Speaker A
Hi YouTube. You might be asking yourself, who is this Italian guy? Why is he sitting on the ground? What's going on with these cables? And why is there a cat? Anyway, welcome to my video about how I spent many hours sitting on
A competitive gamer shares a detailed, tiered guide on optimizing PC performance for gaming, focusing on physical setup, overclocking, and software tweaks.
Key Takeaways
- Proper physical setup, especially cooling and power delivery, is critical for maximizing gaming performance.
- Overclocking hardware yields significant FPS improvements after ensuring physical optimization.
- Software tweaks should be benchmarked carefully; many popular tips are placebo or harmful.
- DIY optimization is accessible and preferable to costly, questionable services.
- Thermal management directly impacts CPU and GPU boost behavior and overall FPS.
What the video covers
- The video is created by a competitive gamer and world record holder focusing on maximizing FPS and minimizing latency.
- It is structured in tiers: physical setup, hardware overclocking, and software optimization.
- Physical setup covers correct RAM slot usage, GPU power delivery, CPU and GPU cooling, thermal paste application, and case airflow management.
- The creator emphasizes the importance of thermal management to prevent throttling and improve boost clocks.
- Overclocking and hardware tweaks provide the majority of performance gains after physical setup.
- Software optimization includes benchmarked tweaks and a discussion on placebo effects of popular but ineffective tweaks.
- The video discourages paying for expensive optimization services, promoting a do-it-yourself approach.
- Specific hardware examples include a 7800 X3D CPU, RTX 4090 GPU, and 64GB of Hynix ADI RAM rated at 6000 CL30.
- The creator shares personal testing results, such as improved performance by mounting the AIO radiator in front and repasting the GPU with PTM 7950.
- The video also covers disabling unused integrated graphics and managing Windows services to reduce resource competition.
Chapters
- 00:00Introduction and Video Purpose
- 03:26Physical Setup: RAM and GPU Power
- 06:52Case Airflow and Cooling Optimization
- 10:05RAM Tuning and Overclocking Basics
- 13:15Advanced CPU Overclocking and Curve Optimizer
- 17:17Disabling Integrated Graphics and DDR5 Nitro Mode
- 23:37GPU Repasting and Stable Overclocking
- 26:17Windows Services and Software Optimization
- 31:00Latency, Frame Timing, and Final Performance Tips
Full Transcript — Download SRT & Markdown
Speaker A
the ground trying to make my computer run as flawlessly as I possibly could because I really care about performance and about performing really well when I play my video games. I'm a competitive gamer, expert player, and world record
Speaker A
holder named Trainers, and I realized that it wasn't quite working the way I wanted it to work. I've recorded this video before because really the focus was about how I set up two computers to work together so that I could have 240
Speaker A
fps VODs with absolutely no extra latency for my gaming computer. And a lot of that was this mess of cables that fix my audio problems. And I realized most people won't care about that. Most people that are seeing what I'm posting
Speaker A
care about the FPS. So, I'm re-recording the video with just a focus on everything I've learned about optimizing actual FPS and performance and latency, and it will all be in this video. Now, I want to give you a little bit of a
Speaker A
disclaimer. I'm never going to be selling you a course or personalized optimization stuff. I don't like that.
Speaker A
So, a lot of people reaching out asking me, they want to pay me to tweak their computer and stuff. I don't like that. I really think that a lot of you guys are paying people that are honestly not very
Speaker A
competent and they're selling you snake oil. I'm sorry to break that to you. And I think this is kind of like easy enough that you should be able to learn it yourself. Do it yourself or pay someone, but a lot less than I think most of
Speaker A
these people are asking you for doing magical voodoo stuff. So, I'm going to try to keep this video simple. It's going to be tiers. The tier zero is making sure that your actual physical setup is correct because that's super
Speaker A
important especially for thermals. Then the next tier will be overclocking and actual hardware tweaking which is where 90% of your gains will be after the tier zero. Tier two is going to be software.
Speaker A
Now, part of the software will be something that I've personally benchmarked and then at the very end I will give you the placebo part which is where I will mention some of these things that maybe you will find on
Speaker A
Twitter from these people that are doing the LinkedIn style like clickbait posts. I will tell you which ones definitely don't work, which ones might work, and therefore I might be using myself and we'll try testing later on, and which
Speaker A
ones just are not [music] something that you should do because they're actively bad. So, let's get started with the physical. So, tier one is the first step in your checklist. You should make sure that your physical components are okay. Obviously, you
Speaker A
can't just upgrade your computer without spending money, but there's still very common mistakes that are made that can be fixed by physically going inside your case and changing things. For reference, my CPU is a 7800 X3D, my GPU is a 1490,
Speaker A
and I've got two sticks of 32 GB Hynix ADI RAM rated at 6,000 CL30 in their EXPO. So, here's a checklist that you should go through to make sure that before we move any further to overclocking or tweaks, your hardware is
Speaker A
performing as expected. Number one, make sure that your RAM is in the correct slots. This is a very common mistake.
Speaker A
You need to make sure that if you're using two sticks to properly use dual channel, you need to use the correct slots. This depends on your motherboard.
Speaker A
So, look it up in your motherboard manual. For me, it's number two and number four. Another thing is make sure your GPU is being fed power properly.
Speaker A
Especially if you have a GPU that has a 12-pin connector and you're using one of the Nvidia adapters that give you 12-pin to 48-pin. If you only use three, it's going to work, but it's going to be
Speaker A
power limited. So, keep that in mind. Number three, make sure that your CPU is cooled properly. Any modern CPU determines its clock speed as a function of load and temperature and voltage. But temperature is really important because it's not just about achieving the
Speaker A
thermal throttle or, you know, the point at which your CPU is going to shut off.
Speaker A
That's like extreme overheating. But in general, if you stay higher, both modern GPUs and CPUs will take that into account when they decide how to boost.
Speaker A
So that means that even if you only average maybe 15 to 20 degrees Celsius higher when you're gaming, even if you're not getting officially thermally throttled, you will still not boost as high. So, make sure that you're cooling everything properly.
Speaker A
This also means replacing your thermal paste or upgrading it. For example, I moved to using a cryo sheet, which is a sheet of graphene, which is a conductive material. So, you have to be careful if you do something like this. That goes
Speaker A
onto the CPU and acts as the thermal connection between your heat sink and the CPU. Based on my testing, it performs better than any of the thermal compounds that I have access to right now. Another thing is if you're using an
Speaker A
AIO, look up the best ways to mount it. Both in ways that won't kill the pump, which means that the pump can't be the highest point in the loop, which is pretty well known. But everyone will tell you to mount it on the top of the case
Speaker A
in exhaust because that's safe for the pump. But I had very noticeable improvements in temperature for the CPU by mounting the AIO in the front and sandwiching it between two sets of fans.
