Hardware Posts · HookRace Blog

New Hardware and Hacks

2016-07-03T00:00:00-04:00

In my latest post I showed some examples of how I ran mostly the same PC hardware over a period of 8 years. Today I finally finished setting up my new PC hardware in my new home, so I can report about what I did differently, my thought process, and some problems I encountered and hacks I did to solve them.

I’m not sure if this article is interesting for anyone, but I had some fun setting the new system up and felt like writing about it, so here we go:

Case

I wanted a smaller case since the old one was a huge ATX tower with way too many useless ports and fans. I still want a pretty performant system, so an Intel NUC is out of the question. I don’t need a GPU, since the most challenging graphics I need is DDNet client, which easily runs at 400 fps at 1920x1080 and 130 fps at 3840x2160 on Intel’s HD Graphics 530 IGP integrated on Skylake CPUs. But it’s still nice to have a PCIe port in case you ever need it. So Mini-ITX it is.

Unfortunately the Streacom DB4 is not available yet, as it would have made for a beautiful silent Mini-ITX cube. Instead I went with the In Win Chopin, which is even smaller, features no GPU port and has a small integrated PSU rated at 150 W. I was a bit worried about the PSU being strong enough, but you can read more about that later in this post. There really aren’t that many good-looking and reasonable Mini-ITX cases.

The idea of the In Win Chopin is to use the CPU fan as the only fan to cool the entire case, pulling air directly from the side, pushing it out of the top. That’s a stark contrast to my old machine with a total of 9 fans whirring inside the case.

CPU Cooler

Of course that means I need a good CPU cooler for the small space. With just 43 mm space for the cooler my choice fell on the Noctua NH-L9i. Unfortunately the manufacturer claims a 91 W Skylake CPU barely runs with the NH-L9i. So let’s check out the CPU and measure some actual temperatures. Even though I switched to a small form factor, another goal of mine is still to keep the noise down to a level where I don’t notice it.

CPU

There are just two 91 W Intel Skylake CPUs and my choice fell on the i7-6700k. Since I never had a failed CPU before and planned to play around a bit with the CPU anyway, thus voiding the warranty, I decided to get a pre-owned one for cheap. Apparently the CPU was unused, but I wouldn’t believe that.

There are much nicer Xeon CPUs, especially if you’re willing to risk the problems of running an engineering sample, but I wouldn’t be able to cool them properly, wouldn’t need all that performance in the end and miss out on the IGP (integrated graphics processor), needing a separate GPU instead, taking up more space.

Unfortunately it turns out that Skylake can only decode 8 bit H.264 and H.265 in hardware, while most sources are switching to 10 bit, including UltraHD Blu-rays and Netflix. If I had the chance I might have waited for Kabylake, Intel’s next CPU release.

At first I installed the i7-6700k with the regular Noctua thermal paste and used it like that for a few days. Unfortunately the system ran louder than I wanted. And changing the fan settings was not a solution since the CPU reached 80°C at load. Clearly I needed a better solution. I already expected this and had some Thermal Grizzly Conductonaut liquid metal thermal paste ready to use.

Liquid metal thermal pastes conduct heat much better than regular thermal pastes, but unfortunately they also conduct electricity, so you have to be careful when applying them. Usually people use this thermal paste to achieve better overclocking, but my goal is opposite, for now I want a more quiet system running at low power. Maybe I’ll overclock if I ever need the extra performance.

So I delidded my new CPU by removing the heat spreader with a razor blade. You need a very careful hand for that as a single scratch into the PCB could ruin your CPU. Of course this also voids the warranty if you have one. I removed the regular thermal paste between the CPU die and the heat spreader and applied the Conductonaut instead. The heat spreader can be reattached with regular silicone from a hardware store. Finally I used the same Conductonaut thermal paste between the heat spreader and the cooler as well.

Finally I dared to start my system again, luckily everything still seemed to work and I couldn’t quite believe the new temperatures: Instead of 80°C at load I now had just 60°C. I was hoping for a 10-15°C improvement, so that was a nice surprise.

