Btw - the Performance and Ultimate power plans on both single and dual CCD allow for throttling down of the cores when not used (see screenshot below of my PC running Ultimate power plan). No extra heat is actually generated - at least from what I can tell with my 9950x3d.
What is different with these 2 power plans is that your cores will never be parked which is just a special type of deep sleep state. Parked cores are not conducive to a high performing PC. Also - parked cores in CCD1 are no longer necessary to get games to run on the vcached cores (CCD0) on a dual CCD x3d cpu** (this was not the case when x3d technology was first released by AMD).
** Assuming a recent bios, AMD chipset drivers and Windows build
Pleased to report that yesterday’s BIOS update was a complete success. Confirmed the OP settings continue from what @TenPatrol recommended to me when the PC was new last Feb so that is done. Changed the power plan to Performance and am flying the 772 now to ORD from EWR. Everything is running beautifully, very smooth from PC boot to cruise at FL400. Feels like a new PC, very happy, thanks all!
Happy to help
The most important thing is always satisfaction - that feeling when everything works perfectly, especially when done thoughtfully and following proven guidance. I’m glad everything is running smoothly and that you’re enjoying the flight!
Well guys I have to take a break away from the machine and report what I have been up to on the OCing front. I appreciate all the comments especially the OCing video recommended by Gimbaxis that really uses the same machine as I.
So for the past two days I have very carefully measured what happens to CineBench on the basic parameters cited in the video.
I have a baseline of 38598 with a SD of 158 after testing which DOCP memory setting is best (It is the “Tweaked” by a slight 100 point margin… but see the baseline SD).
I ran the CO at -15 and had a system crash that required pulling the battery to recover. So I went back to -10 and the system survived and gets a number of 36311. So far so good. I creeped up on -15 by going to -12 and the -13 and saw 36637 and 36428 respectively. Since I started seeing a fall off in performance I was wondering if this was the sweet spot so I decided not to push on to -14.
To my eye I do not see any drops in performance in effective clock frequency in HWINFO 64. I was thinking I would see some “clock stretching before a crash).
Originally I was going to try the ASUS AI solutions and did. There are at least two: one/maybe two in the BIOS and a second in their Armory Crate Software. The BIOS one only resulted in 36154 but I did notice my very hot core #7 got above 95. The one from Amory Crate crashed and had to pull the battery again. Their AI is a dead issue for me now.
Questions:
In the video charts they have an entry with 5 SCA that they do not define, but it seems to help.
I plan to fly this for a bit and see if it is stable at -13. I read that this represents a drop of 3-5mV/step so a total drop of 39-65mV. That feels noticeable to me. Is this a good order of magnitude for an under-volt?
Need a sanity check on moving forward so please check if the following reasoning is correct? I tried to use a Boost Override of +25 with the CO of -10 and the system hard crashed again (battery out). Should I work on individual cores and what to do with the Core that runs hot? Seems in the video they look for clock stretching on individual cores at -30 and then reduce the CO on those cores but I do not see this at -13.
Writing a lot but this is an interesting educational endeavor. Thanks for the help.
My ram can do any bench test and stress test for 24 hours and any game without errors but I can’t use it at 6000mhz with FS24. Only way to run it stable is lowering to 5400mhz.
The trouble with benchmarks is they stress test one compnent at a time, CPU, GPU, VRAM, memory. FS24 puts stress on all of those components simultaneously which probably makes it the best stress test of all.
Well I have been deep in my ASUS BIOS and have a number of findings to report.
As mentioned above with stress tests not all the Curve Optimizations I have been able to achieve have stuck for 24h. I thought I could get it to -13 without a crash but it just took time for the crash/system shutdown to occur. I could get -10 to work but it crashed immediately if I added in a +25 Boost Clock Override.
In reading about similar efforts with the Ryzen 7950X there are reports that individuals cannot get beyond -5. There seem to be individual hot Cores and I can see them on mine (#7). I tried to individually adjust and that failed too.
