Review

ROG Overclocking Guide For Core i7 5960X, 5930K & 5820K

SOURCE : rog.asus.com

The big question for many enthusiasts contemplating the X99 upgrade is, how well does Haswell-E overclock compared to previous gen processors? We'll be answering this question together with how to overclock the processor in this guide.

Frequency Expectations

In our testing to date, the average overclocked frequency for 5960X processors is 4.5GHz. Very good processors will achieve 4.6GHz fully stable with less than 1.30Vcore. Lesser samples achieve 4.4GHz with the same voltage:

Overall for the 5960X we're looking at around 100MHz lower than the (Ivy-E) 4960X, but with two extra cores in tow, which is more than respectable.

For those of you wondering about the K parts. They are easier to OC on air and water due to having fewer cores, thus less heat to contend with and higher voltages are possible. The end result is the possibility of overclocking the 6-core K series CPUs 100~200MHz higher than the 8-core 5960X.

Cooling Requirements

Depending on your ambient temperatures, full-load voltages over 1.25Vcore fall into water-cooling territory (dual-radiator). With triple radiator water-cooling solutions, using up to 1.35Vcore is possible. For air cooling the value is much lower, limiting total overclock, so plan your cooling investment appropriately.

PSU Requirements

For overclocking 5960X processors, we recommend PSUs that can supply a minimum of 30 amps to EPS 12V. At 4.6GHz a 5960X can draw close to 25amps from the EPS12V connector under software load. Minimum recommended PSUs for Haswell-E are upwards of 1,000W if using more than one high performance GPU.

VRM Heatsink Cooling

If running full-load stress tests over 4GHz, active cooling of the VRM heatsink with a fan is advised due to current requirements of the (8-core) 5960X processor. K series (6-core) at ~4.2/4.3GHz and above should also have active cooling, depending on voltage.

Stress Testing

Users should avoid running Prime95 small FFTs on 5960X CPUs when overclocked. Over 4.4GHz, the Prime software pulls 400W of power through the CPU. It is possible this can cause internal degradation of processor components.

We use the ROG Realbench stress test, with the corresponding amount of installed memory selected in app. The stress test is equal or greater to alternatives, using real, open source apps with an oscillating load across the main PC subsystems.

DDR4 Memory

At first glance, the primary timings of DDR4 seem underwhelming. If we look deeper, a shift in specification voltage down into the 1.05V~1.20V region helps make sense of the “looser" primary timings initial DDR4 ICs support at stock. Consider running enthusiast oriented DDR3 kits at similar voltage levels and it would not be surprising to see similar timing ranges, and constrained data throughput rates.

The good news is that most of the DDR4 ICs we've tested to date scale quite well in terms of frequency and timings as we increase voltage. At worst, some ICs stop scaling past 1.40V. At best, we've pushed up to 1.50V (up to 1.80V with LN2) into good Hynix-based modules and seen acceptable returns. Primary timings are still a tad looser than one would like to use, however, there's more to memory performance on modern architectures than primary timings, so we encourage a deeper learning

We’ve used 3GHz G.Skill Ripjaws 4 in the XMP screenshots below

Third Tier Memory Timings Are Key

For a few generations now, the third tier of memory timings – read to read, read to write, write to write and write to read are larger factors in memory performance than the primary timing set. This is due to how efficient the memory controller is at organising data across memory ranks and banks.

Once a row is latched, the following transactions are almost exclusively back to back read and write operations – hence the impact of a 1 clock adjustment to an important “third tier" timing often has more performance impact in memory sensitive applications, than a 1 clock change to any of the primary set. In this regard, DDR4 fares better than DDR3 – in our testing to date, DDR4 retains a tighter set of third tier timings than was possible on DDR3 at similar frequencies and voltages. So in some applications we will see DDR4 win out over its DDR3 counterpart, in spite of the primary timing disadvantage.

Due to some nifty enhancements in its new architecture, DDR4 is also “more stable" than DDR3 at equivalent and higher operating frequencies. New termination schemes with lower simultaneous switching noise, timing concessions for different types bank access, on-die Vref generation and an improved pin-out for balance between power and ground pins all help improve signal integrity.

From the ROG perspective, we're expecting our expertise in trace layout benefit once again on DDR4. Feedback to date is that our motherboards overclock memory very well – improved frequency overhead means a more stable board for the end-user.

• The best all round DDR4 IC for overclocking in our testing to date (Sept 2014) is Hynix. Samsung is a close second. This will change regularly as new ICs and manufacturing processes are designed.
• Most processors will run DDR4-3000 with 32GB of memory at 15-15-15-34-2T easily (with capable memory modules – good Hynix!). 1T Command rate at these timings is also possible, if the memory controller is a good enough. Good memory modules will reach these speeds at 1.35V or less.
• 64GB of memory will run DDR4-2666 at 15-15-15-34 2T easily. 1T command rate is difficult to get “unconditionally" stable at these speeds.
• For speeds over DDR4-2666 try using 1.35~1.40VDIMM as a starting point.
• VCCSA helps stabilize the processor’s memory controller. For the most part the ROG UEFI BIOS left to ‘Auto’ will scale fine for most CPU/memory combinations. Some CPUs may need manual adjustment. and the maximum we have needed to use is 1.15V. However, some CPUs do not respond well to anything over 1.05V­; such processors usually overclock memory fine at stock settings – even though they prefer lower voltage. Start low (default is 0.80V approx.) and work upwards. The relationship between voltage increases to VCCSA and stability are non-linear. There are certain ranges of voltage which may exhibit worse stability, once you go past them (higher or lower) stability improves once again. Try 1.02V for memory speeds above DDR4-2900 as a starting point and work up or down from there.

