"Needy Healing" - Used Parts Overwatch PC Build - PART 5


With the system build complete, it's time to get all of our software installed and get to the business of setting up a stable overclock to see if this old CPU can run Overwatch smoothly.  As this is the first time I am overclocking a PC, I am sure the process of getting everything sorted out will involve a lot of trial and error, but fortunately, we are using core components that did not break the bank!  So, this sets up a great opportunity to be a little bit adventurous and learn as much as I can about the process.


If you want to catch up on the previous segments of this project, they are linked below:

"Needy Healing" - Used Parts Overwatch PC Build - INTRODUCTION
"Needy Healing" - Used Parts Overwatch PC Build - PART 2 (parts hunting)
"Needy Healing" - Used Parts Overwatch PC Build - PART 3 (case hacking)
"Needy Healing" - Used Parts Overwatch PC Build - PART 4 (assembly)

The first thing I need to do is get Windows installed onto the system.  When looking at ways to save cost on the system, I came across this YouTube video that pointed my to kinguin.net which is a selling portal for people selling licenses for various software.  I also double checked a few other tech vloggers, and there seems to be agreement that these OEM Windows licenses are in fact legitimate.  So what is the catch, and why the dramatically lower price?  These "bulk" licenses are intended for system builders and manufacturers, so, there is no technical support for the software.  It's on you and your Google skills to solve any issues that may come up, and the license once activated is tied to your specific motherboard, so you can reinstall it as many times as you need if your system becomes corrupted, your hard drive fails,  or if you simply decide at any point that you want a fresh install. But, once installed and verified the first time, the license is LOCKED to your motherboard, and you cannot transfer it to another computer like the more expensive retail licenses allow.  One perk to the setup is when reinstalling the OS, Microsoft automatically verifies new installations on your motherboard without the need to re-enter any serial numbers or codes.  As far as I can tell, Windows operates and updates perfectly with my license.  I downloaded the installers and software directly from Microsoft, entered the serial numbers and product codes provided via email by the Kinguin vendor upon initial installation, and everything has been solid.

Even though Home Edition probably would have worked just fine for gaming, I decided to install Windows 10 Professional which runs just a bit under $30 from most sellers on Kinguin.  I do not know where this computer will end up, so I didn't want it to be lacking professional features that may be needed in the future.  I made sure to choose a seller with a lot of transactions and very positive feedback on Kinguin.




There are many ways to set up a fresh install of Windows, but here are the steps that I took for this specific build, and it was expedient and works just fine.

The first thing I did was install graphics drivers from Nvidia for the GTX 970.

Next, I needed to update all of the various drivers for the motherboard and peripherals.  I found the fastest way to do this is to use Driver Booster 4 which is free to use in this capacity.


Once installed, running a scan revealed that 24 drivers needed updating, and a single click downloads and updates all of them which for me, sure beats tracking down all of  the latest drivers individually.


With the drivers up to speed, the next item that I like to get sorted out is my fan control.  I scratched my head about this for a while as fan setups can vary significantly from build to build.  There are external modules available, but they all seemed a bit redundant and costly.  The PC should have perfectly good sensors and fan speed regulators built in.  There are two things I like to see in a fan setup.

1)  All or most of the fan power coming directly from the PSU so the Motherboard (especially a used one) does not have unnecessary power loads.

2)  Fine control over fan speed and curves based on real-time internal temperature readings (CPU, GPU, and motherboard)

I came across this interesting YouTube video that explained how to set up to a free fan speed control program called SpeedFan.  This Asus motherboard has built in fan speed control capabilities, but I found it did not allow me to fine tune things as well as I wanted.  I did, however turn these options on in the BIOS just so the fans were not running full tilt when the system boots up.  Downloading and setting up the Speedfan software is not the most intuitive process, so I thought I would detail how I set this up on my particular system, and the information should be useful (if a little different) for most other computers.

Right off the bat, downloading the software is a little wonky as you need to sidestep quite a few adware landmines just to get to the download.

When you get to the SpeedFan website, you want to click on this Download tab at the top menu bar.


And then click on this little highlighted text link in the descriptions avoiding all of the flashing "download here" links.


I accidentally clicked through one of the other prominent "download here" links, and my system started automatically downloading and installing all sorts of drivers.  I freaked out so badly that I reinstalled Windows just to make sure everything started off clean.  Don't be me!

Once SpeedFan is installed and running, you'll notice that changing the Sys, CPU, and Aux percentages does nothing.  But, a lot of information is available.  It has various real time temperature readings as well as tach readings for my CPU fan and System fan.  Some putzing around in the "Configure" menu is in order.
Clicking on the "Configure" button should pop up a separate window with a lot of tabs.  In the first Temperature tab, digging into a few of the main items in the nested tree shows a "Winbond W83667HG-B" chip that apparently spits out some of the critical system temperatures.  I make note of this device.  I have no clue what a W83667HG-B is, but that's what I'm going to be primarily dealing with in SpeedFan for the purposes of temperature reading and fan control.
In order to acquire some control of my system fans, click over to the "Advanced" tab and select the "Winbond W83667HG-B" chip in the drop down menu.