So I have gone back to the latest version of ASUS OC AI. This is only a couple days old and is written up nicely here:
So I did what they suggested and there are a huge number of AI tweaks and readings that are now available in the ASUS BIOS. The most interesting should what they did to individual cores:
Core 7 is treated differently. So it does core by core tweaking.
Anyway, with this OC AI in place CineBench goes from 35598 to 36181 or an improvement of 1.6%. Not a huge benefit but I think this is par for the course.
Again thanks for starting this thread and here is the latest ASUS news.
Question: If you have 4 NVMe’s, are you not limiting the bandwidth of your PCIe 5.0 x16 slot by using the M2_2 slot? Just curious. I was considering this MB but that was a show stopper for me since I also use 4 NVMes.
Yes, I believe on the MSI X870E Carbon, the main upper PCIe slot shares lanes/bandwidth with the lower PCIe slot and M2_2, so if either of those are populated, the main PCIe slot (where you normally put the GPU would be 8 lanes instead of 16.
That reduction in pcie lane bandwidth isn’t a concern even for a 5090…I’ve seen testing done on various posts and YouTube videos that a one notch downgrade like you would get here isn’t anywhere near material (as I recall it was maybe a 1-2% drop in observed GPU performance under max load/stress tests so not a concern in real world scenarios)
Ok thanks. I was wondering how much of an impact it would make because going from 16 to 8 lanes sounds like you would get a much higher performance hit. Sometimes the numbers do not tell the whole story.
@GimbalAxis@tempestornado23 You might think this is a limitation, but I’m about to surprise you - I’ve already bypassed it.
Nope, mate - it doesn’t matter for me, since I’m not using the M2_2 slot
You don’t need to worry about that - I just didn’t install an NVMe drive in the M2_2 slot
Just to clarify - I’m running the fourth M.2 drive through a PCIex4 M.2 adapter. The adapter sits in a PCIe 3 slot.
MSI unfortunately placed the additional PCIe power connector right next to that slot, so the adapter physically blocks it. But honestly, that’s not an issue for me. I don’t need that extra PCIe power connector, even with my current RTX 5090. There are also vertical M.2 adapters available, but since my RTX 5090 works perfectly fine I’m not considering getting one.
That additional power would only really matter if I ever decided to run a second RTX 5090. For a single GPU setup, everything works perfectly fine.
I’m running my 5090 in the top pcie slot and I have a smaller 2nd GPU running in the bottom pcie slot…both are at 4x8 on my x670e board. I needed this setup due to having 8 monitors (3 for the sim screens, 1 for navigraph, 2 for pop out gauges, 1 for say intentions ATC and 1 for hardware monitoring software/misc). I’ve tested the 5090 in this setup and it’s hitting 98%-99% the same performance as when I had just one GPU in the system.
A bit off topic here but just wanted to mention a small investment I made to improve thermals.
I have the 9800X3D with the Arctic Liquid Freezer III Pro 240. My case was not big enough for the 360mm. My CPU temps were not that bad sitting at 46-47C at idle and in the low to mid 80’s with a -15 curve at load but still wanted to try to improve the cooling a bit.
The case came with 3 front RGB fans with a top speed of 1200-1300 rpm that I just set to full at all times.
Decided to swap them out for 3 Arctic P12 Pro A-RGB fans 600-3000 rpm. Best $70 I ever spent. I keep the idle speed at around 1400 rpm and the max speed only at 80% (2400rpm) when CPU hits 80C to help with the noise.
Chassis Fans 3,4 and 5
CPU now idles at 42C and in the high 70’s with CInebench.
@TenPatrol On your Arctic LF3 Pro 420, did you use the single cable or the three way cable? Do you mind posting your bios fan settings for the aio?
I have both the LF3 Pro 420 and Corsair Titan Rx420 already in “inventory”, but leaning towards the LF3 so I won’t have to run iCue on the sim pc. Can use the Titan on a 9950x based build happening in April where I won’t care about running iCue.
