Mr. Fox

If you want to see the difference 2200 FCLK makes, run Y-Cruncher 2.5B and look at the speed improvement compared to 2167.

Hybrid OC (or Dynamic Overclock Switcher aka as "DOS" in ASUS terminology). Basically tune your PBO to the best stable undervolt that gives the highest low core use clock speed for gaming (which is the only thing PBO is actually good for) and them set your per CCD manual overclock and VCore/VID and power switch threshold. If you followed the videos I shared with you previously to tune your PBO then the tedious part is over and the core overclocking is much easier. You will want to set LLC Level 6 or 7. LLC does not apply to PBO, only the manual OC. For the manual overclock switch I have all cores set at 54x for each CCD and 1.275V VCore and VID voltage. I have the switch threshold set for 120A, so it runs PBO clocks until the CPU pulls 120A, then switches to the manual OC. Using the video I shared with you for PBO per-core tuning, this is where I ended up with my per core PBO negative offset (which may not work well for your dual 3D cache CPU, but you can certainly test and see). I do not use Curve Shaper, only Curve Optimizer. Curve Shaper doesn't really do anything for me other than contribute to freezing while watching YouTube videos and web browsing and tuning Curve Optimizer properly is already a pain in the a$$ without complicating matters with Curve Shaper. For illustration, here are the settings I used with the 9950X in the X870E Apex for the hybrid/DOS daily driver settings. I later increased the switch threshold to 120A instead of using the ASUS recommended value (89A) in the "Prediction" window. In case you missed the post with the PBO tuning video, it is posted here. Do the PER CORE tuning (versus per CCD) to maximize your results. Otherwise, per CCD will limit your maximum stable negative offset to your WORST core on each CCD. When you tune the Curve Optimizer offset for the maxium negative offset each core can run stable, all of your cores will boost higher, the CPU will run cooler and be more stable. It's a pain in the butt, but worth the effort if you want the best overall result. Your best cores (that use less voltage) will likely handle less undervolt (negative CO offset) simply because they are already using less voltage by default. If you set all the same on one CCD your best cores will under-perform due to voltage starvation, or your system will be unstable (random lockups in light loads like web browsing and watching YouTube videos).

Unless the core overclocking is worse on the one that runs 2200 FCLK, I would keep that one and send back the one limited to 2167 FCLK. If memory serves me correctly, they are the same SP rating and have a similar V/F curve. The one that cannot run 2200 FCLK would end upl being your weakest link in the system and memory overclocking is the least important performance factor with Ryzen, and even less important with an X3D CPU. Being limited to 2167 tells you on face value that it has a weaker fabric interconnect and is the inferior sample of the two.

I just tested some of the Vallhalla PC Bifrost Alloy thermal paste (company owned by drajav at overclock.net) and it is a good thermal paste. I tested using the SFF living room build that has a 360 AIO. https://www.valhallapc.com/products/bifrost-alloy-thermal-paste

The leaders of the computer tech industry are bad people, with bad intentions, that have no honor and deserve no respect from any of us. The only thing any of them deserve is misfortune and tragedy, and I hope that both are visited upon them corporately and personally.