If you recall, I have the majority of my system fans set up on the "CPU fan" 4 pin header on the motherboard through a PWM fan splitter cable that pulls fan power directly from the power supply, but derives the speed control PWM signal from the motherboard header and splits it to all of the fans.  The splitter also connects one fan's tach signal wire to the motherboard to report RPM for just the single main fan in the splitter.  The tach signal wires for the rest of the fans are not connected to anything.  If all of the tach wires were connected together and piped to the motherboard, we would get corrupted RPM readings because the fans in the splitter will certainly not all be in perfect sync with each other while running.  This setup allows the motherboard to control speed on all of the connected fans as well as receive some feedback, but lets all of the fans be powered directly from the motherboard so I don't have to worry about how many fans I add or how much power they each draw at max speed.

I also have a single fan connected to the "CHA FAN" 4 pin header on the motherboard so I can control an additional fan independently.  The idea was to give the graphics card some priority on one of the main intake fans should it come under temperature load for some reason while the CPU is not stressed.

Because both sets of fans are PWM controlled, I am looking for "PWM" in the drop down menu, and I find three entries of interest to me.  Asus "SmartFan" control is enabled on two of the fan headers in BIOS, so I focus in on "PWM 1" and PWM 2".

I first take a look at "PWM 1" and set it to "Manual" control mode in the drop down menu noting the original configuration (SmartFan III) because after some testing, I will return this setting to original.


I have no clue what the "remember it" checkbox does, but it looked important so I checked it.  Click "OK" to save these settings and go back to the main window and mess with the fan speed percentages to ramp the fans up and down!  Once we confirm that we are controlling fan speeds and verify which header on the motherboard  correlates to the software controller, we can move forward with adjustments.  What we want to do now is turn down the fan speed until the fan stops spinning and determine the minimum value threshold.  This will allow us to set the system to run the fans as slow as possible when idling.  The "Sys" fan starts up at 24% setting.  Below 24%, the fan is stopped.
 The "Sys" fan is now responsive to manual adjustments which corresponds to the single fan connected to the "CHA FAN" header on the motherboard and located at the bottom front intake fan on the case.  This will be our GPU fan.  Going back to the configuration menu, I set "PWM 1" back to the "SmartFan III" setting that it originally had.  I do this so the fan reverts back to the somewhat intelligent BIOS speed profile when SpeedFan is not running (system startup and shutdown).

Next, I change "PWM 2" to "Manual" control to confirm that this is my "CPU FAN" header that controls the other 3 fans in my system (CPU, top front intake, and rear exhaust).

After this setting is locked in, I confirm that "PWM 2" is linked to the "CPU FAN" header on my motherboard  and in my system, controls the main PWM fan splitter.  It is also surprising that the fans start up at 7% setting which could indicate that we have higher voltage or current coming directly from the PSU than from the motherboard's fan header power pin.  The other fan "PWM 1" which is being powered by the motherboard directly needs longer pulse width to start up and runs at a slower max speed.
With my software fan controls identified, I return the "Advanced" tab values to the original values that I found them at so that the ASUS in-BIOS fan controls still operate when SpeedFan is not active.

OK, now it's time to set some custom fan curves to automate the fan controls in the system.  In the "Configure" window, click over to the "Fan Control" tab and check the "Advanced fan control" checkbox.  Then, click on the "Add" button and name the controller curve.  I choose to call this one "CPU fans" to indicate everything on the "CPU FAN" header on the motherboard.


Click on the new fan controller you just made to activate the sub-menu below.  Check the "Controlled speed" checkbox and in the dropdown menu, we will need to choose the fan that we intend to control.  In this case, I select "CPU - CPU from Winbond W83667HG-B @0290 on ISA"

Next, under the "Temperatures" box, click the "Add" button.  We will now select which temperature reading we will use to control the CPU fans.  We have a few options that would likely work, but I'll stick to the main chip that we are focused on and select "CPU - CPU from Winbond W83667HG-B @0290 on ISA".


Once you click "OK", a new entry will appear in the Temperatures box.  Click on this new entry, and a fan curve box should appear.  Drag the curve to the configuration that you would like to use.  Here, we can use the fact that we know the lowest idle speed on our fans is 7%.   I also know that the cpu never idles below 37 degrees Celsius no matter how fast the fans are spinning, so it is pointless to run the fans any faster than minimum when the CPU is under 37 degrees.  So, I set my curve to stay at 7% all the way up to around 40 degrees and then rise to full speed at around 55 degrees.  This should give the system a little bit of room to idle at the quietest noise level possible but ramp up aggressively under load to protect the CPU.  NOTE: when adjusting the curves, right clicking on the control point dots will pop up a little menu that allows you to increment the values up or down by one percent to fine tune the curve.

