Just like a few weeks ago we launched an overclocking guide on AMD Ryzen (socket AM4). This time, I was not going to do less with an Intel X299 Overclocking Guide for the most enthusiastic platform that Intel has released to date. Are you ready to reach 4.8 ~ 5 Ghz? (I.e.Let’s start!
Index of contents
- 1 Intel X299 Overclocking Guide | The “Silicon Lottery”
- 2 What do we need before we start?
- 3 Terminology
- 4 First steps of overclocking
- 5 Stability test
- 5.1 What to do if the equipment is not stable
- 5.2 What to do if the equipment is stable
- 6 We keep climbing
- 7 Advanced Overclocking
- 8 Final steps
Intel X299 Overclocking Guide | The “Silicon Lottery”
A first point that we must take into account when overclocking any processor, is that no two processors are exactly alike, even if they are the same model. The processors are made from thin wafers of silicon, and with manufacturing processes like the current Intel, at 14nm, the transistors are about 70 atoms wide. Therefore, any minimum impurity in the material can drastically worsen the behavior of the chip .
Manufacturers have long been taking advantage of these failed models, using them at lower frequencies, or by disabling some of the worst-performing cores to sell it as a lower processor. For example, AMD manufactures all its Ryzen starting from the same DIE , and Intel on the high-end socket (HEDT) usually does the same.
But it is that even in the same model there are variations, for this same reason. A processor that has come out almost perfect from the process will reach 5 Ghz with very little extra voltage , while one of the “bad guys” will barely raise 200mhz of its base frequency without temperatures soaring. For this reason, it is not useful to look for an overclocking and what voltage is necessary on the internet , since your processor is not the same (not even the same “batch” or BATCH) as the user who is publishing their results.
The optimum overclocking for each chip is obtained by raising the frequency bit by bit , and searching for the lowest possible voltage at each step.
What do we need before we start?
You have to follow these four essential points before entering the world of overclocking:
- To lose the fear of the hangings and blue screenshots . Let’s see a few. And nothing happens.
- Update the motherboard BIOS to the latest available version .
- Clean our cooling, fans and radiators , changing the thermal paste if necessary .
- Download Prime95 , to test stability, and HWInfo64 , to monitor temperatures.
In this guide we will limit ourselves to modifying simple parameters, and we will try to simplify the steps as much as possible. However we will briefly explain some concepts, which will help us understand what we are doing.
- Multiplier / Multiplier / CPU Ratio : It is the ratio between the clock frequency of the processor and that of an external clock (usually the bus clock or BCLK). This means that for each cycle of the bus to which the processor is connected, the processor has made as many cycles as the multiplier has. As its name indicates, multiplying the speed of the BCLK (of series 100Mhz in this platform, and in all the recent ones of intel) by the multiplier gives us the working frequency of the processor.
That is, if we put a multiplier of 40 for all cores, our processor will run at 100 x 40 = 4,000 Mhz = 4Ghz. If we put a multiplier of 41 on the same processor will work at 100 x 41 = 4.100 Mhz = 4.1Ghz,
- BCLK or Base clock: It is the clock to which all the buses of the chipset, the cores of the processor, the memory controller, the buses SATA and PCIE … Unlike the main bus of previous generations, it is not possible to increase it beyond a few Few MHz without having problems, so the usual thing is to keep it to the 100Mhz that it uses of series and to do overclock using only the multiplier.
- CPU Voltage or Core Voltage: Refers to the voltage that receives the processor core as power. It is probably the value that has the most impact on the stability of the equipment, and is a necessary evil. The more voltage, more consumption and heat we will have in the processor, and with an exponential increase (compared to frequency, which is a linear increase that does not worsen efficiency alone). However, when we force components above the frequencies specified by the manufacturer, we often have no choice but to slightly increase the voltage to eliminate the faults we would have if we only increased the frequency . The more we can lower our voltage, both stock and overclocking, the better.
- Offset Voltage: Traditionally a fixed voltage value was set for the processor, but this has the big disadvantage that even without doing anything, the processor is consuming more than necessary (away from its TDP, but wasting a lot of energy as well) . The offset is a value that adds (or subtracts, if we seek to reduce consumption) to the serial voltage of the processor ( VID ) at all times, so that the voltage continues to fall when the processor is at rest, and at full load we have the Voltage we need. By the way, the VID of each unit of the same processor, is different .
- Adaptive Voltage: Same as above, but in this case instead of adding the same value at all times, you have two offset values, one for when the processor is at rest, and another when the turbo boost is active. It allows to slightly improve the standby consumption of overclocked equipment, but it is also more complicated to adjust, since it requires a lot of trial and error, and the rest values are more difficult to test than the turbocharged ones, since with Low load even an unstable system has little chance of failure.
First steps of overclocking
These processors feature a slightly improved version of the Turbo Boost 3.0 Technology that debuted in Haswell-E. This means that when there are two cores or less in use, tasks are assigned to the cores that the board identifies as better (since not all silicon is just as perfect, and some could support higher frequencies) and the turbo frequency Boost is raised to a much higher value than usual . In the case of the Intel Core i9-7900X , this Boost for two cores is 4.5Ghz.
