Haswell-E and its X99 sidekick are easily the most exciting tag team for high-end desktops. The CPU crams up to eight cores into a single socket, and it’s backed by quad channels of cutting-edge DDR4 memory. Haswell-E chips also have up to 40 PCIe Gen3 lanes, providing copious bandwidth for multi-card graphics configs and high-speed SSDs. And then there’s the X99 chipset, which is brimming with USB 3.0 and 6Gbps SATA ports, plus configurable I/O lanes that can be devoted to SATA Express or M.2 storage.
This duo isn’t Intel’s high-end desktop platform for nothing.
Enthusiasts who want to get in on the action will need a CPU from the Core i7-5xxx series and a motherboard based on the X99 chipset. There are dozens of compatible boards from which to choose, including Gigabyte’s X99-UD4, which is one of the more affordable options around. We’ve spent some quality time with the UD4 to learn what it does well—and where it falls short. Read on for the scoop.
From a distance, the UD4 gives off a vaguely Dark Knight vibe. The matte black board and low-key aesthetic should nicely blend into the shadowy confines of most PC enclosures.
Don’t think the subdued gold accents mean this thing is short on bling, though. The UD4 has extra gold plating on the pins in the CPU socket, DIMM slots, and PCIe x16 slots. According to Gigabyte, this 30-µm gold layer is five times thicker than on “standard” designs. The additional material is supposed to improve corrosion resistance and long-term durability, which might appeal to folks who live in especially humid climates or who swap components obsessively. Seems a little gimmicky to me.
The socket is fed by all-digital circuitry based on a new generation of International Rectifier’s PowIRstage hardware. “Server level” Cooper Bussmann chokes dot the landscape along with solid-state capacitors from an unnamed vendor. We’re not experts on the minutiae of base electrical components, so make what you will of those upgrades.
Rows of 288-pin DDR4 memory slots flank the socket, offering a home for up to 64GB of memory. Beware of modules with taller heat spreaders; the memory slots are close enough to the socket to potentially interfere with larger aftermarket CPU coolers. We can’t test every hardware combination for compatibility, so we’ve measured a few key clearances to illustrate how much room is available.
The DIMM slots are no closer to the socket than on the other X99 boards we’ve seen. The clearances for the top PCI Express x16 slot and VRM heatsink are a little tighter, but at least the heatsink is relatively short. Of course, socket clearance issues can be easily avoided by running a closed-loop liquid cooler with a low-profile CPU block. That’s what we recommend for Haswell-E builds.
The expansion slots are partly to blame for the crowded socket. They’re stacked seven tall, yielding one more slot than on the pricier Asus X99 Deluxe we reviewed recently.
All four of the x16 slots have Gen3 connectivity from the CPU. The first two share 16 lanes that can be devoted solely to the top slot or split evenly between them. The third slot has 16 lanes at all times, while the fourth makes do with x8 connectivity.
CrossFire and SLI configurations are supported with up to four cards, though only when paired with CPUs that have all 40 PCIe lanes intact. Intel limits the Core i7-5820K to 28 lanes, which in turn restricts the UD4 to three-way SLI. CrossFire isn’t as picky about bandwidth, so quad-Radeon configs will still work with the runt of the Haswell-E litter.
The spacing provides ample breathing room for larger graphics coolers, including dual-triple-wide setups, which leave the middle x1 slot unobscured. Up to three devices can be stuffed into that gap: one in the x1 slot and two more in the low-slung M.2 sockets to the right. The lower M.2 receptacle has a single-lane PCIe Gen2 link to the X99 chipset; it’s meant for mini Wi-Fi cards. Twice the bandwidth is available to the upper socket, whose dual-lane chipset link is reserved for storage devices.
As in most 9-series Intel boards, the M.2 storage slot shares a reconfigurable chipset link with the SATA Express connector and the dual 6Gbps SATA ports embedded within it. This arrangement is officially endorsed for Z97 and H97 chipsets, but it’s not validated for X99, and Intel warns that some storage-related functions (likely related to its RST drivers) may not work correctly. Motherboard makers seem confident everything will be just fine.
