Видео

The main problem with Ray Tracing isn't technologies. It's the marketing scheme beside it. Ofc there's scalpers and btc, but the sign of bad news was when nvidia first released the rtx 20 series. They inflated the price for something almost nobody cares about. And the greed with those enhanced shadow never stopped. Unless gpu price can go back to what it was when they're linear with raw performance, then Ray tracing will always be considered a scam.

Year its a feature were everyone screams about like PhysX ... thats why i dont like outliers.The same is Tesselation ... its nice but not for that.

the biggest problem with raytracing is the devs not implementing it in good ways to increase image quality since they just want to half ass it to say the game has it. In order for rt to make a big impact the game has to be designed around it from the ground up making it a requirement to run the game but the problem with that approach is that not everyone has the hardware needed to run it making much harder for the devs to choose that approach if not enough people have the gpu's needed to run the game.

I’m not sure RT in cyberpunk can drop by 2x in terms of cpu performance. You need to check if you are gpu limited, which is most likely. Potentially you might be limited with gpu vram as well cause you mentioned you have 10gb and running the game on ultra textures at 4K, cause Ray Tracing requires some extra VRAM. Thanks for the comment!

1. Resolution can affect cpu performance so I don’t see how this “verification” would verify we’re cpu limited. The reason for this is a lot of settings that might be related to resolution (for example in a lot of games LOD levels on the same option are different if you just change resolution). So the difference between resolutions will tell nothing about RT impact on CPU 2. How does 12900K being a better cpu goes against the point that we are cpu limited? The lower cpu you have - the more chances you have to be limited with it 3. How does “clear GPU scaling” is relevant in a cpu scaling test? Of course there is a GPU scaling and 7800XT loses performance from rt, however because we are cpu limited in both cases it just doesn’t matter 4. Why if we are gpu limited despite load sensor showing otherwise, there was no performance difference between applying a single RT effect and multiple RT effects? If we are GPU limited there will always be a significant difference 5. This was NOT a native 1080p resolution in all of the games except crysis remastered. Of course I’ve done a lot of research to find games and scenarios where I can be CPU limited to showcase the phenomenon I’m talking about, in some cases render resolution goes all the way to 360p (Cyberpunk). Please check the reconstruction technique used and what preset it uses.

​@@HardwareLab Hi, thanks for the reply! Love the channel btw, you do testing and experimentation that no one else does! 1. My theory was that RT takes a certain amount of CPU resources regardless of resolution, so if we modify the resolution and compare pure raster versus RT, the performance uplift will be less in the RT case and we can quantify that. This would be indicative of the RT CPU bind. However, I did not know about the LOD levels, and I am unfamiliar with other settings that get affected so allow me to scrap my own idea. 2. This was a point I made playing devil's advocate against myself. I mentioned that the 12900k is better precisely because I know that this is not apples to apples testing with Tom's Hardware's setup. Since he used a better CPU, it could be argued that there was CPU headroom for better GPUs to show higher performance numbers in RT. My point was that even though the 12900K is better, it is not better enough than a "5500x3D" to allow GPU scaling at 1080p RT High settings while the latter is completely CPU limited with no possibility of GPU scaling. 3, 4. So, if we are not GPU limited, then a 7900XT or a 4090 should perform equivalently with a 5500x3D at same settings right? The performance difference when applying multiple RT effects is negligible, as you showed in the video, perhaps because a lot of the calculations for RT is already accounted for by at least one of the other settings, to the extent that the CPU & GPU are capable of. I, perhaps mistakenly, thought GPU limited to mean that if you switch out the GPU with a better one, then you get better performance. I take it you mean that we are only in a GPU limited case if the load sensor is above 98%. However, this definition is somewhat misleading for the purposes of CPU limit testing with FPS comparisons because it would make it appear that RT performance loss is solely due to the CPU being too weak. What if the 7800XT were paired with an extremely good CPU so that the load sensor read 98%? What would be the framerate then? This theoretical max RT framerate of the 7800XT is what the CPU bound RT ON test should be compared to, not the plain raster. The plain raster is CPU bound, but that does not mean that enabling RT is only increasing CPU load which is why we cannot make any conclusions of how severe the RT CPU bind is even in a CPU bound scenario without a GPU bound (or close to GPU bound) equivalent test. RT CPU bind should be quantified as a fraction of performance lost compared to an equivalent scenario where the 7800XT reaches its full RT potential at the specified setting. Of course, at the settings you chose, this would be very difficult. 5. I see, you did not use native 1080p for Spiderman Miles Morales. My bad, did not realize that. You used FSR Performance settings, which should suggest we are in a CPU bind as you claim. However, I do think my argument in my reply above with regards to RT CPU load testing methodology and the conclusions we draw from the data still stands.