Ok, with the main "CPU FAN" controller set up, we can add another fan controller in the "Fan Control" tab for the GPU fan on the motherboard's "AUX FAN" PWM header.


Here, we will check the "Controlled speed" checkbox and select the "Sys-Sys from Winbond W83667HG-B @0290 on ISA" as the fan we want to control.

But, we want to reference the GPU temperature for this fan, so we select "GPU - GPU from GefForce Video Card on PCI" as our temperature reference.


And, we can set the curve however we want.  This particular GPU likes to leave its fans completely off so the idle temp ends up being quite a bit higher than the CPU idle temps.  Taking this into account, I ramp up the fan speed later because if the fans on the GPU card are not spinning, it is pointless to run the case fan fast to try and pull the GPU core temperature down.  It would have little to no effect.  We know that this fan does not spin until 24% PWM setting, so we set that as the minimum value and ramp up from there.  This way, we allow the fan to spin its lowest possible RPM while at idle.  I guess you could set it to 0% and turn off completely as the other case fans in the system may be sufficient at idle, but I choose to leave them on.


On this system, preliminary testing indicates that I will be running up to my thermal limits after overclocking the CPU, so I wanted to make the fans all run full power when the CPU is under load (including this fan!), so one way I can accommodate that is to add a second curve to this fan control.  Simply click the "Add" button under temperatures again.  for this curve, I will select "CPU - CPU from Winbond W83667HG-B @0290 on ISA".


And, I set this fan curve very similar to the "CPU FAN" setting except the idle value is 24% instead of 7% as that is my minimum setting on this header.
Next, change the setting in the "Method" drop down menu from "SUM of speeds" to "MAX of speeds"  This way, instead of taking an average of the two curves, the program will simply take whichever value is higher at an given time and apply that to the fan.  So,  while all the other fans in the computer respond to the CPU temperature, if EITHER the GPU or the CPU start to ramp up, this bottom intake fan will ramp up.


After saving the configuration, go back to the main window and check the "Automatic fan speed" checkbox.  The "Sys" and "CPU" fans should now be automated and respond to the fan curves that we just set!

Next, with our fan curves set, go back into the configuration window and under the "Options" tab, check "start minimized" and "Minimize on close".


With SpeedFan configured, we now need to set the system up to start it up every time the computer turns on.  I tried a few different methods that did not work too well before I found the linked video below:

https://www.youtube.com/watch?v=3krml2Hm8Sk

The video is so concise and thorough that I feel no need to add anything to it or to re-trace the steps here.  Following along exactly as described will get you set up in just a couple minutes.  My only deviation is I set my start delay to 10 seconds instead of the recommended 30 seconds for the login trigger because the SSD allows this build to boot up so fast.

Whew. . . after all that, I maybe should have just bought an external fan control device :)  At this point, the Windows system is pretty much set up and ready to run.  The next step is to download the utilities I need to overclock the system.  Two main programs that I need to set the CPU overclock are:

CPU-Z
Core Temp

Note: on my computer CPU-Z version 1,80 would not display the core speed or multiplier for some reason so I installed the previous version 1.79 which worked great.

For stress testing the CPU and memory to find the best overclock, I installed these programs:

Prime95
AIDA64
Memtest

For overclocking the GPU, I am using:

MSI Afterburner

For stress testing the GPU, I am using

3DMark Firestrike Benchmark
Unigine Heaven Benchmark
Unigine Valley Benchmark

 And finally, the main test and the whole point of the build:

Overwatch Game Installer

For CPU overclocking, I did not have a cohesive methodology as this is my first attempt.  There was so much sporadic trial and error associated with the first attempts that it is hard for me to recall and log all of the steps that I took.

But, as I understand it, overclocking the CPU is a juggling act between 3 main variables:

CPU clock - try to get this as high as possible

Core voltage - as CPU clock increases, voltage needs to increase to support the higher speeds but increasing adds heat and at a certain point can damage the CPU, so try to keep this as low as possible while maintaining the target core clock speed

Temperature - as core clock and voltage increases, so does heat on the CPU.  Try to keep this within "reasonable" limits for the system operating conditions.

