Hello world; kala with A64. Are six cores sufficient for gaming in 2025? Now, this question has no right answer because we are lacking crucial information required to come up with a verdict. That information being the number of threads, the core microarchitecture, the node that the cores are using, the frequency of these cores and also throw in the DDR gen supported.
So we made this little here graph that goes from 2010 to 2025 on the x axis and from a very scientific snoring emoji to a thunder on the Y axis to represent computing power. What is left to add are some straight gains in computing just from dram gen so we have one in 2010 when DDR3 became mainstream, one in 2017 when DDR4 took over and one for DDR5 in 2023. These are all 3 years removed from launch to mitigate the new cpus that still had support for the older standard.
Okay, so let's talk about the first 6-core processor; or better yet, processors because there were 9 that released on the same day and you have those listed now. These were Opterons from AMD on the K10 architecture and used a 45nm node. They were made for the F socket and it's really easy to tell with AMD that these were enterprise chips due to the LGA socket; for the longest time they had LGA for server and PGA or OPGA for consumer until Zen 4 basically; but back to the Opterons and if you're wondering why we are talking about server chips on a gaming video 1. It will soon make make sense and 2. I love going on tangents; so these were released in 2009 with support only for DDR2, clocks up to 2.8GHz and they had really good cache capacity of 128KB L1, 512KB L2; these are per core and 6MB L3 shared; so considering the 4100 from last video had less L cache and that was released in 2022 these values are impressive for 2009.
And now we get to our first consumer 6-core aka the Phenom II X6 which was released in 2010 on the same K10 architecture or k10.5 actually I guess due to the DDR3 support and this was under the hood very similar to the Opterons we talked about earlier so the same cores whith the same L cache but slightly higher clocks up to 3.7 and of course DDR3 support.
Now let's move on the the current day on the opposite side of the graph and look at the latest and greatesat 6-core with the 9600X. This sufficed to say uses a way newer node 4nm for the cores and 6nm for other processes and newer architecture Zen5, boosts up to 5.4GHz, has DDR5 support and has way more L cache 80KB L1, 1MB L2 per core and 32MB L3 shared. So case closed newer DDR, higher clocks and more L cache.
Hower right in the middle of this time frame sits 2017 with the Zen 1 architecture and it's six core - the Ryzen 5 1600. This is, in my opinion the most representative 6-core we could have used to answer the question in the title simply because it sits in the middle of the 6-core evolution and DDR support so what we've started here is a binary search of sorts to find the first 6-core that handles 2025 gaming. This processor is also very interesting because by specs alone it's very similar to the Phenom 2. Same boost clock of 3.7 same L2 cache, slightly lower L1 cache and more L3 cache. Of course it has double the threads so in multithreading it's a whipeout however in single core performance you have to look at the node and the microarchitecture that is uses because all of that performance doesn't come from the memory controller alone.
So the reasons the 1600 wins songlecore by almost triple are: 1. better out of order ececution that is basically the CPUs ability to optimize instruction cycles by not wasting them and executing instrucitions not in the original order but in the order that it decides based on input data and execution units availability. 2. Better speculative execution that is the cpu being better at predicting tasks and executing them beforehand without knowing that they are needed for certain. 3. Larger ROB or re-order buffer that supports points 1 and 2. 4. Introduction of micro-operation cache that being cache for the processor to refference from if it encounters the same low-level instruction. 5. Much better FPU that being the floating point unit that is shipped with all processors since bout 2000 and surprise surprise is used exclusively for operations that involve floating point variables. 6. The L caches and prefetchers might be similar in size but they are much faster storage on the 1600 and finally 7. The newer node allows for lower TDP and in turn less heat so better sustaind clocks and of course more sophisticated and specialised execution units overall. These microarchitectural improvements are mostly what drive gains in single and multi core performance along with others compared to the lesser gains from frequency and cache sizes and combined you get superior IPC that is instructions per cycle and waaay more transistors per die size.