Counting the SATAe connector, the X99-UD4 has 10 Serial ATA ports. Six of those are available for RAID arrays managed by Intel’s drivers, while the remaining four are restricted to IDE or AHCI modes due to limitations associated with the X99 chipset. Drives connected to the non-RAID ports can still participate in arrays managed by the OS or other software, though.
The X99-UD4 lacks auxiliary storage controllers, but it does split the chipset’s USB 3.0 connectivity using a Renesas hub. Four of the rear ports are connected through that hub, while the remaining two—and the two-port internal header—enjoy direct lines to the chipset. Too bad Gigabyte doesn’t identify which of the rear ports are which.
The company does, however, mark one of the USB ports with white trim. This port is meant for Q-Flash Plus, which can update the mobo firmware using little more than a power supply and a thumb drive. The automated process copies the primary firmware to the onboard backup chip before updating the main one.
Apart from the antenna bracket for M.2 Wi-Fi cards, the rear cluster is fairly conventional overall. The GigE jack is backed by an Intel controller, and the audio is fed by a Realtek codec. There’s an amplifier chip for the stereo output, plus a familiar mix of other enhancements, such as audio-specific capacitors, codec shielding, and isolated traces. The left and right output channels are split between different PCB layers to minimize crosstalk, too.
The analog audio output sounds decent to my ears, with no audible hissing or buzzing under load. Realtek’s drivers add surround sound virtualization, but they can’t encode multi-channel audio for digital output, which limits the S/PDIF port to stereo playback and surround content with pre-encoded tracks.
Gigabyte ships the UD4 with a cushioned I/O shield that’s infused with LEDs. The board has extra LEDs, too, and the lighting can be set to flash in time with audio piped through the onboard stereo output. Grab a glow stick, drop a tab of E, and let your dilated pupils take in the show:
I’ve gotta admit, my inner 16-year-old thinks that’s pretty rad. (People actually said “rad” back when I was a teenager—without any irony, even.) The rest of me is happy the lighting can be disabled in the firmware or via Windows software. Pulsing and always-on modes are also available.
In addition to the funky lighting, a handful of other unique twists separate the UD4 from the rest of the pack. The included SATA cables are sheathed in woven housings, just like typical PSU wiring, and there’s a bizarro three-to-one splitter for auxiliary 12V power. Gigabyte also defines a wider boundary around the mounting holes to prevent slipped screwdrivers from damaging traces and onboard components. That little tweak was inspired by the RMA department, which apparently sees a lot of surface damage around the screw holes.
Despite this attention to detail, the UD4 lacks wiring blocks to simply front-panel connections. There’s no way to boot directly into the firmware interface, either. Gigabyte does offer a direct-to-UEFI switch on some of its higher-end X99 boards, but that feature didn’t make the cut for this particular model.
Speaking of the firmware, let’s assess the UD4’s overclocking and tweaking options…
Firmware and software tweaking options
The X99-UD4’s firmware and software interfaces are very similar to those of other 9-series Gigabyte boards, including the Z97-UD5H we reviewed earlier this year. Minor changes have been made here and there, and we encountered a few oddities during testing.
Sadly, one of those quirks tarnishes the firmware’s gorgeous, high-definition interface. This so-called Smart Tweak GUI normally renders at 1080p, but it requires some cooperation from the graphics card, which must expose that resolution to the firmware. The hot-clocked GeForce GTX 680 we use for motherboard testing renders everything correctly at 1080p when using the DVI output, but the maximum firmware resolution drops to 1600×900 over DisplayPort. Weird.
Tweakers can set their own firmware background images, search for individual settings by name, check recently-changed variables, and configure custom tabs with their favorite options. The mouse sensitivity is adjustable, too, and the pointer tracking is pleasantly smooth most of the time. There’s annoying lag and visible hitching when dragging the settings sliders any faster than a snail’s pace, though. At least most variables can also be tweaked via drop-down menus and direct keyboard input.
