tasks on certain cores. So information that the GPU needs may be exclusively processed on the P cores, and all but GPU information may be dedicated to the E cores.
When I was playing the CS2 beta I thought it felt kind of stuttery, so I decided to measure frame times and frame time variance (the variance of the distribution of frame times, basically the frequency of microstutters) and found that, at least on my machine (5800x + 3070), the frame time variance was measurably worse in CS2. You can use CapFrameX to measure this. These were my results just running around Dust2 with bots:
CSGO:
< 2 ms : 97.55%
< 4 ms : 2.42%
< 8 ms : 0.03%
< 12 ms : 0%
> 12 ms : 0%
CS2:
< 2 ms : 53.11%
< 4 ms : 23.2%
< 8 ms : 23.63%
< 12 ms : 0.06%
> 12 ms : 0%
So I don't think I'm imagining it at least, there is a difference in the frametime distribution (for my machine).
You can see the precise thread count for CSGO/CS2 (or any single process) with Process Explorer, which might interest you. If there's more threads, then the game should benefit more from higher core count systems. Even on multicore systems the OS scheduler will move processes around if there's too much load on a single core.
Could it be that CS2's server is more demanding than CS:GO's and causing more of a performance hit, assuming this testing was done on a local server, if so could testing be done with the server running on a different pc, might show a more accurate difference of just the game client?
Seems like this beta only has a DX11 renderer, and uses DXVK to translate it into Vulkan (just like L4D2 and Portal 2 does with -vulkan variable). Some performance loss is expected because it isn't running natively.
I seriously hope they develop/switch to native Vulkan renderer like they did on DotA 2.
DX11 and Vulkan won't make a big difference on Nvidia hardware as their DX11 driver is optimized to a black magic degree. But it'll make lots of difference on AMD hardware.., at least on Windows.
Source 2 has native Vulkan implementation, not DXVK.
I had to modernise my setup for the future smooth 144hz CS2 gameplay, since coupled with RTX 2070 super, Ryzen 5 1600x wasnt enough to provide it during alpha testing. There were a lot of drops, 1% low was something around 90fps at 1290x920 resolution.
Now with the same GPU and Ryzen 5 5600x - the most cost-effective upgrade its something like 2x the fps at the same resolution and all visuals on the highest settings.
I hope valve is going to suck everything from S2 to optimise it for as old equipment as possible, for the sake of all the people that are going to play it despite low performance setup (just like I did, and still do with my shitty internet lol). They live from those people money.
Why Valve just not do some kind of benchmark in game?! So after change of new version, gpu drivers or even cpu/gpu we could test if smth changed about fps.
In CSGO we can only use „timedemo” for that with our 1 min demo with bots but in CS2 it’s not working. Greets, micronn
I'm entirely self taught so I could be way off base but,
My guess on why the e cores lower framerates is due to the p cores having to still wait for the e cores to finish the lower priority operations before being able to process the information for the next frame, and by trying to more efficiently spread the load it's slightly slower for the e cores to complete the work than if the p cores just did it in the middle of everything else.
Basically it's not optimised for the speed of your p cores and just assumes offloading some lower priority work will be faster, but it's not.
The hyper threading thing is interesting, I have no idea how hyperthreading actually works but, I'm guessing hyperthreading leaves a tiny bit of performance for more concurrent operations, so with it disabled the individual cores can put out slightly more iops than with hyperthreading enabled, but less than if cs2 fully utilised 16 threads (likely due to not having enough operations to fill out the threads so some of them end up waiting for other threads to finish what they're doing before they can do their next thing, or perhaps before the engine will hand out the next operation)
Alternatively it could simply be an optimisation problem and there is no tangible benefit to disabling hyperthreading and source 2 just doesn't know how to properly use this specific cpu properly.
question to anyone who would be able to correct me, is there any potential benefit to disabling hyperthreading (stability, a minor single core boost, etc.) or is it purely a compatibility thing?
The thing with Vulkan is that it's so different from DX11 or OpenGL
To put it simple, in DX9/10/11 or OpenGL you are expected to talk to the driver, in Vulkan/DX12 you are expected to write your driver.
Its low level capabilities will give you an edge in performance, only if you invest the time in knowing how a GPU works
It'll be interesting how this changes when we're able to look at the maps that have actually been remade.
I believe Vulkan is slower as it's not fully optimized yet. In Dota 2, which is also Source 2 it can go head to head with DirectX any day.
That may also be the reason why there's no CS2 beta build for Linux yet.
you want all your cores being used actually. the e-cores being used means the p-cores are busy with stuff. the more cores you have the more likely you'll find situations like this where the load is spread across all the available cores and this results in many peaks and valleys on the graphs.
also, the higher the graph the longer the waiting time. cpu usage is determined by how much time it took to complete the work. i can't remember how or where i learned this or if i'm correct at all though, but for digital systems it is vital to measure time as accurately as possible so there might be some truth there.
You give yourself too little credit for how well you chose what to test and what not to test. You anticipated and hit exactly all the questions I would have had around performance in relation to power limits and p/e core allocation.
Could consider making a separate video investigating the same questions with AMD Ryzen, and see whether the scaling goes beyond 8 physical cores. Even better if you do a counterpart to the p/e comparison with the X3D cores - assuming you have one of those lying around.
8 P-Core, No E-Core with HT Off surprised me... I would like to see a how scalable this is at this point, somehow I imagine 16 P-Core would only be a 20% uplift
For implementing Vulkan they should hire ID Software engineers, cause they nailed it with ID Tech 7 (DOOM Eternal engine); Minimum CPU bottleneck with all cores been used + GPU maximum usage !
I played it low settings on my laptop i7-8750H (6 cores; 12 threads) with GTX 1050 (Potato compared to current Gen); Constant 140 FPS in Final Sin arenas without signle freeze or stuttering !!
Or maybe ID makes ID Tech 7 open source and it's late 90s early 20s all over again !!!