My last step was to undervolt the CPU and IGP as far as they would go. You have to be careful about system stability with this. I set up a few different work loads to test stability. This further reduced the load temperature to 55°C. That’s a temperature I can definitely live with and doesn’t even require full fan speed to keep up. At regular light usage the CPU stays at 30-35°C and I can’t even hear the fan. I don’t dare to turn off the fan entirely, since it’s the only active cooling solution for all the other components on the mainboard as well.

Screen

In a few months Planet Earth 2 will air in Ultra HD. Since I’m a huge fan of David Attenborough and nature documentaries in general I clearly need a screen capable of showing it. I also want to use my screen as a television at the same time, so this fits in well.

So the plan was to get a 40” screen with a 3840x2160 resolution, which enables me to work as if I have four 1920x1080 screens, positioned in a 2x2 grid. Luckily this is pretty simple to setup with xmonad, my window manager. Using the LayoutScreens module I can switch between pretending to have a single large screen or 4 small ones. Since xmonad configuration is just Haskell programming more layouts would of course be possible, but so far I’m happy with this setup.

It turns out that there is still a problem here which I can’t quite pinpoint yet: When I use xmonad, mpv, NetWM support, 3840x2160 resolution, and the Intel IGP driver’s TearFree setting, memory leaks and after a few minutes of watching some video all 32 GB of RAM are used up and a process is killed. Changing any single component in this list gets rid of the problem. For now I use DRI3 instead of TearFree, which has slight tearing only in very special cases that I never encounter in regular usage.

But one problem I imagine to have with a large screen is that the left and right edges are so far away that they look distorted. After all when I use two regular screens I rotate them a bit so I can look at them head-on. So instead I wanted a curved screen to correct the distortion and get a similar effect to having the screens rotated. I’m not really convinced of curved screens for a pure TV setup, but as a computer screen they are great.

Luckily the Samsung UE40JU6740 fit all my requirements and was available for cheap since it’s last year’s model. Getting it to display 3840x2160@60Hz with reasonable colors is a bit challenging though. There are even guides about how to properly set these TVs up as computer screens. In the end I have a far better picture than with my old screens, but the input delay of 40 ms is noticeable when playing DDNet. Since I don’t play much I don’t mind and in case I ever need it it’s possible to get it down to a more reasonable 20 ms at the loss of color accuracy.

Mainboard

But now I faced a new challenge: TVs generally don’t support DisplayPort, so in order to get 3840x2160@60Hz HDMI 2.0 is required. That’s another thing Skylake CPUs don’t support, so I had two choices: Get an external converter from DP 1.2 to HDMI 2.0, which Club 3D offers. Or alternatively get the only Mini-ITX mainboard with HDMI 2.0, the ASRock Z170 Gaming-ITX/ac. So now I have a gaming mainboard even though I don’t want to game, oh well.

In the end both the Club 3D adapter as well as the ASRock motherboard have nearly the same chips inside, a MegaChips MCDP28.

Unfortunately the board still has a few issues even after having been released nearly a year ago:

Using the HDMI 2.0 port at 3840x2160@60Hz causes a frame with artifacts every few minutes, both in Windows and Linux. In Linux sometimes the screen goes black entirely when I switch on stereo HDMI audio, while mono works. My assumption is that this is a bandwidth or synchronization problem. So for now I disabled HDMI audio and this seems to fix the image artifacts. I never planned to use the HDMI port for audio since I still have my fancy amplifier and speakers, but it would be nice to get this working.

Another temporary workaround to get working HDMI 2.0 audio on Linux is to user another CRTC manually by running xrandr --output DP2 --crtc 2. But in my experience it’s not 100% reliable and I’m not even sure why it works.

Actually ASRock offers a firmware update for the MegaChips HDMI 2.0 chip, but it doesn’t run on my system and I’m still talking to the support about it.

The next issue was that resuming from S3 sleep was broken, also both in Windows and Linux. I assume this might also have to do with the HDMI 2.0 chip, but I’m not sure yet. So far my only solution is to keep using an older BIOS version (1.8) which does not exhibit this problem.