@avogadroboltzmann7807 2. I still don’t get the point. This is not the way it works. In most games you can find a gpu limited scenario and a cpu limited scenario on any build even with Ray Tracing regardless of what your hardware is and these tests would be valid. 3. No, there will be a difference but probably 4090 would perform slightly worse due to driver overhead and bigger chip. The difference in RT multiple effects can only be negligible on CPU, come on bro, this is not how it works when you are gpu limited, you can watch tests of any gpu in cyberpunk you will find out that adding rt effects significantly reduces gpu performance 4. Yes, when we are GPU limited is when GPU is ~~98% and this is the only case when you can see a major difference between gpus aside from driver overhead. So if, let’s say, 7800XT was switched in any of these tests to 7900XTX the result would be the same. That is a cpu limitation and a lack of gpu limitation. I’m not entirely sure what you mean, but there is no such thing as “being gpu limited partially” or whatever. You’re either limited by the amount of draw calls your cpu can prepare (for whatever reason including RAM and stuff) or you’re limited with the amount of frames your GPU can render. There is absolutely nothing else in your computer. 5. “What if 7800XT was paired with a really good CPU so that the load is 98%, what would be the FPS? The FPS would be higher and we would become GPU limited unless we lower the settings to be cpu limited again. Why this “theoretical max rt framerate is what cpu bound must be compared with”? What does it even mean? Like we are trying to calculate what amount of performance you lose because you don’t have best CPU? Isn’t it absolutely pointless because on different settings and for different cpus this number would be different so it represents nothing? 6. In general I am honestly shocked and I feel kind if hopeless after seeing how people react to this video. For some reason a lot of people don’t understand a very simple concept of bottleneck either from cpu or gpu side and they generate tons of crazy ideas that would end up in rubbish data representing nothing. Guys, it is simple. You are cpu limited = you are measuring cpu bound performance. You want to measure impact of rt on cpu performance -> you compare cpu performance of RT OFF and ON, the difference is the impact. It doesn’t matter that these settings are also impacting GPU because if we are cpu limited in both cases it doesn’t matter, we are looking at the cpu performance and changes in cpu performance. I really was expecting some debates about wether it is all worth it or now, I think some of my results are quite shocking. I wanted people to share ideas why this might happen that much differently in different games, how it affects people with lower cpus etc. Instead of that I’m trying to explain to everyone what is a cpu test because nobody gets it. Maybe go watch our video about bottleneck? Or actually maybe I do need to make another one super short and simple and reply with the link on any comments like this. Anyway, thanks for the comment