The overclocking options are mostly complete, but the memory multiplier tops out at 26.66, which means DDR4-2800 and faster modules require higher base clock speeds to reach their full potential. Plenty of DDR4-2800+ kits are already for sale, so this isn’t just a theoretical limitation. We’ve already tested an Asus X99 board with higher memory multipliers, and that company tells us the required hooks are in the Intel microcode available to mobo makers. However, when we talked to Gigabyte about the 26.66X ceiling, the firm didn’t express any plans to push higher.
While we’re griping, we should also point out that the firmware engages in a little unsolicited CPU overclocking if the user changes the memory speed or invokes an XMP profile. Pretty much all modern mobos do this to some degree, and it’s particularly annoying in Gigabyte’s case, because the firmware displays the correct Turbo multipliers while sneakily using higher ones. Ugh.
The Smart Tweak interface covers overclocking and most other options, but some settings, like advanced peripheral and boot options, are only accessible through the Classic UI pictured above. This alternate interface gets a fresh coat of paint on the UD4, but little else has changed. That’s not necessarily a bad thing; the whole point of the Classic interface is to preserve the familiar environment of old-school BIOSes.
Unfortunately, the fan controls in both the Classic and Smart Tweak interfaces are stuck in the past. Pre-baked and manual profiles are present for all the CPU and system fan headers, but the only configurable option in manual mode is a single value that covers the slope of the fan speed
curve line. If you want more control over how the fans react to temperature changes, you’ll need to use Gigabyte’s System Information Viewer software.
This Windows app maintains a six-point profile for each fan header. All points but the last one can be dragged around with the mouse to create just the right response curve. The integrated calibrator measures the actual speed range for each connected fan, ensuring an accurate profile. There are also configurable alerts tied to temperatures, voltages, fan speeds, and other system variables.
Motherboards come loaded with a dizzying array of software these days, and we simply don’t have time to sample it all. Besides, most of it’s gimmicky fluff that we’d never run on our own systems. But we would install Gigabyte’s EasyTune utility, which brings key overclocking and power-tuning controls to Windows.
The EasyTune interface is clean and functional overall. Newbies should feel right at home with the draggable sliders and drop-down menus. More advanced users may miss the ability to key in values directly, though.
The manual controls are complemented by a handful of pre-defined profiles and an automated overclocking mechanism. We like the idea behind auto-tuner, which seeks higher speeds by slowly increasing the CPU frequency and then testing system stability at each step, much like a savvy enthusiast would. On the next page, we’ll see how well it works—and how far we were able to push our Haswell-E CPU manually.
For the last couple generations, Gigabyte’s auto-overclocking mechanisms have been very heavy-handed with voltages, as if the firm’s engineers expected everyday users to be chilling their CPUs to sub-zero temperatures. The X99-UD4 is much more reasonable on that front. When we ran the auto-tuner, it cranked our Core i7-5960X up to 4.1GHz on just 1.2V. The system was rock solid at that speed, and CPU temperatures hovered around 65°C with our mammoth Cooler Master Nepton 280L water cooler attached.
Our CPU can achieve higher speeds with only a smidgen more voltage, but we’d prefer that an auto-tuner be too conservative rather than too aggressive. We’d also prefer that such mechanisms use higher Turbo limits when fewer cores are engaged, since higher speeds are usually attainable with only one or two active cores. Instead of taking that route, Gigabyte’s utility sets a single multiplier that applies regardless of how many cores are spooled up.
Next, we turned our attention to manual tuning. Our typical M.O. is to slowly increase the CPU multiplier and leave all the other settings at their “auto” defaults. Most motherboard firmware increases voltage and power settings automatically as the CPU speed rises, and we like to know how well that integrated intelligence keeps up.
