I have always thought of mini-ITX motherboards as utilitarian in that they are used only for specific reasons: HTPC, “carputer” or for basic computing tasks in cheaper, smaller footprint than ATX offers. Most people probably never think of them as motherboards for gaming, overclocking or benchmarking with, and I was one of them.
My view changed a few months ago when an online competition required use of a mini-ITX motherboard and a Gigabyte H55M-UD2H mini-ITX board was purchased to compete with. Maybe it was because the board was so inexpensive compared to the decked out “overclocking” boards or maybe I just wasn’t expecting much from such a diminutive motherboard, but was actually surprised by how well it overclocked. Soon after that competition was over, the board was sold, but it wasn’t long after that, another online competition required the use of a mini-ITX board again, so now enters the more recently released Gigabyte H55N-USB3 mini-ITX motherboard and this time it will definitely be kept to take on HTPC duties after the competition.
The H55N-USB3 is an Intel H55 chipset board for socket 1156 Core i3, i5 and i7 processors and is part of Gigabyte’s “Ultra Durable 3” line. That means that it incorporates twice the amount of copper for power and ground layers than traditional designs, 50,000 hour Japanese solid capacitors, Lower RDS(ON) MOSFETs and Ferrite Core chokes.
It’s other notable features are USB3.0, “3x USB Power” and “ON/OFF Charge”. USB3.0 has a theoretical maximum bandwidth of 4.8GB/s, which is 10x as much as USB2.0, and will be a welcomed feature for transferring large files. “3x USB Power” is a feature that can be a major benefit given the high-powered mobile devices being released lately. A standard USB port has a maximum of 500mA, but Gigabyte’s “3x USB Power” allows up to 1500mA (1.5A) on a single USB port, which is beneficial for charging the ever-increasing power needs of mobile devices, especially when using the device and expecting it to charge at the same time. Another feature on the Gigabyte H55N-USB3 is “ON/OFF Charge” which allows charging and quick charging of mobile devices via the USB3.0 ports regardless of whether the motherboard is on, suspended or even off (hence the name “ON/OFF Charge”).
The H55N-USB3 features four internal SATA2 ports (no RAID), internal headers for four USB2.0 ports and a PCIe 2.0 x16 slot. The rear I/O panel contains six USB ports: four USB2.0 and two USB3.0 (blue). When using an Intel Core processor that has an on-board graphics processor (IGP), the motherboard has options for D-SUB, DVI-D and HDMI connections. Rounding out the rear I/O are an eSATA port, PS/2 connector for keyboard or mouse, S/PDIF audio, Gigabit LAN port and standard analog audio ports.
While it’s not expected that many will opt for large aftermarket CPU coolers on a mini-ITX setup for good reason, some may be tempted to squeeze one on. Large aftermarket coolers can have spacing issues even on standard ATX boards and space is really at a premium on mini-ITX. As seen in the photo above, large, tower-style coolers, like the Thermaltake Frio pictured, can be as large as the board itself and will almost certainly obstruct either the memory slots, the PCIe slot or both depending just how big the cooler is. Be very aware of the proximity of the CPU socket to the slots if planning to ditch the stock Intel heat sink in lieu of an aftermarket unit. Also be aware that back plates that may come with large aftermarket coolers will likely not be usable do to SMT placement on the bottom of the board near the mounting holes.
Now that issues with aftermarket coolers have been noted and being fresh out of OEM Intel 1156 coolers, water cooling will be used on the processor for this article so that access to PCIe slot is possible for a discrete graphics card. Setting up this board is quick and easy. If no standalone graphics card will be used, the only hardware to install on the board are the IGP-equipped processor, CPU cooler and the memory modules. Done!
Note that if a standalone video card will be used, the memory modules should be installed first just to make it easier. The lock tabs for the DIMM slots closest to the PCIe slot will not fully open if the PCIe slot is populated with a card that extends past them. While the tabs generally still open enough to install or remove memory modules, it is easier to install the memory without a card in the PCIe slot. Similarly, depending on what CPU cooler is used, it may be difficult to reach the lock tab for the PCIe slot. Neither issue is a serious problem, just part of the reality of using such a small motherboard.
