The miniaturization of PCs is an interesting trend. Of late, PC makers have been able to pack big computing power in boxes smaller than some mid-range graphics cards. We’ve looked at some of those systems, like a pair of Intel’s Next Unit of Computing boxes. The NUCs are missing something, though, and that’s graphics horsepower.
Zotac thinks it has the cure for what ails us in the Zbox Magnus EN970. The Magnus packs a Broadwell CPU, an Nvidia Maxwell-based graphics card, and lots of space for storage into a box not much bigger than a Mac mini.
The GeForce GTX 960 inside the Zbox isn’t based on the standard, fully-enabled GM206 chip of its desktop namesake. Rather, the GTX 960 here is more like the GTX 970M found in gaming-oriented notebooks. This is a cut-down GM204 chip with 1,280 stream processors and a 192-bit bus to 3GB of GDDR5 memory. The chip has a 924MHz base clock and 1038MHz boost speed, and the memory runs at 5000 MT/s—speeds that are eclipsed by the average “real” GTX 960. This chip’s wider engine could make up for some of the clock speed deficit, though.
The mobile Core i5-5200U found in the Zbox Magnus EN970 is built around a pair of Broadwell cores running at 2.2GHz base and 2.7GHz Turbo clocks. This chip has Hyper-Threading enabled, and it’s fed by 3MB of L3 cache. The CPU includes Intel’s HD Graphics 5500 IGP, but the onboard graphics processor isn’t used at all in the EN970.
The Zbox is clad in a mixture of textured and glossy black plastic. The textured parts should deter fingerprints, but the front panel is a smooth, reflective surface that will smudge quickly. There’s a bit of flex in the top and bottom panels, so I wouldn’t want to stack anything on top of the Zbox. The large power button commands attention with an orange LED ring when the system is on. An SD card reader, microphone and headphone jacks, and a pair of USB 3.0 ports round out the front panel.
The power connector is around back. Next to that, the Magnus sports another couple of USB 2.0 ports alongside a pair of USB 3.0 connectors. There’s also a barbershop quartet of HDMI 2.0 ports that can transfer 4K video at 60Hz. These ports can transfer audio, as well. Next up is a pair of Realtek-powered Gigabit Ethernet jacks. The lone antenna header is used by the Realtek 802.11ac Wi-Fi controller. Another antenna is hidden inside the housing for Bluetooth 4.0LE connectivity.
When we had Zotac’s Jacky Huang on a recent episode of The TR Podcast, he told us that using four identical display outputs on the EN970 would allow users to connect four identical displays for consistency in multi-monitor setups. Since Nvidia’s G-Sync variable-refresh-rate technology requires a DisplayPort connection, though, the Zbox can’t support G-Sync operation.
At 8.3” wide by 8” deep and 2.1” tall, the Magnus is a fair bit bigger than either of the NUCs that have passed through our labs earlier this year. The Zbox’s external power supply is nearly big enough to house a NUC on its own, too, at 6″ x 3.3″ x 1.3″.
The EN970 comes in two configurations. The barebones model is the one we’re testing today, and it’s available from Newegg for $799. The EN970 Plus comes with an 8GB SO-DIMM and a 120GB AHCI SSD in the M.2 slot. That configuration goes for $899. Zotac doesn’t include a Windows license with either model, so builders will need to factor that into their budgets, as well.
Now that we’ve toured the Zbox’s exterior, let’s take a look at what Zotac has stuffed inside.
Getting into the Zbox to install memory and storage is an almost tool-free process. After removing a pair of thumb screws and sliding the base plate toward the front of the Zbox, we have full access to the memory slots and storage ports.
The EN970 has a pair of SATA 6Gbps ports and an M.2 slot for storage. The 2.5″ drives get dedicated trays that are held in place with a single thumb screw each. The trays have tabs that line up with the standard screw hole placements, and storage devices just pop right into them. The M.2 slot requires a small, not-finger-friendly screw to hold down the drive, though. A variety of M.2 2280 SSDs are supported by this slot, and AHCI or NVMe drives get two lanes of PCIe 3.0 connectivity.
The stacked SO-DIMM slots support low-voltage DDR3L at speeds of up to 1600 MT/s. This low-power RAM runs at only 1.35 volts, so those buying the barebones Zbox need to be sure to pick up the right memory. To outfit our Zbox, Kingston supplied a pair of 8GB DDR3L-1600 SO-DIMMs and an SSDNow 240GB M.2 drive. Our thanks to Kingston for providing us with this hardware.