On the X99-UD4, we made it to 4.3GHz with multiplier tweaks alone. The CPU ran on 1.248V at that speed, but we needed to up the voltage to 1.3V to get the system stable at 4.4GHz. That ended up being the highest stable speed for our test rig. Although the system booted at 4.5GHz, we couldn’t find a mix of power and voltage settings that didn’t induce thermal throttling or produce BSOD errors under load. We’ve yet to get this particular CPU-and-cooler combo perfectly stable at 4.5GHz on any X99 board.
Motherboards typically influence system power consumption more than performance. We tested that tendency over a five-minute idle period and under a load comprised of Unigine’s Valley graphics benchmark and Cinebench’s multi-core CPU rendering test.
The X99-DU4 consumes 13W less than the X99 Deluxe at idle and 20W less under load. Those are sizable differences, and they seem unrelated to the UD4’s memory frequency. The Gigabyte board has pretty much identical power draw with the RAM dialed back to Haswell-E’s DDR4-2133 default.
The Deluxe is powering more onboard extras, such as auxiliary SATA, USB, and Wi-Fi controllers, but that doesn’t explain why the UD4’s advantage expands under load. Gigabyte’s fancy power regulation circuitry might have an overall efficiency edge.
When they’re equipped with the same components running at the same speeds, motherboards typically have little impact on system performance. There are rare exceptions, so we still run a range of application and peripheral tests on the boards that pass through our labs.
Unfortunately, a strict apples-to-apples comparison escaped us here. We’re using Corsair’s Vengeance LPX DDR4-2800 memory in our Haswell-E test rig, but due to firmware multiplier limitations, we can’t run it at top speed in the X99-UD4 without boosting the base frequency to 105MHz. That might seem like a small increase—it only raises the peak CPU frequency from 3.50GHz to 3.68GHz—but Intel CPUs are typically very sensitive to changes in their base clock. Our test system isn’t entirely stable with a 105MHz base frequency.
The other X99 board in our stable is Asus’ X99 Deluxe, which has the multiplier required to maintain stock CPU clocks while running DDR4-2800 memory. Rather than re-testing the Deluxe to match the slower DDR4-2666 setup we used on the UD4, we decided to compare the two configs to see where the slight difference in memory frequency might affect performance. Here’s how it impacts a targeted memory bandwidth test:
Surprisingly, the bandwidth gap is a little wider than one might expect from the change in memory frequency. Yet the two boards produce identical results in many of our performance tests:
So, yeah, nothing exciting here. The Gigabyte config is a smidgen slower in a few instances, but it actually comes out slightly ahead in x264 encoding. It’s been a while since we took an in-depth look at the impact of memory speed on application and gaming performance. Perhaps it’s worth revisiting the subject with Haswell-E and a wider range of memory speeds. Hmmm.
The results of our peripheral tests are mostly a wash, with the X99-UD4 shadowing its Asus competition virtually throughout. In most cases, the differences are no larger than the run-to-run variances for each board. That said, the Gigabyte has slightly higher random I/O rates with our PCIe SSD but slightly lower random I/O performance with USB 3.0 storage. By far the biggest discrepancy is in USB 3.0 random write speeds: the UD4 hits 147MB/s, while the Deluxe reaches 171MB/s. Sequential throughput is much more relevant for USB storage, though, and the gaps are much smaller in those tests.
Both boards exhibit low DPC latencies during basic music playback and a full system load. They also have similar analog audio signal quality according to RightMark Audio Analyzer’s “loopback” test. We could make you scroll through a bunch of unnecessary bar graphs (again), but there’s little point when the results are basically the same.
I guess graphs can help to break up large swaths of text. So, how about some boot times?
The X99-UD4 goes from power to the Win8 desktop 1.7 seconds faster than the Deluxe. That number may have some significance for a notebook or mobile device, but if you boot your desktop frequently enough for a couple of seconds to matter, you have bigger problems to address. And you should probably just be putting the machine to sleep, a state from which the UD4 awakes almost instantly.
