DCS VR testing 11900K vs 10900K

[PicklePicklePickle]

Hello,

 

I recently did some testing with the i9-11900K and i9-10900K. To keep things as consistent as possible, I used a fixed test case of starting F-16C free flight in Persian Gulf map. The controls are not touched and the VR headset is always in the same place. In this state the aircraft flies in a descent for ~2 minutes and then crashes(into a race track!), so it gives a relatively repeatable test. Of course there are still some variations from the random locations of clouds, etc, but this is assumed to be small.

 

I tested 2 configurations with each processor: a baseline with BIOS defaults and the 'best' overclock configuration that I found for DCS.  I should caveat that I am not an 'expert' overclocker, I'm just sharing what I have found and the few things that have (slightly) improved DCS VR performance.

 

I used fpsVR to capture the metrics over each 2 minute test run, and am using the average FPS reported in the fpsVR History Viewer. I did 5 repeated tests for each of the main configurations to confirm repeatability of the results.

 

General Observations/Notes:

• Since this is a 'relative' test, I figured my specific DCS/VR settings are not super critical as long as the HMD frame rate is not being maxed out in the test case(otherwise FPS metric would be 'clipped'). I am using my personal preference settings which is 200% SteamVR on the Index and PD1.0 in DCS with all DCS AA off.
• The mother board BIOS support for 11900K is beta (Asus BIOS 0704) so my results might be affected by bugs bad optimization for Z590 chipset.
• Hyper threading off consistently gives about 3 fps in DCS.
• On the Index, the 90fps setting gives higher FPS than 80Hz(-9fps), 120Hz(-25fps) or 144hz(-18 fps). I do not understand this, as I would expect 80Hz to give the most consistent performance since the system has the longest time to output a frame.
• The 11900K is better than 10900K in the BIOS default state, but about equal to the 10900K when overclocked, and the 11900K is less consistent.
• The overclocked 10900K is very consistent. Repeated runs yield the same FPS to within <1fps.
• The 11900K has very different behavior in almost all regards(overclocking, etc) and was far less consistent in repeated runs giving a 4-5fps spread
• The 10900K overclocks in an 'intuitive' way, meaning higher clocks give better performance and reducing the number of cores(since DCS only uses 1-2) gives an expected reduction in heat, etc. The limit is when the heat gets close to 90-100C and/or the voltages are getting too high.
• On the 10900K, the best DCS-only overclock I have found is to reduce to 4 cores, and max out clock rate as allowed by voltage/heat(5.4Ghz in my case).
• With the 11900K, it ran 30 degrees colder than the 10900K and doesn't seem to go beyond that. Higher clocks does not equal more performance. It is almost as if it is throttling internally, but the tools(HWInfo, CPUZ, etc) do not show throttling and a static clock. Also, reducing the number of cores unintuitively seems to reduce single-core performance.
• Due to the above, on the 11900K the best DCS overclock I found was just to disable hyperthreading and set a higher ratio limit of 5.4Ghz, leaving everything else in default.
• I should note that my 'best' DCS overclock overheats/throttles with the prime95 small FFT test. My machine is dedicated to DCS so I am overclocking specifically for DCS VR and nothing else, and as long as only DCS is run it runs stable(10900K tested stable over months), but this is DCS stability not general stability.

 

Results:

• i9-10900K
  • Default: 70-74fps
  • Overclock: 85fps
• i9-11900K
  • Default: 80-81fps
  • Overclock: 83-86fps

 

Hardware Specifications:

• Motherboard: Asus Maximus XIII Hero (Z590)
• RAM: 128GB G.Skill (F4-4000C18Q-128GTZR)
• GPU: MSI Ventus RTX3090
• CPU AIO: Cooler Master ML360 (3x 3000RPM Noctua NF-F12 fans at full speed)
• PS: 1.5kW Corsair AX1500i
• SSD: Samsung 980Pro 2TB
• HMD: Valve Index x4 base stations

 

So my take away is that at least at this point, the 11900K isn't really a worthwhile upgrade. However, it is very early so it is possible that someone finds a good overclocking strategy or perhaps some BIOS bugs are addressed that improves performance.