Don't give up on the one that only runs 2167 just yet. If the voltage suggestions I texted to you work you might be able to run 2200 FCLK stable. If they do not work, then I would choose CPU #1 that does run 2200 FCLK at 6400 C28. The 2200 FCLK is more important than the 6600 C30. That is because the FCLK is the frequency that connects the CCDs and IO die and being limited to 2167 is not ideal because it becomes a bottleneck that affect more things adversely than just being able to run a 200 MHz higher RAM frequency. Because these components are physically separate, they must communicate across the CPU package. FCLK dictates how fast data can travel between these chiplets. If FCLK is low, it creates a bottleneck where your memory sends data faster than the internal highway can handle. Conversely, pushing FCLK too high past its physical stability limits (usually a maximum of 2200 MHz on Zen 5 CPUs) results in audio crackling, USB dropouts, system crashes or no boot (code 00). Running high memory frequency on Ryzen has diminishing returns, especially on X3D chips. You are better off focusing on bandwidth, which is why 8000 2:1 is better than 6400 1:1. It allows more data to pass through in the same amount of time due to higher bandwidth. You will find that 6600 1:1, 8000 2:1 have similar benchmark scores, but 8000 2:1 is higher bandwidth and allows you to do a little further with performance. And, 2200 FCLK matters for both configurations. The CPU that can do 6600 C30 1:1 has a very strong IMC and a weak fabric. Stronger fabric trumps stronger IMC unless the IMC is really horrible. There is not going to be much difference between 6600 C30 and 6400 C28, but a 2167 FCLK would hold you back and probably have measurably higher latency. The fabric is sort of like a transmission. If you have a strong engine and a fragile transmission with the wrong gear ratios the strong engine is going to be of limited benefit because it cannot push the performance to the wheels where it matters most.

This isn't what is sounds like on the surface. Interesting take and I think I agree with it.

@Reciever and @tps3443 if you don't already have it, download this modded version of ZenTimings. It was previously shared by @safedisk on HWBOT and OC.net and it has a few features the normal releases do not have, including memory temperature and power, WHEA error counter and CPU die temperature. The sensors for VDDIO typically do not work unless the BIOS is set to Auto for VDDIO, but this modded version may fix that as well. (It fixes it on my B850MPOWER.) This app functions portably. No installation nececessary other than the prompt to install PawnIO on first launch. (PawnIO must be installed or ZenTimings will not work. PawnIO is a replacement for WinRing0.) @tps3443 also check your Nitro settings to confirm the BIOS is automatically assigning them as 1/3/1 and manually correct if not (shown in ZenTimings). Make sure your Bank Refresh Mode is "Normal" (not "Mixed") and manually set it if not. This will make it so only tRFC is used. tRFC2 and tRFCsb/pb are not used in Normal mode. You can set them to "1" in Normal Bank Refresh mode, or not set them at all, and it doesn't matter since their values are ignored. In Mixed mode latency is higher and those values do matter. ZenTimings-v1.42.6.zip

I hope that I can visit some day.

Sounds like your CPU is being voltage-starved based on the behavior you described. If you are running CCD0 55.50 and CCD1 55.25 you're going to need to manually set Actual VRM Core Voltage (set Core VID the same of 10mV higher) for somewhere between 1.375V and 1.425V (depending on your CPU) or drop them down to 54x or 53x depending on what voltage those core ratios need. If you are running PBO with DOS (Dynamic Overclock Switcher) enabled your Curve Optimizer negative offset might be too deep and that can make the transition from low-load PBO to high-load manual overclock is too dramatic. If the CPU has not been delidded it is going to be very difficult to run 54x or higher than 1.275V-1.300V due to overheating. Each CCD is very small and temperature sensitive due to the lack of thermal dissipation surface area. Much more so than Intel. My Ryzen CPUs tend to freeze the OS when overheated rather than thermal shutdown. I don't know why. It's scary that we can't fix stupid. Scary that some people don't even know how stupid they are. The sad part is that stupid people and their bad choices affect everything, including having an effect those of us that are not stupid. We have to deal with the fallout of their stupidity and its contribution to asinine pricing. The options available to the not-stupid people are (a) become stupid like the others, or (b) do without and settle for less. There is no value in the current pricing of GPUs, storage and memory.

Awesome. Nice bump in bandwidth and drop in latency.

BeQuiet PSU is a better thing to fail than an Astral 5090. The petite arson feces connector doesn't care what it ruins.

