Bastard Modding a Define R4

[Gorgon]

I blame it all on @Ben Quigley ...

 

So I got the bug for BOINC (that sounds rude) during the Pentathlon but found my old trusty daily driver with my circa 2012 smokin' e3-1231v3 a little slow for CPU projects these days and the Intel options not a very good value.

 

So I started having buyers remorse for the $99Cdn Ryzen 3 1200 I installed in my $130Cdn Gigabyte GA-X370-Gaming K7. Looking at the options I pulled the trigger on a Ryzen 7 2700 (non X) as it was about $100 less than the X these days on sale. After running it with the stock Wraith Stealth cooler for a week or so I've been disappointed with the 3.4GHz stock all-core turbo I've been getting even with the fancy LED cable connected to the motherboard for some pulsating bling.

 

Turns out for the $100 I saved I can get a lovely chunk of Austrian engineering so a Noctua NH-D15 will be arriving today.

 

I went with the stock D15 as Noctua is now shipping it with the AM4 mounting kit and the AM4 variant just has the AM4 mount so this way I get better future options. I did check the motherboard clearances and it should fit OK but the bog standard G-Skill Ripjaws V DDR4-2400 CL15 4GB dual channel kit will likely collide with the outside NF-A15 fan. The D15S was another option but that one only comes with one fan for the same price.

 

Bastard Mod1 was to rip the shiny red heat spreaders off the DRAM. It's Hynix 4GB modules and though only single rank is not known as very over clockable.

 

The Fractal Design Define R4 that I picked up used for $50 will have to be cracked open to fit the cooler so I'm going to take the opportunity to make it more "S" like by drilling out the rivets that hold the 2-bay 5.25" drive mount. It didn't come with the bay filler plates so they are currently covered with two pieces of Black Gaffer's tape in true LTT style.

 

Once the cage is out, the shavings all vacuumed up, and the components re-installed I'll test the airflow with and without the gaffer's tape and take some measurements to see what the options are for adding more fans or, at least a piece of fine mesh for dust filtration.

 

The goal is to see what effect the D15 has on the stock turbo using PB2, XMP and all the other magical settings then run a not-to-aggressive all core overclock.

 

The last time I overclocked a CPU it was a Pentium 166 (It cracked and I let the magical smoke out).

 

I have the battery charging for my 18V Cordless Hammer Drill in eager anticipation.

[Gorgon]

Unfair competition

[GOTSpectrum]

Not sure how I'm to blame...

 

Insert the frankenfolder....

[Gorgon]

At least you have a case

[GOTSpectrum]

50 minutes ago, Gorgon said:

At least you have a case

It was found in a skip(dumpster) with a dead mouse in it...

 

Also it now weighs 40% less with all the holes I've cut into it and the amount of metal cut out the inside

[TVwazhere]

23 hours ago, Gorgon said:

The Fractal Design Define R4 that I picked up used for $50 will have to be cracked open to fit the cooler so I'm going to take the opportunity to make it more "S" like by drilling out the rivets that hold the 2-bay 5.25" drive mount. It didn't come with the bay filler plates so they are currently covered with two pieces of Black Gaffer's tape in true LTT style.

 

I have the battery charging for my 18V Cordless Hammer Drill in eager anticipation.

[Gorgon]

Out with the old, Look at all that Noctua goodness. 3 140mm iPPC 3000 PWMs, 1 1700rpm 120mm redux and an older Cougar Vortex HDB 120mm PWM on top.

a little TLC for the 1060 6GB. Its been on-line for a year, only a slight puff of dust from the fins and some new paste

In with the new:

the front fan still had to be raised a bit even with the heat spreaders removed from the DDR4-2400. Made sure I connected the fans with the Y-adapter to the CPU header on the motherboard before mounting the cooler and only got a few scrapes from the heat-sink fins while tucking the excess cabling through to the back.

 

Noticed a 14C drop in CPU temperature under 100% load in BOINC from 66C to 52C.