Instead I get a very strange issue after resuming from sleep. In DDNet client I get 40 fps instead of 500 fps after the first sleep cycle. My first idea was that it had something to do with the IGP and power saving, but I couldn’t find a way to pinpoint it. Finally I noticed that some screens in DDNet have far fewer FPS than others. Finally it dawned on me: The more text is shown on screen, the slower it gets.

Nevertheless it had nothing to do with font rendering. Instead the font rendering system in DDNet client, which is inherited directly from Teeworlds, needs to call the gettimeofday syscall A LOT. Checking dmesg then confirmed that after a S3 sleep the Linux kernel clock source switched from TSC to HPET:

Jun 28 22:49:23 al kernel: TSC synchronization [CPU#0 -> CPU#1]:
Jun 28 22:49:23 al kernel: Measured 3072272461 cycles TSC warp between CPUs, turning off TSC clock.
Jun 28 22:49:23 al kernel: tsc: Marking TSC unstable due to check_tsc_sync_source failed
Jun 28 01:17:48 al kernel: clocksource: Switched to clocksource hpet

Manually setting tsc=reliable makes the clock run backwards after a sleep cycle. From what I read it sounds like the BIOS is erroneously changing some TSC registers at suspend or resume. I didn’t find a way to fix these registers again in the Linux kernel.

Fortunately the support quickly gave me a BIOS that is supposed to fix this issue. Unfortunately I can’t test this since every BIOS version newer than 1.8 doesn’t resume from S3 sleep at all for me.

Instead I did the best I could and fixed DDNet client to call clock_gettime two times per frame instead of potentially thousands of times per frame (once for each text glyph that is rendered). Some measurement showed that getting the time with TSC requires 18 ns, while HPET needs 4700 ns on my system, quite a remarkable difference.

Storage

As a nice bonus the mainboard also has an M.2 port on the backside, so I can run a fast Samsung SM951 SSD. Unfortunately being located on the backside of the mainboard means that air circulation is rather bad, so the SSD runs at a temperature of about 50°C.

If this becomes a problem later I can try a thin thermal pad to get contact between the SSD and the metal plate inside the case, thus using it as a cooler.

RAM

For memory I actually just bought the cheapest 32 GB DDR4 kit from Crucial at 2133 MHz. Turns out that you can undervolt the RAM a bit and still overclock it to 2700 MHz. That helps the IGP a bit, but is otherwise not really noticeable. It turned out that you actually need dual channel RAM to achieve 3840x2160@60Hz, so memory bandwidth is actually a concern.

Keyboard

My choices regarding a keyboard were very limited. I wanted a flat keyboard but no rubberdome since they kept annoying me with their inconsistent feeling and bad quality. The only keyboard to fit these requirements was the Cherry MX-Board 3.0. With MX-Blue switches it’s quite loud to type on, but makes my other keyboards feel incredibly cheap.

Mouse

Actually I rarely use a mouse since I feel at home in the terminal and use a few programs with vim-like key bindings, like the Pentadactyl extension for Firefox. Still I sometimes like to have a mouse and I want it to fulfill two purposes:

Regular use next to keyboard, for example for casual gaming
As a remote control from my couch for watching movies

The first is easy to fulfill of course, you could even buy 2 cheap mice from eBay for 1 € including shipping. The second purpose requires a wireless mouse that works well on any surface, a perfect application for the Logitech Anywhere MX2. It even feels a bit like a small version of my old MX518, quite a nice bonus.

To actually use the mouse as a remote control for movie watching I had to set a few mouse button binds in mpv, my video player:

MOUSE_BTN5 run "mixer" "pcm" "-2"
MOUSE_BTN6 run "mixer" "pcm" "+2"
MOUSE_BTN1 cycle sub-visibility
MOUSE_BTN7 add chapter -1
MOUSE_BTN8 add chapter 1

Now I can move through the movie manually or by chapter, switch through audio tracks, enable subtitles and switch through them, change the global system volume, toggle fullscreen, and pause and play. That’s quite enough for me.

Since I barely touch the mouse I expect the battery to last forever. And since my mainboard already supports Bluetooth, I used that instead of the dongle.

Unfortunately you have to extract the Bluetooth firmware from the Windows driver.