@@HardwareLab Hi thanks for the second reply. I apologize if my confusion/ignorance has irritated you in some way. Just trying to understand the data and its implications the best I can. First, I totally understand that the performance drop is entirely due to the CPU since we are in a CPU bottleneck. I was only questioning whether the comparison to pure raster was useful, because I doubt the frame rate would be as high as the raster with RT ON even if we weren't CPU limited. So yes, perf drops by almost half due to CPU in a case like Spiderman Miles Morales. I was only thinking that from the perspective of a buyer, is it worth more to know the total perf dropped due to CPU bottleneck or the perf drop off as compared to what the GPU is capable of maximally at the same settings. If the Radeon card is only capable of 95-100 FPS in Spiderman with RT on, then pairing it with a "5500x3D" and it losing some performance doesn't seem as bad (though it is still a significant drop, I agree). Some might watch this and think that since we in a CPU bind and the FPS drop is so large in some instances that we are severely underutilizing the 7800XT in the RT workload when in reality we are closer to full utilization than that, as suggested by the increased utilization numbers in Afterburner (from low 70s to close to 80%). It is not like we are underutilizing the card by almost half its total possible performance, which is what my impression was as a noob just looking at the numbers at first glance. Remember, not everyone is pouring through every number and setting like you are, and everyone has differing levels of computer hardware knowledge and background. I am not a tester, and I have never owned more than one card before (a non RT card at that), I just like researching this because it is fun. But I see your point and I agree that what you showed is a CPU bottleneck performance drop. So as a pure CPU test, I can agree it is valid. EDIT: Also, (in response to 5) I don't understand why you say that testing against a better or best CPU is pointless. If you can get back to GPU bind scenario with the same settings and only improve the CPU, there is a limit to how much better performance we get due to CPU. The 7800XT will be GPU limited after all. The FPS will have to be limited by what the GPU is capable of, and so we can benchmark CPU perf as a fraction of how much of that theoretical limit it can achieve. So, I see that is a valid comparison, that is when trying to answer the question: Is this CPU strong enough to be paired with a 7800XT? And I get it, the point of the video is not to ask whether the 5500x3D takes full advantage of the 7800XT for RT, it was simply something that I think is worth asking, especially for buyers.

@@avogadroboltzmann7807 1. Sorry if I sounded rude, didn’t mean it 2. This is true, the framerateif we were gpu limited would be different, but it doesn’t mean it affects the cpu limited performance in any way in both cases, because this is how game rendering works. CPU prepares X amount of frames and then GPU renders Y amount of frames. You would see the amount of frames that is lower of these two. So if X < Y then you would see X fps no matter how high or low Y is. That allows us to measure the impact if any settings (including ray tracing) on the cpu and on the gpu independently. 3. The “perf drop off” compared to full GPU utilization can be useful for a buyer if we are talking about some particular components and settings buyer is interested in, but not if we are looking at the Ray Tracing impact on cpu performance as a phenomenon, because the amount of that drop will fully depend on chosen settings and hardware and might be extremely small or extremely large depending on a lot of factors. However the performance drop when enabling RT, that I have measured in this video will always be there no matter the settings, the particular CPU, resolution etc. Sure int might vary from one CPU to another but it will always be there. 4. There are just no CPUs that would be “strong enough” for some particular GPU etc., this is not how it works because it depends on a lot of factors and I really recommend watching our video about bottleneck to understand it better. But generally, it just depends on your game and settings. I can make you a bunch of tests proving that Ryzen 3600 is fully enough for 7800XT and then I can do the opposite and prove that even 5700X3D is not enough. That will only require me change settings in games and chose slightly different games list.

Depends on the game and frame rate you want to achieve

@@antondovydaitis2261 we have a small comment about every game during the test run, however just voicing the numbers in the end of each test… it sounds a bit weird for those who watch the thing, cause you have graph, why should voice repeat it? Anyway, thanks for watching subscribing!

Why do you even care about cpu utilization with locked game? This is the test of the cpu performance with or without Ray Tracing, what guess work? What are you even talking about?

