Gigabyte GA X79-UD5 Main Board Review Part 2: Performance Test Results
by Mark W. Hibben
A Top Performer
In this part of the Gigabyte X79-UD5 main board review, I review the highlights of the performance tests we’ve performed over the past two months. As I discussed in Part 1 of this review, we went through three different BIOS versions in order to get to the latest version, F8, and most of the results presented in this article are based on F8, which performed the best overall. I’ll also show the effect of overclocking on various aspects of computer performance including CPU and graphics performance, as well as power consumption, which we consider important from the standpoint of environmental impact.
Performance Test Results
Passmark’s Performance Test 7 allows the user to download detailed test results from among the many user uploaded test results Passmark has received, and they have a very efficient built in search engine that allows you to find relevant test results. For this review, I downloaded results for a few systems with performance and features roughly comparable to the Technomicon X79 system I described in detail in X79 System Design.
A summary of the key features of our system and the others is shown below:
|System Designation||i7 990x MSI||Generic Core i7 2700K||Generic Core i7 3930K||X79 Marvell RAID (Technomicon 3930K @ 4.375GHz)|
|Processor||Intel Core i7 990X||Intel Core i7 2700K||Intel Core i7 3930K||Intel Core i7 3930K|
|OS||Windows 7 x64||Windows 7 x64||Windows 7 x64||Windows 7 x64|
|Motherboard||MSI MS-7666||Gigabyte Z68X-UD3H||Asus P9X9 Pro||Gigabyte X79-UD5 (F8 BIOS)|
|Base Clock Rate||167 MHz||100 MHz||130 MHz||125 MHz|
|Max. Non-Turbo Processor Clock||4.345 GHz||3.511 GHz||4.161 GHz||4.001 GHz|
|Max. Turbo Processor Clock||NA||3.912 GHz||4.291 GHz||4.376 GHz|
|Total Physical Memory||12.28 GB||16.30 GB||16.36 GB||16.33 GB|
|Primary Disk Drive||Unknown||Intel SSDSA2CW120||WDC WD4000AAKS||2x WDC WD6000HLHX|
|Drive Size||1863 GB||111.7 GB||372.6 GB||488.2 GB|
|Graphics Card||GeForce GTX 580||GeForce GTX 560||GeForce GTX 580||GeForce GTX 580|
|Graphics Memory||3072 MB||2048 MB||1536 MB||1536 MB|
Shown below is the summary of the PT7 test scores for all the basic tests, CPU, Memory, 2D Graphics, 3D Graphics, Disk Speed, CD Speed:
The CPU Mark covers functionality such as integer and floating point math, string sorting, encryption, compression and physics modeling. The 2D Graphics Mark tests performance in 2D drawing and image processing, as well as handling standard Windows user interface items such as menus and scroll bars, and results tend to reflect the speed of the processor more than the graphics card. The 3D Graphics Mark tests performance in 3D animation and specifically tests DirectX 10 performance, and here the results tend to reflect the choice of graphics card more than the processor. Memory Mark and Disk Mark test speed of writing and reading the memory and hard disks respectively. CD Mark tests optical drive read speed, and test results vary considerably based on the type of drive and media (DVD tests much faster than CD), so I don’t put much stock in the CD Mark results in evaluating PC performance.
CPU Mark and Power Consumption Update
CPU Mark performance was covered in detail in our Core i7 3930K review, but I’ll update it here for the new F8 BIOS. In the processor review, I plotted CPU Mark scores as a function of overclock frequency as well as power consumption at the CPU and system levels. As one would expect F8 didn’t affect CPU performance, but it did improve power consumption somewhat at the system level. In the following charts all quantities are normalized to the stock configuration for the F7 BIOS: CPU Mark = 13886.5, CPU maximum power consumption during the CPU Mark test = 113.2 W, and maximum system wall plug (AC) power consumption during the CPU Mark test = 277 W. The first chart (next panel) shows the behavior of the system when overclocking by increasing the Turbo Mode maximum multiplier (pure Turbo Mode).
