PAX South 2016 – Day 1

The gaming club I belong to, the Juans of Gamelon, decided to make a pilgrimage to the Penny Arcade Expo. Except we already live in Seattle and attend PAX Prime every year. So instead we decided to invade the great Republic of Texas for PAX South in San Antonio.

Our flight over left at 11:00 AM PST and we eventually landed in San Antonio at 9:00 PM CT with a layover in Denver. There’s not much to say about the trip over, other than I got some quality time with Undertale and Mario & Luigi: Paper Jam.

Having attended the original two years of PAX and the last three years of PAX Prime I had certain ideas of what PAX was, of what it should be. It wasn’t really any of that, which is kind of nice. Even before getting here there was a marked difference in tone. Individual day passes were still available what must have been a couple months after going on sale; unheard of back in Seattle. (Edit: Friday and Sunday passes are still available!)

When we got to the convention center there was a long line to get in but it moved fast. Once inside and in the exhibit hall there was… space, a lot of space, separating the video games from the tabletop games. In the darken end of the hall with the glowing Twitch and Intel signs there was more missing. Sony, Microsoft, Nintendo, Sega, Square-Enix, Bethesda, EA, Ubi; the place was completely void of AAA studios and publishers with the exception of Capcom. Even the well known indie guys like SuperGiant Behemoth were nowhere to be found. Instead we saw a lot more hardware manufacturers like Intel, Zotac, DXRacer, and Kingston (parading as Hyper X) and small studios making small games. While jarring at first I don’t know that I actually dislike it. PAX Prime is a madhouse; not compared to PAX South but compared to something on the scale of E3. PAX South, on the other hand, felt like a reminder of those first couple years of PAX when it was in Bellevue’s quaint Meydenbauer Center. Do you want to talk to someone that works in the studio of the game you’re looking at? They’re right there. Want to see where the lead developer gets their inspiration from? They’re probably 10 feet away. Want to play a card game called “Poop”? By god it’s right there with no line and a friendly person eager to show you how to flush. It’s like a kid wearing their parent’s clothes and I absolutely adore it.

IMG_20160129_103221707

On the other side of the exhibit hall’s massive divide was the tabletop area, home to the Magic: The Gathering tournaments, various venders and studios, an PAX’s freeplay area where you can borrow virtually any game under the sun and make bitter enemies out of once close friends. This place was huge. Like, original Xbox huge. ‘Day-one patch’ huge. Not just huge, but full of gamers shuffling cards, rolling dice, moving figurines, cooperatively lifting objects with cranes attached to their heads… (yes, that was a thing). It was inspiring, like the shining yellow save points in Undertale. It just felt good.

* (The sound of shouty nerds fills you with determination.)

IMG_20160129_180826558
IMG_20160129_195217208

So despite all of the would-be short-comings of the event I think I actually prefer it this way. It’s like a breath of fresh air after the chaotic holiday season. I can’t really blame the bigger studios for not making an appearance, either. With most studios shipping back in November or pushing back until summer it doesn’t make a whole lot of sense to invest so much money when you’re not in a position to see a return on that investment.

It’s now 3:16 AM on what will be day 2 of PAX South in under 7 hours. Tomorrow’s missions are to get this article posted (since our free hotel wi-fi isn’t connected to the internet), win Zotac’s raffle, and play every indie game I can put my hands on.

Oh, and a side note: Zotac let me hold their AMP! EXTREME GTX 980 Ti, a 2.5-slot tall graphics card that taller, wider, and longer than any card I’ve ever seen. And it costs $650-$700. What in the…

Passive Water Cooling Memorandum

I get inspired by a lot of things. Sometimes I’m inspired by idiotic and impractical things just because I find them fascinating and I want to apply the logic in more reasonable scenarios. After watching the Linus Tech Tips series Whole Room Water Cooling Project I started thinking about impractical ways to water cool my own system while keeping it as quiet as possible. For some reason I decided on using a fish tank to act as both a reservoir and passive radiator, relying on convection and evaporation to provide a powerful one-two punch of cooling.