​@@HardwareLab Warning: long ass comment. So you are looking for the impact that RT has on the CPU specifically.. That is a hard thing to isolate. Take the first game you tested, Cilistol Protocol. With pure raster you weren't CPU limited but you were pretty much GPU limited as the GPU was reading 94+% utilization. When you enabled RT the GPU utilization went down and the CPU's went down as well. So that tells me it isn't the CPU you were finding that had the bottleneck but it was how AMD does their RT or how the game interfaces with AMD's hardware in RT mode... uhh something to do with that anyway. This is why I say when you want to figure out something like RT overhead when enabled you should limit the FPS to something that the GPU can do easily in both raster and RT so you can compare the CPU utilization and it is an isolated metric. However the test in CP2077 you can see the CPU utilization being around 5-10% higher when RT on and the GPU is around 20%+ more utilized. however neither the CPU or GPU were at full utilization (which is good for this test) even if the GPU did spike up to 90% once in the video. This is the kind of game that I think would show what you are trying to say with this video, where as Cilistol Protocol is not. In spider man you can see the CPU is up 15 or so percent and the GPU is up a bit as well with RT enabled which is to be expected, it acts much like CP does. Forespoken shows the CPU utilization going down slightly with RT enabled and this is the game that shows the biggest issue for me. The CPU goes down in utilization slightly but the GPU goes up a a bit. However the frame rate is much lower on the RT enabled run so what are you testing here? That is a pure GPU stress test as you aren't isolating the CPU performance at all. It is 20-30 frames lower on the RT run so even if the CPU is getting used for RT you can't tell as the GPU is just too swamped. GOtG is the same as Forespoken. You aren't able to see the difference in CPU demand from RT on and off as the GPU is hitting 90+% utilization. Anyway I'll comment again on this if you want to talk about it more but I'll end with this. CPU utilization goes up the higher frame rate you are at so if your raster game goes up to 300 FPS and when you turn RT on it drops to 150 FPS you might not even be able to tell that the CPU is being more active when RT is on as the pure raster test will be hammering the CPU as well.

@@Azureskies01 You're thinking exactly along the same lines I was. Please see my comment in the other thread on this video. :)

@@WSS_the_OG Well even if he was to lock the frame rate he would still need to bust out excel to figure out the % change from having RT on and off. Then you would also want to make sure graphics presets didn't fuck with the numbers as well as resolution. It would be A LOT of testing to figure out exactly what the CPU impact is for RT being on as compared to it being off. Something for GN to do, not sure many people would want to be that uhh... in depth.

How does lock fps makes it apples to apples? I mean yes if you’re gpu limited in both cases you will only see the gpu performance, if that what you mean, but the whole purpose of this test was to check how RT affects cpu performance, cpu and gpu performance are two different things

@@HardwareLab I believed your test to be about how RT affects CPU load. If so, would it not be prudent to lock to an FPS that RT and no RT could achieve on the same GPU, and then compare CPU load under both conditions? Unless, as I said, I am missing something. It just seems to me that if we want to test effect of RT on CPU load, we would want to do that under identical conditions, i.e., frame-rate. Because RT creates a very large GPU bottleneck, frame rates decrease sharply. However, a lower frame rate also decreases the amount of workload on the CPU. Because the frame rates differ, how can we compare CPU load when non RT has a higher frame rate, and RT has a lower frame rate? Hence my suggestion of locking frame to isolate that variable, and have an equal comparison. In other words, we might determine if one frame with RT requires more CPU power than one frame without RT. I hope that made sense.

@@WSS_the_OG we are not comparing the load percentage cause this is quite irrelevant to a real performance. We are comparing the FPS that CPU can achieve with and without ray tracing because percentages of load might not tell you about the actual performance you are going to get from the CPU. I just don’t get what exactly is unclear here. You want to know how RT hits cpu performance - you check the game with and without RT, being CPU limited in both case and you compare the FPS values, it’s simple, straightforward and self-explanatory, why would you need to lock fps and compare the load stats, what this numbers are going to tell you? The FPS, on the other hand gives you a performance and you can figure out the difference between them -> the hit on cpu performance

@@WSS_the_OG It is clear he either doesn't know what he is asking (from the video title) or he is just arguing in bad faith. in one of my other comments he even said that if you are system RAM/VRAM bandwidth limited then that is a CPU/GPU core issue and not a RAM speed issue. Soooo I'm not going to try and explain how computers work to him any longer but you want to try more power to you.

