Kingston HyperX LoVo: If I want lightning fast, if I want frugal
Our test configuration included the following elements:
- Motherboard:
- GIGABYTE GA-H55N-USB3
- Processor:
- Intel Core i7 860 2,80 GHz (133 × 21)
- Processor cooler:
- Scythe Grand Kama Cross
- Memory:
- Kingston HyperX LoVo KHX1866C9D3LK2 / 4GX 2 × 2 GB
- Hard disks:
- HITACHI 160 GB SATA2 (HDS721616PLA380)
- Samsung 200 GB SATA2 (SP2004C)
- Video card:
- Gainward 8400 GS 256MB DDR2
- Power supply: Xigmatek NRP-HC1501 1500 W
- Software environment:
- Windows 7 RTM 64 bit Ultimate HUN
- Intel INF 9.1.1.1019
- Realtek HD Audio 2.49 Driver
- Display: ASUS 24T1 TV Monitor
We had a lot of trouble compiling the test configuration, as we definitely wanted to compile an Intel, but not an X58 system. Only one motherboard could be considered with such criteria, the GIGABYTE GA-H55N-USB3, which is still our guest. There wouldn't have been any problems with it, but its extremely small size - which is an advantage anyway - was a disadvantage in our case, as we didn't have a factory LGA 1156 processor cooler. The space reserved for cooling on this motherboard is very small, and we could only try a set of battery, skyscraper-sized proci coolers. Because of this, inserting a full-width video card became virtually impossible (the time came for the low-profile GeForce 8400 GS dusting in its unopened box), and inserting a high-ribbed counter-foot memory and drifting into the forgetful category. Since we only had this in our editorial office, LoVo eventually had to deal with it on its own, but we still tried to put together an interesting and instructive test.
Some photos of how our test machine assembled this way was painted, the sight speaks for itself:
Setting variations for measurements:
- 1333 MHz: By all default settings, without using an XMP profile: As you can see in the pictures, nothing has been pinched, both the CPU clock and the base clock (BCLK) at the factory 133 MHz, and the LoVos at automatically set by the board at 1333 MHz, with timings of 9-9-9-24, operating at 1,5V.
1600 MHz: Using an XMP2 profile: In the second step, we targeted 1600 MHz, which requires the use of a second XMP profile. The setting does not change the clock signal of either the CPU or the BCLK, but the operating voltage of the modules drops from 1,5 V to 1,25 V, the clock signal to 1600 MHz and the timings remain - in principle. According to the official source, 9-9-9-24 also applies to this profile, yet, two test programs also indicated 10-9-9-24, plus with CR2, so we believed them. There is certainly room for correction from the BIOS, but we left everything as set by the XMP2 profile. If we look at the transmission speeds measured at 1333 MHz and 1600 MHz, we cannot record an improvement at 1600 MHz, but a weakening, which is most likely due to the high latency and command rate.
1866 MHz: Using the XMP1 profile: XMP1 is the faster profile, for which the CPU clock is just wide, but the BLCK changes significantly from 133 to 156 MHz, which can already be attributed to factory tuning, but is required to achieve the desired RAM clock. Here, all right, there were the 1866 MHz and 9-9-9-27 timings, the 1,35 V operating voltage specified by the manufacturer. No stability problems were observed during the use of the XMP1 profile, as in all other modes. At transmission speeds, we could already see the effect of this, the values increased nicely, except for writing, which still proved to be the fastest at 1333 MHz with CR1.
Using the 2244 MHz: XMP1 profile, tuning: We thought we wouldn’t let the LoVo set home without tuning in a bit. Using the XMP1 profile, we further exercised the clock signal from 1866 MHz upwards by increasing the BCLK to 187 MHz, while the CPU clock signal was intentionally kept at 2800 MHz. The timings remained 9-9-9-27 and the CR 2, but the operating voltage was increased to the still safe 1,66 V in the BIOS, manually. The result was a tweak of 2244 MHz, which proved to be stable in our short testing, which, however, does not automatically mean that it will run smoothly in the long run. What is certain, on the other hand, is that a more knowledgeable individual could make this or even higher clocks stable, as we only pushed the BCLK up with instant tuning without fine-tuning the smaller settings. This clock, on the other hand, proved to be the fastest in all areas at transfer speeds, copying over 20000 MB / s, which we think is a very good figure in two-channel mode.
Measurements were performed with the settings described above in the following test programs, the results of which are shown in the table:
- Super PI mod 1.5 XS
- WinRAR 3.92 x64
- Cinebench 11.5 x64
- Fritz Chess Benchmark
- Pov-Ray 3.7 beta 38 x64
- Lavalys Everest 5.50 beta
LoVo test | 1333 MHz CL9 | 1600 MHz CL10 | 1866 MHz CL9 | 2244 MHz CL9 |
---|---|---|---|---|
SuperPI 1M | 21,341 | 14,804 | 14,711 | 14,742 |
SuperPI 32M | 1609,174 | 793,371 | 776,722 | 781,587 |
WinRAR | 3029 | 3149 | 3446 | 3720 |
Cinebench | 4,13 | 4,75 | 4,82 | 4,84 |
Fritz B. multiplier | 18,09 | 21,48 | 21,62 | 21,71 |
Fritz B. point | 8684 | 10311 | 10379 | 10420 |
Pov-Ray | 3265 | 3755 | 3768 | 3773 |
E. Queen | 32053 | 32040 | 32136 | 32072 |
E. PhotoWorxx | 33540 | 34103 | 36013 | 37559 |
E. Julia | 11716 | 11714 | 11752 | 11750 |
Let's summarize in a few words what we see in the table! At 1333 MHz, the Super PI ran terribly slowly in both cases, producing a much lower time than expected, we don't know the reason for this. For the other settings, the slowest of the three, i.e. the 1600 MHz profile, was the third, which of course did not cause any surprise. It is all the more interesting that with the same delay values, the program ran faster at 1866 MHz than at 2244 MHz in both cases. In WinRAR, Cinebench, or something in the Fritz Benchmark, the world is restored, and the higher clock speed always wins, regardless of the delay. Then the picture becomes more nuanced again in the Everest measurements: under Queen, 1333 MHz is faster than 1600, while the 1866 setting beats 2244 MHz, interesting. Under PhotoWorxx, the slogan "time above all else" prevails, and in the case of the Julia measurement, the formula set up at Queen can be used again.
Consumption measurements:
Unfortunately, we in the editorial office do not have reliable and accurate tools to make a reliable measurement of consumption, so we describe Kingston's laboratory test:
Test environment:
- Modul P/N: KHX1866C9D3LK2/4GX
- Used profile DDR3-1600 CL9-9-9 @ 1.25V (XMP Profile 2)
- Motherboard: Asus P7P55D Deluxe / AVL SN: SI7906;
- CPU: Intel Lynnfield 860S, 2,53 GHz
- Measuring instrument: Fluke Hydra Data Logger 2625A
- Diagnostics: MemTest86 + v4.0
Load test:
Engineers used special add-on cards to measure the consumption, current, and voltage of each module. With the help of the current and voltage values read out in synchronism, the consumption can be plotted.
Note: Consumption data and temperature values are for reference only, as even this test environment and these devices do not give a completely accurate end result during the measurement.
The temperature sensors were mounted on the surface of the heat sinks of the modules
LoVo memories ran at 1600 MHz from 1,25 V to 1,85 V operating voltage
A specific measurement phase
The maximum idle and average idle values are the same
Kingston Technology measurements
We would have the tests, the evaluation could come!