Troubleshooting Matt's New Gaming Rig

Southern Utah Makerspace
459 N 1300 E St. George, UT

I am a member of Southern Utah Makerspace. We had just paid rent on a new location so that we had a place to meet and make stuff but before moving in we needed liability insurance. I took the advice of our new landlords and contacted Matt Howe as he has provided insurance for them and other tenants before. It took a while for a makerspace to be understood by the underwriters that needed to sign off on the policy so I had also shopped it around a couple other insurance companies but Matt pulled through quicker than the others. When I went to his office to button it all up, I saw a water cooler and a motherboard box in his office. It caught my attention and I soon discovered that Matt had a lot in common with myself and many of the other people in our makerspace: we like fast, graphic capable computers and sometimes those computers don't work. Matt explained his enthusiasm for his new machine and his frustration that the local computer shop was not able to resolve his problem. I explained that a community of technically inclined people can often be a better solution than shelling out dollars to an expert that can do nothing more than shrug it off as a difficult problem. I asked Matt to become a member of the makerspace so that he could experience the magic of the community and make some more techy friends along the way. I told him that I would document the experience so that others might understand where the process goes ugly, and where it finds resolution.

Matt's Rig

AMD FX 9590 CPU

Cooler Master Seidon 120M closed loop liquid cooler

ASUS Crosshair V Formula-Z Motherboard

16GB RAM (4x4GB)

Intel 120GB SSD

1TB Hard Drive

MSI 390X Video Card

Power Supply

Case

Matt's use case was World of Warcraft. It would crash about 45 minutes into playing. Knowing that WOW is not particularly taxing, I suspected that there must be a pretty obvious issue. Since this was his first custom build, it seemed likely that something was not hooked up right or that he had caused some ESD (Electro Static Discharge) damage. A shock so little that cannot be felt can do serious damage to sensitive electronics. Isolating the component would be possible since I have a similar gaming rig, which allows me to swap out parts and narrow it down to the offending component.

My Rig

Intel CPU

Closed Loop CPU Cooler

ASUS Motherboard

12GB RAM (3x4GB)

512GB OCZ Vertex SSD

1TB Hard Drive (RAID mirror)

Gigabyte 290X Windforce

Platinum Rated Corsair Power Supply

Case

Large power connector on motherboard

I first checked that all of the components were installed correctly. He had the fan for the radiator on the CPU cooler connected to the large style molex connector on the motherboard. I had never seen this on a motherboard and it was reverse gender to be plugged in as he had it. It turns out that this can be used to supply additional power to the motherboard. But since it is on the same power rails as the rest of the board, you can also draw power from it. I changed it to use a chassis fan connector and left this additional power for the motherboard disconnected.

Unused 4 pin extra motherboard power

There was also an unused 4 pin connector that I thought was necessary but it turns out that it is optional because the similar 6 pin style extra power connector was being used. Initially, we were very concerned because his modular power supply did not come with one of these cables. It turns out that we were worried for nothing.

I needed a more predictable and automated use case. I don't game much anymore and I had no desire to play WOW to an impending crash. I decided to use 3DMark Basic Edition. It is a benchmark package that can run a while and work the system pretty hard without me having to do more than click start. The basic edition is free and download links can be found at:

http://www.futuremark.com/benchmarks/3dmark?_ga=1.74296913.2128338082.1442750633

My video card has been RMA'd twice before. I really punish it by mining cryptocurrency. Upon the first RMA, I found that using a better power supply really helped it run higher hash rates. That's when I upgraded to the platinum level. Since Matt's computer had a similar video card, I first suspected the power supply, after a switcheroo and 3dMark, the machine locked up just the same as it did with its original power supply.

I pulled just 2 of the 4 RAM modules from the motherboard and it passed. Well, it seems that he either has some bad ram, or a bad ram slot. I didn't use the ram from my computer initially because I thought he was using DDR4 since it is a brand new machine. I thought this was a limitation to my testing process but later discovered that he had DDR3, just like my machine. I swapped the ram sticks, seeing if I could confirm that it was either the ram or the slots. It failed again so I thought it was the ram. I then replaced the original 2 sticks and it failed again. And another test confirmed another failure. I guess the one time pass was a fluke. No combination of ram in the slots would produce another pass, so I concluded that the ram was not the cause.