Even on this mini-ITX board, Gigabyte offers a BIOS with many overclocking options. While options are fewer than one might see on a high-end P55 board, everything that is really needed to overclock the motherboard, CPU, IGP and memory is available. Most non-extreme overclockers may never look at the maximum voltages allowed, but as with their ATX boards, Gigabyte offers some very high voltage limits. Vcore: 1.900V, Vtt: 1.49V, IGP: 1.800V, PCH: 1.500V and DRAM: 2.600V. The memory screen has full timing options available, including a CAS 5 option should lower speed and really tight DDR3 be needed. There is also one hidden option for the DVMT Memory Size, which defaults to MAX. While it can be adjusted, there is little reason to ever want to. To bring up the hidden menu, press CTRL+F1 from the main BIOS screen and a new sub-menu called “Advanced Chipset Features” will be displayed as shown in the screen images above.
This particular motherboard will find its way into a small form factor (SFF) setup for use as an HTPC, but since it’s not needed for “work” just yet, there is some time to play first. As you are about to see, as unassuming as a mini-ITX motherboard may be, the H55N-USB3 can deliver some big overclocks.
|Motherboard:||Gigabyte H55N-USB3 mini-ITX|
|Processor:||Core i5 655K|
|CPU Cooler:||NexXxos XP water block / Single-stage phase|
|Memory:||2x2GB Corsair Dominator GT 2000C7|
|Graphics:||Intel HD – Integrated Graphics Processor / Gigabyte GTX460|
|Solid State Drive:||Kingston V Series 64GB|
|Power Supply:||Gigabyte Odin 1200W|
|Operating System:||Windows XP SP3 (stripped)|
Intel Core i3 and Core i5 dual-core processors are particularly well suited to SFF setups in that they have fewer cores and run cooler than their Core i7 S1156 counterparts. Luckily, a Core i5 655K (unlocked) dual-core processor is on tap and works extremely well with this board. To begin, the IGP on the processor will be used, but since the IGP memory speed is directly linked to the BCLK of the motherboard, the IGP will limit board overclocking at a certain point and it doesn’t take long to find that point.
Keep in mind that scores were not the goal here, but general overclocks that would complete the benchmarks.
It doesn’t get much easier than this for overclocking. 225 BCLK, AUTO everything in BIOS except DRAM voltage/timings and manually set CPU multiplier. It actually boots to the desktop with higher BCLK than this, but overclocking the IGP memory (tied to BCLK) any higher causes a black screen. No amount of any voltage seems to be able to offset the black screen issue. While it can’t be cleared up or made to go higher, it is a nearly a 70% overclock of the IGP memory without having to (or really being able to) do anything to achieve it except raise BCLK.
A quick 3DMark05 was run (because it was already installed), but BCLK had to be dropped to 220 due to the IGP memory overclock. The IGP core was not overclocked for this bench, but reaching the higher default 900MHz IGP core speed of the Core i5 661 should be reasonably easy and the H55N-USB3 BIOS does have options for overclocking the IGP core. Since the IGP is limiting motherboard overclocks and not very good for gaming or 3D benchmark performance in general, it is time to toss in a stand alone graphics card and see what happens.
Dropping an external graphics card into the PCIe x16 slot and turning off the IGP immediately nets 15 more BCLK on AUTO voltages (yes, AUTO Vcore is lower at 240 BCLK than 220) and now boots to the desktop at 240 BCLK, 4.8GHz QPI and 4.32GHz Uncore on the first try. That is a strong ambient-cooled overclock by any measure. For a comparison on 3D processing, 3DMark05 was run with the GTX 460 this time and the same system clocks as before. This simple comparison shows that a mid-range external graphics card performs infinitely better for 3D rendering (3500 Marks on IGP to 35,000 on the external card). While the motherboard and processor would make for a fine SFF gaming rig, as seen by the 3DMark05 scores, a standalone graphics card is pretty much a necessity if the build will be used for serious gaming while the on-board IGP will be fine for just about everything else.
