Mr. Fox

The Ryzen fanbois are not going to like this. Starting from the meat of the video.

Yes, it should be able to do at least 8000 C36 and likely C34. My X870E-E Strix was able to and my MSI B850MPOWER is able to do 8000 C36 with a 48GB kit. It will depend on the IMC and memory modules as well.

Still respectable performance for such an old (and very affordable) GPU. And, DDR3 quad channel is on the heels of a modestly tuned DDR5 system.

What BIOS are you on? If you are on the latest it could also be causing you issues. A number of people (me included) saw reduced stability and performance. This is the best BIOS for me, and it is available for the Dark Hero as well. If you are on a newer BIOS you will need to flash using the renamed file and do the BIOS recovery with the button on the back of the I/O or the motherboard firmware will tell you it is too old. ROG CROSSHAIR X870E DARK HERO TEST BIOS 9961 based on 0701 Do not use older versions of cmo files, they may cause problems with agesa 1300a bios 1. Update AGESA version to 1.3.0.0a 2. Apply new odt rules for 16x2 6000cl26 1.4v and 1.45v kit Download

The first 9950X that I owned maxed out at 2175 FCLK and the 9950X3D that I returned for a refund (a horrible silicon sample) had very high voltage needs, a weak IMC and it would not boot with FCLK at 2175.

Just for Giggles, check the SPD tab in CPU-Z and confirm it shows "1" for Ranks. I have a V-COLOR 6000 C26 32GB kit that is dual rank and it will not boot at 8000. (You can also see memory chips on both sides of the PCB.) Dual Rank DDR5 sucks at overclocking. I never would have wasted money on this crappy V-COLOR memory kit if I had known it was dual rank. The voltages you are testing with are too low for 8000 and won't work. Also, you're going to need about 1.600V for VDD and at least 1.550V for VDDQ. Set them both at 1.600V to start and if that works you can try lowering VDDQ. (MSI and Gigabyte BIOS makes them the same value by default and you have to manually change them to different values, so you could leave them both at 1.600 with no problem.) Leave VDDIO on Auto. 1.400V is way too low for 8000. Auto should take it to roughly 1.475V.

If you are using the latest BIOS you might want to test something older. I think the version numberings are the same on the Apex and Hero. 9961 is one of the best for memory and CPU overclocking. Some of the latest with newest AGESA are crappy and hinder CPU performance. (Edit: Not sure if older BIOS is compatible with 9950X3D2.) 6000 and 6400 being the "sweet spot" for AMD is a hold-over myth from the days that Ryzen motherboards couldn't handle 8000. (They made up that excuse rather than admit that memory overclocking sucked.) For the best gaming experience you should be targeting 8000 with tight timings and latency below 60ns. It doesn't matter a ton with X3D but the higher bandwidth does help with the "AM-Dip" that Jufus the Dufus talks about. Turn off GDM and set Bank Refresh mode to "Normal" (not mixed). Feel free to give these Dynamic OC Switcher and 8200 memory tuned settings a try and see if they will work for your CPU. Compare my V/F curve values (not necessarily the SP rating) to your CPU to see if the voltage needs are close to the same. ZenTimings shows all of the values that matter. Leave the ODTs on Auto. ASUS does a good job of setting correct resistance based on the detected memory modules. Should not need to mess with ODTs unless they are modules ASUS hasn't optimized for in their BIOS tweaks.

You should be getting more than 44K in CBR23 at those clock speeds. Like 46.5 or 47K. This is with dynamic oc switching (MSI version of it) with my 4585PX. Here is with the Apex with DOS enabled.

I am glad you are enjoying it. If you are using PBO it won't get too hot because the voltage stays low and it clocks down to around 5.1 GHz all core in Cinebench and other all-core stress tests. If you use Dynamic OC Switcher or set a manual OC and manual voltage (necessary for fixed core clocks) it will run much hotter. If you set 55x on your preferred CCD and 54x on the lesser CCD with enough voltage to hold those clocks in Cinebench, your temps will go way up and you will have about a 48K or 49K score in Cinebench R23. PBO works great for gaming and light workloads. SP118 is about average (probably the most common SP rating) and a totally decent silicon sample that is worth keeping if it will run 2200 FCLK stable. As I mentioned before, the SP rating changes based on the CPU temperature at the time the BIOS is flashed. If you were running chilled water and flashed the BIOS and it hold SP118 that is good. If you got the CPU toasty (like a Cinebench 10 minute stress test) and immediately flashed the BIOS it might show 119 or even 120. The per core voltage shown on the same screen as the individual core SP ratings is the more reliable measurement on AMD because of the fluctuating SP ratings that vary based on temperature at the time of flashing.