Which brings us to the rest of the system. We still have to talk about the motherboard, RAM, thermal solution and GPU. We'll start with the most straightforward ones that being the motherboard and GPU. The motherboard we had on hand and built the system on is a MSI B350m Bazooka so nothing crazy special; we have PCIe gen 3 support but that is also dictated by the CPU , semi-decent cooling for the vcore vrms that are 4 phase, I think this is only a 4 layer design since at that time in 2017 only the crazy oc motherboards had 6 layers on B350 but I could be wrong since I haven't found definitive evidence that it is indeed 4 layer and not 6 layer; finally it comes with all the bare minimum oc quality of life features that you would expect so we have LED and audio debug with a straightforward bios with profile saving and safe mode all of which we'll be glad to have for the OC chapter. Next for the GPU since we are benchmarking CPU performance it will have to be something that bottlenecks it; the only one we had on hand was a 6900xt but you could get away with something less powerfull to create a bottlenck here and bare in mind we are already dropping some performance by having a PCIe 4 card go through PCIe 3 interface. I've put it here for anyone that cares the profile we are running for the GPU even though it is of very little inportance since running it stock does the job anyways and finally regarding gpu we also had to change the psu to accomodate it so we're moving on from a corsair VS650 that doesn't have enough PCIe connectors to a slightly better Deepcool PK750D that can fill out the GPU's 8 and 6 pin connectors.
Okay now we have to talk about how we are dispersing the heat generated by the CPU. Now this is an A64 production and we love to delid these Ryzens and that is just what we did so we started by hacking at the rubber adhesive that supports the ihs with a razor but this time around unlike the 2200G we did a while back it didn't really get us that far so we moved on prematurely to the vice and this method with the vice basically works by pushing the IHS from a side and pivoting with the substraight on the opposite side. Now in order to do this successfully you have to be careful about a few things: 1. The vice should not touch the inside of the substraight when pushing the IHS 2. Always watch out for the pins on these PGA Ryzens and 3. This is the most important one: before you begin always know how your CPU  looks like underneath the ihs so you can avoid hitting SMDs with it during the proccess. Now if the CPU you are delidding has paste TIM from fabric and not indium you can get away with other simpler methods and if you have a CPU with indium and SMDs very close to the IHS on all sides you are better of running the risk of heating it to about 150-160C and going on from there. The 1600 we have only allows us to push from the top or the bottom only since the sides have SMDs hugging the IHS. Here we ended up making a small mistake; that being not protecting the substraight in any way from the vice so our pivot side got heavily chipped from the iron and pressure pushing against it. Finally now cleaning the indium is pretty straightforward you basically want to remove as much as you can with a sharp razor or 300+ grain sandpaper if you are brave enough and the rest you can remove by etching or more feasably by means of galinstan or generic liquid metal. Now that we have a delidded CPU first thing to do is to try and boot to bios with it to make sure it's still alive and here you have to take into account that some motherboards won't boot without a cpu_fan connection so make sure to account for that if you fail to boot first time around and if all goes well you should be left with a working delidded Ryzen that you now have to come up with a cooling solution for since it's much shorter than AM4 coolers expect basically. So the best option is to go with a radiator block that has it's own mounting mechanism and swappable screws you want the factory screws to be removable from the block basically so you can go with longer or shorter ones to accomodate the height difference and also you want your contact plate ideally to not go over the socket itself as that will be higher than the CPU silicon and also I forgot to mention make sure that the SMDs themselves are not taller than the silicon before you choose to direct-die. An example that fits these criterion are most Thermalright AM4 coolers; this time around we had a King V2 Mini; bet you didn't expect direct die with an 18 euro cooler that we had to swap the original mounting screws with case stand-offs because they needed to be slightly longer; this was a happier solution because depending on block you might have to go with something custom to fit. Another thing with this Thermalright is that it comes with copper nickel plating so it's ready to go for liquid metal TIM applications but for now we are sticking to the GD-900 chinesium that's 4.5 W/mK and we'll swap later on.
Now since we are talking about radiator blocks I wanted to go a bit into contact plate implementations since it's extemely relevant to this experiment as we are moving away from the ihs that is used to spread the heat from the confined silicon to a larger surface area and that is basically what allows modern cooler to implement a stupid number of heatpipes. So let me explain the most common plate design you see is basically this one with the heatpipes being shaved off to form a flat contact point and here all is good with an IHS and thermal paste because every heatpipe gets to be used for a certain sector of the ihs but when you reduce the heat source to a more confined area you get some heatpipes doing basically nothing. This also is quite a bad design because you can't really get the flattest surface by shaving off the pipes and trying to fill in the gaps with aluminium that of course is where the paste comes in but there will still be imperfections. The next implementations is actually what we have here with the contact plate being a solid block of copper or copper nickel and the heatpipes going trough it. This allows for perfect contact area so you can use liquid metal with it unless it's aluminium and also slightly better dispersion of heat from the plate to the pipes so it basically does the same job as the IHS. Before I move on the the next one I need to explain how a heatpipe works; it's a very simple principle you basically have a hollowed copper rod filled with usually H2O for room temperature implementations but you can put whatever liquid in there for other ambiental use cases and on contact with your heat generator that H2O vaporises and raises towards your finstack or away from the heat source where it undergoes liquefaction by cooling and repeats this cycle all over. Now that we know how a heatpipe works guess what we also know how a vapor chamber works because they are the same thing with minor differences obviously one being the geometry and the others regard the wick structure. And this ties into the next cooling evolution that is swapping the contact plate with a vapor chamber that is effectively a wider and much shallower heatpipe and having subsequecent standard heatpipes attached to it. Now this is a game changer for cooling because again if you remove the IHS this time around the system doesn't care because this vapor chamber is king at spreading heat from a smaller hotspot to all the other classic heatpipes so this just outright removes this benefit of the heatspredder because it does the job much better and it just leaves you thermally bottleneck by it if you dont delid because of Fourier's law. So this is where we are at in the PC cooler contact plate lore as of now and logically the next inovaton would be somehow making your heatpipes that go to the fin stack go directly through the vapour chamber itself if possible.
