Another i9-9900k build

[LRW 42]

I've completed my first pass of setting up my new build.  I thought I'd pass along my experiences so far.

CPU:   i9-9900k
Mobo:  Asus Maximus XI Code
SDRAM: G.Skill TridentZ DDR4 4400 18-19-19-39 16 GB Kit
PSU:   Seasonic Prime Ultra 850 80+ Titanium
SSD:   Samsung 970 EVO M.2 1TB PCIe (2)
GPU:   EVGA GTX 1080 Ti SC Black 11GB
Screen:Samsung UN40KU6290 4K @ 30Hz
AIO:   EVGA CLC 280

Case: Lian-Li OC11 Dynamic

tl;dr:  I'm running all cores at 5.0GHZ, HT off.  Still in early testing of P3D v4.3 

I took the GPU and AIO from my old build.  I probably would have bought a 2080 Ti if EVGA had any in stock when I was purchasing the parts.  Now I'm glad I don't have it yet.  By keeping the 1080 Ti, it gives me a better comparison with my 4-core 3570k @ 4.5 GHz.

I also took the AIO from my old PC.  I'm going to replace it with an Asus Ryujin 360, which includes a fan to provide air flow over the VRMs.  Like most of the components I wanted it wasn't available when I ordered my other parts, but I ordered one from Amazon (sold & shipped by) that should arrive on Monday.

The first issue that came up in the new build was that I couldn't boot with the TridentZ at 4400 MHz in XMP Modes I or II.  Asus includes the TridentZ 4400 in their QVL, although with the 32GB kit.  I was able to get it to 4000 MHz  17-17-17-37 at 1.35V.  I knew this was a possibility when I bought it, so I wasn't surprised.  I haven't tried moving the sticks from A2-B2 to A1-B1.  Not my highest priority right now and I don't know that it would make much difference anyway.  I might try it with a future BIOS update.

Speaking of which, I updated the BIOS to 0602.

My first tests were at stock settings (including the RAM at 2133 since XMP wouldn't boot).  When running Prime95 I quickly learned about PL1, PL2, and tau after it would run at 4.7 on all cores for about 100 seconds, at which point it would throttle down gradually to about 4.1 on all cores.  This would bring the power back to TDP, which is 95W for the i9-9900k.  Power Limit 1 defaults to TDP.  PL2 = PL1 * 1.25, or 118.75W.  

These can be configured in BIOS under the Extreme Tweaker/Internal CPU Power Management menu:
PL1 = Long Duration Package Power Limit (max = 4095)
tau = Package Power Time Window (max = 127)
PL2 = Short Duration Package Power Limit (max = 4095)

Following the videos of Der8auer and Steve Burke for OC'ing the 9900k, I set those values to max.

See Steve Burke's explanation of how different mobo manufacturers do their own thing with power settings https://www.youtube.com/watch?v=IBrumDWpl-c&t=696s.  According to Gamers Nexus, Asus is the only board maker that follows Intel guidance for power levels and duration at stock settings (but not with MCE on).

I also tried out the Asus AI Overclocking to see what it would do but quickly abandoned it when it set the RAM voltage to 1.6 to achieve 3600 MHz.

I started off my manual overclocking with HT enabled.  Cinebench R15 seemed to be a quick way to see if my settings were way off, because it would hang or crash almost immediately and a full CPU run took less than 30 seconds.  I had to abandon Prime95 after I disabled the power limits because the CPU would hit thermal limits almost immediately with small FFT values.  I would have thermal throttling on 2 or 3 cores, plus the CPU package.  Temps would hit 101 on the CPU package and upper 90's on the cores.  I started running the Asus RealBench 2.43 for 15 minutes for stress testing.  At first, RealBench would cause the 9900k to thermal throttling 2 or 3 cores after 5 minutes or so.  That turned out to be because I didn't have the radiator fans ramping up properly.  I tested with the AIO pump at minimum speed (~1950) and maximum (which varies on mine from 2880-2940 RPM), which only varied 1-2C on any core.  With the fans set properly RealBench could run 15 minutes with no thermal throttling, maxing at 87C.  However, Prime95 still hit the thermal limits.  I installed 3DMARK and ran Time Spy aifter seeing Rob's mention of it.  It ran to completion and scored 10,077 with Graphics 9,802 (single 1080 Ti) and CPU 11,983 (HT on @ 5.0 GHz).