The integrated memory controllers (IMCs) on the socket 1156 processors aren’t always the strongest for overclocking. Add to that, an imposed Vtt limit and some choices need to be made when it comes to overclocking socket 1156 processors. Since just about everything must be handled by the IMC, trade-offs generally need to be made between BCLK, CPU clocks, memory clocks and timings, QPI and Uncore. To get one or more higher/tighter, one or more of the others may need to be lower/looser. Sub-ambient cooling helps what can be done, but there is no real cure. So, for ambient benchmarks, the focus here was mostly on high BCLK (and thus, high Uncore) and high QPI rather than CPU clocks and memory. While this memory has no problem running CAS 6 at these speeds, its left at CAS 7 to make things a little easier and that was one of the trade-offs necessary as previously mentioned.
Note that the Intel Core i3/i5 Clarkdale processors do not have Quick Path Interconnect (QPI), but CPU-Z shows their DMI bus speed under the “QPI” field and will be referred to as “QPI” because of that.
Here we have a small round of the usual benchmarks, beginning with the most difficult, wPrime 1024M. This is not a very easy bench to complete and generates a lot of heat, especially since this particular 655K processor requires a decent amount of Vcore when Hyperthreading is enabled, so this was actually a very nice overclock considering ambient cooling, high BCLK, high QPI and high Uncore. For the shorter 32M test (right), clocks were raised to 245 BCLK, 4900MHz CPU, 4900MHz QPI and 4410MHz Uncore.
Next on the list is 3DMark06, which again, is not a particularly easy benchmark to complete by any means, but is able to be completed at the same high clocks as the wPrime 1024M and with slightly lower Vcore.
With the “more difficult” benchmarks out of the way, Hyperthreading can now be turned off to save some stress on the IMC. Given how the multipliers, memory and BCLK land, BCLK was dropped slightly to 235 for 3DMark05, but the CPU multiplier was increased to 21x for a CPU overclock of 4935MHz. QPI remains at a very strong 4700MHz and Uncore is at 4230MHz.
Yes, it’s “only” SuperPi 32M and while not a particularly demanding benchmark as far as they go, it is not necessarily easy to complete given such a strain on the IMC with high overclocks . The fact that we’re seeing over 5100MHz CPU speed, 4860MHz QPI and 4374MHz Uncore benchable at 243 BCLK on water cooling is very nice.
Just about the easiest of all benchmarks, SuperPi 1M borders on just being a screen shot of the maximum overclock a setup will run. BCLK was bumped to 245, giving a CPU clock of 5145MHz, QPI at 4.9GHz and Uncore at 4410MHz. Over 5.1GHz CPU and 4.9 GHz QPI on water cooling is impressive, even if only for a short bench.
Of course, the motherboard and processor had to be put under sub-ambient cooling because that’s what I do! It’s a little tight, but the evaporator mount does fit (as does the mount for cascade, the dry ice tube and liquid nitrogen tube). The single stage phase unit used here is an older unit and by “extreme cooling” standards, is actually not very cold (-50C on the evap, warmer on the cores), nor does it have a lot of capacity. Gains were made, though cold or not, the issue remains of finding a happy median between everything there is to overclock and as CPU clocks go up, the others become more and more difficult to keep high and/or tight. Be that as it may, some very high overclocks are possible on this setup, so even having to give up a bit of something still means it stays at a fairly high clock.
Going back to the most difficult benchmark used here, wPrime 1024M, there is nearly a 500MHz gain on the processor (5292MHz), BCLK is up to 252 and Uncore is up to 4536MHz, but QPI had to be dropped to 4536MHz in order to pass this test because of low cooling capacity and rising temperatures during the benchmark. On this memory multiplier (8x), this was the first time the memory speed had gone past the 2000MHz barrier (7-8-7-20-1T timings), which is excellent given the stress on the IMC. Again, for the shorter wPrime 32M test, BCLK was raised slightly further to 254, giving 5334MHz CPU speed, 4572MHz QPI, a matching 4572MHz Uncore and 2032MHz 7-8-7-20 memory.
Back to 3DMark06 again, in the ever-present balancing act, QPI was raised to 5214MHz, but to achieve that, BCLK was lowered to 237 and memory was kept at it’s loose 7-8-7-20 timings. Final CPU clock is 5214MHz and Uncore is down to 4266MHz with the reduction in BCLK, but is over 4200MHz Uncore really that low?