@@HardwareLab How are you calculating the impact of Ray Tracing on CPU performance from the data you gathered? Is the metric you are using FPS? If so, there are so many variables that lead the FPS metric. Obviously turning on RT has a large performance impact (to FPS), but how are then determining how the CPU is affected specifically from the resultant FPS? How are you reverse engineering that with so many variables at play?

It’s GPU utilization…. If it’s under ~~ 98% you’re CPU limited. Also a good proof of that is the test where we disable some RT effects and fps doesn’t change. It happens because BVH doesn’t change, but the gpu load sensor changes since there’s more ray tracing involved, so this is also a good proof we’re cpu limited

@@HardwareLab I see thanks.

@@HardwareLab This is incorrect, you can be GPU cache limited, GPU memory speed limited, CPU cache limited, system memory speed limit, system memory timing limited, and a host of other things. Just because the GPU core isn't hitting (or even if it is) 99% utilization on MSI afterburner doesn't automatically mean you are in a situation where the CPU the limiting factor. An example of this (GPU VRAM speed limited) I can start up RE4 remake on my 7900XT, put to to 4k with all high settings blah blah and run default speeds on VRAM (2500mhz). It will run at 99% GPU utilization but if I go into AMD's drivers and increase the VRAM speed to 2700mhz now I'm getting 5-10fps more and my GPU is still at 99% utilization. In essence I just found that in that particular game my GPU is memory frequency limited even though the core said it was at full utilization.

@@Azureskies01When you are gpu vram limited you’re still gpu limited. When you are ram / cpu cache etc. limited (not gpu and not hard drive) you are cpu limited. That’s exactly what people mean when they say cpu limited. I don’t see how this is relevant to that discussion

@@HardwareLab Because when you are VRAM or system RAM (bandwidth) limited you are NOT CPU or GPU core limited. You are wrong, That is how it is relevant. Granted there is only so much a CPU or GPU will benefit from faster RAM speed but as we can see with CPUs and system RAM that limit hasn't generally been reached on DDR5 (at 1080p) until like 8000 MT/s which not many IMC (integrated memory controllers) can even run. As far as GPU cores go there is also a limit to how fast VRAM can get to benefit the core in games... and it is also game dependent so finding those specifics out would be a headache as it would require many tests per game on the same GPU architecture. GN has had lots of videos going over system RAM speed and performance gains as well as if you look at the 7900 GRE it is a perfect example of a GPU core being held back by limited memory bandwidth at stock. Also note that Ryzen generally doesn't see as much of a performance increase in RAM speed (as compared to intel) because Ryzen has an Infinity Fabric / I/O bottleneck... Which gets solved by in large part by the X3D chips not needing to access system RAM as much leading to massive gains in some games without the need of clocking the CPU core any faster. What I'm trying to say with all this is... testing computer components is complicated and time consuming.

It does depend on the game, and the GPU however (and of course, your monitor's resolution). If it's a lower end RTX (60 series or lower), then yes, RT drops frame rates to unacceptable levels in modern titles. I'm guessing you aren't a fan of DLSS or Frame Gen either? I don't love those features myself, but I do use them in certain titles. I run a 4090 at 1440p with a 170 Hz monitor, and even that card can be brought to its knees in some newer RT titles. You can offset this with DLSS and FG, but this comes at the cost of image quality. Pick your poison; high FPS or high image quality. :)

This is the real question.

All zen 3 is 4.0 except APU 5X00G that still using 3.0

@@Lu5ck no, 3d cpus are locked

Although generally you’re right, this is not always the case. For example, LOD levels may be related to render resolution (as happens in a lot of games). Moreover, we were not using native 1440p, in a most of the games we used a lower render resolution achieved through reconstruction techniques, so as long as we’re not gpu limited the comparison is fair (cyberpunk is our fault, i agee)

@@HardwareLab well,i am maybe nitpick...but it would be bit different results with less stressed gpu. Agreed on real life scenarios...and good idea to test rumored cpu btw! I'm not sure myself is it worth to buy any x3d cpu...only thing i see as good improvement is 1% lows in games...