For what it’s worth, X99 systems boot 10-15 seconds slower than their Z97 counterparts. This sluggishness seems to be related to the extra initialization time required by the platform’s glorious excess. By the time X99 rigs get around to displaying the POST screen, Z97 systems are already sitting at the desktop.
We’ve now covered the most important aspects of the X99-UD4. The following page has more detailed specifications in addition to specifics on our test systems and methods, but the giant tables and nerdy details are a little dry, so we won’t blame you for skipping to the conclusion. Or you could bypass that agonizing decision and pay what you want to become a subscriber, which gets you a nifty single-page view. Just sayin’.
Here’s the X99-UD4’s full spec sheet in case we missed anything.
|Platform||Intel X99, socket LGA2011-v3|
|DIMM slots||8 DDR4, 64GB max|
|Expansion slots||4 PCIe 3.0 x16 via CPU
3 PCIe 2.0 x1 via X99
1 M.2 PCIe 2.0 x1 via X99
|Storage I/O||1 M.2 type 2242-2280 via X99
1 SATA Express via X99
6 SATA RAID 6Gbps via X99
4 SATA 6Gbps via X99
|Audio||8-channel HD via Realtek ALC1150
Surround virtualization via Realtek drivers
|Wireless||Add via M.2 PCIe slot|
|Ports||2 USB 3.0 via X99
4 USB 3.0 via X99 and Renesas uPD720210 hub
2 USB 3.0 via internal headers and X99
4 USB 2.0 via X99
4 USB 2.0 via internal headers and X99
1 Gigabit Ethernet via Intel I218-V
1 analog front/headphone out (amplified)
4 configurable analog ports
1 digital S/PDIF out
|Overclocking||All/per-core Turbo multiplier: 12-80X
Uncore ratio: 12-80X
CPU gear: 1X, 1.25X, 1.66X, 2.5X
Base clock: 80-133MHz
DRAM ratio: 8-26.66X
CPU voltage: 0.5-1.7V
CPU input override voltage: 1.0-2.7V
CPU ring voltage: 0.8-1.6V
System agent voltage offset: -0.3 – +0.4V
DRAM A/B, C/D voltage: 1.0-2.0V
DRAM A/B, C/D VPP voltage: 2.0-3.0V
DRAM A/B, C/D termination voltage: 0.75-1.66V
PCH core voltage: 0.65-1.3V
PCH I/O voltage: 1.05-1.9V
|Fan control||2 x CPU, 3 x SYS:
Normal, silent, full-speed profiles
Manual profile with one PWM/temp value per fan
The PCIe configurations are for 40-lane CPUs. With a 28-lane chip, the lane assignments are x16/x0/x8/x4 or x8/x8/x8/x4.
Our testing methods
If you’ve made it this far, you may be curious about what the rest of our test system looks like. Here it is:
We used the following configurations for testing.
|Motherboard||Asus X99 Deluxe||Gigabyte X99-UD4|
|Platform hub||Intel X99||Intel X99|
|Chipset drivers||Chipset: 10.0
|Audio||Realtek ALC1150||Realtek ALC1150|
|Memory size||16GB (4
|Memory type||Corsair Vengeance LPX DDR4
SDRAM at 2800MHz
|Corsair Vengeance LPX DDR4
SDRAM at 2666MHz
GeForce GTX 680 DirectCU II with 340.52 drivers
Force Series GT 120GB
Samsung 840 Pro 256GB
Windows 8.1 Pro x64
Thanks to Intel, Corsair, Samsung, and Asus for providing the hardware used in our test systems. And thanks to the motherboard makers for providing those.
We used the following versions of our test applications:
- 7-Zip 9.20 64-bit
- TrueCrypt 7.1a
- Chrome 34.0.1847.131
- x264 r2431
- DiRT Showdown demo
- CrystalDiskMark 3.0.3b
- FRAPS 3.5.99
- RightMark Audio Analyzer 6.2.5
- Cinebench R15
- Unigine Valley 1.0
Some further notes on our test methods:
- All testing was conducted with motherboard power-saving options enabled. These features can sometimes lead to slightly slower performance, particularly in peripheral tests that don’t cause the CPU to kick into high gear. We’d rather get a sense of motherboard performance with real-world configurations, though; we’re not as interested in comparing contrived setups with popular features disabled.