Power Consumption

Initially I was worried about the 150 W PSU of the In Win Chopin. But after some measurements my concerns turned out to be unnecessary. Since I undervolt my CPU instead of overclocking it, the maximum I managed to achieve by putting high loads on CPU and IGP at the same time was 90 W, measured at the wall. At regular usage the system needs about 16 W, which I consider to be a pretty fine result for such a powerful machine.

Final Result

The final build assembled in its new environment, where I’m sitting right now on a Sunday morning to write this:

Closeup of the computer case:

Broken Hardware, Fixes and Hacks over 8 Years

2016-06-21T00:00:00-04:00

After reading the feedback of my recent article about running DDNet, I noticed that people found it interesting how I’m trying to minimize money and resources. I also noticed that I had been doing something similar with my personal computing hardware setup for an even longer time.

I’ve mostly been using the same hardware for personal computation purposes over the last 8 years. In this article I want to talk about some of the hardware I’ve been using, how it broke and how I fixed the problems or worked around them. My goal was to be frugal about hardware, to keep using the same hardware for a long time and repair it when possible instead of simply buying new hardware.

The reason for this post is that I’m moving and abandoning my old hardware setup. There may be some interesting tales in here. Depending on how you value your time it is probably cheaper to simply buy new stuff instead of repairing old ones, but I consider it a fun activity and a much more rewarding experience.

Whether you call it planned obsolescence or just cheap manufacturing, it’s a fact that electronic hardware tends to break rather quickly. Most reviews don’t take any note of this and so it is difficult to find out which hardware to buy if you value durability. The best I can do is tell you what you should and should not have bought 8 years ago.

For each of these topics there are probably much more detailed reports and guides online, so I will keep it short and just give an overview. If you’re interested in fixing your own hardware, a search engine of your choice is your friend.

Tools used

The tools I use are absolutely simple: Mostly I just used a Weller WECP-20 for soldering, regular screwdrivers and some other, older hardware to steal replacement parts from. Nothing fancy and I don’t claim to do anything special with it.

Liquid-Crystal Displays

Let’s start with what in my experience was the most common problem in consumer electronics, resulting in their failure: broken capacitors

Or at least it was a major problem for me, probably because much of my failing hardware was produced around the time of the capacitor plague.

On the right picture you can see difference between a good capacitor (left) and a bulging capacitor (right).

I used a BENQ FP91GP display for the entire time. At some point I also added a used Samsung SyncMaster P2250.

Both of them failed at some point, stopping to display anything and making high-pitched sounds instead. So of course the first thing I did was open them up. In both of them I found the exact same thing: Bulging capacitors

When you open up hardware you have to be careful about electric shocks. I never received one, but especially power supply units (PSUs), which were the broken parts inside of my displays, can store a high amount of energy for a long time after disconnecting them from power.

I didn’t want to buy new capacitors, even though they are available for low prices and you can even buy a full repair kit online for commonly failing displays. Instead I went through even older hardware in the house and scavenged still functioning capacitors. Mostly you have to look out for three things:

New capacitor has correct characteristics
New capacitor fits into space of the old one
Install the new capacitor the right way or more hardware might blow up (negative stripe on capacitor to white mark on board)

Once I found replacement capacitors I desoldered the broken ones and replaced them. Surprisingly these capacitors from TVs from the 80s still work just fine. My plan was to buy brand-new capacitors once they fail, but it never happened.

In hindsight it’s certainly curious to repair old hardware with parts from even older hardware.

Power Supply Units

The same story didn’t just happen with the PSUs inside of both of my displays, but also with the power supply unit of my computer. I fixed it in the same way and the PSU still works to this day. But since my LC Power LC6420 was never very efficient I bought a replacement anyway, the more efficient HuntKey Jumper 300G.

Unfortunately the new PSU turned out to be too loud for my taste, so I opened it up immediately and noticed that the fan was fixed to 12 V. I reconnected it to 5 V instead and tested that the PSU does not overheat with my system even at high load. I wouldn’t recommend doing this since it immediately voids the warranty and the PSU can overheat easily. Ideally just buy a modern PSU which controls its fan properly based on the internal temperature.