The video cards in the 2 machines are similar but not exactly the same. They both use the catalyst drivers so a quick detection and reboot is all that is needed to swap them. When using his video card in my machine, it passed every time. When using my card in his machine, it failed every time. The video card was not the cause. I also played with the overclocking in order to reduce the speed of the card and to pump up the fan speed to 100% all of the time. Temperatures never got high but I was unable to see the temp while running the 3dMark app because it runs in fullscreen. I ran some other non-fullscreen apps to watch the temp of the graphics cards and the CPU and it all seemed rather low. I should have kept closer track for the sake of this post but I did not.

I then realized that the ram in the new machine was DDR3 so I thought it might be that all 4 sticks of his ram were bad. When I placed my ram in his motherboard, it still failed. At this point, I thought it must be the motherboard that is causing the issue. Matt sent it back to newegg they sent him a replacement.

A week later and we had the new motherboard. And it still failed. My confidence was kicked and I began to think we would not figure it out.

Next on the hit list was the SSD. A few years ago, I had a Crucial M4 256GB SSD that would lock up after it had operated for a certain amount of hours (about a year and a half, if I remember) and it would lock up consistently after that time about every 15 minutes (which might be wrong but I cannot find the exact reference). It seemed odd that a problem like this would still be around since SSD have become very popular but I thought I'd check into it and, of course, there were no newer firmware updates available. Even less likely, but still possible, it might be a problem that has not yet been identified and fixed by new firmware. So I decided to clone the SSD to the hard drive so that I could boot and test from the hard drive while the SSD was completely disconnected. In the old days, I would have used ghost. But it has been a while and since I did not want to steal or buy ghost, I looked around for an open source tool. I seem to remember using partimage (or something like that) long ago but I found and used clonezilla instead. It did the trick although I needed to use the command line and remove the GPT info on the SSD since it was not copyable to the hard drive (I have no idea why). I followed the directions on screen. GPM (General Purpose Mouse) was running on this fine little linux distribution so I could highlight the command with the mouse, right click to pop it in the clipboard, and then click to paste the command. You can download clonezilla for free at: http://clonezilla.org/downloads.php

Using the hard drive for the computer did not fix the issue. It still locked up as usual. I thought I might be on to something because the lockup usually seemed to occur in between the tests, when it was loading the scene. It seemed that it would make sense that the disk was most in use while loading the scene but I was wrong again.

Well, that does not leave much left to suspect besides the CPU. I have an AMD APU in my FreeNAS server. It has served me well, but it is not doing any hardcore graphics. I have not used another AMD CPU since the K6... a long time ago. I generally will never suspect a CPU as the cause, but I could not think of any culprits. A quick search in google for AMD FX 9590 lockup returned the following page on Tom's Hardware:
http://www.tomshardware.com/answers/id-2191595/stabilize-9590-freezing-completely.html

This seemed to describe my experience exactly. I also corroborated the information in a few more pages that resulted from the same search.

CPU Ration Before

CPU Ratio After

Advanced CPU Settings Before 

Advanced CPU Settings After

So I only changed 3 settings from the default:

1. Changed the CPU ratio from Auto to 20.0. Auto for this CPU should be 23, which would clock it at 4.7GHz. By changing the ratio, it only clocks to 4GHz. That's painful especially when the 9590 is sold as "Black Edition" chip that should be capable of overclocking.
2. Change the Cool 'n' Quiet to Always Enabled. This probably doesn't make much difference, but it was suggested and a quiet PC is a luxury.
3. Changed the Core C6 State to Disabled. While I understand that this has to do with power management, I didn't take the time to really understand it. Many people said that things worked when they turned it off, so I turned it off. The C1E was already disabled in the default profile.

UPDATE:
I changed the CPU ratio back to auto and it passes all of the tests. I guess it is that Core C6 state that caused all the problems.


If you'd like to get some insurance from a guy that you can gab about gaming with too, please contact:
Matt Howe Insurance
(435) 674-8771
321 Mall Drive
Building R
St. George, UT 84790

Tell him that Ben Engel sent ya.