WooHoo! my CPU is now turboing up to 3425MHz on all cores rather than 3393MHx so an extra 32Hz!

 

The plan for tomorrow is to drill out the rivets and remove the 5.25" bay assembly then to reassemble installing the 2060 I moved to the spare rig then test airflow with and without the Gaffers tape over the bay opening then to start on the overclocking.

 

Leaving it to fold & BOINC overnight and off to get some quality couch time.

[Gorgon]

While waiting for a WU to finish before I start disassembly I'll make some observations on airflow.

 

Replacing the NF-A14 that was in the rear with an iPPC 3000 definitely improved cooling. Looking at the Gaffer's tape over the 5.25" bays it has a discernible inward bowing and using a tissue test on the top vent position that is not populated with a fan I observed the tissue sticking to the grill rather than floating indicating that the system is negatively pressurized and drawing air in.

 

That's impressive and gives a indication just how much impedance to airflow the front grill of the Define R4 produces even with 2 of the better high static pressure fans on the market.

 

2 Noctua NF-A14 iPPC 3000 PWM as front intake

1 Noctua NF-P12 Redux 1700 120mm PWM as bottom intake (to provide more air for lower GPU)

1 Noctua NF-A14 iPPC 3000 PWM as rear exhaust

1 Cougar Vortex CF-V12 HDB 120mm PWM fan as top exhaust

 

Subjectively, opening the front door on the chassis results in higher noise initially but once the PCIe x16 Thermistor  on the mainboard that is used for temperature control cools down a bit and causes the fans to ramp down the noise is about the same level.

 

Perhaps Fractal is on to something with the Meshify ... now if they didn't have that foam dust filter that would be a pain to clean and the power supply shroud that prevents the bottom fan mount from being useful ... but that is another project for the Define C that I'm  using as a reference build.

[TVwazhere]

1 hour ago, Gorgon said:

That's impressive and gives a indication just how much impedance to airflow the front grill of the Define R4 produces even with 2 of the better high static pressure fans on the market.

 

Perhaps Fractal is on to something with the Meshify .

Makes sense, as similar testing has concluded a similar result

Cases like the Define R series and be quiet! case's are better suited for removing higher frequency noises like coil whine, small high RPM fans, pump noises or any other high pitched frequency. The low hum of fans can often be mitigated in other ways, like reducing RPM's. 

[Gorgon]

5.25" Bay removed. In the past I've used a 1/16" tungsten bit to drill out the center of pop rivets and it's been a PITA with the odd snapped bit and having to use side cutters to cut the base off the pop rivet then pull it out. This handy video:

suggested using a  1/8" inch bit and just letting the pop rivets spin and you can see he tilts the drill a bit which helps push through. The other tip he has is to use a pair of needle-nose pliers with the drill in reverse to remove the pop rivet flanges that get wrapped around the bit.

 

Here's the case sans the bays

Managed to get the motherboard out with the CPU cooler still attached with the help of a pair of long needle-nose pliers.

While the board was out put a macro lens on the camera to photograph a few things such as the Super I/O chip peeking out under the Chipset cooler.

and some b-roll

A quick vacuuming of the metal shavings and accumulated dust and the components were reinstalled.

Replaced the Cougar 120mm fan at the top with a NF-A14 just to keep things all Noctua. Can you tell I'm a Fan Boy?

 

It was a little more tricky getting the Motherboard back in but plugging in the eATX cable with the board out and wiggling it into place with the two top screws in the board and the one at the rear middle made life easier. Connecting the rear fan proved the most challenging but again the long needle-nose pliers came to the rescue.

 

Installed the cover and booted the system and ran the Linux version of prime95 with the small FFT torture test for 20 minutes as a sanity check.

 

Though the chassis fans were hardly spinning due to being tied to the thermistor under the upper PCIe x16 slot the CPU passed the torture test and settled at 55C (23C Ambient, VRM: 52C) and the all-core turbo was 3290MHz.

 

I should be able to add a 120mm fan at the front top with a little bit of dremmel & drilling but I'll likely only do that if the CPU cooler appears starved for air after being over-clocked.