Since my current "main" computer is running a non-overclocked i5 at 3.8 GHz, my hope was to get this system to 4.0 GHz or above, so that was my target for this X3460 build.  The methodology I used for working up to a working overclock is:

1) tap "Delete" key while system is booting to enter BIOS and make adjustments
2) save/load settings and boot system into Windows
3) open CPU-Z to observe current CPU state
4) open up Core Temp to monitor temperatures
5) run Prime95 stress test while observing temperatures and system stability
6) system either runs or crashes, log results and return to step 1

This cycle is repeated over and over again until a balance between desired performance and operating temperatures is reached.  The tricky part is figuring out which BIOS settings should be made at any given time in the process to keep moving in the desired direction!  And, each individual CPU responds differently to a whole host of BIOS settings.

Right off the bat, I was frustrated in my overclocking by seemingly random blue screens.  No matter what I was doing, the system would crash completely randomly for no apparent reason.  It could be idling, it could be under running prime95 stress test, I could be surfing the internet, I could be in game.  It would stay up long enough for me to run tests and benchmarks, but crash at the weirdest times.  Sometimes the system would run for hours and sometimes minutes.  This caused me to go around in circles in the BIOS settings quite a bit as I was seemingly able to overclock, but never stable even at settings that were well below the CPU's capability.  I reached out for help on the overclock.net web forum, and two suggestions were made that would explain the symptoms:

1) Power supply instability
2) DRAM instability

I did not observe weirdness or fluctuation in the PSU operating voltages, but it is a used unit, so that could be a possibility.  Forum members confirmed that my specific power supply is relatively decent topology, so I looked to the memory to see if that was an issue.  I was running the memory at manufacturer recommended timing specification (9-9-9-27), voltage specifications (1.65V), and clock speeds ~1600MHz (rated at 1860MHz).  So I thought I was safe, but one of the forum members recommended stress testing the memory.

I downloaded Memtest and opened up 6 separate 2GB sessions simultaneously.  Multiple instances of the program are recommended to cover the installed RAM, and I have 16GB in this build.  Memtest should run flawlessly indefinitely without error, and my system threw up sporadic errors!  Incidentally, the errors seemed to generally correlate with the frequency of my blue screen crashes.  I understand next to nothing about RAM timing settings, so in my frantic Google searches, I came to understand that VERY generally:

- lower numbers for RAM timing mean faster operation and less stability
- higher numbers mean slower operation and more stability
- lower operating voltages indicate better quality RAM

I had always assumed that RAM is RAM, and it's all the same except for how much you have.  So, the fact that my specific PNY XLR8 modules are rated at 1.65V is not a good thing.  That means these are factory rated at an overclock already to achieve the "stock" performance specifications.  The factory recommended timing rating of 9-9-9-27 is also a bit "looser" than other brands.  But, there might be a reason I got 32 GB of the stuff for $80 on Craigslist!  For no other reason than looking at a bunch of other people's RAM timing settings on the internet and wanting to "loosen" my RAM timing settings incrementally, I set my timing to:

9-10-9-27

and ran Memtest again.  The RAM was running at 1600MHz still which is under the 1860MHz rating.  And to my amazement after wresting for days with an unstable system, the RAM seemed stable according the Memtest for a few hours.  The system also stopped randomly crashing to blue screen.



I learned from this little incident to make sure system memory is running stable (even if you are not overclocking it) before moving forward with working up a CPU overclock.  For someone coming in new, the random crashes really threw off my confidence that I was moving anything in the right direction and resulted in "random stabbing" at the BIOS settings instead of reasoned, methodical adjustments.

With the memory hurdle resolved, I was able to hunt and peck my way to a 4GHz overclock on the x3460 chip with the following BIOS settings

AI Overclock Tuner - MANUAL
OC From CPU Level Up - AUTO
CPU Ratio - 20 
Speedstep - DISABLED
Xtreme Phase Power - ENABLED
BCLK Frequency - 200
PCIE Frequency - 102
DRAM frequency - DDR3-1600MHz
QPI frequency - 6407 MHz
ASUS/3rd Party UI Priority - ASUS UTILITY
OC Tuner - TURBO PROFILE

DRAM - 9-10-9-27
Timing Mode - 2N

CPU Voltage Mode - MANUAL
Fixed Voltage (CPU) - 1.3625
IMC Voltage - 1.35
DRAM Voltage - 1.650
CPU PLL - 1.9625
PCH Voltage - AUTO
DRAM Data Ref Volt Channel A - AUTO
DRAM Data Ref Volt Channel B - AUTO
Load Line Calibration - ENABLED
CPU Spread Spectrum - AUTO
PCIE Spread Spectrum - AUTO


The Orange colored entries represent settings that I changed from the default BIOS settings.  Instead of talking about things I don't necessarily understand thoroughly, I found this guide to be helpful in explaining what a lot of these settings are and a much more concise methodology than I would be able to describe to working up a stable overclock on a very similar motherboard to the one used in this project.

http://www.masterslair.com/vcore-vtt-dram-pll-pch-voltages-explained-core-i7-i5-i3

While I was successful in working up to a 4GHz overclock, the temperatures under Prime95 stress test were touching 95 degrees which is more than I'm comfortable with so I would either need to upgrade my cooling or lower my settings or both.  For this specific chip, I believe I have found the point in the overclock curve where I reach significantly diminishing speed returns for significant increases in core voltage.