The second chart shows the behavior of the system when overclocking via our so-called “Hybrid Mode” where the base clock is raised to 125 MHz. This increases the non-Turbo Mode maximum clock rate (the rate all processor cores can be operated at to 4.0 GHz), and then the maximum Turbo Mode clock multiplier is set to give the Turbo Mode maximum frequency as plotted in the graph. For instance, for a Turbo Mode maximum clock of 4.5 GHz, the maximum multiplier is set to 36 in the BIOS.
Although the F8 BIOS had no effect on processor performance as measured by CPU Mark, it did have the effect of slightly reducing power consumption at the processor and at the system level. We still regard 4.375 GHz in Hybrid Mode to be the sweet spot in overclocking, with peak power levels of 140 and 312 Watts recorded during the CPU Mark test for the CPU and system respectively (F8 BIOS). Given that the system power consumption only rises by 12.6% for the 23.5% CPU Mark performance increase, we consider this an environmentally responsible performance trade-off.
Peak CPU operating temperature during the CPU Mark test was a relatively cool 61 C, well below the Intel specified maximum temperature of 72.6 C. Although the peak power dissipation of 140 W for the CPU is 10 W over the Intel Thermal Design Point of 130 W, we are confident that this overclocking mode is sustainable over the long term with negligible risk to the processor or the motherboard. The reader’s results may vary, depending on the effectiveness of the processor cooling approach.
The X79-UD5 supports overclocking of memory manually or using the Extreme Memory Profile (XMP) supported by the 3930K. We outfitted our X79 system with G. Skill XMP memory rated at 2133 MHz. The BIOS has presets for XMP, and by selecting “Profile 1” we painlessly achieved 2133 MHz. The memory clock multiplier is completely independent of the core multipliers, so we operated the memory in XMP mode at 2133 MHz for all of the stock and OC settings with the base clock at 100 MHz. For Hybrid Mode with the base clock at 125 MHz, the closest we could get to the 2133 MHz limit without exceeding it was 2000 MHz (16x), so this was used for all the Hybrid Mode settings. Exceeding the 2133 MHz rating in XMP mode generally produced BIOS startup failure using the board XMP presets. Shown on the next panel are the details of the Memory Mark test results for the X79 system (overclocked at the Sweet Spot to 4.375 GHz) and the other comparison systems. The Memory Mark tests allocating small amounts of external RAM, reading from the processor cache memory, reading from external RAM (Read Uncached), writing to external memory, and allocating large blocks of external RAM.
Memory Mark performance does scale with maximum clock frequency, even for a fixed XMP clock rate, so there’s a memory speed up to be expected with overclocking. The chart below shows the improvement in the Memory Mark for both the pure Turbo and Hybrid overclock modes. Both Turbo Mode and Hybrid Mode Memory Mark scores are normalized by the stock (max. frequency = 3.8 GHz) Memory Mark score of 3987.9.
The slightly lower performance for Hybrid Mode at 4.25 GHz may be due in part to the lower memory frequency of 2.0 GHz.
2D Graphics Performance
PT7 tests various 2D graphics operations including drawing basic solid and transparent shapes, drawing more complex vector based images, rendering fonts and text, drawing various Windows GUI elements, and performing various image processing tasks such as filtering and rendering. In general, we find considerable variability in these results depending on the processor architecture, whereas there isn’t much dependence on choice of graphics card, as these operations mostly depend on the integer arithmetic speed of the processor. Shown on the next panel are the details of the 2D Graphics Mark results for the comparison systems.
2D Graphics Mark scores also show a dependence on maximum clock rate, with Hybrid Mode doing somewhat better than Pure Turbo Mode at moderate overclock frequencies. In the chart below, the Hybrid Mode and Turbo Mode 2D Graphics Mark scores are normalized to the stock score of 453.3.
3D Graphics Performance
3D Graphics Mark tests performance in 3D rendering and animation where 3 of the 4 tests depend both on processor and graphics card performance, and the fourth test, a DX10 based simulation, depends mostly on the performance of the graphics card.