Now, getting to dram for the system, we used G.SKILL Ripjaws 5 3200 CL 16 18 18 38 at 1.35v 2 8G sticks for dual channel that we'll overclock in a bit and here you could go with a 3600 kit and underclock that to get better timings then we ended up with. Now I will put on screen this ram's label and we can easily get the specifications from that. Here we can see that this is 8 gigabits density per chip so we have 8 chips on a single side and empty space on the the other side of the stick. Next is the manufacturer; here 2 means SK Hynix and finally the revision so that is C and if we put it together we can identify this ram as H8C.
So with the system components listed and assembled it's time for overclocking. I'll start with the cpu but before that I wanted to make a quick representation of time invested to reward and other quick notes regarding system OC. Before that quick disclaimer: if you are on a platform 2017 or newer you WILL NOT run any risk of frying your components when doing this trust me I've accidentally maxed out SoC voltages to the point that my system crashed when plugging or unplugging USB devices (yes, that is a thing) and nothing more happened. If you somehow mess up your bios fell free to reset cmos and go back to your prior stable configuration. Now you can obviously do overclocking on older components the only thing I'll say is you better have a remote idea of what you are doing just try to not go into it blind since back then especially motherboards were less reliable; looking at you cheapo AM3 high side mosfets; so you should know that. Okay so back to the quick diagram I wanted to discuss. So your gains from OC here I'm talking about gaming like 70 to 80 percent of that will come from your GPU gains this is for gaming and from that 100% GPU gain on newer cards about 10% are from unlocking you power limit, 60% are from vram overclock, 15% from core boost overclock so the maximum frequency your GPU core will be able to hit and the last 15% from core undervolt which helps your sustained clocks under high loads however most games might not count as high load. Next your CPU depending on game might contribute with an extra 20 to 30 percent up to 40% weighted if your game catalog consists of mostly games similar to say City Skylines or other CPU intensive application. So these are the big two and if you're wondering where does dram OC fit in here well think of that like filling in the cracks so your 1 to 5 percents in average but on the other hand it'll make a bigger difference in your frame drops think 1% or 0.1% lows. So with that set let's talk about about difficulty and the time needed to finnesse each component. First is the GPU here difficulty is low and time needed is also low so pretty straight forward other things that will enable or help your GPU OC are your PSU if you're already running low on watts with the stock configuration you will not be able to unlock power limit for example also case airflow here is very important; new GPUs already come aggressively binned from factory so most of the time you'll be thermally limited on core so watch your temperatures. Next on our scale is CPU; now back in the day it used to be much simpler with it but now they come with gizmos like PBO and curve optimizer that require much more time and understanding to tune so I'm going to CPU medium to high difficulty and medium to high time. Here the most important part is to know what you want out of your silicone so you might want to start with curve optimizer and optimize for voltage or optimize for single core performance or for all-cores to reach crazy boost and so on. As for what will help you most overclock well on the first and second place are your cooler and motherboard I don't think I have to explain why but again make sure beforehand your PSU can handle extra load and as with the GPU you want your case airflow to be adequate if you are on air cooling. And now for the dram here there's no two ways about it high difficulty and high time requirements. The only redeeming factor for your ram OC is that it doesn't put more load on your PSU unlike CPU and GPU and the only supporting actor here will be your motherboard and maybe again general airflow if youre pushing stupid frequencies and yeah that's about it.