Yeah that was kind of miserable day. I was in DC and Richmond all week. I had direct flights from Phoenix into Dulles and back. I had to return the rental car at noon to avoid adding an extra day. I already had to wait until 5:00 PM (expected) for my flight home, but the added 3-hour delay due to a mechanical issue made it that much worse. Had they not been direct non-stop 5 hour flights I might have ended up spending the night at the airport and being put on a standby due to missed connections. So, it could have turned out a lot worse than it did. Yes it'll be something new and fun to play with while we wait and see what happens with Nova and Zen 6. I expect the overall performance to be remarkably similar to my AMD builds. From all I have seen as an observer they trades blows and come out close to the same in most benchmarks, one randomly beating the other depending on the benchmark or game being run.

Finally landed in Phoenix so I should be home in about an hour. LONG day. I arrived at Dulles at 12 noon EST.

Having a nice Friday sitting for 8 hours at the Dulles Airport waiting for a flight home. *not*

Considering that the Apex motherboard was already slightly discounted having the free NVMe included made it seem like a no-brainer.

Um... uh. Ok. The NVMe was free. I guess that makes it ok.

I have to drive past Microcenter on my 1 hour trek home from the airport at the end of each business trip. I have had to exercise a lot of self-restraint to not go by on the way home and pick one up, along with a Z890 motherboard. The only thing that has held me back is the motherboards that I think I would find suitable are stupid expensive. I'm headed home tomorrow night and I already was checking the Microcenter website last night before going to sleep to see if there was something that I couldn't live without in terms of motherboard options.

Naughty kitty.

The Asus SP rating on AMD is less reliable because it can vary by temperature when the calculation is performed. It's kind of weird. The hotter the CPU is at the time the calculation is performed increases the SP value. I have only seen a change in the SP rating by 1 with my CPUs. My 9950x has an SP rating of 119 it goes down to 118 if I flash the BIOS when the system is totally cold. It may be different for the 9950x3d2 but 118-120 is about average. My 4585PX is SP120. However, the VF curve doesn't fluctuate as much. You can look at the Prediction area at the lower right to see what voltage value is displaying for a manual core ratio. Try setting a fixed all core ratio of 55.00 and look for a voltage value of 1.400V or less in the Prediction box and see if it will boot with the FCLK set to 2200. If it will not boot with 2200 FCLK I wouldn't want to keep it Use that first video that I shared with you the other day in this thread (added below) for tuning a per core PBO Curve Optimizer. Stronger cores will have a lower voltage value by default and will accept a smaller negative offset then the weaker cores. You want to see as little variance between core quality as possible when you look at the VF curve page in the BIOS. Being able and to set a deeper negative offset on individual cores that have a higher default voltage value on the SP rating page isn't a good thing to see. The wider the SP rating variance is on each core in a CCD is less desirable.

I hope that your hard work and faithfulness is recognized by your employer and pays off in compensation equal to the time and effort. It generally does when you are working for a respectable organization. Good companies know that good employees are hard to replace and need to be retained. It's sad how things are turning to crap for our hobby. It's hard to get excited about anything from AMD, Intel or NVIDIA. Not only because of abusive, predatory pricing. Everything is riddled with compromises and mediocrity in one or more areas. Hardly anything is awesome. Even the extremely retarded price of a 5090 gets you a GPU that has been deliberately castrated in terms of power and voltage controls needed for overclocking excellence, a dangerously inept power delivery mechanism (12VHPWR) and dishonest, unreliable warranty handling by the OEMs and retailers. It applies to DDR5 and NVMe failures as well as GPU failures. "Sorry, we took your money, sold you a defective parts, but you're on you own now because we make more money selling the same parts to AI data centers for 150-300% more than what you paid for them." Now even Best Buy is selling "open box" GPUs with no core and memory. Clearly they did not bother testing anything before putting it on the shelf. This video from Alex should help the victimized customer get a refund. I'm surprised it wasn't handled and made its way into a video like this.

A 9950X3D2 will likely degrade your gaming and overclocking experience. It is generally not good at either task. It is interesting to see people buying them anyway, despite the reality of the situation. It is the Ryzen equivalent of a Quadro GPU. Not the right tool for the job.

Tony is fixing borked DDR5 modules now. I need to get myself one of those programmers. I do not use RGB software anymore for changing RAM LEDs to white because it destroys the SPD. I've had several DDR5 kits corrupted by it.