 

You can get a pre-made magnetic dust cover for the bay opening but I'll probably just go ghetto and chop one of the Define C top filters I have lying around.

 

The horde is now home and grumbling about Pizza so I'll run some WCG and fold a WU until I can get back to it.

[Gorgon]

OK. The teens have Pizza and are glued to their devices.

 

References:

• AnandTech: Overclocking Ryzen GPUs
• Tweaktown: Ryzen 2000 Overclocking
• LinuxConfig: Overclocking Ryzen
• TechPoweredUp: Ryzen Memory Tweaking

 

Linux doesn't have all the shiny GUI tools that Windows has for overclocking Ryzen but when you get right down to it all you really need is:

• CPU Utilization Monitoring
• CPU & VRM Thermal Monitoring
• Vcore Voltage Monitoring
• System Power Monitoring

and a stress testing tool.

 

I'll be using the lm-sensors package to monitor the Thermals and Vcore Voltage and Htop to verify utilization. mPrime will be used to torture test the CPU. The system power will be measured at the output of the UPS the system is connected to.

 

The process I'll use is:

1. Disable Cool n' Quiet and other "features" that can impact raw all-core performance.
2. Initially set Vcore to 1.000V and the CPU multiplier to 30 for 3GHz.
3. run a 15-20 minute stress test and if it passes increase the Multiplier by 1 and repeat the stress test.
4. If the stress test fails then I'll increase Vcore by 10 to 25mV and re-test until it passes.

This should result in a table showing the minimum Vcore for a given multiplier and hence the lowest power consumption at that Multiplier.

[Gorgon]

Made it to 4.0GHz at 1.29375v Vcore with Load-Line Calibration at "Normal" The CPU was at 74C and the VRM at 77C with the CPU fans at 1328/1500rpm and a Total System Power Draw of 180W.

 

At 4.1GHz/1.3750v I could get the system to run mPrime but the CPU was at 85C and the VRM at 84C on 22C Ambient with a total system draw of 230W and the NF-A15 HS fans were maxxed out at 1433rpm so I reached the cooling capacity of the NH-D15.

 

This was only running mPrime and with only one GPU in the system so the chassis "System" temperature sensors were both at 29C wheras normally during Folding they're at 34/37C respectively, so a delta T of 5-8C. On the flip-side without the GPUs running full-bore the chassis fans are barely spinning.

 

So I'm going to run the system for a few hours with a GPU folding and WCG on 14/16 threads to get a better idea if 4.0GHz is a practical 24x7 all-core speed or if 3.9GHz would be a better option.

 

3.9GHz came in stable at 1.23125v Vcore with the CPU and VRM at 68/67C at 23C Ambient and the CPU fans at 1231/1500rpm and 161W Total System Power Draw.

 

The Idle Total System Power Draw is 40W which includes 5W for the GPU and the Power Supply loss between 8 and 12W.

So at 4.0GHz that's a 17.8% increase in performance for an 133% increase in Power.

[Gorgon]

Still chugging along at 4.0GHz with the 1060 6GB running at a 140W power-limit. CPU cooler and VRM definitely likes the increased airflow or having one less core pegged and their running at 71/62C respectively. Waiting for the 2060 to finish a WU then I'll re-install it and we'll see how the thermals are and start tweaking the fan curves a bit.

 

BTW installed the 2060 in my spare case to keep on folding while I was setting the overclock and normally I run it with the side off as it's a Fractal Core 1100 and even with a 120mm Pressure optimized fan in the front and a 3200rpm 92mm fan in the back it tends to cook cards likely due to the really awful front ventilation. Had a Noctua 120mm iPPC3000 PWM fall out of the recent fan upgrades so I thought  I'd give it a try. Works like a charm. Keeping the GPU at 58C running full-tilt. Impressive.

[imreloadin]

What monitoring software is that?

[Gorgon]

@imreloadin

I'm using Zabbix. Free for 50 clients.