The Maximus mobo has an AIO pump header, which I used.  The header can be used with 4-pin (PWM) or 3-pin (DC) connectors.  The EVGA CLC 280 has a 3-pin connector.  I could configure the pump in BIOS to run at 3 different speeds, depending on the reading of the parameter being monitoring (which defaults to CPU).  Since I could only control the fans from the EVGA Flow Control app, I disabled BIOS monitoring of the AIO and set the pump speed and radiator fans in the EVGA app.

I currently have all cores running at 5.0GHz.  I won't try pushing it higher until I get the Ryujin AIO on Monday and see what my temps are.  

My FS hardware:

• MFG Crosswind rudder pedals
• PFC Cirrus desktop yoke, Saab model
• JetMax 737 TQ
• TrackIR - Disabled to get more consistent results between runs

I use 0.70 zoom with wide angle enabled.  All P3D settings below.

I have two standard test flights that I use, both in the PMDG 737 NGX.  The first one is a circuit at KLAX from RW25R to RW25L that takes about 20 minutes. Here's the route:

25R
DOCKR
WEILR
SHAEF
PEVEE
SLI
GAATE
HUNDA
ILS25L

I have terrain settings at maximum (except for water, which is set to High. With Ultra my RTSS stats wouldn't appear on the screen).  My first run was with default scenery and the fps stayed at 30fps for the duration with smooth animation.  Adding Orbx Global didn't affect it, but OpenLC NA introduced a multi-second freeze when almost over Terminal Island at Long Beach.  There were also micro-stutters occassionally, especially when turning.  Adding Orbx SoCal increased the stuttering.  I installed FSDT KLAX, which didn't affect it much except for one extra pause on long final.  I also ran FSDT KLAX without Orbx OpenLC and SoCal.  Some stutters that I used to think were from FSDT turned out to be from the Orbx sceneries.  The only real effect on maintaining 30 ftps came when I enabled AI traffic; MT6 set at 25% caused my frame rate to drop to about 25 while on the ground at KLAX and on final.  I've barely scratched the surface on P3D testing.  I did one quick check at FlightBeam KMSP at night with dynamic lighting.  There was no AI traffic.  It stayed at 30fps, even with external views.  But ImagineSim ZSPD with AI traffic at night killed my frames down to 12.  I didn't try any other settings to try to pinpoint it.  Still lots to do.  Overall, I will be able to run with much higher settings than previously.  I just don't know exactly what that will be yet.

Edited November 21, 2018 by LRW
Photo update

[mpw8679 428]

Thanks for the excellent write up.  It’s a shame even with a 9900k one will still experience the scenery loading pauses.  I hope they address this problem soon.

[BrettT 63]

If you don't mind me asking, what sort of CPU temps are you getting?

[LRW 42]

1 hour ago, BrettT said:

If you don't mind me asking, what sort of CPU temps are you getting?

Good question.  I captured stats for most of my KLAX circuits. Here are a couple of examples.

Orbx Global Only

               Core #0           Core #1          Core #2         Core #3           Core #4          Core #5           Core #6           Core #7

Min                   42                   41                   40                   41                   37                   38                   39                   37

Avg                   65                   53                   62                   61                   61                   60                   59                   57

Max                   76                   73                   84                   82                   86                   82                   83                   81

 

Orbx OpenLC & SoCal

               Core #0           Core #1          Core #2          Core #3          Core #4          Core #5           Core #6          Core #7

Min                   34                   36                   35                   34                   35                   33                   36                   34

Avg                   61                   50                   57                   55                   55                   54                   54                   52

Max                   70                   69                   78                   77                   79                   76                   76                   77

 

I'll update it after I install the Ryujin 360 AIO.

Edited November 21, 2018 by LRW

[Ted Striker 338]

Thanks for taking the time to post the great summary of your build and P3D performance experience Larry. My current system is very similar to your old one. When you described the P3D performance on your new system I thought, "oh oh, this is not good, this is what I get on my current system." Well that was until I saw your settings which are mostly maxed out. My settings are mostly medium to low.  I know your have a lot to test and adjust right now but I would appreciate it if when you have the time you would let us know how much you need to relax your settings to eliminate the stutters and get smooth flight.