Speaker A
I'll talk about it later when I touch on my case. Number three, same thing for the GPU. Make sure that it's not being thermally throttled. And GPUs are even more sensitive. Nvidia has an algorithm and it's pretty clear to observe that as
Speaker A
temperature goes up, your maximum boost goes down proportionally. So, the lower you can keep your GPU in terms of temperature while you're gaming, the more FPS you will get. So clearly you can do that much with GPUs compared to a
Speaker A
CPU, but you can repaste it. I did it and I put PTM 7950, which is a phase shifting compound that's super popular now because it's cheap and it works great. And that had a massive effect on my thermals. So repasting a GPU is
Speaker A
definitely more of a technical thing than repasting a CPU. So be careful if you decide to do it. Watch some videos before you attempt it. And of course, there's some risk that you're going to mess it up. So be careful. But I have to
Speaker A
mention it because the gains are meaningful. Now number five, and this is often ignored, but the airflow in your case will help both your GPU and CPU and the other components like your RAM, your VRM stay in the right thermal range to
Speaker A
perform. And you kind of need to experiment and figure out for your case what works best. I tested a lot of configurations. And for me, the key components were the AIO radiator mounted in the front, as I mentioned earlier,
Speaker A
with cold air from outside going through three sets of case fans being pushed to the radiator and then into the case that gives plenty of cooling to the CPU and also a lot of positive pressure. So, the GPU has almost ambient temperature air
Speaker A
to cool itself. Then the exhaust for me that worked great were at the top and at the back. And ultimately, I tested with and without the glass panel on, and the GPU temperature is way lower when I don't use my glass panel. So, for me, no
Speaker A
glass panel is better. But, of course, you're going to have more dust if you do that. Now, you should really take the time to test this out for yourself in your case, in your CPU, your GPU, because it's pointless for you to be
Speaker A
overclocking or tweaking things if you're leaving most of the gains in FPS on the table by not optimizing your thermals. And I'm talking serious gains.
Speaker A
Just changing this improved my overall performance by two-digit percentage points. But now assuming that you fixed your physical. Let's move on to overclocking and optimizing your hardware. Tier two. Okay guys, tier two hardware optimization. We st
Speaker A
bias. We're going to be optimizing three pieces of hardware, the most important ones in this case. CPU, RAM, and GPU.
Speaker A
We're not really going to be doing the GPU here, but we'll be focusing on CPU and RAM because the GPU is usually done in the operating system. Now, before we start, please update your BIOS. Even recently, I was helping a friend
Speaker A
optimize his computer. I was just trying to see if I could achieve the same results that I achieve on my machine on his, which we did. But we realized his BIOS was really old. And we tried first specifically as an experiment not to
Speaker A
update it to see if it would change something. And first of all, some of the good options that I need to tune wener even there because they were added later on in the new AGSA versions of the BIOS
Speaker A
is also AMD. But even just the timings for the RAM, for example, weren't as stable on older versions of the BIOS.
Speaker A
So, please update your BIOS. I know it almost sounds like your parents telling you to make sure that you wear a sweater when it's cold outside, but it's not.
Speaker A
It's not like, "Mom, I'm not going to get sick." No, you will get more instability if you don't update your BIOS based on my testing. Unless you you have a faulty BIOS version that's the latest, but I doubt that's the case. So,
Speaker A
please update your BIOS before you do any of this because then you might realize that oh, for example, maybe even instabilities that you were getting with your base system are not there anymore.
Speaker A
Now, usually when you open your BIOS, depending on your motherboard, you will see something like this, which is a simplified version of your actual BIOS.
Speaker A
I suggest to go to the advanced mode of your BIOS, so that you can get more settings. This is mine, for example, on the ASUS machine. In my case, I have AMD and my RAMs have expo profiles saved
Speaker A
inside them. Now, what is Expo? It's basically a preset profile that tunes voltages, frequency, and timings automatically uh for values that are known to work well with that given stick. If you have Intel, it's called XMPP. You can use sticks that only have
Speaker A
XMPP profiles on AMD and vice versa. But that means that you probably won't be able to use any of the automatic tuning, but you will have to do the timings yourself for which I highly recommend starting from a video from build zoid
Speaker A
called build zoid easy timings for die. Heinix is a type of stick that you can have which is probably what you have if you're watching this video. If not, you will have to look up good timings yourself or or try yourself. I will say
Speaker A
this, I know most people just go like enable XP and forget about it, but I got really big gains from tuning my RAM.
Speaker A
We're talking like 10% more FPS compared to Expo, [music] at least depending on the scenario. Um, especially when you're CPU bound like most competitive games are. So, I know it's a pain in the ass.
Speaker A
I know most people say it doesn't matter. And maybe for most people, it really doesn't because you don't care about the 10% extra FPS, but I am an extremely competitive gamer. I wanted the edge. So, I went through the effort.
Speaker A
This is not going to be a guide on how to tune your RAM. I'm also not an expert. I highly recommend you watch someone like Bill Zoid or a thousand people that are more competent than me when it comes to actually tuning your
Speaker A
RAM. I did it on my own. I'll show you the numbers that I managed to get, but I highly recommend that you go through the effort of doing it. It's also really hard to pay someone to do it for
Speaker A
yourself because unless you have a setup like this where you can stream your BIOS, it's just very inconvenient to have someone doing it for you because it's a really long back and forth process. So I recommend you you learn
Speaker A
how to do it. Um but start from your expo and make sure it works and then move to some someone else's timings like build zoid and then move from there.
Speaker A
Memory frequency part of your of your RAM like it's the big number that determines the mega transfers per second. FK frequency is something AMD related that means infinity fabric clock. I'm not going to explain all the details here because I already am
Speaker A
re-recording this for the second time because I was going too long on the first one. But TLDDR this number depends in a way with the one you want depends in a way from your memory frequency because there is such a value that it is
Speaker A
synchronized with it depending on your memory controller on your CPU and uh another value called UCLK D1 mode which we'll talk about in a little bit but basically there's a value of this that minimizes latency and then as you
Speaker A
increase you're going to increase your latency but also increase your bandwidth and to overcome the fact that they're not synchronized anymore. You need to get to a certain value that is now better for both latency and bandwidth.
Speaker A
So just increasing this might not give you the best results unless you increase it by enough to overcome the loss in latency by not being synchronized anymore. Now I'm not talking input lag.
Speaker A
I'm talking memory latency which is a value measuring nanconds. Remember there's a million nconds in a millisecond. Okay? So this is not like oh you lost three nonds. No human will be able to perceive three nanoseconds.
Speaker A
Okay, we're probably able to determine stuff in the milliseconds. I'm now pretty confident that we are, but not nanconds. I'm sorry, guys. So, when I say latency, that's a value that implies a lot of other things. And lower RAM
Speaker A
latency will improve your overall performance, but we're not talking milliseconds here. But anyway, in my specific case, I am lucky because I can go all the way up to 12 to 2200 at 6,000 MHz, which is really good. I actually
Speaker A
get a little higher than 2200 in real in the real world when I run my PC because my BCLK is slightly higher than 100.0.
Speaker A
But please, this value BCLK, don't touch it unless you know what you're doing because you can absolutely corrupt all your data in your NVME, which I did, and I had to reinstall my Windows, which is a pain in the ass. And you can also not
Speaker A
manage to boot because your GPU, which runs on the same PCIe bus that shares this base clock, might not work. You might have USB problems. Just be very careful if you if you tweak your BCLK.
