A while back I picked up a Dell Optiplex 755 from RE-PC, a local computer recycler, for $25 with the intention of making it a dedicated retro PC gaming rig that could run DOS games without the need for DOS Box. Games like Rainbow Six, Quake III Arena, and Nightmare Creatures run beautifully with all the visual settings maxed out at 1080p and the system seems to be damn near asleep while doing it.
Fast forward to November 25th when OzTalksHW uploaded this video documenting his build of the “OzBox”, a $160-ish gaming build.
“…if you’re thinking about building your own OzBox then definitely tweet at me and use the hashtag #OzBox because I really want to see you guys’ creations.” Challenge accepted.
Our starting points are pretty similar. They are small form factor computers, originally designed for business tasks and general home computing with very little in terms of upgradeability. The key is that they both have a single unused PCI-Express x16 slot.
|Base System||Dell Optiplex 755||HP Compaq 6000 Pro|
|CPU||Core 2 Duo E6550 @ 2.33 GHz||Core 2 Duo E6300 @ 2.8 GHz|
|RAM||2 GB DDR2 800 MHz||4 GB DDR3 1333 MHz|
|Storage||80 GB SATA||250 GB SATA|
The OzBox comes with a newer, faster CPU, double the memory, and three times the storage for an extra $30 (or 2.5x the price, if you want to make it sound more sensational). To be fair, that $30 different is entirely the shipping cost from the eBay auction, so if you could find a similar deal locally that would be the way to go.
For $20 you can buy a Core 2 Quad E6700 at 2.66 GHz. For another $10 you can pick-up a 2x2GB kit of RAM to replace the 2x1GB sticks it comes with, or even suppliment the RAM your system comes with. I happened to have a 2x2GB kit laying around, so I replaced the old memory with the new stuff.
Our video card of “choice” (that is, out of the very limited selection of low-profile graphics cards available, this is the one I felt like spending my money on) is a nVidia GT 730 by PNY. I picked this 2GB model up for $54, so depending on what upgrades you need we’re looking at $79-109 before tax.
Falling at the Starting Line
My original goal was to keep Windows XP on this machine, and that’s how I started the testings. Unfortunately, running XP meant I’m limited to 4 gigs of RAM (the board supports 8), spotty driver support, questionable-at-best security, and, most importantly, most modern games and benchmarking utilities simply won’t run. This limited me to testing older titles which didn’t reflect what a “gaming PC” should be able to play, so after countless software crashes and failed benchmarking attempts I eventually caved and installed Windows 7. Depending on wether or not your computer came with Windows 7 installed, or it came with the OEM license stuck to the top or back, of if you need to purchase a whole new copy, this could add some amount of cost to the build.
I couldn’t decide if I wanted to test across an array of resolutions or target a common low resolution like 1280×720. Eventually, I decided on 1280×960 for a couple of reasons. First, it seems to be the resolution of choice for pro Counter-Strike players, so that’s the resolution I wanted to test there. I also thought it would be ideal to use the same resolution across all the games to get more comparable results. I also did all my benchmarking with four gigs of memory installed. Originally I was going to test with two gigs installed, then four, but it seemed like a huge hassle when the cost to upgrade (assuming you don’t have some laying around) is so little.
For my first round of testing, I went with the PC game I play the most often: Counter-Strike. Unsurprisingly, Global Offense had the lowest average frame rate at 62.4 FPS. 90th percentile frame times were 20.8 ms (48 FPS).
Going back to CS 1.6, I saw an expected boost in performance, but also an unexpected boost in erratic frame times. Average frame rate was 168.9 FPS, 90th percentile frame times were 7.2 ms (139 FPS), and a frame time deviation of 39%. Ideally, we would see the individual frames bunched as close as possible to the average (which we see with Source and Global Offensive), rather than scattered across the chart.
I was most impressed with CS: Source. Here I saw the highest average FPS between the three games, 212.3, the lowest 90th percentile frame time with 5.9 ms (169.5 FPS), and the lowest frame time deviation at 22%. This gave the best sense of responsiveness and fluid gameplay out of the three.
Unreal Tournament has always been a game that combined incredible graphics with fast gameplay and blazing frame rates. I remember being absolutely floored by Unreal Tournament 2003 on my AMD AthlonXP 1000+ and nVidia MX440, then again by Unreal Tournament 3 on the high-end machines at work back in 2007. So how do these titles hold up on our budget hardware?
UT 2004 had an average frame rate of 146.2 FPS and a 90th percentile frame time of 8.6 ms (116.3 FPS). Despite the modest frame rate and low video settings the game still looked great and was an absolute blast to play again.
I had my doubts about how well Unreal Tournament 3 would run, but those were soon laid to rest. With an average frame rate of 74 FPS, I was concerned it would dip below 60 FPS, but the 90th percentile frame time was 8.6 ms (64.9 FPS) it managed to stay consistently playable. Most surprising was the frame time deviation which was only 14%, meaning the frame times were very consistent.
It shouldn’t have surprised me as much as it did, but Left 4 Dead runs amazingly on this setup. Since there are tight corridors and large outdoor areas flooded with zombies I figured I should measure the performance of both scenarios, which were really quite similar. Indoors saw an average frame rate of 101.3 FPS with a 90th percentile at 12.9 ms (77.5 FPS). Outdoors, while being swarmed by zombies, the frame rate averaged 100.1 FPS with the 90th percentile at 14.3 ms (69.9 FPS). My playthrough of the first mission was very smooth with no noticeable frame drops, stutters, or other performance issues. Then again, it’s an 8-year-old title at this point, so the impressive performance does make sense.