With a 4GHz successful overclock, I decided to run a few preliminary tests and benchmarks.


And, running Overwatch at mid/high settings revealed that this pile of spare parts would actually be able to run respectably at 144fps!  While this was always a pie in the sky goal of mine, for this build, I did not actually expect to achieve it so seeing these numbers in a real game with 11 other live players was very satisfying.  In-game performance was also very smooth.  I was also extremely happy to see the CPU running at 50-65 percent where my current LGA 1150 i5 setup is pegged at 100% and CPU limited in Overwatch.  As preliminary tests go, this was pretty good, but Overwatch even though running for the most part was still unstable and would crash every once in a while.



Up to this point, everything has been a bit frantic and disjointed, so for the purposes of maximizing my learning about overclocking, I decided to take advantage of the cheap prices on these used chips, and buy a few more samples to get a hands-on feel for the differences between different chips.  I figured if I am going to play the "silicon lottery", having a few more rolls at the dice could not possibly hurt.  In obscure forum posts from years ago, I found that some people believe the earlier X3440 xeon CPU's overclocked better than the X3460 and X3470 CPU's.  Looking at the technical specifications, these earlier chips did not seem to have different specifications besides the slightly lower stock core clock value which doesn't much matter when overclocking.  I found a seller on Ebay and made an offer of $18 each on two X3440 CPU's, and the seller accepted.

Because there are so many tweak variables in BIOS for overclocking, I decided to narrow down my test parameters to compare the 3 chips.  I decided to leave everything the same in my BIOS settings as many of my "accessory" voltages (IMC, PLL) were already near maximum safe settings.  I would dial back my CPU core voltage to 1.3V and then test each chip adjusting only the BCLK (base clock) frequency setting and see which chip achieves the highest value and note the temperatures along the way.

One of the benefits to the old Thermalright Ultra 120 cpu cooler that I am using is the mounting system allows for CPU changes WITHOUT needing to access the back side of the motherboard which is perfect for my build because the older case does not have an access hole in the motherboard tray.  I start by disconnecting the CPU fan so I don't have to cut the zip ties holding the cable to the main bundle.  The wire clips holding the fan proved very convenient for easy fan removal and installation.


Next, two "spring screws" release the heatsink.



One of the brackets interferes with the CPU latch, so one screw needs to be removed and the bracket can pivot away for clearance.



then, the CPU can be removed, cleaned and swapped.



With a slightly better plan going into the test, I was able to get some interesting results and get a feel for the different chips, and record some of the data in a useful manner.


x3460 1.3V overclock
BCLK1703426.34 MHzbootsstable68-65 degrees
BCLK1803611.04 MHzbootsstable71-67 degrees
BCLK1903809.35 MHzbootsstable74-70 degrees
BCLK2004012.67 MHzbootsprime95 errors test stopped
BLCK1953910.29 MHzbootsprime95 error worker #2 test stopped
BLCK1943893.11 MHzbootsprime95 error worker #6 test stopped
BLCK1933870.20 MHzbootsstable75-71 degrees
x3440 #1 CPU 1.3V overclock
BCLK1703426.34 MHzbootsstable77-75 degrees
BCLK1803611.04 MHzbootsstable80-73 degrees
BCLK1903809.35 MHzbootsstable83-75 degrees
BCLK2004012.67 MHzbootsstable87-78 degrees
BCLK2054130.79 MHzbootsstable88-79 degrees
BCLK2104213.12 MHzbootsblue screen
BCLK2094190.22 MHzbootsblue screen
BCLK2084173.03 MHzbootsblue screen87-78 degrees
BCLK2074151.56 MHzbootsstable90-80 degrees
x3440 #2 CPU 1.3V overclock
BCLK1703426.34 MHzbootsstable69-65 degrees
BCLK1803611.04 MHzbootsstable72-68 degrees
BCLK1903808.35 MHzbootsstable75-70 degrees
BCLK2004012.67 MHzbootsblue screen
BCLK1953910.29 MHzbootsstable77-72 degrees
BCLK1963931.06 MHzbootsblue screen78-72 degrees
BCLK1973967.57 MHzbootsblue screen79-73 degrees

Generally, running this test gave me a bit of feel for how different chips heat up when clocks increase (core voltage remained at 1.3V for the entire test).  It also allowed me to become a bit more efficient in my workup methodology selecting a "safe" start point and moving higher by steps of 10.  Then, if I hit a wall, I could split the different by a step of -5 to the last stable setting.