Thanks again,

Ted

 

Edited November 21, 2018 by Ted Striker
grammer

[martin-w 3,234]

Quote

However, Prime95 still hit the thermal limits.

 

 

Prime is vicious. Which version of Prime? Try the version without AVX. I would recommend to set an AVX offset as well. Auto wont work. I suggest set to 3. It will hit high temps running a stress test that runs AVX with no off set in place.

[martin-w 3,234]

Awesome looking case Larry... but  how does the airflow works on that Lin Li model? Whats the rear of the case like? Is it positive pressure, in-taking air into the CPU/GPU side?

[LRW 42]

2 hours ago, martin-w said:

 

 

Prime is vicious. Which version of Prime? Try the version without AVX. I would recommend to set an AVX offset as well. Auto wont work. I suggest set to 3. It will hit high temps running a stress test that runs AVX with no off set in place.

Manually setting AVX offset to 3 made all the difference.  That ran the cores at 4.7.  I let the test run for 30 minutes with small FFTs and the max core temps ranged from 77 to 84 with CPU Package 84.  The Prime95 version is 29.4 build 8.  I'll see what my temps do when I get the Ryujin 360 next week.  Big thanks for that advice!!

[martin-w 3,234]

33 minutes ago, LRW said:

Manually setting AVX offset to 3 made all the difference.  That ran the cores at 4.7.  I let the test run for 30 minutes with small FFTs and the max core temps ranged from 77 to 84 with CPU Package 84.  The Prime95 version is 29.4 build 8.  I'll see what my temps do when I get the Ryujin 360 next week.  Big thanks for that advice!!

 

It's a pleasure and an honour. 😁

That asus AIO looks interesting. One of the reasons I'm a fan of the D15 is because the centrally mounted fan, 150mm overlaps the heat-sink and blasts lots of air across the VRM's. Something that's lacking with AIO's. So great that asus have addressed this. 

[LRW 42]

2 hours ago, martin-w said:

Awesome looking case Larry... but  how does the airflow works on that Lin Li model? Whats the rear of the case like? Is it positive pressure, in-taking air into the CPU/GPU side?

The back has a small grill that's about 7x14cm.  It has  the slightest outflow, even under load when the radiator fans are high.  The bottom of the case can take 3 x 120mm fans, so that's a large open grill at the moment.  There's definitely air coming out of the bottom under a load when the intake fans are at high speed, ~1500 rpm.  I can add 3 more fans to the bottom if I need them for better GPU cooling, but I'll have to add a Corsair Commander to power and control all 6 fans and to provide an extra USB header for the LED controller of the additional fans.

[martin-w 3,234]

Doubt you'll need anymore fans to be honest. Looks an awesome machine.

[LRW 42]

15 minutes ago, martin-w said:

Doubt you'll need anymore fans to be honest. Looks an awesome machine.

It's an adventure of exploration.  I'm having a lot of fun with it.

[Bob Scott 7,716]

3 hours ago, martin-w said:

Prime is vicious. Which version of Prime? Try the version without AVX. I would recommend to set an AVX offset as well. Auto wont work. I suggest set to 3. It will hit high temps running a stress test that runs AVX with no off set in place.

You can disable AVX in current versions of Prime95 by running it briefly one time, which creates the config files, then edit local.txt, adding these three lines:

CpuSupportsAVX=0
CpuSupportsAVX2=0
CpuSupportsAVX512F=0

Nice build.

Cheers

[martin-w 3,234]

4 hours ago, LRW said:

Manually setting AVX offset to 3 made all the difference.  That ran the cores at 4.7.  I let the test run for 30 minutes with small FFTs and the max core temps ranged from 77 to 84 with CPU Package 84.  The Prime95 version is 29.4 build 8.  I'll see what my temps do when I get the Ryujin 360 next week.  Big thanks for that advice!!

 

 

[J0nx 48]

I replaced my 5ghz OC 8700k with a 5ghz OC 9900k over the weekend and didn't see much improvement in any sims. Maybe 1-2 fps. Mostly switched due to way better multi-thread performance that I needed. I'm still getting about 30fps sitting at Jeffersonville with all orbx related products enabled and p3d sliders about 3/4 on average at 5760x1080. I'd say that those with at least a well overclocked 6700k or greater will be disappointed if they are expecting any improvement in this old sim any time soon with the current gen hardware.