Edited April 5, 2021 by PicklePicklePickle

[Nahemoth]

Thanks for the info, very interesting.

 

What is your 10900K overclock configuration?

[PicklePicklePickle]

Unfortunately, one of the only messups during this whole endeavor was that I expected the BIOS profile to be around after I changed CPU's. I should have saved it before, the 10900K profile got blasted due to the required BIOS update for the 11th gen. However, I can tell you what I remember and how I arrived at it, it is nothing special.

 

I started with defaults, and then tuning the CPU first, and then did the memory last. Otherwise, it is hard to tell if it is crashing due to memory or CPU, and profiling the memory is a bit ambiguous.

 

I used HWMonitor and CPUZ, as well as prime95. As I said, prime95 with the smallFFT is pretty much the most stress you can put on a CPU and it doesn't model how DCS runs. I find it is a useful tool to learn where the threshold is where prime95 starts throttling, but for a DCS-optimized system I am pretty much guaranteed prime95 SmallFFT will hit 100C and throttle.

 

I set (Asus BIOS):

 

All fans to full speed, note I upgraded the fans on the 360mm water cooler which made a significant difference in temperatures.

 

MCE to enabled

 

Hyperthreading off

 

Load-line calibration to level 4

 

CPU Core ratio: Sync all cores and ratio to 54 - This is what I ended up with, but I started at 49 and slowly increased. In retrospect, what we want is 1 core to run fast for DCS, and the others for the rest of the background stuff. However, I wasn't sure how to assign DCS to a specific core reliably so I just sync'ed all of them. If we could assign DCS to one core reliably, then we could potentially run 1 core higher, maybe 5.5GHz with the same cooling.

 

CPU Voltage to manual and started with something like 1.3V - This is what I started with, I can't remember where i left it but essentially increase it very slightly if it crashes. Be aware of the voltage limits, I believe you can fry the CPU with this setting. It is also my understanding that too high a voltage here can lead to shortening the life of the CPU, but in my case with this system all I care is running DCS well so the CPU is sorta consumable anyway. I'm pretty sure I ended somewhere around 1.45V.

 

I used HWMonitor to look at the CPU temperatures while stressing with CPUZ and DCS, and aimed to keep it <=90C which is 10 degrees below throttling, but it ended well below that. The stopper was the voltage. When I tried going to 5.5Ghz it looked like I had to increase the voltage quite a lot to avoid crashing and it was getting to the level where a lot of folks thought it would reduce the life of the CPU.

 

For the memory, with the GSkill memory I have, it is supposed to run at 4000MHz, so i started by loading the XMP profile. However, it crashed and so I reduced the speed to 2666MHz and slowly worked my way up. I found that it ended up at 3800MHz. So the memory does not meet the advertised speed but I didn't feel like returning it.

 

I'm probably missing stuff but that is the general procedure I followed.

Edited April 5, 2021 by PicklePicklePickle

[DerekSpeare]

I am building an 11700k system right now...I have to keep my 1080ti card for lack of 3090's   This is replacing the 9600k system and I am hoping for some improvement with the FPS.  I'd have gotten the 11900k but there is not much difference than that of the 11700k (for me).  This thread gives me some encouragement that I might get a few more FPS

Edited April 5, 2021 by DerekSpeare

[jib]

I have just build a i5 11600K system but I cannot get near to maxing out the CPU because my RTX 2060 maxes out long before the CPU (I Can only run medium Texture settings and no AA in VR but in 4K flat screen can run most things maxed out and still hit 60fps) . I am going to start saving and wait until 3080's are back down to reasonable prices or try to snag a 3080Ti at MSRP when they first launch.

Edited April 5, 2021 by jib

[Nahemoth]

12 hours ago, PicklePicklePickle said:

Unfortunately, one of the only messups during this whole endeavor was that I expected the BIOS profile to be around after I changed CPU's. I should have saved it before, the 10900K profile got blasted due to the required BIOS update for the 11th gen. However, I can tell you what I remember and how I arrived at it, it is nothing special.

 

I started with defaults, and then tuning the CPU first, and then did the memory last. Otherwise, it is hard to tell if it is crashing due to memory or CPU, and profiling the memory is a bit ambiguous.