- DiRT Showdown was tested with ultra detail settings, 4X MSAA, and a 1920×1200 display resolution. We used Fraps to log a 60-second snippet of gameplay from the demo’s first race. To offset the fact that our gameplay sequence can’t be repeated exactly, we ran this test five times on each system.
- Power consumption was measured at the wall socket for the complete system, sans monitor and speakers, using a Watts Up Pro power meter. The full-load test combined Cinebench’s multithreaded CPU rendering test with the Unigine Valley DirectX 11 demo running in a 1280×720 window. We reported the peak power consumption during the Cinebench run. Our idle measurement represents the low over a five-minute period sitting at the Windows desktop.
- The Force GT 120GB SSD was used as the system drive for all tests. The Samsung 850 Series 512GB was connected as secondary storage to test Serial ATA and USB performance, the latter through a USAP-compatible Thermaltake BlacX 5G docking station. The Samsung SSD was secure-erased before each test that involved it. The Corsair drive was also wiped before we loaded our system image.
- Ethernet performance was tested using a remote rig based on an Asus P8P67 Deluxe motherboard with an Intel 82579 Gigabit Ethernet controller. A single Cat 6 Ethernet cable connected that system to each motherboard.
- Analog audio signal quality was tested using RMAA’s “loopback” test, which pipes front-channel output through the board’s line input. We tested while the system was loaded with Cinebench’s multithreaded rendering test, the Unigine Valley benchmark, and a CrystalDiskMark 4KB random I/O test running on the Samsung SSD attached via USB 3.0.
The tests and methods we employed are publicly available and reproducible. All tests except power consumption were run at least three times. Unless otherwise indicated, we reported the median result for each test. If you have questions about our methods, hit our forums to talk with us about them.
Gigabyte’s X99-UD4 has a lot of appealing characteristics marred by a few rough edges. The board takes full advantage of the bounty of bandwidth and I/O built into Haswell-E and its chipset companion. There are enough PCIe x16 slots—and space between them—for four-way, double-wide CrossFire and SLI configs. DDR4 memory slots abound, and the storage options are varied. M.2 connectivity even extends to Wi-Fi cards, which can be run alongside mini SSDs.
Although the UD4 isn’t any faster than other X99 boards, it seems to be particularly power-efficient, and overclocking on the thing is a breeze. Some nice extras are included, too, like sheathed cables, a cushioned I/O shield, and rave-worthy lighting that dances to music. It would be nice if Gigabyte could extend lighting control to generic headers for modders to connect their own LED arrays. And it’s even nicer that the lighting can already be tweaked and disabled with ease.
Gigabyte’s accompanying Windows utilities are generally solid, but the firmware could be more polished. The antiquated fan speed controls pale in comparison to those available in the Windows software—and in the firmware on competing X99 boards. Also, the UEFI’s cap on memory multipliers complicates running DDR4-2800 or faster modules. Haswell-E is technically only rated for DDR4-2133 speeds, though, and there isn’t necessarily a lot to be gained from much faster RAM.
Those might seem like minor flaws, but the fact is that most modern motherboards work pretty well. The differences boil down to the details. Gigabyte gets a lot of the little things right on the UD4, but it also misses some low-hanging fruit that could improve the overall experience for users. There’s no excuse for omitting port blocks or some kind of improved front-panel wiring harness on a motherboard targeted at people who build their own systems.
Considering the price tag, the X99-UD4’s less desirable traits aren’t severe enough to ruin the overall product. Newegg is charging only $244 for the UD4 right now, which is much less than the going rate for other X99 boards with eight DIMM slots and four-way SLI support. Even though the UD4’s priorities might not be aligned with the needs of every enthusiast, the value proposition is undeniable for those who want to tap into all the goodness Haswell-E offers.