Project Logs

Collapse

• A quick measured runtime test

  ServError • 10/27/2019 at 19:54 • 0 comments

  With an inexpensive power meter finally in hand, I did a quick single battery runtime test (dual battery test to follow).

  The pack uses ideally 7000mah cells. I decided to do a run down test bringing the cells down to about 3 volts under load, even though I know I have the headroom to go further. This test cheats a bit by not using the VR headset, which shaves off maybe about 4 watts of added draw from the hardware and potentially a bit from increased CPU/GPU usage.

  The pack is recharged, I use the watt meter's balancing feature to passively even out the cells a bit further, and then use my charger to see if the pack needs any topping off (not really). All cells are brought to 3.6 volts which is the max I'm planning on using for the life of these packs. I could go to 3.65 per cell but don't feel that it's worth the minimal gains.

  Lets load up No Man's Sky, a title that has taken way too much of my VR time, and set it to coast around a planet to approximate the load of normal gameplay.

  I hit exactly 5Ah at 1 hour in, as well as my 3 volt cutoff.

  Shutting the PC down, my cells returned to a 3.2 volt rest and the gas gauge believes that about a third of the pack life is still there (there isn't, but I'm not planning on trusting this gas gauge anyway).

  Putting the pack back on the charger, the pack takes about 4950mah to return to full charge, verifying my readings above.

  Next up will be the dual battery test, which with the distributed load should hopefully net me a nice little bonus rather than just doubling my runtime. Stay tuned.

• This isn't even my final form...or is it?

  ServError • 10/11/2019 at 21:20 • 0 comments

  As time went on and I used my bulky backpack, I kept in touch with a company I've mentioned once before in this build log; HDPlex. They were working on releasing a new "direct plug" power supply and after many delays, they finally started taking orders. Direct plug supplies are ones that sit directly on top of the 24 pin ATX power connector, meaning that this thing is insanely tiny for a unit that includes an 8 pin GPU connector. The downside though is that it's only rated for 200 watts continuously, and 220 watts peak. This means that if my custom CPU/GPU frequency voltage curves somehow get wiped out, I start coming very close to the limit. I trust their power ratings though based on previous user feedback, and decide that I'll add a power meter at a later stage to safeguard against any unusual scenarios.

  I quickly made a parallel adapter for my batteries, as its been on the back burner for too long.

  And now for the interesting bit... Around this time IMAX decided to shut down their VR business and clear out their VR backpacks. This gave me an opportunity to pick up a zotac VR Go for pretty cheap, though it lacked batteries and a charger. I played with the backpack for a bit, mocked its inferior specs (laptop 1070 and i7 6700T), and parted it out, keeping only one part.

  The backpack frame.

  So the time came to start over on the plywood base and make a setup compatible with the zotac's mounting system. I made two versions, one that allowed the mounting studs to slip directly into the plywood, and one that used some 3D printed mounts I mocked up that would screw into the plywood. Because of the hardware that would be attached through the plywood, the 3D printed route made more sense.

  I laid out the components on this new base for a test fit. The idea in this build was to put the GPU vertically to shave off a huge amount of the original width. One PCIe extension later and...

  It's tight, but it works. I'm using a GPU wall mount adapter from Thingiverse (link pending), but needed to make a brace for the rear of the card to prevent movement. I used the metal backplate as a wedge for the brace I eventually modeled.

  Time for a quick power on test with the new power supply.

  More Success! Time to take it all apart and start staining. The extra piece off to the left is a small riser for the DC barrel jack.

  I took the existing aluminum frame and reduced its width to accommodate the new form factor. Sadly the project was 5ish months old at this point, and I wasn't being as careful with my cuts as I should have been. The acrylic was damaged in a couple of spots, and the bracket holes needed to be redone as things weren't aligning as they should have. I also added a "floor" to the frame for the batteries to sit on.

  And a battery test fit.

  It's tight, but it makes it. Time to secure the "floor".

  And the final assembly with a new mount hole for the power switch.

  The Zotac backpack frame snaps in and then locks. It feels very secure and has shoulder straps, a chest strap, and waist strap. Big upgrade.

  With headset.

  There's plenty of clearance from the person's head, and comfort is greatly improved from the original backpack.

  Time for a real test. I clear the deck, set up a new playspace in the headset, and set up a camera to take some action shots.

  Now the only thing left to do was to box it up and find some willing test subjects.

  Time to make good use of office space. Coworkers make great testers.

  All in all a great success.  Thanks for joining me on this build and tune in for future projects.

• Safety and upgrades, upgrades, upgrades

  ServError • 10/11/2019 at 20:27 • 0 comments

  Now came the point where I needed to work out how to secure the backpack from outside elements, while also adding some convenience for the user. A trip to the hardware store started as a plan to buy two long aluminum bars to bend to the shape I wanted, but trouble finding the type of aluminum I wanted and concerns with the bending process led to me buying 90 degree bent aluminum bar and some straight aluminum bar, along with all the nuts, bolts, and washers I would need. With this approach, I could cut the bent aluminum bar to the width of the straight bar and screw everything together.