All that separates us from the Zbox’s naughty bits are some screws and a warranty-controlling seal. Five screws hold the motherboard in place, and one of them is under that seal. I also had to remove the six smaller screws that mount the backplate to the enclosure to break the motherboard free from its bonds. After that, the whole system slips right out.
The Zbox stays cool with a pair of blowers mounted to a copper-and-aluminum cooler. One blows through the rear of the system, while the other breathes through the left side. We’ll examine the Zbox’s thermal performance and noise levels in more detail in a bit.
Once the cooler is out of the way, we can see something interesting: the GeForce GTX 960 in the EN970 resides on an MXM card. Huang tells us that using MXM graphics cards is more for Zotac’s benefit than the end-user’s, though. Swapping out the MXM module would allow Zotac to sell a Radeon-based version of the Magnus, and the expansion card also makes its easy for Zotac to build a Zbox using whatever graphics chips Nvidia and AMD have in the pipeline.
Now that we’ve taken the Zbox EN970 all the way apart, let’s put ‘er back together and see how this system performs.
Our testing methods
We’ll be presenting the results we gathered with the Zbox Magnus EN970 alongside those we collected for the mini-PCs and AMD APUs in our NUC5i7RYH review. We’ll also compare the Zbox’s performance to several CPUs from our Core i7-6700K review. We used the same applications and testing methods as in the Broadwell-based NUC review. The Core i3-4360, Pentium G3258, and AMD APUs were tested with their integrated graphics solutions. The other systems were tested with a GeForce GTX 980 graphics card. Thanks to Zotac for supplying the Zbox Magnus EN970 for review.
Staying with the same applications and conditions allows us to put the EN970’s performance in a broader context. The EN970’s unusual pairing of a mobile CPU and discrete graphics should make for interesting comparisons to a wide range of systems. In particular, the Core i5-5250U found in the NUC5i5RYK has the same 2.7GHz Turbo frequency as the Zbox’s Core i5-5200U, but a much lower base frequency of 1.6GHz. For the most part, we expect the two CPUs to perform similarly, assuming there’s plenty of thermal headroom.
To make the graphs easier to read, we’ve highlighted the mobile-Broadwell-based systems in orange.
Memory subsystem performance
There aren’t any surprises here. Since they’re both equipped with DDR3-1600 memory, the Zbox and the Core i5 NUC are neck-and-neck.
The productivity benchmarks are broken up into three cateogories: web, compression, and video encoding.
While the Zbox came in dead last in SunSpider, the bottom end of the graph is packed closely together. Zotac’s mighty mite makes something of a comeback in Kraken, though. Despite the Zbox’s relatively poor showing here, I found browsing the web using Chrome to be plenty fast and responsive.
In our 7-Zip and video encoding tests, the Zbox Magnus EN970 runs neck-and-neck with the NUC5i5RYK. Since the Zbox’s Core i5-5200U (along with the mobile CPUs found in the NUC5i5RYK and NUC5i7RYH) only has a pair of cores with Hyper-Threading, it can fall behind in tasks that can use lots of threads. 7-Zip especially seems to use all the resources it can get, and the Zbox stumbles there.
Obviously, not all Broadwells are created equal. In Handbrake, the desktop-Broadwell Core i7-5775C system completes the job in roughly a third of the time the Zbox does. The Core i7-5775C has a 37% higher maximum Turbo clock speed and a much more generous 65W thermal envelope compared to our 15W, dual-core mobile Broadwell variant, but I think the fact it has four cores explains most of the gap. x264 encoding tells a similar tale, and the gap between the two Broadwell-based systems is even bigger there.
3D rendering and OpenCL
The Cinebench rendering test uses the CPU power of a system to draw a complex scene as fast as possible, while LuxMark uses OpenCL to put both the CPU and the graphics card to work. The LuxMark “vendor ICD” results use only the CPU cores via OpenCL with support for SIMD instruction set extensions like SSE and AVX. The “graphics only” results use OpenCL to accomplish the same work on the GPU or IGP.
Our full-fat desktop systems were tested with a GeForce GTX 980 graphics card, so it’ll be interesting to see just how close the GeForce GTX 960 in the Zbox Magnus EN970 can get to the full-power version of the GM204.