Speaker A
Now, in my BIOS, you can see we have DAM timing control, which is where you actually go tweak your RAM timings. It's a very important part of the of the tuning. The only thing I would recommend is if you get really slow boot times,
Speaker A
you have to turn on memory context restore, which will automatically turn on power down enabled. In a lot of BIOSes, you need them both to be enabled or you will get black screens if you try enabled and disabled. Some people will
Speaker A
tell you that you don't want this because it will make your memory less stable because it's not going to be training your RAM every time it boots, but it's going to reuse known uh values and maybe there's like a slightly
Speaker A
different voltage from your power on your wall and it needs to retrain. I personally also this power down enabled used to give uh maybe still does slightly gives a little higher latency reading in ADA 64 which is a software
Speaker A
used to test latency and therefore people were like don't turn power down enable on because you get much lower latency because your RAM has to wake up because it allows part of your RAM to go to sleep. In the real world this isn't
Speaker A
meaningful, okay? Because your your game isn't constantly going to use different parts of your RAM. So it's not going to incur in this extra latency anyway. Yes, it might be measurable by something like ADA because it's going to be trying to
Speaker A
randomly sample for latency, but it's not meaningful. And I would much rather take like 10-second boot times than 1 minute every time. So I recommend using memory context store and power down enable, especially when you're going to be tuning stuff like your CPU and GPU.
Speaker A
It makes it much faster to cycle through different tests. So because you don't have to wait a minute. So I personally use MCR MPD on UCLK V1 mode. You have to make sure that it's Ucl equals MCLK which is the memory controller on the
Speaker A
CPU is synchronized with the frequency on your uh RAM on on the memory controller for your RAM. But if you're using very very high frequency on your RAM 7600 and higher apparently it's going to be realistically impossible for
Speaker A
your ULK on your CPU to run that fast. So you need to use this UCLK equals MCLK divided by 2. your memory controller on your CPU is going to be running slower.
Speaker A
Now, if you're going to be using 7600 or 8,000 MHz on your RAM, you probably already know what this is about. So, you probably already have it set to MELK divided by two. But just be careful because I know some people have like
Speaker A
RAMs whose expo is 7600, then they set the expo, then they get boot problems or instability, then they lower their frequency uh by manually. The expo probably said Uclals MCLK divided by 2 and then they end up with the UCL the
Speaker A
memory controller on the CPU going way slower than it should uh add low memory frequency and that loses like makes you lose a lot a lot of performance. So if you're in the 6,000 to 6400 MHz frequency based on my testing
Speaker A
obviously you want your CLK equals MLK gear down mode. The old way of doing this was to find the values at which you would be disabled because you get uh lower latency uh even though you give up stability. Personally, based on all my
Speaker A
testing, I would much rather have enabled for the extra stability even though maybe there's like a little extra latency, but uh it allows me to go deeper in all my other timings.
Speaker A
Now, again, I said you have to do the timing tweaking yourself. So, I'll show you my timings are stable for my sticks.
Speaker A
my 32x2 Heinix ADI dual rank sticks and I'm sure there's a lot of juice left to squeeze here. I'm not an expert RAM tuner and it's the most complicated part, but I'm pretty happy with my current results and they're very, very
Speaker A
stable. So, that's what I'm keeping for now. I'm having fun tuning, so I might go deeper. I'll just show them to you.
Speaker A
You can just pause the video if you want to try them yourself on your RAM if you have similar sticks. But this is it. I'm not going to talk about each individual timing because honestly even the experienced overclockers are
Speaker A
finally admitting then that you can sure try to use logic behind these. To me the reality is that you need to like just keep trying to find which ones work best because you can really just logic your way out of these timings. If you don't
Speaker A
know what any of these are just use expo or use the video that I'm going to link in the description from Bill Zoid. Now the next very important thing for tuning your CPU is precision boost overdrive.
Speaker A
There's an AMD feature that allows your CPU to boost its clock, which is the most important part of the CPU depending on load. So, how much load is happening on the CPU, temperature, and voltage.
Speaker A
You can if you don't turn it on, your CPU just won't give you all the the the juice that it can give you, unless you're doing a different type of overclock where you just raise the clock all the time. But for my 7800 XTD, you
Speaker A
really can do it. It's completely locked. There's no multiplier that you can change. You can't really tweak the voltages manually. So, the only way to get maximum performance out of a 7800 XD is PBO. Some people will be thinking,
Speaker A
well, there's another way. There's ELK, the asynchronous base clock uh tweaking on some motherboards. Yes, a few of you that bought those motherboards that have it because you are experienced overclockers will have a way to get even more performance. Most of us normal
Speaker A
humans will have a normal motherboard and therefore our only way is PBLO. Now I've tested medium load boost. It actually uh gives me a little more performance. So I use it enabled precision boost overdrive enabled because if you actually read for me uh
Speaker A
let me move my face well. I guess this works for me. It's basically saying that it allows the processor to run beyond the defined values as in the limits defined by default by AMD to the limits of the board. You can't go higher than
Speaker A
the limits of the board. So what it's basically saying is that by using enabled it raises the limits to as high as they can go. Enhancement manual AMD eco mode even though enhancement sounds like it should be better than enabled
Speaker A
actually aren't. Enhancement basically like changes the limits for yourself based on some temp temperature goal.
Speaker A
Enable just raises them to as high as they can go. So use it to enabled.
Speaker A
Overdrive scaler. Real quick, this value here is used internally by the PBO algorithm. It defines how long it can stay at the boost voltage and it will make your CPU degrade faster the higher the number goes. A lot of people say to
Speaker A
stay at 2x. I haven't really measured the big of a difference and I'm okay with my CPU degrading faster. I don't really trust that the risk is that large, so I keep it at 10x. But experiment for yourself. This value
Speaker A
here, on some CPUs, it will actually help it boost higher at the maximum level of boost from the PBO. On the 7,800 XPD, it can't go any higher than 50/50. So, it does absolutely nothing.
Speaker A
Same thing here. Doesn't I can't really tweak how high it's going to boost. Now for the curve optimizer which is the actual important part of this. This is what will allow your AMD CPU to boost higher and maintain its boosts higher.