  This was also about the time where I started to feel like I wasn't pushing the limits of my build enough. The i5-6600T processor was limited to a fairly low total power, and the new generation of Nvidia cards included some offerings that used a similar amount of power as the GTX 1060 while offering some new VR tech that Oculus had leveraged specifically for the Rift S. Sounds good to me!

  A new GTX 1660Ti that was very briefly on sale for $230 caught my eye and seemed like the perfect candidate. It was rated to use 10 watts more than the 1060, but I was planning on downtuning it anyway. Along with this I got a decent deal on an i7 7700. The non T variant meant it used a hell of a lot more power. Using overclocking utilities meant I could limit the frequencies and voltages of both to a territory I was comfortable with. Keep in mind, you can shave a lot of power consumption off of computer components and lose only minimal performance. The amount of power expended to obtain that last 5-10% of performance is not worth it in lower power setups.
  

  I picked out suitable bolt hold locations for the plywood and cut and drilled my 90 degree aluminum bar and flat bar to make the main frame rails.

  The flex in the center was significant and called for the addition of a cross member.

  With the frame, the unit is finally able to sit upright.

  I picked up some acrylic to close off the open sections. Bending it using the frame itself proved doable, if a bit sketchy at some parts.

  Success? Survey says....

  Yes! I didn't specifically intend to buy a GPU with LED lighting, but here it really shines (sorry again). 

  Not shown: Adding furniture feet to the bottom of the frame to avoid scraping the floor, and countless other bits.

  Time to try it on.

  It works, but it's honestly pretty wide. A few months pass and I use the backpack in this format, but eventually the release of some new hardware and my acquisition of something special encourages me to pursue the backpacks next form.

• I'm an engineer, not a woodworker

  ServError • 10/11/2019 at 19:54 • 0 comments

  With a power supply for my backpack picked out, it was time to make it mount to the backpack. Horizontal space was limited after laying out the main components, so I cut out and glued in a riser that the PSU could screw in to.

  Out of impatience, I started testing the polystain I had picked out. As it turns out, I should've mixed it before doing this for significantly better results. You can also see the ITX hole pattern drilled out from a previous test mounting.

  Next it seemed reasonable to modify my design and scrap the piece of wood which sat at the bottom section of the backpack. It would need some unusual mounting to the main piece to work, and increased the thickness of the pack quite a bit. Instead I made another "riser" which the batteries could sit on.

  There are two slits above and below the riser for tie down straps.

  As I was going riser happy, this seemed like a great time to add yet another one for the power button.

  Sneak preview mounting of the board and GPU.

  Thinking about how my fabric backpack and straps would mount to the plywood, I opted to add adhesive-backed velcro in the areas of the backpack that could be easily reinforced with nylon. Not shown: painstakingly threading through adhesive-backed velcro around the perimeter of these strips.

  In the same vein, I didn't want to trust the velcro adhesive on the plywood veneer either, and so reinforced it with staples.

  In this front view you can see the linseed oil I tested on the front. I wasn't thrilled with the result and opted to try that polystain again (mixing it properly this time).

  First coat of stain

  Second coat of stain

  Much better! Today was also the day I picked up the Rift S (about a week after release).

  I needed to do some haphazard last-minute reinforcement of the shoulder straps on the backpack to ensure they wouldn't separate from the back and make a very expensive mess. I also sewed the top black strap into a loop for future utility.

  Time to put everything together and put it on!

  I loaded up some Dead and Buried and kicked the butts of some folks who didn't seem quite sure how to play (sorry). Performance was nice and solid, and the new guardian system setup is very convenient. I walked into my kitchen while still wearing the headset, set up a new playspace, and messed around some more before calling it a night. As a side note, this headset is way more convenient for long haired folks than the original Rift. The Rift often forced my hair onto my face as I put it on, whereas this one just clamps to my face without any sliding motion.

  As I went to put the backpack down, my decision to skip the bottom plywood in favor of a riser proved problematic. The backpack could not sit upright, and instead needed careful handling to set it down flat.

  In the next segment, I'll start adding a metal frame and going a bit upgrade crazy.

• Off the shelf parts save your sanity

  ServError • 10/11/2019 at 18:52 • 0 comments

  While considering the obstacles I encountered on my first build, I revisited my design goals to see if anything needed changing.

  I was looking for:

  • A motherboard that could handle a wide input voltage (preferably 18-22V)
  • Only needing one extra regulator for the GPU
  • A light total weight
  • And obviously a backpack form factor

  Revisiting my GPU implementation with the micro PC, I realized that the total cost of the micro PC and PCIe adapter were not actually better than a used or refurbished ITX motherboard that's a couple of generations old. An off the shelf motherboard wouldn't give me trouble through undocumented BIOS security features, and ideally might also be able to supply 75 watts to a full sized PCIe port while using on board regulators (though my mind was pretty firmly set on 120 watt GPUs anyhow). Also, the use of an M.2 slot and adapter relegated me to 4 lanes of PCIe 3.0, which while nothing to sneeze at, was not as good as a native 16 lane connection.