The CPU-bound Cinebench and Luxmark’s Vendor ICD test show the same story as before. Anything that can use a lot of threads will suffer on a dual-core system like the Zbox. The Core i5-5200U generally matches the Core i5-5250U, but trails the desktop CPUs by a wide margin.
The Magnus righteously trounces the mini-PC competition and the lower-end desktop systems that rely on integrated graphics, though. The cut-down GM204 obviously can’t keep up with the fully-enabled GeForce GTX 980, but it still made a strong showing in the graphics-only portion of the test.
Our graphics test rigs are outfitted with Haswell-E processors at the very highest end of Intel’s lineup. Testing against even the results from the GeForce GTX 950 didn’t quite seem fair—the Zbox’s fixed graphics configuration means that there’s no easy way to tell if a game is being held back by the CPU.
On the other hand, all the mini-PCs we’ve reviewed in recent months use integrated graphics. They won’t pose a threat when it comes to gaming performance. Have a look at the Thief benchmark from the NUC5i7RYH review with the Zbox’s numbers added in. All of the systems were tested at 720p with the Normal graphics preset. The sole 1080p test below, made especially for the Zbox, was run with the Very High preset.
Every gaming test would probably look something like this, because integrated graphics just can’t keep up with the EN970’s discrete chip. Instead of going further with the benchmarks, we’ll subjectively examine how the system handles some triple-A blockbuster titles.
The Zbox’s unusual all-HDMI output configuration didn’t play well with my main monitor, though. My display is limited to 1080p over HDMI, so testing at higher resolutions seemed impossible at first. Thankfully, Nvidia’s Dynamic Super Resolution tech saved my bacon. All of the quoted frame rates below are from Fraps’ in-game overlay.
1080p is, coincidentally, the same native resolution found in most TVs. That target resolution coupled with the Zbox’s modest footprint makes Zotac’s little PC a prime candidate for an HTPC or maybe even a roll-your-own Steam machine. Zotac has such a machine, the similar-looking SN970, and the two mini PCs seem to have a lot in common.
Toss in a wireless Xbox 360 controller and some sort of small Bluetooth keyboard-and-mouse combination like the Hausbell Mini H7 from our peripherals guide, and we should have the makings of an excellent living room computer. 4K is here for some folks, too, and the Zbox should be ready thanks to its HDMI 2.0 ports.
The first stop for my gaming tests was DiRT Showdown. I was able to crank up this game’s resolution to 2560×1440 with DSR, and I could do so using 4x multi-sampled antialiasing and the Ultra graphics preset. My racing experience was perfectly smooth, and Fraps reported well over 60 FPS with vsync disabled. I also got a steady 60 FPS with vsync on.
Blizzard’s games tend to run well even on lower-end hardware, but they can look plenty nice with all the details cranked on high-end systems. StarCraft II and Diablo III both ran admirably with antialiasing and detail options cranked to their maximums. Again, the graphics horsepower of the EN970 and DSR let me turn the resolution up to 1440p.
After that it was time for Tomb Raider, which played flawlessly with all of the settings maxed out. This time, I was limited to 1920×1080—turning up DSR didn’t make for playable framerates. With the Ultimate preset at 1080p, I saw anywhere from 45 FPS to well over 60 FPS. Turning off TressFX bumped up the frame rate further, but it wasn’t necessary for a playable experience. The game just looks better with it turned on, too.
My final stop was Grand Theft Auto V. I started with the same settings we used in the GeForce GTX 950 review. Fraps showed the frame rate to be in the mid-50s most of the time. The first time I ran the game on the EN970, GTA V notified me that the CPU did not meet the minimum requirements, but that didn’t stop the game from running smoothly throughout. Still, that warning makes me wonder whether GTA V’s performance is CPU-limited on the Zbox.
As a test, I enabled Nvidia’s fancy PCSS shadows and set the texture detail to Very High, and the framerate didn’t budge, which seemed to confirm my concerns. Changing other graphics options started to send frame rates into the ditch, so this was the first game I tried that couldn’t be completely maxed out at 1920×1080. Still, GTA V looked great and played very well.
The fact that every game I played—even something as intensive as GTA V—ran well on the Zbox at 1080p means that the Zbox does indeed make for an easy drop-in HTPC. Running older titles with DSR gives them an extra sheen that only super-sampled images can provide, and newer titles don’t pose much of a threat to its performance.