DRAM Voltage and POST

When overclocking DRAM, one needs to bear in mind that during a first POST after full shutdown, or an AC power cycle, the memory controller performs driver calibration, de-skewing of signal lines and a memory test. The memory test uses a data pattern designed to create a large current swing on the DRAM bus. As the memory is overclocked, these operations are progressively more difficult to pass – tougher than some stress tests in fact. In comparison, a warm reset of the system may only require light retraining to account for temperature drift.

The caveat to tuning DRAM voltage is that while memory is stable during a warm reset at 1.30V, a system power down or power cycle may be another matter. Often the training process from a full system shutdown requires a higher DRAM voltage to pass. For this reason we have created an additional setting for DRAM voltage on our boards.

Within the Extreme Tweaker menu:

DRAM Voltage AB and DRAM Voltage CD.

These set the DRAM voltage level for the POST process.

Within the DRAM timing page there are two additional settings for DRAM voltage labelled Eventual DRAM voltage relating to each pair of channels respectively.

This allows setting a different voltage, which is applied after the POST process completes. By default the Eventual DRAM voltage will automatically sync with the DRAM Voltage setting (both are the same). If we enter a value here, then it will be applied after the POST process completes. The upshot of this allows us to set a higher DRAM voltage to help pass training from power cycle, while we can apply our “OS stable" voltage in the eventual setting so that we are not running more voltage than required.

The Uncore Story

Overclocking the cache frequency is not essential as processor core frequency dominates overall performance. However, if one wishes to experiment with cache overclocking – the cache frequency (AKA Uncore) can help boost some benchmark scores if run close to the processor core frequency.

Good processor samples can achieve 4.6GHz Uncore frequency in tandem with CPU core frequency at the same value. Depending upon the processor sample, this may require cache voltage in the region of 1.35~1.45V.

From a methodology standpoint, it is wise to leave cache frequency at default whilst determining the maximum processor core frequency. Once the maximum processor core frequency has been found, one can start overclocking the cache. Following this method helps to isolate possible causes of instability for troubleshooting purposes.

We may need to apply a small boost to Vcore and VCCSA when the cache frequency is overclocked. This is because a faster cache ratio will increase the amount of data over the bus thanks to faster L3 cache access times.

5-Way Optimization

The latest version of AI Suite and DIP 5 includes a few cool enhancements for 5-Way optimization.

Within the advanced options, we have added options to select AVX stress testing, fixed Vcore (one can set Vcore to any value from 1.1V~1.5V) and tuning from the processor's default ratio instead of starting from a higher ratio. The latter is useful in cases where the user only wishes to apply a mild overclock or has a CPU that does not overclock very well.

It is worthwhile running the 5-Way optimization to quickly gauge how well a CPU will overclock – even if you wish to overclock manually.

Manual Overclocking

1) Set the CPU multiplier ratio and strap (if needed). Aim for 4.5GHz with 1.30Vcore. If the processor sample is good it will make it to the desktop. If not, reduce the CPU frequency by 100MHz (with the same Vcore). Run a stress test of choice – we use the ROG Realbench stress test with the corresponding amount of installed memory selected in app. If 4.4GHz isn't stable with 1.30V, you will need to reduce the CPU core frequency further.

2) If the stress test passes, apply XMP for the memory kit and repeat the stress test again. For most CPU samples there should be no need to make any VCCSA, DRAM voltage or memory timing adjustments- the stress test should pass. If it does not, then try tuning VCCSA; start with a manual voltage of 1.02V and work up or down.

3) Assuming steps 1&2 go through, you may evaluate if there is room to push the processor further. Check load temperatures – I generally aim to keep load temps lower than 80 Celsius. This helps system stability under all software loads.

4) You may wish to experiment with memory overclocking at this stage. Leaving timings on Auto is advised to start with – our auto rules will adjust them to cater for most memory modules that can be overclocked. The primary timings can be tightened manually, after evaluating initial stability at our auto values first.

The maximum DRAM voltage we use here in the lab for 24/7 table systems is 1.40V. Some modules may not scale with more voltage anyway. Good Hynix can handle higher voltage, but long term effects of running such voltage levels are unknown at this point.

Running higher DRAM frequencies can impact processor core stability and/or require increased Vcore to ensure core stability.

5) If the CPU and memory are stable you may wish to try Uncore overclocking. Good CPU samples can achieve 4.6GHz Uncore with 1.35V Cache Voltage. Obviously if the CPU core frequency is only 4.4GHz then one should aim for a cache frequency of the same value – no need to set cache frequency higher than core frequency.

Vcore, DRAM voltage and VCCSA may need tuning to help stabilize the processor when the CPU, DRAM and Uncore are overclocked.

6) Enjoy yourself!

SOURCE : rog.asus.com

Comments