CPUs don’t relate to GPUs in that way, here’s more detailed video about it ruclips.net/video/3vWcTq60PdA/видео.htmlsi=U6ajOZudvv54QXkB

rx 6600 will be fine at 60 fps 1080pon most games( with fsr or lossless scaling), 144 fps on valorant/csgo. normaly r5 3600 will be good but if you want to play switch game smoothly then choose r5 5600.

If you didn't know, 5600X3Ds are basically rebranded 5800X3Ds that doesn't pass the requirements to be 5800X3D (can't reach certain clock speed, unstable or dead cores, etc). Hence the stock is limited and somehow they chose Microcenter to sell those.

@@evilzinabyssranger5695 somebody skipped the introduction…

No this is a 5700X3D but 2 cores and 2 threads disabled and downclocked from 4.1 GHz to 4 GHz.

Depends on how many cores the game use, majority of games at least the ones i play stress core 0 and 1 while making little use of the others cores. You can still get bottlenecked at 60% cpu usage if the workload isnt equal within all the cores.

Most games overall will hammer one or two cores/threads so even if you're not seeing full CPU utilization you may be bottlenecked by the CPU, a decent enough measure to know you're CPU bottlenecked in a game is not getting 95/100% GPU utilization. Still if the performance is good enough for you there's no need to worry, there's always a bottleneck somewhere anyways. But if you want to upgrade a R7 5700X3D will surely sufice and be at least 25% faster at the low end (Games that don't benefit from the massive L3 Cache) and sometimes almost double the performance in other games, overall a great CPU and pairs very well with Mid High GPUs (Even High End if you play at 4K).

lmao damn and i thought 5600x with my 7900 gre was a slight bottleneck, dont you notice any lower performance? or have any type of stuttering etc with your combo?

If your gpu is not 100% utilized, ure cpu limited. Every game is different, you can be cpu bottlenecked in 1 game and not bottlenecked in another. The job of a cpu is to fully utilize the gpu. Cpu (master) is responsible to whip the gpu (slave) to work. The master need to be strong enough to keep whipping the slave to make sure it works 100%.

​@@sc9433 fucking hell... the analogy...

Yeah AM4 is a great deal for ultrabudget builds right now. Pair it with an rx 6600 for around $200 new and you have a great 1080p machine for around $500-600.

@@maxpower4859 1. Some settings affect cpu performance, so it’s better to keep them high. Resolution might affect cpu performance if, let’s say, in the game LOD levels are related to resolution. Also, it is not 1440p in most of the tested games, it is 1440p with some sort of reconstruction, so render resolution is lower 2. Oh I’m sorry I didn’t know it’s an old game that can’t utilize modern CPUs… If only it was the whole point of adding this game? To represent the old games behavior to show the differences between the cpus and test differences scenarios so whoever watches that gets a bigger picture?

@@atifahnafhossain90 It shouldn’t be pci 3, there’s no reason for it

No..

its the same as long as he isnt gpu limited

@@Duuh_Eazy he is gpu limited in almost every test tho and still gets different results. I doubt he actually got it, when noone got it yet

@@jlmpc8733 he literally isn't gpu limited in any of the scenarios he showed, he obviously doesn't have the processor, he just simulated it taking the 5700x3d and disabling cores and locking clocks

Its a fake video. 5500x3d hasn't even officially released 😂

@@daotpeke he disabled the cores of a 5700x3d, doesnt take a genius

The recording is there to demonstrate the system behavior, the numbers are there on the graphs in the end

@@HardwareLab It's difficult to accurately evaluate system behavior in-the-moment while watching the gameplay footage if the FPS is constantly changing every second. Seeing average FPS and average 1% lows, 0.1% lows, etc. during the run makes viewing the run actually useful. Otherwise it's best for us to just skip it entirely to the end result.

@@ThePlantoparadise if you think two numbers can represent a system behavior then skip, but they don’t, the frametime graph and fps in relation to whatever is on the screen does

no if you do your apartment will catch fire

yes