Originally, X3440 #1 worried me because of the high temperatures right at the gate at relatively low base clock settings.  It started out a full 10 degrees hotter than the original X3460 chip, but it ended up clocking higher than the other 2 chips by a significant margin. 

X3440 #2 seems to track very closely with my original X3460 chip for temperatures but achieved a slightly higher stable base clock speed.

So, from my limited test samples, the rumors that I read no the internet were true.  Both X3440's at least from this test appear to have more overclock potential than the X3460.

After reviewing the multi chip test results, I decided to swap the X3460 CPU for X3440 #2.  I am hoping this chip overclocks slightly better while shaving $16 off of the overall parts cost.  X3440 #1 seems to be the thoroughbred of the group but I feel the chip's high temperatures might require more drastic cooling measures that would alter the scope of this project too far.

With a chip swap determined and some data to work with, I decided that my temps especially under Prime95 were higher than I wanted to see.  So, I decided to try an upgrade my cooling without breaking the bank.  I found an older Cooler Master V8 cooler (without fan) on Craigslist for $20 that I thought could help.  I figured more heat pipes, more fin surface area, and a faster fan couldn't hurt.



 On a trip down to Southern California, I discovered a cool store that buys corporate tech excess, store return merchandise, retail overstock inventory, and other bulk batches of computer parts and mostly sells to computer stores and various vendors, but just started to dabble in retail operations and opened a very small storefront with a few local Craigslist ads bringing some visibility.  All sales are cash only often with no guarantees on "as is" items, but for someone who is a bit risk tolerant looking for some bargains, it was a treasure trove of great parts (if they work), and I did go a little bit crazy.  Among my purchases were a few Cooler Master T4 coolers and a Hyper 212 EVO (without mounting hardware) all listed at $5 each.  I scavenged the PWM fan from one of these coolers because I figured at a rated 2000 RPM 81 CFM, it may outpace the Arctic F12's 1350 rpm 75 CFM rating.  At this point, quiet operation at load was completely out the window with this overclock.  I needed as much airflow as I could get so I was only concerned about max specs.

For people interested in checking out this interesting computer parts seller, the contact information is copied below:

2300 S. Fairview Street
Santa Ana, CA 92704
(714) 200-9334

In the back of my mind, I had lingering doubts about the current Thermalright Ultra 120 cooler because of the curve I discovered on the heatsink/cpu mating surface, so having installed and removed this cooler a few times, I noted the thermal paste spread which shows the point of contact.




After removing the Thermalright cooler, I assembled and installed the Cooler Master V8.  this cooler does seem better machined and the mounting system is very robust although it requires access from the back side of the Motherboard for installation.


I noted that it locks down tight an does not wiggle around like the Thermalright and Hyper 212 Evo coolers but due to the larger size, it is a bit more of an ordeal to get set.


I also took the opportunity to move the case panel to motherboard wiring out from behind the motherboard because I was picking up noise in the headphone wires likely from running right along and directly in contact with the bottom of the motherboard.


And right off the bat, the temperatures were just horrible!  I was running 4 degrees hotter than with the Thermalright Ultra 120 installed.  I was immediately thankful that this mistake only costed me $20 for the tuition.  Researching deeper online revealed that in somewhat controlled tests, the Ultra 120 does indeed outperform the Cooler Master V8 by a significant margin.  The Hyper 212 EVO should beat the Ultra 120 by a few degrees, and the Cryorig H7 should be better still by about 3-4 degrees.

There may be installation issues to contend with as the V8 has a very flat machined mating surface and the Ultra 120 has a pretty defined convex shape.  One shape may favor more thermal paste than the other, but after removing the V8, I found the thermal paste coverage to be just about perfect.  My decision would be whether to reinstall the Ultra 120 and add an additional fan to try and get 1-2 degrees better or move to a Hyper 212 EVO or Cryorig H7 both of which would add cost.  I decided to go for push pull on the cheapest option whihc is the Thermalright Ultra 120 that I've been using all along.  One of the problems I had trying to figure out push pull on the Ultra 120 was I could not find extra fan clips for this cooler anywhere including the manufacturer.


After staring at it for a bit, I visualized a zip tie solution that may not be completely horrible.  I would deploy the two original clips that I had one on each fan.





Then, I zip tie the other side together through the fan's screw holes.


As undignified as this solution is, I figure I'd try to clean it up as best I can.


And, in the end, I was able to stabilize the overclock to 201 Base Clock  at 1.35625V core voltage to achieve 4.031GHz running 88-80 degrees in Prime95.  This setup also ran AIDA64 stress test for 8 hours at 85-79 degrees.  Removing the side cover of the computer case drops temperatures a solid 5 degrees.  I guess in the end that's the cheapest cooling upgrade!  These were my workups to the overclock.