Speaker A
Now the curve it's optimizing curve optimizer is the voltage frequency curve. It's saying at a given voltage what frequency can the CPU be at? And you can shift with an offset so that say that by default at one volt your CPU is
Speaker A
at four. You can shift it so that it is at four at like 0.9. Therefore, if it gets to one, it's going to be higher than four. So it's a way to overclock by undervolting because by undervolting it's going to be colder and then the the
Speaker A
overgorithm will allow it to be at a higher voltage at the same temperature and therefore that or even the same voltage now but that voltage will have a higher clock connected to it. So you can do it per core or all cores. As I
Speaker A
mentioned, maybe you can squeeze more by doing it per core. And you have to use a negative value if you want that a lower voltage will correlate to a higher frequency which is what we want. So it's technically an undervolt uh even though
Speaker A
it's much more complicated than that. [music] And the value is just a number which is used internally to determine how many millolts per frequency it will shift. I'm pretty sure this number is is a little more complic in a little more
Speaker A
complicated way than just like undervolting inside the algorithm for PBO, but whatever. Now, how low you can go depends on your CPU itself, your probably your motherboard, the quality of the of the power from your supply, the RAM. It's just complicated. You have
Speaker A
to try and see what value like how low you can go. They will still allow you to boost pretty high. Now I can honestly I get pretty much the same performance for any number lower than 20 because of a
Speaker A
different setting elsewhere that I will show you. So my CPU is still able to boost 50/50 which is the highest it can go 5.05 GHz on each core that is running my game which is what matters with no
Speaker A
clock stretching which means that it's actually at that clock and effectively. Now why am I at minus 27? because I'm kind of seeing how far I can go with it still being stable.
Speaker A
Realistically, I think minus 23 is where I've noticed I have pull stability even like in synthetic benchmarks and testing and my CPU still has like imperceptible differences in games. So, I have 27 now, but I went as low as 32, but that
Speaker A
absolutely does give me some instability or even 35. Same thing, some instability in some cases. This one you have to try yourself. I would start probably from 15 or 20 negative all cores which will allow you to boost much higher and you
Speaker A
will notice differences and then go from there. Go lower and see when you start getting instabilities.
Speaker A
If you have a newer 9xxx uh chip, you can also use this really cool thing called curve shaper, which basically allows you to define different values of the curve optimizer per different situations, which is really nice because any CPU gets less stable at lower
Speaker A
voltages and the boost depends on load. So an issue with P with a basic curve optimizer is that when you don't have load, the boost will be really low. So in the voltage frequency curve, you will be really far to the left. And what that
Speaker A
causes is that you will go so low in voltage even though you would be okay when you're not playing being at a higher voltage that it will crash when idle but not when under load which is kind of annoying. So the curve shaper
Speaker A
seems to help with that but it doesn't work on the 7800 XD on my motherboard at least. So sadly I can't test it but it's uh something that makes perfect sense to me. Now these settings here I use level
Speaker A
two enabled disabled. Why? Core tuning configuration for gaming according to AMD is changing some settings of the prefetcher so that it's better for gaming. Even though some people will tell you to use legacy because they measure a lower latency in ADA 64, that
Speaker A
to me is just an artifact of the preetching. But if this gives better FPS, that's what matters, not just like nanoseconds difference in a synthetic benchmark, which doesn't mean that it's actually real in the in the actual usage
Speaker A
of the of the CPU. So I suggest using level two core performance boost you need enabled or it's not actually going to let the PBO algorithm work. Turbo game mode disabled because enabling basically makes your CPU uh an 8 core
Speaker A
8thread CPU no matter what it is which I think is just pointless unless you are so lazy that you don't want to do affinities in the operating system. But you can achieve this with just changing the affinities in the operating system
Speaker A
without actually killing the rest of the cores which might be used for background stuff. So I don't recommend turning turbo game mode on even though they named it terribly. DG+ VRM. This one is kind of risky. I'll go really quick. If
Speaker A
you want to change stuff here like I did, know that you might risk prying your CPU if there's a transient very high voltage and just degrade it faster.
Speaker A
I'm okay with the risk. I'm not going to explain these here at the bottom. The TLDDR is that they're going to prioritize the stability of the voltage over thermal and allow the CPU to pull more current. This one, honestly, I'm
Speaker A
pretty sure does absolutely nothing on a 7900 XCD because it's going to get limited way before it needs to pull more current. But this one does do something.
Speaker A
So, I'm going to talk about this really quick because a lot of people have a lot of opinions about CPU line calibration.
Speaker A
What this does is basically when your CPU is suddenly loaded, the voltage drops. It's called V droop. And CPU load line calibration is an algorithm that tries to flatten this curve so that when load is applied, it increases the
Speaker A
voltage accordingly so that it can keep it more and more uh flat. This is risky because what happens is that by the time the voltage starts dropping and the CPU adds it, this there's a little bit of latency. So what can happen is also that
Speaker A
a much higher voltage than what is reported is for a brief period of time fed to your CPU which could kill it.
Speaker A
Some people think that it's related to some famous like 7800 XD like AM5 deaths on some ASRock motherboards. Personally I'm okay with the risk because it does increase performance by a lot and I think I figure out why. First of all uh
Speaker A
the actual value depends on your motherboard. you have to look up which value is the flattest B droop. So the one where the motherboard is feeding more voltage and therefore more risk of frying your CPU. But what what happens
Speaker A
is that remember how earlier with PBO we shifted the curve so that at a given voltage it can boost higher that is a value that's used internally to the CPU as we've mentioned and it will and I so
Speaker A
some people are like you are giving voltage back from with the lowline calibration that you're taking away with PBO so that makes no sense but no because I tried and even at the same value of PBO as you increase load line
Speaker A
calibration and therefore the voltage increases the actual boost increases too. So it seems like PBO shifts the curve and then it's almost like load line calibration is yes giving the voltage back but still at the shifted curve. So it's allowing the higher I go
Speaker A
the more flat I go the higher it allows my CPU to boost even at a fixed PBO. So eight allows me to feed more voltage and keep the CPU more boost like get it to boost higher uh even at lower PBO. For
Speaker A
me that's okay because my thermals are are great and my CPU never really overheats. But of course an effect of this apart from the risk of attrction voltage by killing your CPU is that it will effectively allow your CPU to boost
Speaker A
higher at a higher voltage or or actually to it will allow your CPU to stay at a higher voltage and therefore boost higher so it will increase thermals. So you need to be careful because you're undervolting on one side
Speaker A
and then giving that voltage back and it will boost high and it will generate more heat and of course it's pointless if you end up hitting a thermal throttle. So you need to make sure that you take all these things in
Speaker A
consideration. But this one, no one really talks about it or they tell you not to use the maximum. But this one is massive in terms of actual performance gains. So I'm using level eight because again I'm okay with the risk and my CPU
Speaker A
is really well cooled. This one is just if you want to be doing benchmarks and it changes some of the internal settings to do to get more numbers in benchmarks.
Speaker A
I never changed this little thing here. Apparently, if you disable clock spread spectrum, which is a feature that allows your CPU not to stay at exactly its base clock, but go slightly up and down all the time to minimize electromagnetic
Speaker A
interference. Apparently, this helps with stability, which makes logical sense to me. I haven't tested it a lot, but it seems like it's true. So, I have it disabled. This one, I don't really know. I'm testing because I don't no one
Speaker A
really this isn't really documented, so I'm trying to figure out what it does. So, I've been testing it. That's why it's enabled. But, I don't recommend actually changing this one. For flex, I would leave it alone. It basically
Speaker A
changes the algorithm for how the different CC the different CCDs which are the sets of cores in multiCCD CPUs or that might have some cores with the X3D which is the actual level three cache or not are prioritized differently
Speaker A
according to values I will leave it alone cuz you can honestly probably do a lot of that work manually in process lasso later in the tier 2 level that we're going to see later the software level now the voltages I'm not going to
Speaker A
really spend much time here I'll tell you SOC is really important because the higher you go, the higher you can have your ULK be stable, which means you can go higher memory frequency. But the higher you go, the less stable the FCLK
Speaker A
gets, and also the more heat and power you draw, and also the more dangerous it gets for your CPU durability. This one you kind of have to play with yourself, figure them out. Some of these you will find in the build zoid, the ones for the
Speaker A
DRAM. This I'm showing you what I have, but I'm not going to spend much time talking about them. Now, other there's just a couple more important things.