You really need to get the memory kit that I recommended if you want to have great results on AM5. Anything else is going to be a crap shoot. All of the AM5 memory overclocking record holders are using what I recommended and ASUS has built their firmware around it for unrivaled results. If you get something else YMMV and may not be as good as you might want. I don't know of anyone getting good results on a Ryzen build using Corsair memory kits. You're going down a totally different rabbit hole with AMD and I'd hate to see you make it more frustrating than it will be otherwise by making a wrong choice on the RAM. AM5 also copes much better with 2x16GB than it does with 2x24 or 2x32 kits. Overall the platform is finicky and requires more effort while yielding smaller rewards. It's not horrible by any means, but it is not on the level you are used to in terms of ownership experience. The pricing of memory and SSDs right now it off the rails and beyond stupid. I paid about $300 for this G.SKILL 6000 CL26 kit and now even the trashiest crap DDR5 memory kits with Micron and Samsung ICs are selling for 2-3 times what they should be. I have a V-Color Hynix 6000 CL26 kit and it performs like garbage and sucks at overclocking. It's back in the box and will only be used if necessary due to a hardware failure. Pay the extra $50-$100 for the recommended G.SKILL kit and avoid regrets. No. I would not purchase a 9950X3D2 CPU. There would be zero benefit for me. Total waste of money from what I can tell so far. They do not play games better (maybe actually a little worse honestly) and unless you get super lucky with a god bin sample they are worse at overclocking. The CPUs with 3D cache also have a small memory and L3 cache latency penalty. X3D2 is very niche with no practical application for gaming. My favorite Ryzen CPU is 9950X (non 3D). All Ryzen CPUs are literally designed to boost as far as they can with 95°C as a threshold. One or two CCDs and regardless of core count they hit 95°C under max load and only delidding can tame them. They are power, current and multiplier limited processors. The CCDs (cores dies) and IHS are small and have minimal surface available for good heat dissipation. The only meaningful benefit to going AMD is socket longevity and a longer CPU upgrade path that does not require replacing the motherboard. Otherwise, I would say go with Z890 and 270K Plus. If I could do a straight across trade for that without spending any money I would. Rumors are that Zen 6 is going to be awesome though. It might be but nobody knows for sure yet. I know that this commentary probably all comes across as negative. Let me balance that by confirming that I do not hate my AM5 systems at all. They are decent and I have learned how to like them, but my expectations going into it were higher than they should have been and I am just being very frank with you about what to expect. All of the former Intel enthusiasts on overclock.net that made the switch to X870 from Z790 that I have communicated with echo the same. By no means bad, but rather mediocre overall. If you go into it with low expectations it will be easier than if you go in expecting it to be everything you loved about rocking an Intel system. If you go into it listening to AMD fanboy fables you're going to quickly get pissed off and frustrated while you learn how to make it work to the best of its ability. The performance tuning gains are smaller and require more effort. The X570/AM4 experience was an utter dumpster fire for me. That has not been the case with AM5/X870E/Zen 5. If you decide to go with a Ryzen build, I highly recommend watching both of these videos. The first is for CPU tuning. Very useful. The second is some memory tuning and other myth-busting that goes against what many of the fanboy gamers will tell you. This guy knows what he is talking about.

LOL... my WTB thread was 1 year ago today... 6/18/2025. FreakFreak is hunting unicorns. He posted this today. I wish him better luck than I had (which was none). What he is asking for is even more rare that what I was looking for. An X3D with a strong IMC and high core clocking capacity is even more unlikely. What he is asking for is also less relevant... X3D CPUs are largely memory clock agnostic in terms of the effect on performance. The benefit of memory bandwidth is mostly negated by the cache.

100%. I have not seen them priced that low for a LONG time. (I paid like $300 before the AI-retarded nonsense ruined everything.) @tps3443 better get them before they are gone or go back to $1200+ again. Edit: Good find Brother @electrosoft. Those are the "H" variant with 1.450V but the advice is still the same... 2x16GB. The Amazon listing appears to include typos or conflicting info as to whether they are G or H but the voltage indicates H. The G variant is 1.350 or 1.400V with the same timings. To add to my previous comments about older Intel XMP kits on Ryzen, the default SPD timings are not good for Ryzen as I mentioned, but the ODT values are also wonky and don't work well. So, you are chasing not only Ryzen-friendly timings but also different ODTs for Ryzen versus Intel. ASUS Crosshair mobo firmware sets all of the right ODTs specifically for these 6000 CL26 G.SKILL kits.