I think that MSFS benefits most from stability, and if there is extra room, a bit of extra clock cycles can help, a bit. I don’t think there is any real risk, as long as you check temps, and have decent cooling (this is why it’s a good idea to undervolt a bit, especially if adding to the boost frequency). Having PBO enabled will shut the CPU down if it hits any of the limits (Temperature, Power/watts, or Current/amps), but you still want to stay away from those limits just for stability and longevity. Up to 200 MHz doesn’t add a lot, but just check temps after changing settings.
With default settings for BIOS for the 9800X3D, I think it runs with a bit more voltage than most individual units need, and this means that when MSFS spikes, it might push temps and or power momentarily closer to the PBO safety limits (mostly that’s no problem). But there is room to undervolt modestly (eg All Core -10, or -15 on the negative curve offset), and this does create extra headroom to reduce temps, or allow for increasing boost frequency, or both, for those moments when MSFS hits the CPU and requires boost.
So I think it’s worth trying a bit of boost clock if you are curious to see what it does (it goes in increments of 25 MHz up to +200) if you are Main Thread limited, and provided you check temperatures and apply a modest undervolt if needed. It would just help a bit as you suggest. Probably more important is making sure EXPO is enabled, and if available try a High Efficiency Mode preset to tighten the memory subtimings a bit. That has impact. But as with all these settings, you need to try on your system to know how it behaves.
I’m pushing my set up a bit because I’m driving 3 screens, and want increased TLOD.
1 Like
Thanks for the information, appreciate it.
1 Like
This is an epic thread and am considering eventually upgrading my system (Ryzen 7950 and 5080) but I am thinking again since I have had nowhere near these problems and am running the 787-10 at a solid 72 FPS. Might want to stay with the devil I know.
Please note the 787-10 may be the most modern of the Boeing airliners but the MSFS implementation has a long list of bugs that may drive you crazy especially the clarity of the display… which never really got clear for me until the last NVIDIA driver. Many prefer other airliners…
Beautiful display(s), but I have come to prefer VR even in the 787 and the cockpit display is fine again with the latest driver.
Clear skys.
1 Like
The biggest benefit is most likely going to be with the 1% lows as your cpu will be able to absorb a few more of the stutters. Following a scientific methodology to tweak PBO settings as @GimbalAxis has suggested will ensure the risk is low.
I’m actually going to try to test if the boost clock helps, as I’ve just done a clean install. I may open a different thread for tuning experiments, but here’s some interesting info about how MSFS uses resources…
I've just done a complete reset, and am testing default stock settings:
- Erased my PC (I keep installers and backup on 2nd SSD)
- New BIOS, Windows 11 25H2, new Chipset Driver, latest NVIDIA 591.86.
- Stock BIOS settings for 9800X3D, no PBO or EXPO yet.
- Stock settings for RTX 5090, no undervolting yet
- Load MSFS, as if for first time, Graphics are:
- 4K 120Hz (single monitor for testing)
- DLAA at 100 (310.5.2 with preset K automatically applied) I’d normally use TAA but decided to test DLAA
- Max Frame Rate 60
- No Frame Gen
- Ultra Preset for testing with no changes
Using HWInfo, I’ve opened a few of the key metrics as graphs for a short flight - a simple circuit in the C172 from KTEB, flying around NYC for 30 min or so.
2 things stand out:
- The spikes in the CPU Temp whenever I change camera view, whether looking to the sides in the cockpit, or switching to external or drone view, there is a massive CPU spike. This is of course the MSFS function of jettisoning assets that leave your view, and processing what you’ve turned to look at
- The amount of data moving around in RAM, VRAM, and Virtual Memory. 42 GB of Virtual Memory allocated, but not sure how it’s used. I wish it was in RAM and not on disk. I have a fast SSD but I worry about the amount of wear. I’m going to re-do the flight with the SSD info open so I can see the read/write amounts.
So the boost clock may help in those moments of high demand, but the spikes show how it’s good to have a bit of headroom. This is just the C172, I’ve yet to test an airliner.
Since I was doing a full reset and clean install of MSFS, I decided to finally try to figure out how to Benchmark test using MSFS. It was a lot of work, but also a lot of fun.
TL/DR here's what I found:
- Enable EXPO. My G-Skill RAM is 5600 MHz at stock, and 6000 MHz with EXPO
- Enable PBO. Auto is good, but Advanced lets you access Curve Optimizer, and a modest undervolt of -15 reduces CPU power draw by 5 to 10%.