Suffice it to say that we found nothing to indicate that the motherboard or processor were inhibiting graphics performance either in single card or dual card SLI operation.
Disk Drive Performance
The comparison systems show a wide range of disk drive performance and commensurate with the drive approach. The 2700K system employs an Intel solid state SATA 2.0 (3Gb/s) of 120 GB capacity. The 990X system uses a RAID 0 system of 1.8 TB capacity but unknown drives and interface. The MSI board hosting the 990X could interface the drives either through the ICH 10 (SATA 2.0) or Marvell 9128 (SATA 3.0). The generic 3930K system uses a conventional Western Digital Caviar 7200 RPM Drive with SATA 2.0 and 400 GB capacity.
Disk performance has turned out to be independent of processor maximum clock rate on our X79 system, but not of configuration. For this review, we explored a number of different combinations of BIOS and drive configuration. We also explored the effect, if any, of overclocking, and finding none, only the stock results are included in the table and chart below.
|PT7 Designation (File Name)||BIOS||Drive Configuration|
|X79IntelNonRAIDF3Stock||F3||Single Velociraptor to PCH SATA 3.0 port|
|X79IntelRAIDF3Stock||F3||Dual Velociraptors to PCH SATA 3.0 ports|
|X79IntelRAIDF8Stock||F8||Dual Velociraptors to PCH SATA 3.0 ports|
|X79MarvellRAIDF3Stock||F3||Dual Velociraptors to Marvell SATA 3.0 ports|
|X79MarvellRAIDF7Stock||F7||Dual Velociraptors to Marvell SATA 3.0 ports|
|X79MarvellRAIDF8Stock||F8||Dual Velociraptors to Marvell SATA 3.0 ports|
The non-RAID results demonstrate the basic speed of the Western Digital ‘Raptor WD6000 drives with 134 MB/s sequential read and 102.8 MB/s sequential write speeds. The Intel X79 PCH based RAID performed well, but was a little imbalanced between read and write speeds, being 235 MB/s and 150 MB/s respectively in final F8 configuration. We regard the Marvell based RAID as the better performer overall with 225 MB/s read speed and 234 MB/s write speed in final F8 configuration.
The much lower write speed for Marvell based RAID under the F7 BIOS of 151 MB/s shows the performance penalty exacted by that version of the BIOS.
Final LAN Performance
PT7 also has a LAN performance test utility that requires two computers connected via a network in order to run the test. During 60 second tests the X79-UD5 averaged 935 Mb/sec in receive mode and 911 Mb/sec in transmit mode over our gigabit Ethernet network. Plots of transmission or reception rate are shown below and on the next panel.
The data rate glitch that can be seen in the Received Data plot was produced by the transmitting system used for the test.
We also tested data throughput using PT7 when connecting to our network using the Gigabyte supplied Wireless N card. Our network uses the Linksys WRVS-4400N Wireless Router operating at 2.4 GHz in N-only mode with 40 MHz bandwidth enabled. Security was set to WPA-2 with connection control. At a distance of 8.5 M from the router we averaged 53.2 Mb/s data rate in transmission and 42.8 Mb/s in reception at the X79 system, respectable numbers for a single band system. The Gigabyte card also incorporates a Bluetooth radio, and we found we could connect Bluetooth devices to the X79 system without difficulty.
Although it was a long and somewhat difficult process to get to this point, the final state of the X79 system and the performance of the Gigabyte X79-UD5 motherboard (running the F8 BIOS) are extremely impressive. The system doesn’t merely test out fast, it feels fast in a fluidly powerful way. This computational power can be enjoyed without guilt, since the system consumes relatively little electrical power at idle. And idle is where the processor tends to sit most of the time it’s on, being able to polish off mundane computational chores in the blink of an eye. For demanding tasks, the system rises to the occasion like no other, providing performance that will only be exceeded by future LGA 2011 systems. I expect the Gigabyte X79-UD5 board to provide many years of service, as it will surely be the recipient of one of those future LGA 2011 processors.