Speaker A
First of all, you need to make sure that you have resize bar above 4G decoding enabled. This is really good because it allows games that Nvidia wide lists if you're using Nvidia to use resizable bar, which is basically it can address
Speaker A
the memory in the GPU directly from the CPU in larger chunks than what would be otherwise possible. So, just just turn it on. It's probably going to be on by default. So, just wanted to mention that one. I highly recommend you turn off
Speaker A
your integrated graphics unless you're using it for something because I did notice very big stability differences in mine, especially Steam. It's a known problem that I found on the Steam forums. Steam would actually crash my computer sometimes when I was
Speaker A
overclocking because of the integrated graphics being on. Super weird. So, I disabled it and it fixed it. So, I recommend disable if you're not using it because why not?
Speaker A
Then this one, experiment yourself. Some people will tell you that it's more stable with C states on. Some people will say that it's more stable with C states off. I honestly again don't notice a difference. And I've never
Speaker A
noticed my CPU actually entering like C6 or any of the sleep states no matter what, which is kind of weird, but that also probably means that this isn't doing really much on my CPU, even though a lot of people will tell you this does
Speaker A
a lot. This doesn't change whether your CPU is in C 0 or one which is basically the like super like focus high performance or still like high performance but not uh focus it still goes 01 anyway so this doesn't control
Speaker A
that so I'm not exactly sure now IO MMU if you disable this you will get more performance especially in something like CS but first of all it's a security feature so a lot of the things I will disable you need to know what you're
Speaker A
doing if you have very important stuff on your PC it's your own risk I'm okay with it But just know that all of the virtualization features allow for bad viruses to get even deeper in your system and get access to more stuff
Speaker A
because there's no virtualization. The drivers for a malicious attacker now get control over your memory CPU uh a lot more directly but it does give you extra performance but face it won't allow you to play. So depends on what you you play
Speaker A
on what you you what you're doing. you will get more performance if you turn IO MMU off, but you won't be able to play Face It. So, right now it's disabled for me because I haven't played Face It last
Speaker A
few days. It's really quick for me to enable disable it. I kind of go back and forth depending on whether I'm playing Face It or not. Another thing on the same topic, I am MMU. There's also SVM, which is CPU virtualization. This also
Speaker A
gives me more performance when it's turned off. It also turns off memory integrity in uh Windows, which is a security feature. you won't be able to play face it if if you turn this off, but you get more performance. So, up to
Speaker A
you if you want SPM off. Now, another quick thing here is this one was actually meaningful for me. The LDR, you need to make sure that your GPU is in the correct slot obviously, and it was for me, but my motherboard shares that
Speaker A
PCIe lane with an NVME slot that was close to it and also another PCIe lane.
Speaker A
So basically, if I had anything in those two, instead of being PCIe Gen 5 x16, it switched to PCIe uh Gen 4 8X, which is only like a 2 to 3% difference in my in a 4090. But, you know, I was trying to
Speaker A
push as high as I as I could. So, I removed the NVME. Um, and you can even force it to be X16 like this, but I know auto mode will go X16 for me because I've got nothing on M.23. 2_3
Speaker A
and PCI x16_2 which is the PCI e slot just under it and the NVME. So be careful because you might be using your super OP 1590 maybe at like PCI gen 48X because you you put an NVME in the wrong
Speaker A
slot that you didn't realize. So use something like GPU Z to check the actual like bandwidth of your PCI lane for your GPU because you might be leaving a couple percentage points of performance on the table. Last thing to do here, if
Speaker A
you go in the AMD overclocking area, which will have a lot of the settings that we've seen in the AI tweaker thing, only change them in one of the two because who knows which one is going to be prioritized if you change it in both
Speaker A
is, but you will find this in build zoid stuff too. The nitro mode for DDR5.
Speaker A
Again, I'm not going to spend much time talking about these, but know that as per usual, you get more performance, but less stability if you tweak these. Uh, I just wanted to show you what I have because a lot of people just want to
Speaker A
know what I'm currently running. Apart from that, I haven't changed anything else in my BIOS. [music] So, you've seen everything. Quick recap. These are my voltages.
Speaker A
This is my VRM setup. I have clock spread spectrum off. This is my core tunings.
Speaker A
I'm running slightly higher BCLK, but don't do it. It's honestly doesn't really change anything to me either. I'm just trying to figure out what the limit is for me. I'm using 6,000 even though it's a 6118 because of the VCLK 2200 MHz
Speaker A
Infinity Fabric. PBO is at the limits of the motherboard with 10x overdrive scalar and minus 27 all chorus curve optimizer and my DRAM is running uh with MCR and PD enabled UCLK equals MCLK obviously gear down mod enabled and
Speaker A
you've seen the timings earlier but I will just quickly scroll through them again in case you want to stop and use this as a [music] as um uh timestamp reference then in advanced my only differences from talk are that I have
Speaker A
right now SVM off just for the benchmarks of course I have rebar turned on then I have PCIe I made sure that it's on 16x my integrated graphics is disabled [music] my IO MMU is disabled and cate is disabled and my RAM
Speaker A
nitro looks like this I also have a custom fan curve which I will quickly show it's pretty basic it's probably way easier to see like this. But basically, my CPU will uh ramp up to 100 when it hits 42 C.
Speaker A
So when my CPU when my CPU temperature hits 42C, it will go from about 30% to 100. My pump is always at 100% because the changes in its speed were bothering me in terms of sound. And my chassis
Speaker A
fans are all at 100% all the time because they're really quiet and I want a nice air flow in general. So that's all for my BIOS. So CPU and RAM. Now let's quickly go check my GPU in the
Speaker A
operating system. Okay. So now for the last part of the tier one, which is the hardware optimization. Let's talk about optimizing my GPU. Um, which implied of course overclocking it and a little else which means overclocking. And in my
Speaker A
case, I also flashed a different bias, which up to you if you want to risk it.
Speaker A
My GPU has a dual bias, so it was very safe for me. Um, but there's always a risk there. Um, I am pretty sure that there's risks involved with even just explaining how to do it based on the
Speaker A
fact that no one seems to explain it. But you can look it up. There's a GitHub for a program called NV FlashK, which allows you to flash BIOSes that aren't compatible with your GPU. And then you can find BIOS on tech power up.