- GPU: A modest undervolt on an RTX 5000 series card really reduces power draw and heat, also by roughly 5 to 10%
- Tightening Memory Subtimings has a HUGE benefit. See the note earlier in this thread. I’m guessing because MSFS moves so much data around, timings really matter.
- Amount of RAM and VRAM usage seems to benefit from tighter subtimings.
- Adding extra Boost Clock helps a bit, and if combined with a modest undervolt, it would still run with less voltage and heat than stock.
Here was my process
I created 2 flight plans, both 20 minute circuits:
- Default A330, KJFK/31L GLDMN CASLE DPK KJFK/DPK.I31L
- Default C172, KTEB/06 JULEB RINNG HUDSN LBRTY LAYDE DANDY KTEB/06
I flew both flights under 4 tests:
- Test 1 - Stock
- Test 2 - EXPO, PBO, and undervolt CPU and GPU
- Test 3 - Tighten Memory Subtimings
- Test 4 - Add Boost Clock
My MSFS settings were:
- 4K (TV is 120Hz, VRR enabled)
- DLAA Preset K 310.5.2
- No Max Frames or Frame Gen
- Preset Ultra with no changes, so default Traffic was on
- Flights loaded with “Few Clouds” preset at noon.
- I have 3 screens, but used only 1 for this test
I opened the MSFS Dev Mode FPS overlay and HWInfo, and took screen shots at roughly the same points in each flight, and below are the numbers:
Test Setup
A330 at KJFK
C172 at KTEB
2 Likes
@GimbalAxis Ever since I got my RTX 5090, I’ve been thinking about trying a GPU water block myself. Do you know how that affects the warranty? Have you checked it before doing it?
Yes I did think about it… My understanding of the warranty (and I’m not a lawyer) is basically that they leave the warranty vague enough to allow you put a water block, implying that as long as the card is not damaged physically, and any issue is clearly the result of manufacturing defect rather than external cause, you have a hope of recourse… I assume it’s also vague enough for them to have an out if they want to deny warranty. I’m sure their lawyers have crafted it so they could deny warranty even if you don’t take it apart…
So it’s a risk I took based mostly on it being a fairly common practice, with established manufacturers and hardware content makers’ videos. But I can’t say either way if it’s risk free from a warranty perspective. I took photos of every step and kept all the parts labeled in the original box should I have to put it back together.
I went with Alphacool for pretty much everything, and it worked out. Taking the Asus TUF fan / heatsink assembly apart was the hard part - the thermal paste really has suction - that was a bit uncomfortable. After that, it was really about going slowly, reading each step, dry run, back off if not sure about a step…
Even though I did all this, I would play Devil’s Advocate and question if you really need it? How are temperatures with the native fans? For example - yesterday I updated DLSS using DLSS Swapper and I swear things are running better. Not sure if they are cooler, but perhaps if as MSFS becomes more optimized, all this becomes less necessary? (The question I ask when looking at my expenses list…)
1 Like
Good point mate. I need to think about that myself - with FPS capped at 60 in almost all my games and no GPU overclocking, it might turn out that a water block isn’t really necessary.
Don’t get me wrong - I’m glad I built mine, but it was partly out of fear, and partly determination to really make sure heat was not a roadblock to performance for 3 screen flight sim. The trend I saw at the time was components demanding more and more power and generating more heat, so I reacted to that (arguably over-reacted…) but I also wanted the challenge.
I also really enjoy those kinds of challenges. Makes me wonder if it’s worth doing just for the challenge itself, especially since I swap CPUs quite often…
To me the greatest advantage of liquid cooling my 3090 Ti furnace is getting heat out of the case.
98% utilization pulling nearly 450W generated a lot of it. I’m happier having it heat up a radiator on the top of the case where fans can push the heat into my room more efficiently.
2 Likes
I don’t think you mentioned earlier that you water-cooled your GPU - or maybe I missed it? Either way, I’m curious: did you only water-cool the GPU, or is your CPU in the loop as well?
CPU is air-cooled with a Dark Rock Pro 4. Heatsink has two fans in a push-pull pointing straight back at the rear case exhaust fan.
I might add it to the loop, but so far I haven’t needed any additional CPU cooling. The fan curves are pretty mild, and unless I’m running Prime95 the CPU stays around 60°C (such as in flight.)