Instead of jumping into the deep end I started with a very simple setup, opting to cool only the CPU since I wasn’t sure exactly how effective the tank would cool the water. The loop starts in the tank, flows through a water-feature pump submerged in the tank, though a soft tube to the water block, then back into the tank. The dimensions of the tank where a huge unknown since the volume determined how long the water would take to heat up and the surface area would determine it’s cooling characteristics. After a lot of research and goofing around in Excel I came up with this chart to determine how long it would take to heat a given volume of water with various heat loads (that is, how many components I added to the cooling loop and if they’re at idle or full load).

tank size

The processor I was using at the time, an AMD Phenom II X4 965, had a maximum heat output of 140 watts. Given that, we can use a range of tank sizes to determine how long it would take to heat the water. For example, it would take 6.6 hours to raise 10 gallons of water 30 degrees Fahrenheit, or from a normal room temperature to around 100 degrees, my target maximum water temperature. Given that a 100% workload across all four CPU cores for over 6 hours is not a reasonable workload I figured this was a good size, providing cooling under a worst-case scenario.

What the chart above doesn’t take into account is cooling; it only addresses heating the water. This system has essentially three sources of cooling: radiation through the sides of the container, surface radiation, and evaporation. Based on testing I did the tank was able to radiate 48-70 watts, depending on temperature, through the glass sides when the top was sealed. According to this chart a 30 degree Fahrenheit difference between water temperature and ambient temperature produces 130 watts of cooling per square foot. However, the tank I’m using has a surface area of 1.2 square feet, producing a total heat loss of 156 watts. That puts our total at 204~226 watts of cooling meaning the water should never actually reach 30 degrees above ambient. So far everything looks good, but how does it work in practice?

prime95

To test the real cooling performance of this system I ran Prime95 for five hours then let it idle overnight (I didn’t take measurements while I was asleep so there are no data points). We can see the average difference between CPU temperature and water temperature is 9.2 degrees Celsius which has more to do with the water block’s performance than anything. More importantly, the peak CPU temperature came in at 38 degrees Celsius. To put that in perspective, on the all-in-one water cooling solution that was previously on this same CPU I was seeing temperatures in the high 50s under gaming load which is much less stressful than Prime95 and on standard air cooling you might see high 60s, low 70s, or even higher if you’re overclocking. Compared to that, 38 degrees is absolutely frigid.

case and tank

At this point we know it works both on paper and in practice, but what is it like to live with? Let’s start with the bad parts.

It isn’t silent. Seems insane for a passive system but it really isn’t that simple. It’s quiet, sure, but only sometimes. The pump is cheap and meant to be used outdoors. The pump itself vibrates, meaning I can’t use its suction cup feet to mount to to the walls of floor of the tank as pictured above because it transfers that vibration straight to the glass and into the room. My solution has been to position it so that it sort of floats in the water, not contacting anything but the water. It still makes sound but you really only hear it when the rest of the system is off. Speaking of which, there are still other system components that generate noise. The biggest offender was my dual GTX 560 Ti SLi setup which, under load, sounded like four jet engines in a screaming match. When those were idle the case fans are the source of noise. Not much, but when the rest of the system is so quiet little noises become moderate noises. Basically by removing the one fan cooling my CPU I did little if anything to quiet the rest of my system down at all.

With evaporation being responsible for half my system cooling a lot of water evaporates. It takes a long time but the occasional trips to the drug store for more distilled water are a bit inconvenient. Summer time makes the system operate at a higher overall temperature, too, which increases the evaporation that much more.

It’s not all that pretty. I had bought the blue glass rocks to spice up the tank and had some plans to make it look nicer but I just never bothered. It takes up a huge amount of space and cleaning it between refills isn’t fun.

It’s not all bad, though, the cooling performance is insane and it cost half of what a typical water cooling loop would. In all this system should cost around $130 USD which is about the price of a radiator and reservoir alone.

If your goal is silent computing this isn’t the way to go. High quality air cooling can do much better with little to no maintenance. Traditional water cooling loops can achieve similar performance with quieter results assuming you’ve invested in high quality fans. That said, if your goal is to build a cheap water cooling loop that works well then I can’t think of a better solution.

 

GTX 560 Ti SLi vs GTX 970 SC

970

The comparison no one asked for but here it is. I’ve always been fascinated by nVidia’s SLi technology, shotgunning multiple videocards to multiply graphics performance. Of course, it’s not a perfect technology. You’ll see the same or better performance investing the same money into a single better graphics card, performance doesn’t scale 1:1, and the difference between minimum and average frame rate typically grows as you add cards (citation). So why do it? For me, I was adding a second card after I had owned the first one for a while. Prices dropped and and it became reasonable to invest a few extra dollars to increase my system’s performance. Later I was simply given two identical cards that were a whole generation newer, so I had no real reason to run a single card. Now that I’ve been gifted a GTX 970 SC it’s time to compare multiple older cards to a single current one.