Speaker A
um because they have a little nice database. Why did I do that? Because my 4090 by default would only allow me to overvolt it up to 1 070 m volt. Originally 490s could boost could be pushed all the way to 1.1 volt,
Speaker A
but then uh GPUs that were made later on were prevented from doing it. So if you flash a particularly old BIOS, you can get back to 1.1 molt, which allow you to get a little extra clock at higher
Speaker A
temperature. Obviously, uh I remember that I also repasted my GPU with PTM 7950 or really I wouldn't have been able to squeeze any extra juice from overclocking. This is my uh everyday stable overclock where I can run OCT 3D
Speaker A
Extreme and get no memory errors u ever. Basically, I ran it for hours. Now, when I got the 3D Mark scores, I wasn't using this BIOS, but I was using a different BIOS that allowed that allows me to lift
Speaker A
the power limit at all. The power limit for the BIOS I'm using right now, which is the one that I use daily, is 600 W, which is the safe value for the connector, the infamous connector from Nvidia, the 12vt um uh 16 pin connector
Speaker A
that is known to melt. I just for the benchmark I used a 100vt extreme overclock modified BIOS that allowed me to lift the power limit to a 1000 watt and I've seen it transiently hit like 750 which is not safe. So don't
Speaker A
do it unless you know what you're doing and you okay with risking a little fire on your connector. Um and in case you want to know the overclock uh that I use on 3D Mark was this plus 205 plus 79 70
Speaker A
uh 1700 which got me to 12.2 2 GHz on the memory and it allows this to boost up to 3.1 GHz um at a lower temperature.
Speaker A
Of course, my fans were spinning at 100% and I started from about 28C uh on the GPU when I was doing the benchmarks. But my daily uh which really is basically the same performance overclock that I use is this. I don't lock my boost so
Speaker A
that I allow it to get to lower power levels. But I do use Nvidia preferred maximum performance which means that it will never really go to the like 200 300 uh hertz values, MHz values um that it would get. It doesn't really change
Speaker A
much, but it allows you to be a little cooler at idle. Uh, of course, I use the maximum power and GPU temp limits, but they're never hit in real gaming. And I use 12.002 002 GHz memory clock which is a plus 1500 on my
Speaker A
specific BIOS compared to its base memory values. So 140 on the clock so that it boosts to uh about about three GHz on the clock. Of course voltage I let it go all the way up to 1.1 volt.
Speaker A
Maximum power limits 15 under clock. And then I have a custom fan curve similar to the one you saw in the BIOS for my CPU where at idle like now it's only running at 30% so that my PC is super
Speaker A
quiet and then if it gets any higher than 40C they spin up to 100% and stay there. That way I don't let it just hover at like 70C, but when I'm gaming, for example, in Call of Duty, it's about
Speaker A
45 C, which lets it be at 3 GHz all the time as I'm gaming. Um, with maximum stability. The software here is Precision X1. I've also used MSI After Burner a lot. Um, I just kind of like Precision more. I like the UI. Um, and
Speaker A
it seems like it applies the voltage more consistently even though I can't tweak the curve like I can with MSI After Burner. So, of course, this will depend on what GPU you have, how comfortable you are, your thermals, but
Speaker A
I just wanted to show what I did, what I use every day, and what I use for the 3D Mark course. Okay, here we are. Tier three, the tip of the iceberg, Windows optimization. There's a good chance that
Speaker A
you're here hoping that I'll give you the next holy grail registry tweak that's going to improve your latency or whatever. And I will have to disappoint you. I fell for the same trap believing that there was some sort of tweak and
Speaker A
that some of the things I was trying the popular ones were actually doing something I could perceive until I decided, hey, I want to be rigorous. I need to make sure that whatever is there I can measure. So I built my measuring
Speaker A
tools which I've discussed on Twitter, Discord, and I'm going to make more videos about. But this video will be insanely long if I give you the details of my measuring devices and me measuring methodologies.
Speaker A
Just know that absolutely none of the registry tweaks or the win32 separation or the interrupt affinities and policies or trying to optimize ISR and DPC and manually dealing with your interrupts did anything.
Speaker A
All of them consistently change my click to photon latency by zero. I could reproduce to the 0.01 01 millisecond precision with 100% consistency, zero effect from any of these tweaks. There's things that you can do in Windows to improve your
Speaker A
latency, which I will explain, but they're very specific and they're not the tweaks that you think they are. So, let's define a ground rule based on my testing, and I'm open to someone proving me wrong. If a tweak has no effect on
Speaker A
frame timing, so average FPS, 1% FPS, something that you can easily measure, then they do not affect latency. Again, I will gladly be proven wrong. But everything I've tested shows that if a latency difference is measured, it is
Speaker A
only thanks to changes in frame time. So unless you see your FPS increasing, which you don't need knowledge or research to test, it's very easy to test. You're probably not getting any latency change from a tweak. So that
Speaker A
said, clearly I've shown you these charts before and the tuned OS has a difference in FPS. How do you achieve that? There are five main things you can do. I'm going to keep them simple.
Speaker A
Everyone has a lot of nuance. I can't cover it all in a single video that's already this long, but it's what I'm doing to show you the larger numbers.
Speaker A
The most obvious one is you don't want stuff running on your operating system competing for resources with your game.
Speaker A
This will cause a measurable difference in your average and especially low percentile FPS. So, you need to decrease the number of background things that are happening. So, you need to debloat your Windows. I do not recommend you install
Speaker A
a custom Windows image. That's an insane risk. I don't see why anyone would do it unless you are kind of like in the Linux space where you're going to be checking the [ __ ] image yourself which no one
Speaker A
is doing. You don't even know where the image comes from. Please, it's just it's just not wise to do. just if you want reinstall your Windows so that you start from the default level of blo in your windows and then there's some open
Speaker A
publicly available tools like wind toys or Windows optimizer which are honestly unnecessary if you're willing to do some stuff yourself manually that are going to remove some of the more obvious bloated services that maybe I don't know your motherboard tried to inject like
Speaker A
Asus motherboards tried to get you to install Armory Crate when you uh started with Windows If you remove those and you can easily just open your task manager and check what's using your CPU. No magic, no voodoo involved. If you debloat your
Speaker A
Windows so that when your game is running, there's no other apps competing for any meaningful amount of CPU time.
Speaker A
There's not going to be a change in your FPS. So I debloated largely manually by going through my process list and my services and finding stuff that was actually running and I could see running and removing it. And that had an effect
Speaker A
on overall frame timing across long periods of time because the times where that process would have stolen resources by running in the background was now not happening anymore. I will add this bit here but I do not recommend it. There
Speaker A
are two main Windows important services that actively compete for resources and are noticeable for games and these are Microsoft Defender and the Windows Search Indexer.