---- 1.35V overclock
BCLK2004012.67 MHzbootsstable91-83 degreesCM V8 cooler
BCLK2004012.67 MHzbootsstable84-78 degreesThermalright Ultra 120 push pull
BCLK2014031.28 MHzbootsblue screen ~20 min84-78 degrees
BCLK2024052.76 MHzbootssystem lockup84-78 degrees
----1.35625V overclock
BCLK2024052.76 MHzbootsblue screen86-79 degrees
BCLK2014031.28 MHz

The BIOS settings that I ended up with were:

AI Overclock Tuner - MANUAL
OC From CPU Level Up - AUTO
CPU Ratio - 20 
Speedstep - DISABLED
Xtreme Phase Power - ENABLED
BCLK Frequency - 201
PCIE Frequency - 105
DRAM frequency - DDR3-1600MHz
QPI frequency - 6407 MHz
ASUS/3rd Party UI Priority - ASUS UTILITY
OC Tuner - TURBO PROFILE

DRAM - 9-10-9-27
Timing Mode - 2N

CPU Voltage Mode - MANUAL
Fixed Voltage (CPU) - 1.3625
IMC Voltage - 1.35625
DRAM Voltage - 1.650
CPU PLL - 1.90
PCH Voltage - AUTO
DRAM Data Ref Volt Channel A - AUTO
DRAM Data Ref Volt Channel B - AUTO
Load Line Calibration - ENABLED
CPU Spread Spectrum - DISABLED
PCIE Spread Spectrum - DISABLED

After I had this overclock set and apparently stable, my testing in Overwatch resulted in a couple nights of pure frustration as the game continuously crashed while I was trying to connect with friends online.  During the chaos, I was frantically trying to figure out the problem between freezes and reboots.  Was it a the memory problem returning?  Should I loosen the timings more?  Should I increase DRAM voltage?  Was it a PCIE issue?  Should I lower the PCIE multiplier?  What about increasing IMC voltage?  I immediately flattened out the GPU to stock settings, and after taking a step back for a while and thinking the problem through more methodically, I figured the most straight-forward solution would be to either increase voltage (adding heat that my cooling setup is already marginal on), or to decrease the overclock to Base Clock 200 4.012GHz which is a bit disappointing because it does not "beat" the original X3460 chip, but stabilizes the game which is the most important thing.  So, in the end, not being able to tolerate the extra heat that raising the core voltage more would generate, I reluctantly dropped the base clock down to 200 (4.012GHz) with a core voltage of 1.3625V and called the CPU overclock finished.  At this level, with no graphics overclock, Overwatch finally stabilized.

With the CPU overclock set and running stable in Prime95, AIDA64, and Overwatch, I turned my attention to finding the GPU overclock.  Since for whatever reason, this system would work fine in Unigen Heaven and Valley on an overclock setting and then fail FireStrike, I decided to simply use Firestrike to work up my overclock.

The memory overclock on this card was pretty insane. . . it ran all the way up to +870 which is miles above what I read are the average GTX 970 results (+500).  So, even though it passes tests, including Overwatch, I am cautious about accepting these.  But, it is great to know my card's memory bats a significant margin above average.


mem 399gpu 204overwatch worksfirestrike won't run
mem 520gpu 0firestrike pass
mem 550gpu 0firestrike pass10119
mem 570 gpu 0firestrike pass10098
mem 600 gpu 0firestrike pass
mem 620gpu 0firestrike pass10171
mem 660gpu 0firestrike pass10156
mem 680gpu 0firestrike pass101512nd try 9865
mem 700gpu 0firestrike pass9875
mem 720gpu 0firestrike pass9878
mem 730gpu 0 firestrike pass9920
mem 750gpu 0firestrike pass9911
mem 770gpu 0firestrike pass9904
mem 790gpu 0firestrike pass9913
mem 810gpu 0firestrike pass9936
mem 830gpu 0firestrike pass9961
mem 850gpu 0firestrike pass9966
mem 870gpu 0firestrike pass9942overwatch ok10235 after reboot
mem 890gpu 0glitches!

I also note after continuous runs, some sort of system resource thing happening in the middle of the testing that dropped the Firestrike scores significantly, but I was testing for stability, so as long as the system continued from there with somewhat intuitive results as I worked upwards, I was ok to continue.