The 16GB G.SKILL modules are definitely better for Ryzen than 24GB sticks, especially on ASUS. Try the Strix X870E-E memory settings I used in this video. Bear in mind that X3D CPUs are always going to have a memory latency penalty, usually 1 to 3 ns. Set the "Core tunings Configuration for gaming" to Legacy and that should bring it down a bit. If the 8000 C34 settings are too tight, try the 8000 C36 settings that I use on the B850MPOWER. There is almost no difference in memory benchmarks between the two. Too small to matter.

Absolutely, 100% greed motivated. It is a predatory maneuver to bankroll a fanatical and maniacal agenda that is not necessary and would fail on its own merits (advancing AI at an illogical pace). It is a plan destined for failure for the same reasons that taxing the rich into oblivion always fails. You eventually run out of other people's money. Then the house of cards collapses. AI is not bad. It is good and useful. The people and companies involved with freakish and obsessive development of it are the problem. Bad company always corrupts good morals. In some ways it resembles the nuclear arms race. You have wicked nations that would abuse it for nefarious purposes, which requires decisive action to keep them in check through the threat of mutually assured destruction.

Awesome. I am glad it was nothing serious with the 9070 XT. The pricing of GPUs and memory is totally inexcusable. It might have been initially due to supply and demand, but it is so far off the rails now that I no longer believe that and think it is part of a fabricated fraud scheme that has been meticulously engineered by exceedingly wicked people, all of whom deserve unspeakable misfortune and destruction to be visited upon them... tragedy on an order of magnitude that makes a Mortal Kombat "finish him" move look like a love pat. Your plan is a smart one. Maximize what you have. I am happy and proud that we are to blame for your wanton disregard for the frugality and functional limitations of air cooling. I've tried to save a few bucks and learn to like it, but in the end it always sucks for more than one reason, and always the same reasons.

He is using a profile I tuned for the 9950X3D2 and I bet he would be happy to export the CMO file, zip it and upload it here if you wanted to tinker with it to see if it works well for yours.

Good news. Yes, SP118 is good for 9950X3D2. That is what mine was that I sold to @Reciever and the 9950X that I sold to @damric at oc.net was as well. SP118 is average or a 9950X (non-3D). My 4585PX is SP119, which is better than average for 9950X3D/4585PX (which are typically something like 115.). All of them will boot 8400 2:1 and run 2200 FCLK and run 8000 2:1 100% stable with 2200 FCLK very easily.

You will need a hardware KVM to switch between two computers on a shared monitor. That will switch display output and mouse/keyboard from one to the other mechanically. That software only works for sharing a keyboard and mouse on two running computers. You designate one as host and it works over LAN. I have not tested it in gaming to see if there is any lag on the client PC, but it functions seamlessly in normal use. You can get a hardware KVM on Amazon for a fairly reasonable price. Just pay close attention to the refresh rate supported. Some of the dirt cheap ones do not support a high refresh rate above 1080p. I have two (one at each desk) and one supports three DisplayPort monitors at 4K 240Hz and the other supports two HDMI at the same resolution and refresh rate. It was such a pain without the KVM having to manually switch between the two. Here is a link to the dual display KVM that I use.

Awesome setup brother. I'm super happy for you and glad the beast ended up with a good home. That's really strange about the 9070 XT. Have you tested it in another system to see if the behavior changes? I am wondering if maybe something with adding the waterblock went wrong. The DIMM.2 is Gen 4 and I believe it runs from one of the two chipsets, not the CPU. If you check the manual, I think there are some toggle switches on the motherboard to do something with the PCIe slots. I don't remember and never used those features, but it might be worth investigating. It might be something as simple as changing a toggle switch to a different position.

This is a pretty awesome AI video. All of the action movie heavy hitters are in the totally fake cast. It is also terrifying to see at the same time. I truly believew that people are going to end up being falsely accused and prosecuted for crimes that are caught on video created by AI that they never commited.

This is a pretty awesome AI video. All of the action movie heavy hitters are in the totally fake cast. It is also terrifying to see at the same time. I truly believew that people are going to end up being falsely accused and prosecuted for crimes that are caught on video created by AI that they never commited.