Speaker A
These are two important services in your operating system. So, I do not recommend anyone turn these off, but I'm not going to claim that having them on or off will never make a difference. Obviously, it is more important for your operating system to
Speaker A
make sure that you're not about to run a malware or actively running malware than it is for you to get at the highest possible frame time. So, if you disable the Windows offender completely or at least surgically disable the real-time
Speaker A
protection, the malware analyzer, and you disable the Windows search indexer, but your searches are going to suffer even those in start. So, again, I do not I do not recommend it. you will have a lower chance of your FPS being affected
Speaker A
by these background services. I'm not recommending you do it. I did it for testing and could measure changes by removing Defender completely through Windows 11 Pro group policy settings, which I do not recommend you do. But I want to be fully honest and say that
Speaker A
those do affect your frame times. But the risk versus reward unless you're trying to like get a high score on a benchmark or something in terms of FPS benchmark, not aim training benchmark are to me not worth it. Number two tweak
Speaker A
that actually works is I could consistently statistically meaningfully reproduce using any sort of VSYNC or G-Sync implementation. So that it means also adaptive sync, free sync, whatever technology name you want to use. That same concept of trying to synchronize
Speaker A
your refresh rate with your frames, add latency. Even if you do all the correct steps, you use the minus 3 FPS limit.
Speaker A
What I'm not going to mention these because those are if you're interested in having a great G-Sync experience, which I recommend unless you're trying to chase that lowest latency possible.
Speaker A
But I could easily measure and reproduce an increase in latency. We're talking about 1 2 3 milliseconds on my testing framework if you use any of the synchronization algorithms and systems.
Speaker A
So if we're talking about purely optimizing for latency and when I say latency I mean input lag I suggest that you don't use bsync or g-sync. Third thing, and this is a universal truth, a pretty easy to understand concept, I
Speaker A
would say, do not lock your frame rate to your monitor hertz unless your game specifically doesn't play well with uncapped frame rate. Because I'll give you a quick example. Say that you are on a 60 Hz monitor. That means that you
Speaker A
will see a frame on screen every 16.6 milliseconds. If you get 60fps and you don't have any synchronization, that means that you have a chance of seeing a frame that is up to 16.6 milliseconds old after 16.6 milliseconds. But if you
Speaker A
say have infinite FPS at 60 Hz, that means that every 16.6 milliseconds, the frame visible on screen will be fresher than if you were getting lower frame rates. So, even if you are already achieving your monitor's refresh rate,
Speaker A
more FPS will always decrease input latency because you will see the same amount of frames on screen per second, but they will be closer to now in actual time. Some games do play poorly with uncapped FPS. So, test on your game. If,
Speaker A
for example, the price to pay for locking your FPS is getting way worse frame times on average or getting really bad dips, then it's not worth it, obviously. But you can measure the expected change yourself. You take your
Speaker A
FPS and you do 1 divided by FPS. That gives you the frame time between two frames. And that's the theoretical maximum amount of time that can be elapsed since that frame was generated when it's shown on screen. FYI for the
Speaker A
very technical people, this is a large simplification because that's not how actual rendering works. We would need to talk about frame buffer cues, Nvidia reflex, the presentation model. It's more complicated than that, but in general, it's true. If you don't change
Speaker A
any other variable, higher FPS means lower latency. And this is super easy to reproduce and super easy to calculate by how much changing frame rate increases your latency.
Speaker A
So if you can don't cap your FPS, let them run higher than your monitor refresh rate. Number four, I know that you probably think full screen borderless is the same in modern day as full screen exclusive. And that is true if a certain
Speaker A
set of conditions are fulfilled. The safer approach unless you have the technical knowhow to verify that those conditions are fulfilled is to use full screen exclusive. For example, I have a 540 Hz monitor on a 4090 which I have
Speaker A
and any other 4000 series Nvidia GPU. to implement 540 htz through the SC. It's going to end up using two display heads inside the GPU. That means that the display acts as a tiled display inside Windows, which means that mo doesn't
Speaker A
work. Which means that I don't get the flip presentation model optimizations that have made for a certain amount of people in a specific setting. full screen borderless equal to full screen exclusive. So in my setup, which isn't that uncommon, and there are other ways
Speaker A
that could force your MP not to be on, I have an very easy 100% reproducible on any game, decrease in frame time, and of course an increase in input lag if I use full screen borderless. Same thing happens if I use full screen optimize
Speaker A
and I use overlays like the Steam overlay. So in my case I have to use full screen exclusive on my main gaming monitor and I have to avoid using overlays.
Speaker A
Now, there are again cases in which full screen borderless is equivalent to full screen exclusive. But unless you are 1,000% sure that your MO clip presentation model optimizations from Windows are working correctly, it's safer for me to say just play full
Speaker A
screen exclusive so that you don't risk not having the flip presentation model working. Number five, and honestly the most tweakingish of the changes that is surprising that is kind of surprising that it works is and you need to test this on your own
Speaker A
CPU for my CPU and a lot of other CPUs that I've seen. If I prevent my games from running on core zero, physical core zero, which means that if you have hyperthreading on would be for my architecture, logical cores 0 and one,
Speaker A
if I prevent my game from running on logical cores 0 and one, I get increases in both average and 1% FPS up to 8% average and 22% 1% in CS2, which is insane, and you wouldn't expect that to
Speaker A
be the And I've tried figuring out why. And I think the reason is just that most legacy code probably including a lot of low-level kernel stuff in Windows tends to spawn its thread on core zero and especially the low-level Windows stuff
Speaker A
will obviously have a higher priority level compared to a user space process like your game and therefore probably end up preempting your game thread more often on core zero.
Speaker A
This is super easy to test. Open your game, open your task manager, go to the details tab, find your game process, right click, select the affinity, and try removing core 0 and one. Or if your architecture defines the two logical
Speaker A
threads for your core zero as something like zero and maybe 14. This one you have to look up yourself.
Speaker A
Disable those. basically disable both logical cores from the physical core zero and see if something changes. It is important that you disable both of them or you will actually make things worse.
Speaker A
If you do that correctly, for me in every game that I've tested, this increases both average and 1% FPS. So, this is a little bit magical, but again, easily justifiable, easily testable. You don't need to do research. You don't
Speaker A
need to measure latency with a complex electronical uh end to end latency tool. Just change it and see if you have FPS changes. If your FPS changes, then it affected your latency. Now, if you change it from the task manager, you
Speaker A
have to do it every time your process starts. So, up to you what you use, but I've been using process lasso because it's basically just a tool that allows you to define rules that then apply these affinities manually. and I just
Speaker A
applied it to my steam.exe because these affinities are inherited by processes spawned by a specific parent process in most cases. You need to check your game.
Speaker A
So, I just set up a rule that my steam.exe only runs on my cores that are not the physical core zero and I don't have to think about it anymore. Now, someone would say they're using something that process lasso introduces overhead. Sure,
Speaker A
you have something running in the background to check if a process started. I could not measure any impact on frame time or latency, of course. So, I'm okay with process loss. So, I think it's a pretty cool software. Up to you,
Speaker A
but this one was very measurable to me, so I had to mention it. And it's the only real tweak that actually did anything. And that's it. Literally nothing else I tried had any measurable effect. I'm going to mention now briefly
Speaker A
because I said I would two changes that are probably in the placebo territory, but that might maybe theoretically do something. I'll be quick because I don't think you should waste actual time doing this yourself. One is that I bought a
Speaker A
PCIe card that uses an as media ASM 3142 XHCI controller. Again, can't go into details. this video will be too long, but just know that it's basically going to make it so that your mouse packets are handled by a different piece of
Speaker A
hardware that will only be handling those and will pull only your mouse, which is the XHCI controller on the card and also using a different driver because I specifically installed as media's implementation of the USB driver and not using the Windows default one.