Case temps, drive temps, and MOS temps are all really good. Part of that is due to having a whole bunch of fans (3 front, 3 side ducted over the MOBO, 2 radiator exhaust, and 1 rear exhaust. And the P/S fan, of course. Most of it I’m convinced is due to getting the GPU heat out of the case.
1 Like
If it helps anyone thinking of taking the plunge, a few things I learned:
-
Loop order doesn’t matter, at all. It’s the volume of the water block to absorb and transfer heat to the volume of liquid that matters. It’s about relative mass - the mass of the CPU and GPU itself are small compared to the mass of the blocks and the liquid.
-
I used 16/10 (outside mm / inside mm) flexible tubing, which seems to be the common size for a wide range of connectors
-
Plan how you will drain and fill before designing loop… I put in 2 disconnects (before and after the components), so I could isolate / drain just the GPU and CPU, leaving the rads and pump full / intact. That already paid for itself when I changed my SSD. I only had to drain about 3/4 of a cup instead of the whole loop, which in my case is about 1.8 L. Also, since my pump has a virtual flow meter that requires calibration by cutting flow, the disconnects make that easier.
-
I choose radiators with ports on both ends and all the sides so I had options of how to connect, and could install a drain at the lowest point for if/when I ever need to drain the loop.
-
I choose 45mm thick rads to have more surface area for cooling
-
I choose 140mm diameter fans, and the Arctic P14 Max are built well and cheap. What I like is the blades are constructed like spokes in a wheel - there is an outer rim they connect to which spins with the blades keeping them stable and rigid
-
I used the recommended coolant, no colors or additives, and so far after … I forget, 6, 7 months, no issues.
-
Research exactly how the pump and fans will connect, be powered, and the software you will use to manage them… map out every last detail of cable type, length, motherboard headers, etc. before buying stuff. In my case, I got a “Quadro” hub which is the same Aquacomputer brand as my pump, this way I can link fan and pump speeds to a simple Algo based on Water, CPU and GPU temps (you have to come up with the logic in the software yourself though…) So I’m using software, not BIOS or Windows or Motherboard for cooling. The pump and Quadro have the settings saved in them, so even if the software isn’t running (like at boot time, or booting to BIOS), they fallback to a default to make sure there is cooling no matter what.
-
Like everything, software is never simple. But the Aquacomputer Aquasuite (which is used for the D5 Pump, and fan control) does work well once figured out. All of these companies had a set back when Microsoft Defender began blocking Winring0 (used for low level access to sensor data), but the installer now adds exceptions to Defender.
-
Remember that some components like the chipset on the motherboard, and RAM, still need airflow for cooling, so make sure even if water temp is low you keep fans going at a speed that will keep everything that isn’t water cooled happy.
-
Water blocks are heavy - usually solid copper. The coolant is also heavy. The weight adds up, and think about if or how you will move the PC around. I bought frictionless vinyl tape and put it on the bottom of the feet of the case, so I can easily slide the PC. Also, think about supporting the GPU. I used an “UpHere” bracket which leverages off the spare PCIe slots on the case, instead of resting on the fans or rad below.
2 Likes
Thanks for such a detailed and thorough post - really valuable information.
I just want to point out once again that BIOS settings do matter. We know this, but unfortunately there are people on forum who ignore it and claim that default BIOS settings are fine – they’re not.
For example, in a custom loop, the coolant temperature is important. CPU and GPU temperatures have dropped, so what should really concern us is the temperature of the coolant itself, which shouldn’t exceed 60C. Some manufacturers even recommend keeping it below 50C. This depends not only on PC case, radiator size and how the custom loop is built, but also on fan speeds - if they are incorrectly set in the BIOS, the coolant temperature can rise above 60C.
1 Like
Very true. I control my front, rear, and side fans with an iCue Commander.
But my pump and radiator fans are plugged into the MOBO, controlled by temp curves based on a temp sensor on the reservoir.
I had a discussion in another thread with someone who keeps their pump running at 100% all the time. I prefer to slow it down when possible.
1 Like
Good that you mentioned that. For example, on MSI motherboards the default BIOS settings have Fan Fail Warning off. Other protections are also disabled by default, like System Power Fault Protection, System Fan Fail, Fan Fail Warning Notification and Pump Fan Fail - I don’t think I need to explain what that means or the potential consequences.
So if someone posts that the default BIOS settings are fine, please consider it as just informational and not necessarily relevant for your decisions.
1 Like