I replaced the dual core E6550 with a Q6700 quad core processor (a surprisingly simple task in this machine), which is the best CPU this motherboard supports. In addition to the extra cores we also get a 333 MHz clock speed increase, so even single-threaded games should see a performance boost.
CS 1.6 saw a huge boost to average FPS, jumping from 168.9 to 211.0, and more importantly, the average frame time deviation (how far away each frame was compared to the average) dropped from 39% to 12%. That means that, rather than having a frame rate that consistently jumps up and down, it stays stable throughout gameplay.
CS: Source had its average frame rate drop from 212.3 to 181.5 while its frame time deviation swelled from 22% to 35%. Not sure why this happened, but it was consistently happening.
CS: GO got a modest increase from 62.4 FPS to 75.3 FPS with minor reduction in frame time deviation.
Both Unreal Tournament games saw virtually no change whatsoever, which leaves me to believe that the game is being bottlenecked by the video card.
L4D got a substantial 13.2 FPS gain while dropping it’s frame time deviation from 27% down to 12%.
This was originally going to be the end of my benchmarking since the selection of modern games that could still run on Windows XP was limited, but I decided to install Windows 7 and see what this hardware was really capable of.
Quad Core on Windows 7
Now that we can install pretty much whatever we want (that will fit on the measly 80 GB drive), it’s time to really put the hardware through its paces.
When I loaded 3DMark the recommended benchmark was Firestrike Extreme, which made me audibly laugh. After selecting the standard Firestrike test, which I already knew would be too much for the system to handle, it came back with a score of 639.
A more fair test would be SkyDiver which came back with a score of 2,639, exactly 2,000 points higher. Still not great, but at least it’s a real score this time.
CS 1.6 saw another boost to average FPS, hitting 232.9 on average. CS: Source continued to drop, this time hitting 153.0, down from 212.3 with a slower dual core CPU on Windows XP. I just can’t wrap my mind around this. Maybe it’s servers, maybe it’s something hardware or operating system related, I have no idea. CS: GO managed nearly identical results with an 115.4 average FPS and nearly identical 90th percentile frame time.
UT 2004 also saw a drop in average frame rate, down from 142.6 to 109.2 with the 64-bit patch. Without the patch, the average frame rate was 100.1. This might be something OS-related, but considering how rarely I play UT 2004 and how little that extra 40 FPS actually matters, I’m just going to leave it alone. UT3 saw no notable change.
L4D, unsurprisingly, was nearly identical to the previous results under Windows XP. Average frame rate grew from 113.3 to 115.4 which is well within the margin of error.
Finally, we get to look at some results from new games that didn’t run on XP.
I wasn’t expecting much out of DiRT 3. It’s a great-looking modern racing title heavy on physics. I lowered the resolution to 1280×720 and ran three benchmarks. The first was with all visual options turned to their lowest settings or completely disabled. The second was the “medium” preset, and the last was with the “high” preset.
This result floored me. I didn’t know the puny hardware inside this little case was capable of playing modern titles like this. Granted, it is at a low resolution and moderate graphics settings, but for $100, that’s not too shabby.
Rocket League is another game that shocked me with how well it performed. With all the visual options low or disabled I saw a respectable 58.6 FPS average with 90% of the frame times being at or above 20.2 (49.5 FPS). Leaving the Render Detail on “High Performance” while turning the Render Quality to “High Quality” resulted in a pleasing image that ran at 43.5 frames per second. While mid-to-low 40s might not usually be an idea frame rate, I found that, with Rocket League, it was plenty for knocking the ball around in the standard 3v3 game type and I didn’t feel like I was limited by the computer’s performance at all.
The last game I tested was 2013’s Tomb Raider, which… It ran, and seems playable, but only with the “Low” graphics preset at 720p and with motion blur and screen effects disabled. It’s possible to play the game at the “Normal” preset, but with frame rates down into the 20s it makes for an unpleasant experience.
Did we accomplish our goal of spending about $100 to play PC games? Yes, absolutely. Is it a good experience? No, not really. You’re better off buying an Xbox 360 or PS3 than trying to built an ultra-budget gaming PC, but if money is really tight and you just need to play those PC-only titles like Counter-Strike: Global Offensive, League of Legends, or DOTA2 this is certainly a possible solution.
Compared to the OzBox this built is based off of, how did we do? Well, it’s hard to compare directly. We have different games, so our benchmarks are going to be different. The OxBox hardware is better, there’s no doubt about that; the GTX 750Ti he picked out for his build costs as much as our whole system did, if not more. Comparing a roughly $100 PC to a $180 PC doesn’t exactly make sense, so I would say this version of the build is for people who want to play older titles or some newer titles if budgets are limited. Ozi’s original version would aim more toward the casual gamer who wants the option to play modern titles either at a low resolution with pretty visuals turned up or a high resolution with lower graphics settings, but still maintaining a 60+ FPS target.
I’ve ordered some parts for a follow-up article, seeing just how far we can push the limits of this compact gaming rig (possibly making it not so compact), so check back for updates.