To figure out how thirsty the Zbox is for electrons, I measured power consumption at the wall socket for the system only—the monitor and speakers were left out of the results. For context, we’ll take a look at a sampling of the systems we’ve already seen, along with a new entrant: Microsoft’s Xbox One. The Zbox Magnus EN970’s performance has proven Zotac’s box to be a prime candidate for an HTPC, after all. My Xbox One is running the most recent preview build of its “new experience” dashboard software. While it’s possible that the final release of this software may result in lower power consumption at idle for the Xbox, I wouldn’t expect miracles, especially under load.
Due to the GeForce GTX 960 inside, the Magnus draws a fair bit more power at idle than any of the other mini-PCs we’ve tested. Its draw with the CPU loaded closely matches that of the NUC5i7RYH, which has a 28W Broadwell chip. We couldn’t test the Xbox One under load using the same benchmark, but it uses more power at idle than all of the Intel systems.
The GeForce isn’t used at all in x264 encoding, so it wasn’t being stressed here. How much power does the Zbox use when gaming? It’s not really fair to compare this box to a fully decked-out Core i7-5960X system, since that CPU is a 130W part all by itself. In the living room, an Xbox is a more likely competitor. We fired up GTA V on both systems and tracked peak power consumption.
The Xbox One has a custom AMD SoC that has an eight-core Jaguar CPU and Radeon GCN graphics. It uses more power than the Zbox while gaming, yet the higher frame rate and detail settings on the Zbox put the Xbox to shame. GTA V is locked to 30 frames per second on the console. In fact, turning on v-sync on the PC lowered power consumption even further to 80W.
Next, I tested how well the Zbox’s cooling system works. To crank up the heat, I used a combination of Furmark‘s graphics burn-in test and Prime95‘s Blend torture test. Together, these applications should keep both the processor and graphics card fully loaded, and they’ll generate as much heat as possible.
For temperature testing, I logged the CPU clock speed and temperatures with Intel’s Power Gadget 3.0. For graphics, GPU-Z logged the chip’s clock speed and temperature along with the GDDR5 memory’s speeds. These graphs ended up being pretty boring to look at, because I believe the Zbox’s fan profile has been tuned to keep CPU and graphics card temperatures under 80° C, no matter what. Both chips plateaued at 79 degrees while gaming and didn’t budge throughout the rest of the test.
Despite those temperatures, neither the CPU or the graphics card throttled under load. That’s quite an achievement considering the graphics horsepower packed into the diminuitive Zbox. It is worth noting that the Core i5-5200U has a maximum Turbo speed of 2.7GHz, but when both cores were busy, it never clocked higher than 2.5 GHz. On the other hand, the GeForce GTX 960 in the Zbox Magnus EN970 hardly ever moved off its boost speed of 1,038 MHz.
Overall, Zotac appears to have done a heroic job with the cooling system of the Zbox, since both major processors inside can run near their maximum speeds without a hitch. That’s good news for smooth frame rates.
As we saw when we took it apart, the Magnus uses two blowers to cool the CPU and graphics card. One exhausts air from the left side of the system and the other from the rear. To measure fan noise, I used an iPhone 6S Plus with the dB Meter app from the iOS App Store. According to the app, my test environment had a noise floor of approximately 29 dBA. Noise testing was performed at idle and under a typical gaming load with Grand Theft Auto V. I measured noise levels 6″ from each system.
I tested three systems: the Zbox, my Core i5-6600K-based desktop PC, and the Xbox One that we used for power testing. For reference, my PC is built inside a Fractal Design Define R4 case with its stock fan configuration, an EVGA GeForce GTX 970 with an ACX 2.0 cooler like the one on the company’s GeForce GTX 980 Ti, a Cooler Master Hyper 212 Evo, and a Seasonic S12II 620W power supply. I set the Define R4’s fan controller to its medium setting for this test. I ran each system for 15 minutes at idle and for 15 minutes while gaming before making any measurements.
At idle, I can definitely feel a slight fan vibration when I place my hand on the Zbox, but even when I put my ear an inch or so from its vents, I can’t hear a sound. That quiet running comes at a price, though. Power Gadget registers idle CPU temperatures around 54° C. Another oddity is that even after I put it into sleep mode, the Zbox remains strangely warm to the touch. I’m not sure why that is, since the system shouldn’t be active at all while it’s asleep.