Speaker A
This technically could make it so that there's more free real estate for your mouse interrupts not to collide with anything else or be moderated through something called IMOD, which is interrupt moderation. But if you stop to actually think about it and
Speaker A
do some math and some statistical analysis or some testing, which I've done, it's actually way more likely that it does absolutely nothing. And I could measure no statistical difference between using my CPU connected USB, which is another thing that I tested not
Speaker A
to be meaningful, motherboard connected USB, or this PCIe as media card with its own XHCI controller and its own driver. So maybe depending on how many peripherals you have connected to a specific XHCI driver on your hub on your motherboard, using
Speaker A
your mouse on its specific PCIe card could maybe do something. I couldn't find a situation in which it could be measurable, but it's only €23.
Speaker A
If you are seeing issues with, for example, running 8k Hz even after you do everything else that I've mentioned, maybe try spending €20 on a PCIe USB expansion card. Up to you. The number two placebo probably tweak is trying to
Speaker A
change your interrupt affinity mask and priority. The priority itself, I'm pretty sure absolutely does nothing when you have MSI enabled, but again, can't go in details because it will be too long. But changing your affinity mask does do something easily reproducible.
Speaker A
If I assign my for example PCIe XHCI controller to have an affinity mask for its interrupts to for example only handle interrupts on cores 14 and 15.
Speaker A
Then I can see that the DPCs which is part of the whole interrupt pipeline ESR causes the DPC the DPC which is the actual handling of the interrupt happens on those cores. So I can see a world where making sure that the popular ones
Speaker A
are usually tuning your GPU interrupt affinity mask and your mouse actually your mouse's controller interrupt affinity mask. I can see a world where depending on what's going on, this makes it so that your mouse doesn't have to compete for interrupts
Speaker A
with something else. Same with your GPU. But again, I could find no real world scenario that I could run and measure any difference. So again technically it makes sense and you can measure differences if you use ISR and DPC analysis tools like the Windows
Speaker A
performant analyzer toolkit. You will see these changes do something at the interrupt level. But if you stop to actually do the math, you will realize that again, unless there's a very obvious visible problem, the likelihood of that affecting input lag is basically
Speaker A
zero. If you're really interested, go lurk my Twitter and you will find a number of threads where I went through the math and shown why statistically it is extremely unlikely that optimizing your interrupts at the ISR and DPC level
Speaker A
actually affects input lag. But again, maybe in some specific case, these two tweaks might do something. If you have nothing better to do with your time and you also want to waste time like I did, try them out for yourself, make sure
Speaker A
that you're not being placebo into thinking they're doing something and let me know if you can measure any change.
Speaker A
And now that we're done with the placebo part of the Windows optimization as well, we're done. I can tell you there's a very high chance that anything you can tell me to try, I've already tried. I've tried Win32 priority separation. I've
Speaker A
tried the hidden power plant settings that you can tweak. I've tried all the registry keys are supposed to play with the power saving features on your GPU and CPU. If you're actually playing a game running a DirectX app, I could not
Speaker A
measure any difference down to the 0.1 millisecond. If you're confident that you have the holy grail of tweaks, the registry secret key that will cause anything to be measurable in some rigorous statistically significant way, please let me know so that [music] I can
Speaker A
openly and happily claim that I found the one tweak that works. But unless you can prove that something that you're tweaking will consistently statistically probably cause two frames that to be different versus the scenario in which you didn't
Speaker A
have the tweak then it is logically impossible to claim that that has a perceivable effect by human. And I'm not saying because it happens faster than a human. I'm saying that unless it's happening so that the frame that is
Speaker A
being sent to your screen is actually different because the effect of the thing for example the interrupt from the mouse caused it to be visibly different.
Speaker A
What is being shown to the user is exactly the same and therefore there is 0% chance that a user could even feel the difference even assuming they had infinite discerning capabilities in terms of frequency or latency. So unless
Speaker A
someone can tell me what it is that can cause that, I will keep saying that unless you see a natural change in frame time, you are not going to see a change in latency. And just as a last thing,
Speaker A
let me clarify why I'm saying that if there's a change in frame time, you will see a change in latency. This is very easy to understand. Frame rates are usually under a,000 FPS. When you reach 1,000 FPS, you now have things happening
Speaker A
at 1 millisecond frequency or higher. When you're below that, which is where the vast majority of your gameplay will be, everything is in the greater than 1 millisecond order of magnitude. So, for example, if your average FPS goes from
Speaker A
400 to 250 average FPS, that means that the time that elapses between two consecutive frames decreases by 1.5 milliseconds, and that's measurable, but that's also very easy to calculate. You don't need magical tweaks to figure that out. So, higher FPS is always going to
Speaker A
be lower latency at a at an order of magnitude much greater than anything these tweaks ever claim to do. So, if you want to focus on something, focus on ways in which you can increase your FPS and share them with the world because
Speaker A
they will probably not be the secret 100 registry tweaks that these people are trying to sell you. So, that's it. This is my optimization video. I wasted an ungodly amount of time on it. I recorded it multiple times. I wanted to be direct
Speaker A
but also accurate. And I'm sure there must be some inaccuracies considering considering how much I yapped about this. So, please let me know so that I can have a pin comment where I clarify anything that I wasn't accurate or I was
Speaker A
actively wrong about. And if you actually watched everything up to this point, thank you. I do not expect many people will do it, but I am super grateful. And I'm also super grateful for everyone on Twitter and Discord that
Speaker A
was actually open to discussing these topics without preconceived notions, without their bias. I think that was super helpful. And I'm also thankful to all the people that probably felt like their business was being challenged. all the tweakers that actively tried being
Speaker A
negative in my comments and even threatening me because they fueled my will to actually go through build my electronics and do all the testing which took many many days and of course the biggest thank you to my patreons whose
Speaker A
names you will see on screen now I still am shocked that anyone is willing to pay considering I haven't really put much behind the payw wall but considering how much I've learned in my research Here, I'm planning on putting a lot of very
Speaker A
technical content behind the payw wall. For example, the specifics of how I built my electronics down to my STDM32 code, my setup for its ADC, my setup for its timer, things that most people probably won't care about, but maybe the
Speaker A
few crazy people will care about. And since they took me so much to figure out, I feel like it's fair for me to put behind the payw wall. Of course, the results of my testing will always be public, and I'll make more videos about
Speaker A
it. Now that I'm finally done with this video, I'll also go back to actually reviewing preiprols, which I have a lot of backlog for, and trying to get world records and scores in aim trainers, which I still want to do. So stay tuned.
Speaker A
I'll see you soon. Ciao.
Topics:PC optimizationgaming performanceFPS boostoverclockingthermal managementCPU coolingGPU repastinghardware tweakingsoftware optimizationcompetitive gaming
![[YBA] frosti — Transcript](https://i.ytimg.com/vi/KAH0o3xSL3w/maxresdefault.jpg)
