Testing methodology
Comparison Memory |
||||
|---|---|---|---|---|
| Crucial Ballistx 2,400 |
Crucial Ballistix Elite 2,667
|
Corsair Vengeance LPX 2,133 |
G.Skill Ripjaws 4 3,000 | |
| Model | BLS4C4G4D240FSA |
BLE4C4G4D26AFEA
|
CMK16GX4M4A2133C13
|
F4-3000C15Q-16GRR |
| Capacity | 16GB (4x4GB) |
16GB (4x4GB) |
16GB (4x4GB) |
16GB (4x4GB) |
| Speed | 2,400MHz |
2,666MHz |
2,133MHz |
3,000MHz |
| Timings | 16-16-16-39-2T |
16-17-17-35-2T |
13-15-15-28-2T |
15-15-15-35-2T |
| Voltage | 1.20V |
1.20V |
1.20V |
1.35V |
| Price* | £200* |
£265* |
£250* |
£350* |
| Cost per GB* | £12.5 |
£16.56 |
£15.62 |
£21.88 |
| *Approximate, correct at time of writing | ||||
Test Platform |
||||
| CPU | Intel Core i7-5960X @ 4.0GHz | |||
| Motherboard | Asus X99-Deluxe (1103 BIOS) | |||
| Storage Device | Crucial MX100 512GB SSD | |||
| Graphics Card | EVGA GeForce GTX 780 Ti (344.75 drivers) | |||
| Power Supply | Corsair AX760i | |||
| Operating System | Windows 8.1 64-bit | |||
Benchmarks |
||||
| HEXUS.PiFast | Our number-crunching benchmark stresses a single core by calculating Pi to 10m places | |||
| CINEBENCH R15 | Using Cinebench's multi-CPU render, this cross-platform benchmark stresses all cores | |||
| HandBrake 0.9.9.1 | Free-to-use video encoder that stresses all CPU cores (64-bit) | |||
| AIDA64 v5.00.3308 beta | Memory analysis tool supporting Haswell CPUs | |||
| 3DMark | DX11, Fire Strike default test | |||
| Middle-earth: Shadow of Mordor | DX11, 1,920x1,080, 2,560x1,440, and 3,840x2,160, very high quality | |||
| Tomb Raider | DX11, 1,920x1,080, 2,560x1,440, and 3,840x2,160, ultra quality | |||
Notes
Comparing memory modules of differing speeds on an X99 platform is not as straightforward as it should be.
Motherboards tend to offer a maximum memory frequency of 2,666MHz when using the default 100MHz base clock. This is fine for memory like the Crucial 2,666MHz pack, but the only method of achieving higher speeds, for comparison modules, even if loftier frequencies are available in the BIOS, is to increase the base clock by inputting XMP parameters.
We want to lock the CPU speed to a a certain frequency and then see what effect memory timings and frequency have on our benchmarks, rather than what effect the CPU imposes. The way to do this is to select a CPU speed - 4.0GHz in our case - and toggle the other parameters to fit.
The Asus X99-Deluxe, updated to the latest BIOS, is able to run 2,133MHz and 2,400MHz and 2,666MHz speeds with the CPU set to 4.0GHz (40x100MHz, synced cores). We also want to see if investing in higher-speed DDR4 is worth it, so we use the 3,000MHz-rated G.Skill F4-3000C15Q pack. Obtaining the correct memory frequency via XMP 2.0 requires the motherboard to set the base clock to 125MHz and all-core CPU multiplier to a capped 32x, resulting in the same 4.0GHz final speed and 3.0GHz cache frequency.
Gaming benchmarks now use two games - Middle-earth: Shadow of Mordor, and Tomb Raider - that span three resolutions. The premise here is to see if memory-bandwidth gains translate well when the onus shifts on to the GeForce GTX 780 Ti graphics card.
We've also thrown in numbers from a Corsair DDR4-2,133 quad-channel pack that's imbued with tighter timings. Is spending more money on memory worth it on a very efficient quad-channel architecture present on X99? Let's find out if this is the case when operating the Core i7-5960X CPU at 4.0GHz. Benchmarks are run three times and the results graphed up on the following pages.
Overclocking
We've chosen three arbitrary speeds and timings in order to evaluate the overclocking potential of the Crucial modules. Voltage was increased to 1.35V.
| 2,666MHz (15-15-15-35-2T) |
2,800MHz (15-15-15-35-2T) |
3,000MHz (16-16-16-38-2T) |
|
|---|---|---|---|
| Crucial @ 1.35V |
 |
|
 |
The results are very similar to those obtained on the cheaper Ballistix Sport modules we took a look at recently. However, testing further shows that this 2,666MHz pack is able to lower the CAS latency rating to 12 cycles at 2,666MHz whereas the Sport manages 14 cycles.