Computer Case

I wasn’t just annoyed with the noise of my PSU though. In the same vein I also switched all 7 case fans of my NZXT Zero from 12 V to 5 V, thus reducing the amount of noise significantly. The fix is simple and does not require any soldering. Instead I used a molex extension cable for the fans and switched its 12 V line to 5 V

The trick here is to remove the pins from the molex cable with a small screwdriver, and switch them around so that 12 V and 5 V are switched, which is quite simple. I just had to make sure that my computer stays cool enough, but with 9 fans in total running inside of it that was no problem, even in the nearly noiseless state of running at 5 V.

The CPU cooler can be controlled by fancontrol on Linux and only spins up when the load is high, otherwise the CPU can be cooled passively, just using the slight flow of air from the case fans.

Random Access Memory

Finally let’s get to an entirely different problem, for which I didn’t even suspect a hardware fault: I’ve been using Gentoo for nearly the entire time, which means I ended up compiling my own programs and libraries. Compiling a single large-scale C++ library failed again and again, at seemingly random positions. The g++ compiler always complained about internal compiler errors. At first I assumed this was a bug in the GNU Compiler Collection, so I switched to other versions, but got the same error in each of them. At some point I noticed that the memory usage increases to multiple GBs when compiling this library and that’s when it hit me: My main memory might be defective and with most programs I just don’t notice.

I booted into Memtest86+ to run the memory benchmark and, lo and behold, a single byte of my RAM was broken. These were my choices on how to fix this:

Remove the RAM module, but then I don’t get the speed benefit of dual channel and am missing out on 2 GB of RAM
Buy a replacement module, but that costs money
Tell the Linux kernel to ignore the broken part of the memory and use the rest of the module

Of course I went with choice 3. A simple Linux kernel parameter memmap=1$0x0007cec2d74 in GRUB marks 1 byte at the address 0x0007cec2d74 as reserved ($), preventing the system from using it as regular main memory. A reboot later everything worked again and I could compile the C++ library. It’s possible that further parts of this memory module will break in the future and you have to repeat this process to disable them as well, but so far this has not happened.

Working around hardware problems with software is cool. And it’s not like I notice a single byte of memory missing.

Graphics Processing Units

My graphics cards like to break. I don’t really play much except for some 2D games, so a good GPU never mattered to me. The GeForce 8500 GT started to show graphics errors after a few years, even while booting, which led me to believe that the problem was on the hardware side and not software related.

Putting the GPU into an oven and baking it for a while fixed the problem for a few days by resoldering cracked solder points, but then it returned. This is probably related to the switch to lead-free solder in 2006, which is more brittle.

I didn’t want to keep baking my GPU, so I used another GPU I had lying around, which someone else had thrown away: A Radeon HD 4350

Mouse

The Logitech MX518 did its job great for the entire time. About 1-2 years ago the left mouse button stopped working reliably though. When keeping it pressed it sometimes released on its own at unpredictable moments. That’s a huge problem while playing something like DDNet.

The first step was of course to open up the mouse. What I saw was that the plastic of the mouse buttons was worn out to the point that it could barely press down the actual button inside of the mouse. I reinforced both mouse buttons with a small piece of thin cardboard. Since then the mouse has not made any further problems and the buttons work absolutely reliably.

Keyboard

I’m using a cheap Logitech Ultra Flat-X keyboard, which doesn’t feel terribly different from a good Thinkpad keyboard. I replaced the green LEDs with blue ones, but that was just an aesthetic change and was easily done with some more soldering.

Unfortunately dust and dirt always accumulate under the keys of a keyboard. If this keeps going for long enough some keys become unusable, so I took out all the key caps every year or so to wash them and clean the space below them. Unfortunately the scissor mechanism on cheap keyboards is quite fragile, so you have to be careful not to break your keys. I don’t need a numpad, so I used the numpad keys as replacements.

After a few years of usage the key labels wear off. So instead I sanded the key labels off entirely, leaving clear key caps which look much better even after many more years of usage. Unfortunately by now the key caps have become so thin that they are starting to break apart, so it’s definitely time for a new keyboard.