Next, I attempted to overclock the GPU with my max memory overclock in place which hit a brick wall on Firestrike.


mem 870gpu 150no go
mem 870gpu 80artifacts lost picture10568
mem 870gpu 40artifacts10399
mem 870gpu 20artifacts10023

Knowing from online data that my memory overclock is outside of norm, I randomly selected a "very good" VRAM overclock number and began working up a stable GPU overclock.  Notice I am getting a bit more efficient with the test values I select.  Instead of straight increment increases, I am making larger jumps and then splitting the results in half up or down depending on whether the test passed or failed to quickly zone in on the max stable value.


mem 700gpu 150no go
mem 700gpu 20firestrike pass10237
mem 700gpu 100firestrike pass10507
mem 700gpu 125no go
mem 700gpu 112firestrike pass10540
mem 700gpu 118no go
mem 700gpu 115no go
mem 700gpu 113no go

With my GPU overclock settled at +112 GPU and +700 Memory, I achieved the highest Fire Strike score thus far on the machine at 10,540.


 But, as luck would have it, Overwatch continuously crashed.  In order to be Overwatch stable, I had to drop the GPU all the way down to +85.  With that drop, the final Overwatch stable score is 10,458, and while I'm disappointed that I had to drop the GPU overclock, I can certainly live with that score on a stable system.




As for Overwatch performance and stability, I have played a few solid sessions with no more hints of instability.  The overall frame rates differ between the different maps, but I have included some of the actual in-game graphs that show the variations from map to map.



My in-game graphics settings are show below, and they are certainly FPS optimized, but we are nowhere near the point of "suffering" here.  The game looks quite good with these settings and runs beautifully smooth.


Cinebench yields a CPU score of 629 and an OpenGL frame rate of 89.81 fps.


I noted earlier that the USB 3.0 and SATA III 6G implementation on this motherboard is a bit clunky and activating either of these features reduces the main pcie slot to 8x.  Up until now, I have not activated either of these features, but I wanted to see if graphics performance would suffer from running at 8x.  From what I have read online, a GTX 970 should not be able to saturate the entire 8x bandwidth much less 16x, but I engaged the 6G "powerup" feature in BIOS and ran firestrike again to see if there would be any deviations.


The score came back 4 points lower which for all intents and purposes is the same.

And with that, I will likely continue to tweak on the settings here and there, but this build is pretty much finished.  I feel the system has a great look and quite a decent upgrade path.  It can accept a larger CPU cooler, one more SSD and 3 additional hard drives via tool-less mounts.  A 4th drive could easily be installed on the lower tray without much difficulty at all.  Motherboard and CPU can be updated at some time in the future even though the current overclock holds up quite well in Overwatch and I would suspect most other modern titles.  I also think the GTX 970 still has some good legs for modern games.

At a final overall system parts cost of $390.42 (including the OS).  I feel this PC hits a really nice price performance proposition.  While not as miraculous as many other budget examples online in terms of overall cost, I think one would be hard pressed to do much better without freebies and impossible bargains.  The performance target of 144hz gaming was ambitious for the budget, and I think this system really steps up and games hard exceeding a lot of expectations.  I am not sure if it will perform well in streaming, but from a pure player's perspective, with most professionals running low settings all around, I think this computer would serve an up and coming serious gamer quite well and I intend to find it a good home!

If you've read this far, thanks for sticking it out with me through this build!  I had a great time shopping, case hacking, parts swapping, and overclocking this system, and I learned a ton of new things without too much breaking of parts!  But, if I were to break things in the learning process, this would certainly be a better place to do so than on premium components.  Old "left-over" parts definitely allowed me to be a little bit brave and push some boundaries.  Luckily, I did not blow anything up this time around.

If you missed any of the build segments, you can catch them in the links below:

"Needy Healing" - Used Parts Overwatch PC Build - INTRODUCTION
"Needy Healing" - Used Parts Overwatch PC Build - PART 2 (parts hunting)
"Needy Healing" - Used Parts Overwatch PC Build - PART 3 (case hacking)
"Needy Healing" - Used Parts Overwatch PC Build - PART 4 (assembly)
"Needy Healing" - Used Parts Overwatch PC Build - PART 5 (setup and overclock)

UPDATE:

After stabilizing the system, I decided to try and activate the USB 3.0 and SATA 6G functionality, and unfortunately, I was unable to get USB 3.0 to function correctly on my external SSD, but changed the internal SSD setting from IDE to AHCI and enabled SATA 6G as the previous Firestrike test indicated no significant performance loss on this graphics card by the reduced PCIE bandwidth.  I was able to see about 120MB/s increase in max disk read speeds, but the interesting thing is a dramatic increase in Overwatch FPS upon initial test.  I was playing a hero I don't normally play (Reinhardt), so that could account for it, but FPS was pegged at over 200fps for half a game on Hanamura.  I will need to play more to confirm, but I have no real explanation for it.


. . . and the next round on the Lijiang Tower map confirmed that there were no miracles happening.  Probably one of the lower frame rates I've recorded.


No comments:

Post a Comment