  Doing some research, I found a few companies that made hard-to-find motherboards that fit the bill (at least more or less). Some had DC barrel jacks but only x4 PCIe ports. Some had a 4 pin molex and offered a bit more power. In every instance, the boards were either incredibly hard to find, overpriced, or had shortcomings that didn't really sit well with me.

  The time came to reconsider regular motherboards with 24 pin PSU connectors. I picked out a used MSI B250i ITX board and populated with the micro PC's CPU, GPU, and SSD, then grabbed some ADATA ram that was rated for speeds well above what I needed (AKA undervolting potential). The motherboard also has an M.2 slot on the underside which gives me the option to lose the 2.5 SSD and any mounting it would need. Everything worked easily, unsurprisingly. 

  Initially I expected that powering this setup was going to be a hassle. Checking out the current DC-DC power supply market, I found a variety of direct plug or relatively small units, mainly from companies like HDPlex and PicoPSU (and a few others). I knew my total power draw was not especially high (sub 200 watts easily at this point), but I decided that my first foray into this space would be with a ~500 watt unit, just as a safety margin. I ordered a PicoPSU model (a couple actually) and added a large jumper to bridge the two regulator circuits for use with a single power source. The nice 12 volt rails on these PSUs means I can scrap the separate buck regulator idea and run the entire system through these instead.

  Around this time the batteries I was waiting on finally arrived. I made up two packs of 6 cell, 7000mah LiFePo4 batteries. Why LiFePo4? At the pack size I was working with, I knew that my load was well within the tolerances to keep the pack healthy. The number of charge cycles offered by this chemistry also puts others to shame. Lastly, the cells tend to be very stable, and convincing someone to strap batteries to themselves is a little easier when you can confidently say "these ones don't explode or ignite!".
  

  I assembled the packs and got them welded up, then attached balance leads and an XT60 cable for use with my RC charger. Finally it was time for the first battery operated test!

  Benchmarks abound. Time to run an actual VR title (though in 2D mode) and make sure my runtime is reasonable.

  45 minutes later, and the pack still wasn't discharged. I was happy to declare the pack a success. Now with my hardware validated, I could return to the backpack aspect of the build.

• Early Prototyping

  ServError • 10/11/2019 at 17:17 • 0 comments

  Step back in time to April 2019. Oculus had revealed two new headsets soon to reach consumers; The PC powered Rift S and the standalone Quest.

  Unlike previous products, both flaunted the ability to track reliably without the use of external markers or sensors. The Rift S, while a nice upgrade from the original, did not garner the attention of the Quest with its promise of a healthy game library without a tether or computer.

  I couldn't bring myself to abandon my game library and simulators though, and decided that the only option was clearly to build my own untethered experience around the yet-to-be-released Rift S.

  Having seen some early attempts at consumer VR backpacks from the likes of MSI, Zotac, and HP, I set out to build my own upgradeable backpack computer that left me in control of components, batteries, etc.

  First came the actual backpack straps and padding. I grabbed a well padded but damaged backpack and got to work cutting most of it away from the only portions I cared for.

  Next, the internal padding of the back and bottom section shown served as a good template for some plywood cuts which would serve as my frame.

  
  Not the prettiest, but definitely functional for my needs. I test fit the plywood into the backpack and found it snug enough to stay together on its own. My quality inspector also came by to verify my work so far.

  Now it was time to pick out some parts. I originally thought to use a 12v buck regulator which would feed off the battery pack and into the GPU and possibly the main system. I ordered a 240w unit and set it aside.

  Looking at what I already had at my disposal, I pulled an HP micro computer with a 35 watt i5-6600t processor.  After adding some more ram, an SSD, and a WLAN card with matching antennas, it looked like I had a pretty good candidate.

  Four mounting holes on the bottom of this unit were going to make this a very easy build....

  Now for the graphics card. A used GTX 1060 with a 120 watt power draw was an easy pick considering their low price (~$100 used). How do you get a full sized GPU into a micro PC you ask? Expensive and overspecialized adapters of course!

  ADT-Link makes various nice products, one of which convert from internal M.2 pcie ports to full size 16x ports and is especially suited for graphics cards. Removing an extra, optional serial port on the micro PC (a serial port, really HP?) left me with just enough room to route the cabling to the adapter board.
  

  With the major components all together, how about a quick power on test? I grab a test display, spare power supply, and various cables to hook everything up.

  And......total failure. Playing with the various jumpers and a few learned tricks about using external GPUs finally gets me some signs of life.

  However....

  The graphics card isn't initializing properly. It looks like the boot process on this HP has some safeguards that really hate fun. I check for any possible workarounds but sadly don't see any.

  With my first build an outright failure, I decide to cheer myself up by making a quick "stupid" build. One Odroid H2 later and...

  Success! Well, aside from the fact that the Odroid has nowhere near the power necessary for this project. Just for giggles, lets see what the compatibility tool has to say.

  We traded one checkmark for another. Only thing left to do with this ridiculous build is to benchmark it and move on.

  Look dad! I'm almost a gaming computer! Honestly it's not too far off from mininum spec. Maybe I could've actually played some very simple demos, but the point is to beat out what the Quest would've offered, not match it.

  In the next log, we move on to more "standard" (and sane) components and try again.

View all 6 project logs