What a good video. Incredible efficiency after undervolting, losing 10% gaming performance with a 50% reduction in power draw.
The value of the 13900K for just gaming is terrible, though, just as it always is for the i9s. This is a productivity CPU.
But the i5s and i7s will offer amazing value for gaming. I think AMD will have to launch the X3D versions at the price of the regular ones.
Yeah der8auer does great content in general buildzoid as well really cover the finer details that others often don't get into much or rarely cover that pretty routinely are like and now I'll show you how I go about trying to nearly break the hardware for a 0.1% better benchmark result. The suggestions by der8auer on min/max power limits were very close to what I had suggested for 13600K at 85w/170w. They are probably pretty spot on mostly too for the 13900K model I can see where he would've settle around those figures for power limits for the hardware in mind and just general expectations around it that are reasonable compromise.
I think I'd probably play with it a good bit if I ended up getting a 13900K chip 102w and 153w might work really well or 68w and 187w. They might work better or worse and depending on the individual users usage and expectations of the chip which der8auer explains in the video and fully agree upon he's really quite level headed more than you might expect for a high level enthusiast. He's usually pretty well spot on with his analysis which is refreshing.
I think you've got a degree of subjectivity on ideal sweet spots for min/max power limits for a given chip and der8auer really tried to convey that to the audience to digest. Where I mentioned above it might work better in regard to 13900K and applicable as well to 13600K, but maybe slightly different figures to aim for I think they would yield a bit differing results in key area's. It's kind of a case of do you want to prioritize a bit more base frequency or boost frequency depending on which you pick and there are things to consider around that with thermal limitations in mind which is sort of where I think 102w and 153 might work better or would in situations depending on a persons cooling setup however 68w and 187w in other instances.
I think in der8auer's case with cooling he used 68w and 187w might help slightly over the 90w and 180w overall it adds up to a bit less wattage average which is good since it's throttling anyway and also allows for higher thermal boost wattage and lower idle however 102w and 153w might work better in such scenario if 68w with 187w just exacerbates the throttling more and cause too uneven frame time variance at the same time.
Looking at the frame time variances in Steve's results in FarCry 6 for example they are pretty stretched and I think that's a result of the low base and high boost frequency arrangement of Raptor Lake design which in some instances will result in more pronounced erratic behavior like that and probably temperature related in part. In Gamer Gandalf's instance I'd be curious what raising and lowering the LLC settings and comparing does. Lowering the LLC could smooth out the erratic behavior a bit because it slightly undervolts under load scenario's like gaming that helps with temperatures and is easier as well on motherboard VRM's so overall you get a bit of a efficiency gain and lower thermals, but you have to be more careful about instability if voltage drops too much however you have less voltage overshoot with a lower LLC and that can cause really bad voltage spikes and lower efficiency. Basically worse frame time variance of the voltage delivery by the VRM's.
Someone on TPU for Alder Lake had a post power limits and efficiency at different wattage figures scaled across the same Blender workload and at the time it looked like 65w look like kind of the peak sweet spot for minimum power limit for the older architecture. A few things have changed and gone up since with Raptor Lake in terms of frequencies so around 80w to 95w minimum being ideal seems about right with the higher frequencies involved. Much better overall performance for wattage draw though relative to previous generation proving once again how much better the changes to E cores are and improves to IPC of both core types along with better cache design.
Intel actually change the cache structure pretty close to how I thought they might make changes to balance considerations around cache misses between the P core and E core die types in regard to L1 and L2 caches of each. Combined with the processor scheduling and a shared L3 cache you can kind of do foreground/background role assignment between either to optimize base frequency and boost frequencies of each die type optimally short duration high boost ST and long duration low boost MT being the general premise of P core and E core design nature and it appears like Intel has tried to do just that. You could also reverse that role structure to much like you can reverse background and foreground process scheduling in window with time slices.
I am pretty interested in the 13600K in particular though the DDR4 board options aren't as appealing on features unfortunately. I really like more full feature workstation boards more like Aero D z670 and the new z790 ProArt. I'm not a big of cut down micro ATX and ITX boards. I don't really have a big issue with micro ATX though I feel like in the modern era they've gone downhill a bit on designs in terms of PCIE slot functionality to incorporate in board M.2 slots. Perhaps if they start going back to them have a reasonable amount of full length slots then combine that with better rear I/O USB4/TB4 ports on the rear in place of a truckload of onboard M.2's I'll take a look at them again more. I think micro ATX and ITX could stand to have more rear I/O USB4/TB4 ports to make them a bit more versatile. Early micro ATX boards I didn't feel like I was really sacrificing anything crucial, but modern ones I can't quite say the same about.