Coming back to the CPU OC, the first configuration tested is stock where the CPU stayed at a solid 3.4GHz during load and we got 375 multi and 60 single scores in cinebench R24 and 3252 in 3DMark CPU profile all thread and we saw slightly under 50C in 3DMark and slightly over in Cinebench. Moving on to PBO unlocked; these are basically a set of limits of how many watts your CPU can pull from the socket and how many amps the vrms can deliver during boost and sustained load and we also applied a small undervolt of minus 112.5mV while keeping the cores on 3.4. I also need to do a little bit of explaining as to what that undervolt means because it's not an override, this MSI board allows us to set an offset to all-core voltage a bit like curve optimizer would and from my tests this offset option is much more stable than a similar override value; and so with that we got 407 and 59 in Cinebench and 3243 in 3DMark. Now these results are peculiar because 1. we saw single core drop that might be just error considering we are still on auto 3.4 or it was an effect of the undervolt and the CPU decided to drop a bit from 3.4. and 2. We saw lower 3DMark performance that could also be error or as we continue on with the tests we will see that 3DMark CPU profile max threads is not very representative or very consistent with the score it spits out. Moving on, here we saw slightly higher temperatures and again spoiler alert the cpu gets slightly warmer during Cinebench than it does during 3DMark CPU profile. Next is PBO 3.6GHz set frequency and set voltage to 1.225V and we saw 418 in Cinebench and here I lost the single core data and 3DMark shot up to 3449 points and we are closing in on the 60C mark on temperature. Next is 3.8 with a voltage offset of plus 100mV and this translates to about 1.315V here we saw 447 with 62 in Cinebench and 3638 in 3DMark with temperatures of 71 to 73 celcius in R24 and 68 to 71 in 3DMark also peak draw was about 100W. Next is the final frequency we reached on GD900 that is 3.85 with plus 137.5mV that's about 1.4V and we got 448 and 62 in Cinebench and 3681 in 3DMark with temperatures ranging from 77 to 81 degrees and peak draw being 105W and the VRMs were toasty for this one we got to about 53 degrees on the radiator after that, seeing that we are thermally limited we switched up the paste for LT-128 liquid metal that guess what is has 128 W/mK and we put a better fan from ID Cooling for the radiator intake and moved the Thermalright fan to the back for exhaust such that we have a push pull system set up. After that we pushed 3.9Ghz with plus 150mV that makes about 1.445V in total and we got 451 with 63 in R24 and 3733 in 3DMark and that also got us rank 23 global 8-thread CPU profile at the time and we got temperatures of 66 to 67 in Cinebench and 65 to 67 in 3DMark with the same peak draw of 105W. Final configuration before the CPU required stupid voltage was 3.925GHz with massive plus 187.5mV that is 1.47V and that scored 459 with 64 in Cinebench and somehow much lower score of 3683 in 3DMark with temperatures of 71 to 73 in R24 and 67 to 70 in 3DMark and a peak draw recorded at 107W. So to conclude this chapter i'd say the CPU scales very well up to 3.8 on multi-threading with diminishing returns above that apart from gains on the single-core scores somehow and this particular sample we had here was quite voltage hungry above 3.9; that might also be because we are running 70 degrees but overall I expected a bit higher peak frequency from this 1600 even on the cheap 18 euro radiator that you might still be able to achieve if you lower TDC and EDC but I just didn't bother spending more time fidling with the PBO at this point.
Moving on to the ram and here I wanted to have a chapter explaining how it works and what each timing does but this video is already long enough and we'll probably do a standalone one just for this but long story short DDR is basically double data rate because you read on the clock edge that allows it double readings and it writes on a separate 90degree offset clock. You have your basic memory chip that takes commands from your sophisticated CPU memory controller and also has some of the timings in storage but not all and these timings are basically; well deadlines you give your memory chip or you data to be ready for these various commands so you're interested mainly in read, write, refresh and precharge. Almost all newer DDR4 sticks will be either 8gbit or 16gbit density with an x8 bus unless you are running an enterprise level server or some laptops come with different configurations but that's another different can of worms. So going into dram OC what you need to know beforehand is, well a few things but most important is to know if you have independent clock domains for your infinity fabric or you are stuck to 1:1:1 gearing and that is very simple to find out basically if you are on Zen2 with the chiplets or 3000 series you have access to the FCLK and UCLK otherwise the MEMCLK will dictate your other internal clocks and that is why you don't see Zen or Zen+ on 3600 memory and here MEMCLK is your clock rate for the memory so your standard divided by two, FCLK is the infinity fabric clock and UCLK is the CPU memory controller clock. Ideally you want 1:1:1 ratio. With this 1600 we got rather lucky this time around as it is stable at up to 3466 and that is exactly what we'll use for the OC. Before we started, we disabled disabled power down mode and spread spectrum this ensures that your dram will not go idle and your BCLK or base clock remains stable. Next are the voltages we started from the upper value we are comfortable with that being 1.5volt for dram, 1.15V SoC that is for the memory controller and other components apart from the core and VDDP at 1V that is the interface that converts memory controller signals for the memory chip and shortly after we decided to move it back to 0.9V seeing that it did not impact stability here. Now depending on platform you might also have other VDD voltages that might be specific to the I/O or infinity fabric so fell free to learn about and optimise them aswell. Now we set our ProcODT to 68.6ohms this is the on die termination that helps with stability regarding frequency so if you run a high frequency this will help to diminish signal undershoot and overshoot and the value you'll end up on is also heavily affected by motherboard design. With that out of the way let me show you the final timings we landed on so 13 17 17 17 35 for the main timings that wil actually run tCL at 14 because of gear down mode and here you can se the rest of the secondary and turn-around timings we got; again tCWL and tRTP will be rounded up by gear down mode. We haven't touched the Rtt values because to put it frankly we're not sure what these affect and require more testing and research from our side and finally this is also stable with CAD bus timings on zero and CAD bus impedance at the minimum 20 ohm value. After that we wanted to see if we can get away with less voltage and we did for SoC we managed to drop down to 1.1125V; unfortunately DRAM voltage below 1.5V was unstable. Finally we saw a considerable drop in ProcODT down to 43.6ohm stable.