The system the cards are being tested in consists of a water-cooled AMD FX-8350 overclocked to 4.415 GHz, 16 GB of DDR3 at 2427.9 MHz, and an ASUS M5A99X-EVO motherboard packed inside an NXZT H440 case with stock cooling. The dual GTX 560 Ti cards are from Palit (I hadn’t heard of them either) while the GTX 970 SC is from EVGA.

First up is the most standard test, 3D Mark Firestrike. Thanks to consistent settings across all systems this is the most reliable test for comparing performance, though it’s arguably not “real-world” performance. The highest score I was able to manage utilizing the dual GTX 560 Ti configuration was 5332. With the single GTX 970 SC my score jumped 62.9% to 8685. While this is a pretty huge jump I wasn’t completely satisfied since my score was still slightly below the “Oculus Rift spec” of 9271. To combat this I moved the card up to the top PCI-e slot, figuring the lower slots might be limited to x8 speed. Moving the card up brought my Firestrike score up to 8854. This could be because of the added bandwidth afforded by the x16 slot or it could be the result of, well, any number of things. I didn’t move the card down to see if that brought the scores back down, but it was at least an interesting change.

Next was the 3D Mark Sky Diver test. I scored 16,082 with the dual 560s and 24,612 with the single 970; a gain of 53%. It makes sense that the jump here wouldn’t be as big as we saw with Firestrike since it’s using older rendering technology and throwing memory and DirectX 12 support at it doesn’t really help here.

3D Mark, 2x GTX 560 Ti vs GTX 970 SC.
3D Mark, 2x GTX 560 Ti vs GTX 970 SC.

The most impressive by far test was Tomb Raider. Using the default settings on “ultra”, a resolution of 1680×1050 @ 120 Hz and V-sync off I got a minimum frame rate of 90, maximum of 140, and an average of 116.4 frames per second. With the new setup those figures increased to 130.1, 198, and 165.7 frames per second for a gain of 44.6%, 41.4%, and 42.4% respectively. That in itself isn’t impressive but when I changed the settings to “ultimate” (identical to “ultra” but enables TressFX for rendering hair as strands rather than a single mass) the frame rates fell through the floor on the old setup. Minimum, maximum, and average frame rates were 3.6, 35, and 13. The minimum frame rate was under four frames per second and the average a meager 13. With the new system we saw a minimum of 82, maximum of 132, and an average of 107.9 frames per second. That equates to a gain of 2,177.8%, 277.1%, and 730%.

Min, max, and avg. frame rates in Tomb Raider.
Min, max, and avg. frame rates in Tomb Raider.

One benchmark came back with kind of the opposite result of Tomb Raider. Cinebench’s OpenGL test came back with 89.57 points with the 560s and 82.32 points with the 970, a gain of only 3.1%. I also ran the CPU test, just for giggles, and that resulted in a gain of 4.7%. Initially I thought that maybe the test only ran on the CPU using software but the reporting in Cinebench also tells you what GPU is being used so I’ll have to investigate.

Going back to trying to match or surpass the Oculus Rift spec score, when I ran Firestrike Extreme and Firestrike Ultra (neither of which could run on the 560s due to only having 1 GB of memory) I somehow got better results. In Extreme I beat the Oculus Rift score 5,136 to 4,926, and in Ultra I beat it 2,746 to 2,596. Other users are supposedly getting over 10,000 points in the standard Firestrike test using the same CPU and similar GPUs so I think there’s still some work to do.

Another observation, outside of performance, is noise. My 560s and my 460s before that both started to sound like aircraft taking off when they started to sweat, even after cleaning and reapplying thermal compound. I’ve supposedly thrown some pretty demanding processes at this new 970 and it never seems to break a sweat. In fact I can never tell if the fans are even spinning (which they don’t at idle). After running a benchmark the temperature drops from an already-cool low-70s to low 50s within seconds. I’m honestly baffled at how this is even possible but I’m not complaining; I’m just not sure if the 970 is even being taxed at this point.

Conclusion: Shockingly, a single modern, high-end video card is better than two old ones taped together.