As with 3DMark06, the BCLK was kept lower in order to keep the QPI over 5GHz for 3DMark05. Since Hyperthreading can be turned off for this benchmark, the CPU clock was able to be raised. Final clocks are 235 BCLK, 5405MHz CPU, 5170MHz QPI, 4230MHz Uncore and memory is down to 1880MHz, but with slightly tighter timings of 7-7-6-20.
As mentioned before, SuperPi 32M is actually not a particularly easy benchmark to finish with high BCLK and high everything else too. BCLK was kept relatively low at 238 with CPU multiplier of 23x for 5475MHz core speed. QPI is still high at 5236MHz with Uncore at 4285MHz and a reasonable 1904MHz memory and 7-7-6-20 timings. That is not to say higher BCLK and memory wouldn’t be possible with lower QPI, but part of the fun (to me anyway) of this setup is finding the combinations that produce the best scores.
One can always count on SuperPi 1M to be done at about the highest clocks a system will run and this one is no exception. While the CPU clocks could be higher for this bench, high BCLK was the goal and it was raised to a whopping 260 for 5460MHz on the processor, 5.2 GHz QPI, 4680MHz Uncore and memory at 2080MHz 7-8-7-20.
Since the single stage phase unit isn’t cold enough to get much more BCLK (or more likely Uncore), one stick of memory was removed and timings left on AUTO to reduce the stress on the IMC. That allowed for a very high 275BCLK, 5500MHz QPI, 4950MHz Uncore with 2200MHz memory at 9-9-9-24. Impressive, and that is just the highest it achieved on the single stage phase cooling. Believe it or not, it has gone even higher. This setup has been under dry ice briefly some time ago and if you think the setup overclocks well just on the water cooling or the single stage, not going even a whole lot colder than the single-stage yielded some amazing results.
Note that this CPU-Z validation is still on dual-channel memory, and on really fast and tight dual-channel memory to boot (2280MHz 7-8-7-20). Though it’s not seen in the validation, Uncore speed is a huge 5130MHz. Now Uncore may limit even higher BCLK when going colder on the cascade or LN2, but with some slacked up memory, single-channel, this setup stands every chance of hitting the magical 300 BCLK limit and we’ll see if it can do just that in the coming weeks.
As a wrap up, here is a chart of the all the overclocks for the benchmarks above. As stated before, there is no free lunch on S1156. In order to get higher clocks on one part of the system, something else generally has to be reduced and the trade-off made for increasing CPU clocks are easier to see in a chart format. Those trade-offs are what make benching this setup somewhat interesting: finding one of the many possible overclock combinations that will yield the best score in each benchmark.
|Water cooling||Single-stage Phase|
|Max Overclock (Single Stage)||2200MHz|
|Max Overclock (Dry Ice)||2280MHz|
The Gigabyte H55N-USB3 defies the general assumption that mini-ITX don’t overclock well. Seeing what is possible on this board, most would have to agree that we are looking at a small motherboard that can, in fact, overclock extremely well. There are plans for this particular board to be put into a case for HTPC duty, but it has quickly become one of my favorite boards to play with and that is why you are seeing a lot of screenshots and benchmarks here rather than pictures of a SFF install with it right now. The actual overclocking abilities of the setup seemed much more interesting to talk about than pictures of it installed in a case for now, but if you are looking for an install with the Gigabyte H55N-USB3, MIAHALLEN has an article up on overclockers.com that you will want to check out. It is his SFF gaming build using a Gigabyte H55N-USB3 and also includes his air-cooled overclocking results.
Most people will probably still continue to buy mini-ITX boards for lower cost, lower power, lower heat, compact space uses and the Gigabyte H55N-USB3 would do extremely well in that capacity, but it’s not your average install it, leave default settings and forget it mini-ITX board if you don’t want it to be. If planning on using a processor with an on-board graphics processor, the H55N-USB3 has connections to cover the oldest to the newest monitors (and HD TVs) while features like USB3.0, 3x USB Power and ON/OFF Charge give it something extra. When using a discrete graphics card, the H55N-USB3 would also do extremely well for a small gaming machine that can actually be overclocked, and, as demonstrated, overclocked very, very well.