Under Linux I use lm-sensors to get the mainboard sensor data and a plugin for GPU monitoring. Works for Windows as well. I'm recording power data via SNMP from the UPSes which have network interfaces but you can also poll this from a connected host via the USB drivers.

I have a script for collecting PPD data from the FAHClient

 

A bit of work to get going and to maintain but it lets me keep an eye on things from the couch.

[TVwazhere]

On 6/8/2019 at 2:16 AM, Gorgon said:

So at 4.0GHz that's a 17.8% increase in performance for an 133% increase in Power. 

That's crazy!

[Gorgon]

System locked up at 21:55 last evening. Nothing in the logs and hardware monitoring did not show high temperatures or abnormal voltages. Running about 375W on a Corsair HX650 so power likely was not and issue.

 

Restarted the system (GUI locked up) and changed the overclock from 4.0 to 3.9GHz and Vcore to 1.29v and resumed processing.

 

System locked up again at 5:55. Went back to "Auto" for CPU multiplier and Vcore to isolate if the issue is the CPU overclock. If it runs OK over the day I'll try 3.9GHz again with a "titch" more voltage.

[Gorgon]

Ran with stock settings for the day with no freezes.

 

Did a little more modding isolating the CPU from GPUs. Introducing the Custom CPU Frozen Pizza Box Shroud version 0.2:

Note the addition of a thermal probe in the CPU supply side.

Adjusted the rear exhaust fan to a less aggressive BIOS fan curve controlled by the CPU to balance with airflow with the fans on the CPU Cooler.

Removed the top exhaust fan so the front and top could provide intake to the CPU. Did some empirical testing with a tissue to test airflow. Covered both top vents and saw the CPU Cooler intake air temperature start rising from 1-2C above ambient indicating exhaust from the cooler was recirculating so uncovered just the rear-most vent and inlet temperatures went back down.

Removed the bottom supply fan so the GPU section is now just getting supply from the 2 Noctua NF-A14 iPPC3000 PWM fans tied to the on-board thermal senor on the upper PCIex16 slot and is exhausting air out the holes in the back of the chassis where the slot filler plates were removed.

I'm happy with the temperatures.

Split the difference between the 1.23125 and 1.29375 Vcore for a 3.9 and 4.0GHz stable and ran:

Multiplier: 39; Vcore: 1.2625v; Cstates: Disabled; CPB: Disabled; Cool n' Quiet: Disabled. All cores running at 3894-3899MHz; CPU: 67C; VRM: 60C; Ambient: 23C for 7 hours now. I'm leaving it running while at work today and it will hopefully be stable.

Total system draw was 131W running mPrime on all threads. Currentlt running 14 threads WCG leaving 2 threads for feeding GPUs for F@H.

The Upper 1060 6GB is power-limited to 110/140W for now and the 2060 to 128/217W. Note that these are the AIB maximum draws not the observed maximums of about 128 and 150W respectively.

Total system draw is 386.25W averaged over 6 hours.

My run-time for World Community Grid are still longer than those reported by my e3-1231v3 running at 3.6GHz possibly due to differences in IPS between Intel and AMD and/or supported extensions.

[Gorgon]

Running just fine at 3.9. Switched over from WCG to F@H last night on the CPU to get ready for the Sprint and successfully completed a WU in a respectable time. I'm back on WCG until the event so I wont have an estimate on PPD for 14 cores until then as it will take a while to get an average over several different WUs as they can be wildly variable.

 

As a side note, also switched from WCG to F@H the 2 spare threads on the other two rigs and the 6 threads on my Daily Driver to F@H to test them in advance and, of course, the Xeon on the Daily Driver got assigned something brutal that took 12 hours where the Pentium Golds on the other two rigs got 7 hour WUs.

 

So the modding project was a success albeit an unexpected one as it made me realize that I needed to treat the CPU and GPUs as separate "Zones" for cooling and removing that HDD cage made that easier.

 

The other conclusion was spending a little more time thinking about the airflow in the chassis enabled me to cool the system the same or better using less fans.