Okaaay, finally moving on to gaming starting off with a classic, a bangar if you will Counter Strike 2 and ladies and gentlemen you can get a competitive experience from an eight year lord CPU running on last-gen platform with last-gen ram because here we measured 225 average fps with 120 fps 1%lows on the dust 2 benchmark which excerts a higher load than normal play with competitive settings in 1080p. So to all the posers preaching that this game is unoptimised or valve don't care about performance or that you need 9800X3D and nothing else runs this game I say skill issue.
Next, we have a game notorious for it's abysmal CPU optimization and performance that is No Man's Sky and here to be honest I expected it to put on more of a fight but regadless of setting preset we saw averages betwnn 90 an 100 fps with 65 or above 1% lows so what can I say I expected it to perform much worst considering this was released just before our CPU and there were talks even at that time that performance was lackluster CPU-wise.
Next on my list here I have Microsoft Flight Sim 24 to benchmark but I seem to be lacking the video footage and framerates data; oh I know why that is; because I refuse to install Microsoft's XBOX spyware on my machine so unfortunatelly this one was a skip. Shame..
Moving on with another CPU intensive game we tested Mount and Blade 2 Bannerlord on a mix of settings that will hopefully hammer the CPU and we got 90 fps on average and 50fps 1%lows this was in the built-in benchmark that is meant to represent a larger battle from your late game.
Next we have the first game that made the system struggle that is Cyberpunk2077 where in the benchmark we got 110fps on average with 62 1%lows with any settings ranging from low to ultra and only maximum raytracing moved the bottleneck from the CPU to the GPU. But one quick thing about this game's benchmark is that it's shit and that is because CD Pprojekt in their clownfiesta decided to only account for a medium GPU load and a minimal CPU load when designing it. And now in game we can see that we drop down to 48fps on average and 27 fps 1% lows and that minimum requiremnt of a 1600 starts to make sense. Why not simulate in intensive load in the benchmark to begin with.
And finally let's move on to the game that broke our 6-core that being City Skylines 2 but we had to give it a little bit of help so we loaded a 100K population save file, set the simulation speed to the maximum and set simulation speed priority over framerate in the settings. This brought us down to a crawling 18fps on average with 11fps 1% lows. This games says that it should run on a 2600X as a CPU minimum requirement and I say that a 2600X will not save this apalling performance. I don't think you'll be seeing 30fps here even on a 2600X you'd surely have to move to a Zen2 6-core at least to save this.
Okay so back to our question are 6 cores enough for gaiming in 2025. Finally we can give an answer and our verdict is: more than 90% of times yes and the rest of the percentages go out to whoever is still running an AM3 six-core and to CPU intensive games like we have seen here from Cyberpunk and City Skylines.
That will be it for today; a bit of a longer one this time around but it had to be in order to explain how we went about it and the system. If you want suggestions for futher improvements for these AM4 6-cores consider the following: 1. use a 1600X instead as that is supposed to be better binned and you should be able to reach higher frequencies during OC. 2. Try using downclocked 3600 ram that should enable you to achieve lower timings. 3. Do the overcloking and testing while using a better cooler preferebly one that is not cheaper then our 18euro one and 4. Consider a high end X chipset overclocker's motherboard to conduct your test on. And that's about it from us if you have other suggestions leave them in the comments. It's been Kala with A64 like and subscribe if you enjoyed and if you have any questions leave a comment. Thank you for watching.