 

I used HWMonitor and CPUZ, as well as prime95. As I said, prime95 with the smallFFT is pretty much the most stress you can put on a CPU and it doesn't model how DCS runs. I find it is a useful tool to learn where the threshold is where prime95 starts throttling, but for a DCS-optimized system I am pretty much guaranteed prime95 SmallFFT will hit 100C and throttle.

 

I set (Asus BIOS):

 

All fans to full speed, note I upgraded the fans on the 360mm water cooler which made a significant difference in temperatures.

 

MCE to enabled

 

Hyperthreading off

 

Load-line calibration to level 4

 

CPU Core ratio: Sync all cores and ratio to 54 - This is what I ended up with, but I started at 49 and slowly increased. In retrospect, what we want is 1 core to run fast for DCS, and the others for the rest of the background stuff. However, I wasn't sure how to assign DCS to a specific core reliably so I just sync'ed all of them. If we could assign DCS to one core reliably, then we could potentially run 1 core higher, maybe 5.5GHz with the same cooling.

 

CPU Voltage to manual and started with something like 1.3V - This is what I started with, I can't remember where i left it but essentially increase it very slightly if it crashes. Be aware of the voltage limits, I believe you can fry the CPU with this setting. It is also my understanding that too high a voltage here can lead to shortening the life of the CPU, but in my case with this system all I care is running DCS well so the CPU is sorta consumable anyway. I'm pretty sure I ended somewhere around 1.45V.

 

I used HWMonitor to look at the CPU temperatures while stressing with CPUZ and DCS, and aimed to keep it <=90C which is 10 degrees below throttling, but it ended well below that. The stopper was the voltage. When I tried going to 5.5Ghz it looked like I had to increase the voltage quite a lot to avoid crashing and it was getting to the level where a lot of folks thought it would reduce the life of the CPU.

 

For the memory, with the GSkill memory I have, it is supposed to run at 4000MHz, so i started by loading the XMP profile. However, it crashed and so I reduced the speed to 2666MHz and slowly worked my way up. I found that it ended up at 3800MHz. So the memory does not meet the advertised speed but I didn't feel like returning it.

 

I'm probably missing stuff but that is the general procedure I followed.

 

 

Thank you for the response. So you limit the 10900K for using just 4 cores at maximum speed. I will do some testing.

[GunSlingerAUS]

Thanks for this fantastic post - excellent testing methodology and description. I've actually just purchased an 11900KF, as I'm currently running a 9700K @ 5GHz. Couldn't get my hands on a Ryzen, so figured this will have to do. I get the feeling I won't be hanging on to this CPU anywhere near as long as the 9700K though...

[dburne]

My i9 9900k running at 5.2 GHz on all cores is still doing very well for me, think I will wait for another generation or two of motherboards and processors.

[Gman109]

On 4/6/2021 at 4:51 AM, dburne said:

My i9 9900k running at 5.2 GHz on all cores is still doing very well for me, think I will wait for another generation or two of motherboards and processors.

 

IMO a wise choice.  I have 3 PCs with 2 dedicated to DCS, 2 10900k and one 5950x (not using it for DCS atm), all using various 3090 GPUs.  I found that comparing my previous 3 9900k based PCs from a year+ ago to the 2 10900ks, the performance jump in DCS is very, very minimal.  I had planned on upgrading the 10900ks to the new 11900k, but based on the great above OP and various other tests I've seen on Youtube and other online sources, I've decided to wait for Alder Lake which hopefully will come this year.

 

The increase in performance, again just IMO, from even the 9900k much less the 10900k, to the 11900k, is simply not worth it.  Not so much the $$ involved either for me, but the general PITA of rebuilding, reinstalling, and setting up new PCs with all the various programs one needs to install and set up each time you rebuild.  

 

I'm hoping Alder Lake brings good things.....hoping...

[DavePastry]

why do you have 4x basestations? from my understanding and anecdotal use of the lighthouse system since 2016 2x base stations give you perfect 360 degree tracking coverage? 

[PicklePicklePickle]

No reason other than they come in pairs and in my VR extravaganza I ended up with 4. I have this cludgy but easy mounting method where I just pinch them with ceiling tiles, so it was simple to mount all of them.