@pz123456 wrote:
I guess we dont pay 2000$ to run stuff on stock settings right?
Actually that's EXACTLY what you do.
In early days of computing overclocking was exactly meant to do the opposite. The intention was to spend less on a cheap rig and then operate it outside specifications to reach performance of a more expensive rig. So you could run a Pentium 60MHz at 90MHz adding more voltage and cooling. Risking stability or even hardware failure. But if you're lucky in the silicon lottery you might end up with a cheap system providing the same performance as a system which is nearly double the price.
Also back then a Pentium 90 and a Pentium 60 might have been manufactured on the same wafer right next to each other and just during test it was declared 60 or 90 MHz based on validation results. So it might either just have failed the upper specs closely or it might have barely met the lower specs. So it's a lottery of some kind. If you are lucky the lower spec CPU might just have failed because it got slightly too hot and therefore cannot be operated with the stock cooler. Slamming another cooler on it mitigates this problem and therefore operating at higher spec might be acceptable.
Nowadays things are different. Manufacturers are in a more competitive environment and start to use those "safety margins" in their products for their own purpose. Even "hand"-selecting silicon (i9-13900KS for example) is done by manufacturers to provide the last bit of performance uplift. AMD AM5 CPUs will also always try to operate at maximum performance at the edge of physical limits. Since the introduction of boot clocks these boots were typically used to make use of every headroom you got in thermals including cooling an chip power distribution. This is why applying power-savings like PBO curve optimizer with negative values is potentially increasing the performance. As power consumption lowers the chips will automatically keep boost clocks for longer and therefore perform better. No need to change fixed multipliers and similar parameters as in the past. There is no real "overclocking" headroom any more. Manufacturers simply found technical possibilities to make use of these safety margins themselves.So the overclocking scene is more about trying to optimize or force the hardware to shift the limits of this headroom. So overclocking focusing on increased power targets and similar options as well as improved cooling.
Despite the hunt for better benchmark scores these optimizations are of course totally killing the efficiency. Best seen with top of the line CPUs like 7950X and i9-13900KS. They do operate way beyond their sweet spot. My 7950X is performing about 38000 CB23 score at stock and burning about 250W on full load. Running it with PBO 105W power limit lowering power consumption to about 170W (roughly 66%) still delivering about 36500 CB23 score (roughly 96%). So from the PBO optimized standpoint (170W, 36500 CB23) it's taking a 47% increase of power to get 4% of performance uplift (250W, 38000 CB23). And I haven't seen any extreme overclocker results going way beyond that except for extreme systems using liquid nitrogen or similar. Nobody is asking then if those systems can be used in day-by-day operation or about power efficiency as the target is always the highest possible clock - even if it's just for a split-second.
Having said that there is of course still some (small) headroom in safety margins provided by chip manufacturers. The goal is of course still to provide a 100% stable platform when operated at stock specs. You certainly don't want OEM systems to randomly crash, garble scientific data in the field or just randomly reset or bluescreen. Having said that you can still eat up those margins to gain a bit of performance at risk of g getting an unstable system. And this is exactly what people here do and they call it "overclocking" - which is sort of true but you have to be realistic here. All you do is to remove safety margins.
The uplift you gain typically is within a single-digit percentage between a fully tuned and a stock system if you still want your system to be usable in a day-by-day basis. You might go beyond that and even get it POST and do a stable CB23 run to get a nice screenshot of your tune to complete a benchmark before it crashes again. You always need to consider that most people (including famous overclockers) hunting for a benchmark score. I guess none of them is operating such a system for daily use. Even the article-writings, video-editing and post-processing/publishing is typically not done on overclocked systems.
Stability also depends on the workloads. I know system configurations for my rig which perfectly boot into windows and complete CB23 with no issue. However they are crashing with an MCE immediately if I try to run the Firefox installer or on other specific workloads. I was operating my system at settings where it seemed to be stable but occasionally I got Firefox tabs crashing. Sure I can blame this on Firefox and I can claim bluescreens to be Microsofts fault and blame it on bad drivers or bad code. The truth however is that a bluescreen often is related to bad memory lines caused by out-of-spec hardware.
Gamers might accept some instabilities as they typically always run the same games for quite a while. So as long as their favorite game does not crash they are happy. Even if it occasionally does nobody will be able to tell if it is crap code (which unfortunately is the case in many games) or denuvo going nuts or any other reason. Most people are not able to hunt down a bug to even find out if it's faulty hardware or out-of-spec-hardware or a real software bug.
When I see some people here reporting to run a 7950X with PBO curve optimizer -30 I am quite sure this system will not be stable under all conditions. From what I researched and my own experience there is barely any 7950X out there which can do more than -20 while being stable on all workloads. Yes I can perfectly boot mine with -32 but occasionally getting hard system resets during some workloads (my workloads typically being VirtualBox, Windows, Linux, Software development, some gaming, 3D modeling etc.). So I found it to be stable up to -18 only. Same goes for memory. Some aggressive timings will just boot fine and I will see some occasional Firefox tab crashes or program/game failures but only very occasionally seeing some bluescreen. So I can of course blame it on the software. Or just accept that I need to tune down my settings a little to complete a night of memtest86+ without any single error.
Things are not black-and-white any more. In early days you would change your Pentium 60 multipler from 60MHz x1 to 60MHz x1.5 and either it would boot or it won't boot at all. Nowadays you might get a bootable system which occasionally facing internal errors. Some of which might be correctable and some won't and case crashes and reboots.
At the end you have to ask yourself the question if you want to get the last 2-5% performance gain and potentially sacrificing stability. But one thing is for sure. It does not make sense to buy a $2000 righ just for overclocking just to gain another 2% of performance. In this case you would be better off investing this money into a dual-socket system or (if not already looking for top-of-the-line) just step up to the next higher tier hardware (e.g. go from 7600X to 7900X).
I am aware that this is probably not what one wants to hear in an overclocking forum. Specifically not from ASUS as they want to sell you $1500 "overclocking"-mainboards which provide features only handy for extreme (liquid-nitrogen level) overclocking. I personally went for one of the most cheap X670 (non-E) boards and not missing out on anything. Sure some extreme boards might have more board layers and therefore could (potentially, no warranties) be stable at slightly higher out-of-spec configuration but still you won't gain substantially.
Personally I am more disappointed at the mainboard lineup these days. I mean you cannot even find any entry-level mainboard with POST-code display - even when speaking about $250 entry level. Seriously a feature which is almost for free for the manufacturer. But not as fancy as 20 ARGB headers. And this is what they sell nowadays. Overclocking boards which feature OLED screens, RGB all over and fancy looking heatsinks. Sure heatsinks at least have a value unless they are purely decoreative and I would rather prefer functional ones with actuall dissipation fins rather than flat ares to print advertisements on it. Overclockers need stable and beefy hardware to run out-of-spec hardware. So for sure good VRM is a nice plus, but let's face it... 24 or more 60A stages are just overkill despite extreme liquid nitrogen cooling. And if I want to do this kind of extreme overclocking I am not looking for fancy OLED but rather want to have a bare-bones board. So I would put this in a fancy "home gamer" line for people not really willing to do any kind of serious overclocking but rather make things look fancy only.
So actual overclocking in that sense is dead. Long live the efficiency tuning and optimization.