The other systems are quieter under load than the Zbox, but they have much more volume and bigger, slower-spinning fans. I’m pretty sure my PC’s quieter operation is entirely attributable to the beefy dual-fan ACX 2.0 cooler on its GeForce graphics card. The Xbox is cooled by a single, relatively large system fan, and its cooling system seems to be overkill. GTA V barely caused the Xbox’s fan to spin up at all. Given that context, the Zbox’s noise levels are pretty stellar for its diminutive size.
Graphs can’t completely describe the character of the Zbox’s fans, though. To help with that, I’ve taken a video that shows the sound of the fans three ways: at idle, while playing GTA V, and during our thermal torture tests. You’ll see the same dB Meter app measuring noise levels, too. The average decibel level we used for reporting is clearly visible in the video.
The idle sounds are just background noise. My camera evidently compresses the audio captured by the microphone, and the system is so quiet at idle that the background noise gets amplified. The system isn’t audible until it’s running a game, and even then the fans produce a gentle “whoosh.” At worst, the fans do sound like a tiny hair dryer when the Zbox is under the intense stresses of Prime95 and Furmark at once. Still, the Zbox is able to keep temperatures in check even under full load. Impressively, it takes two extreme benchmarking applications to push the sound pressure levels into the “annoying” range.
The relatively mild-mannered noise character of the Zbox stands in stark contrast to another mini-PC with discrete graphics that we’ve tested: Gigabyte’s Brix Gaming. That green machine got intolerably loud when running all-out. The decibel levels we measured in the Brix review may not be directly comparable to those of the Zbox because of the different equipment used, but even accounting for that difference, the amount of noise created by the Zbox while gaming is far tamer than the Gigabyte box’s howl under load.
The Zbox is somewhat larger than the Brix, to be fair, but we wondered what a slightly larger system with a beefier cooling system would be like in the conclusion of our Brix review. Now we have the answer to that question, and it seems that slightly increasing the footprint of tiny gaming systems like this is well worth it.
The Magnus EN970’s concept is easy to appreciate as soon as the system pops out of its box. Zotac’s new baby is a compact PC with plenty of graphics firepower backed up with a decent amount of CPU grunt. The Zbox can push pixels at 2560×1440 with DSR in some less-demanding games while delivering smooth experiences at 1080p with more graphically-intense titles—even with the eye candy turned up. That level of performance is pretty amazing considering this system’s diminutive size and modest power requirements.
Even so, I can’t help but wonder whether the Zbox’s Core i5-5200U CPU might become a bottleneck at times. The Zbox didn’t struggle with any of the games I threw at it during the course of my review, but a beefier processor would make me feel better about this system’s gaming chops over the long run. For the most part, though, the 15W Broadwell chip provided an experience that felt like that of a full-sized desktop. Despite its back-of-the-pack web scores in our benchmarks, web browsing and watching full-screen YouTube videos in the HTML5 player were both great experiences on the Zbox.
Zotac doesn’t make our ears ring to deliver the EN970’s graphics firepower, either. That’s an improvement this category of PC desperately needed. The Zbox’s fans are quiet enough under normal loads that the system won’t be a distraction to movie watchers or people trying to sleep in the same room. Since the Zbox can run Windows 10, it’s potentially more useful in the living room than the more locked-down Xbox One, and Zotac’s little system has enough graphics horsepower to put Microsoft’s current console to shame.
The Zbox’s pint-size power doesn’t come cheap. The $799 starting price for the barebones EN970 could cause some sticker shock even before one factors in the cost of storage of the required DDR3L-1600 SO-DIMMs. If you have enough room on your desk or under your TV, it’s possible to build a faster Mini-ITX system for about the same price, particularly something with more CPU grunt.
Builders who want to stay with something the size of the Zbox can spend a little more for a ready-to-go SN970 Steam box. That system’s quad-core Core i5-6400T should add quite a bit more CPU power at the cost of a little more power consumption and potentially more heat and noise.
Mini-ITX systems are still quite a bit larger than this Zbox, though, and there’s no guarantee that moving to a bigger case would even make for a quieter system. All told, our complaints are minor niggles. System builders who are short on space and need of plenty of graphics horsepower would do well to check out Zotac’s little engine that can.