Printer

This is the oldest piece of hardware in the list so far. This HP LaserJet 4L has been in continuous use for about 23 years. The only serious thing you need to worry about with it is to have fresh toner cartridges if you plan to print extensively. Luckily I got 10 full toner cartridges on eBay for 1 €, taking care of this problem.

The only part of the printer I had to replace was the pickup roller, which became unable to pull in any pages. I tried cleaning it, but in the end decided to just buy a cheap replacement part, which is easy to find in online shops, even 23 years after the introduction of the printer.

Headphones

My headphones were not really broken, but rather are a chimera. The base are cheap Superlux HD-330 headphones (29 €), a not-so-subtle clone of the Beyerdynamic DT-770 Pro (137 €).

Nevertheless they still sound fine to me, but I don’t claim to be an expert. Since I didn’t like the cable I used the old nylon cable from my previous headphones and soldered them in as replacements. A little bit of soldering gets you quite far.

Instead of using the included fake leather ear pads I bought the original ear pads from Beyerdynamic (21 €), which cost nearly as much as the rest of the headphones.

Amplifier

Finally here’s the oldest part of my setup: A Marantz PM710DC amplifier from 1980 and HECO Superior 700 speakers from around 1985. The speakers still sound just fine. The amplifier on the other hand is showing its age:

The analog potentiometers are filled with dust and you hear a crackling sound when adjusting the volume. Blowing on them with some compressed air and moving the potentiometer a few times entirely removes the crackling for me. In more serious cases potentiometers can also be cleaned with contact cleaner or simply be replaced with some soldering.

Desk Light

The IKEA Global work lamp is quite useful, as it can be attached to many surfaces with a clamp and can be moved flexibly. Unfortunately all the flexible movement kept weakening the plastic connection between lamp and metal arm. In the end the plastic simply snapped off. My plastic glue didn’t work with this kind of plastic, so I decided to work around the problem instead.

Using a few pieces of paper as a buffer the lamp can be inserted into the clamp directly. Now I have the lamp fixed directly above the desk instead of being able to move it around freely, but that’s not a big problem because the entire desk is still well-lit.

Network Cables

I used to have lots of old Ethernet cables, some working only up to 100 Mbit/s, others not working at all at times. Instead of buying many new Ethernet cables, I opted to buy a 100 meter roll of high quality CAT5e Ethernet cable, a hundred RJ45 jacks and a crimper tool. Finally I can build as many cables as I need with the exact lengths I prefer.

Laptops

Initially I had a used Thinkpad T43p from 2005. After a few years of running Gentoo on it and compiling too many packages the fan broke. Fortunately replacement fans are easy to find and opening up and repairing old Thinkpads is a blessing. Printed on the bottom of the laptop you are instructed how to open which part of the laptop. There are official guides for replacing any part of the laptop.

Since 2010 I’m running a used Thinkpad x200s. The only problem it ever had was a broken battery. But since the laptop features replaceable batteries installing a cheap replacement battery from eBay is a matter of seconds.

Home Servers

Broken Thinkpads make good servers. I had an old, very broken Thinkpad T42 lying around. The screen did not work and it could only be powered from the battery, while it didn’t run on AC power at all and couldn’t charge the battery either. My solution was to remove the screen entirely and build a small adapter with some soldering to connect AC power to the battery connection directly, leaving out the battery.

This worked as a home server for a few years, until the performance of such an old Pentium M system got too low and I switched to a totally passive Q1900-ITX system in 2014.

Conclusion

All in all I would say that the hardware held up reasonably well and when it broke it was usually possible to fix it without replacing the entire device. More repairable hardware would still be a huge bonus, especially since the industry has been moving in the opposite direction with ever smaller and more compacted devices.

Maybe someone who read this post found it interesting and will have fun fixing their own hardware (or trying to) once it starts malfunctioning.

I wrote this up on the exact hardware described in this post, while waiting for my new hardware to arrive in the next days. I hope build quality and repairability of PC hardware haven’t gone down in the last years. But only time will tell.

Discussion on Hacker News and r/Hardware.