Before we start, let’s comment on the equipment we used:
- Corsair Obsidian 900D.
- Intel Core i9-7900X.
- Asus Strix X299-E ROG.
- 16 GB of DDR4 memory.
- Nvidia GTX 1080 Ti with liquid cooling block.
- Custom liquid cooler for heatsink.
- 512 GB Samsung 950 PRO SSD Disk.
The first adjustment that we will make in the BIOS is to select the XMP mode to be able to modify values (manual also serves us, but with this our RAM will be configured to its frequencies, voltage and latencies automatically)
If we have chosen Manual mode, we check that the voltage and frequency of our RAM is that indicated by the manufacturer. In our case, 2666Mhz and 1.2V. We should also check for latencies.
A first test that is interesting to do, is to see if we have been lucky and all cores hold this frequency reserved only the two best, with the voltage of stock. In our i9-7900X (we remember that each unit is different) this has been so. For this we will change the following parameters.
- CPU Core Ratio: From Auto to Sync All Colors
- Core Ratio Limit (which in this mode is the only option enabled): 45 (43 in the case of the 7800X and 7820X)
We see that with this also has increased the frequency for instructions AVX and AVX-512 to 4.5Ghz. These instructions are especially demanding for the processor, and it is very likely that very few units can make the 4.5Ghz with the voltage of stock when these instructions are a considerable part of the work. Luckily we have an option to lower the frequencies when these instructions are executed, so for this point do not bother us, we will lower the multiplier 5 points when these instructions are active. That is, we will change the following:
- AVX Instruction Core Ratio Negative Offset: 5
- AVX-512 Instruction Core Ratio Negative Offset: 5
With this we have (45-5) * 100) = 4000Mhz (4Ghz) frequency with AVX instructions, which is the normal behavior of stock. When we raise the processor to 4.6Ghz, the frequency in AVX will be (46-5) * 100) = 4100Mhz.
The task should already be finished. But we must do more. We are going to change the voltage to offset (or adaptive, if you prefer to configure also the voltage without boost), since we do not want the plate to raise automatically if it is not stable. By the way, this feature will facilitate the following steps.
As we do not want to increase the voltage, but only change the mode of operation, we put as offset the minimum value that allows the plate, which will normally be 0.001V, a very low value that we can despise (it is approximately one thousandth part of the total). If we have chosen offset, simply change a value, which is the voltage added to the processor at all times. If we have chosen adaptive voltage, we will have two values, one to add at all times, and another to add only in turbo mode, and we must change the two, from Auto to 0.001V.
We will also set up a curve for our fans. Depending on the cooling we have, we will choose a higher curve (better for overclocking, especially heavier tests) or a quieter curve. In the case of Asus, we will see it this way:
In short, we want to find the highest possible frequency with the lowest possible voltage, and that the cooling is not an impediment . For stability tests we are not concerned about noise, then we can use a quieter profile.
We have increased the frequency of our processor (or lowered the voltage, later), it is the time to test if the equipment is stable, or on the contrary we must increase the voltage. These tests will be carried out with short tests, of approximately 10 minutes , to lose the minimum time and to change values of agile form. This process will be repeated by raising the voltage until the test passes, always watching that the temperatures do not go to very high values. If we remember the previous step, we have chosen a lower frequency (500 MHz in our case) when the processor executes AVX and AVX-512 instructions , so we must check the stability with and without them .
You should know that we are not responsible for misuse when overclocking . All you do is under your responsibility . We do not recommend to exceed 90ºC for each core (without opening the CPU), although its delid the processor to lower temperatures drastically (you will lose the warranty with it), there are stores that already sell it with this service Or even there Users who do it in different forums.
A complete stability test should last for several hours, but we will only do a test of this type at the end of the process, since it would be very slow to invest 8 hours to test every 100mhz. You can also replace this test by using the PC in a normal way, and if any application fails or the PC hangs assume that it is the fault of the overclock and raise the voltage. We do not recommend this method as the equipment could hang at a bad time and lead to data loss, but it is a valid and less aggressive way with the components.
We are going to use Prime95 to test stability, being free, open source and very flexible . There are a variety of programs, if you feel more comfortable with things like IntelBurnTest, Lynx, or OCCT can be used for stability testing without problem, following the guide in the rest of the steps. It is recommended that the same program allows you to enable or disable the AVX instructions, just do not work double .
Since in this socket our processor will probably have more than 6 cores, we will leave a single window for all threads, or we will not be able to read the results of that the main window will be. This is done only for convenience (can be omitted without problem) and just do it once.
To start the test, we will select the next option
In the window that appears we marked the option Blend, which tests in depth both the processor and the RAM , and often detect faults quickly. The number of threads should be correctly marked, in processors with HT twice the physical colors of our processor (in the capture we see 20, for a 7900X)
Nothing to accept is likely to accelerate the fans . It is normal, since the heat generated in this test is superior to any realistic load. The test will run until it finds an error or until we stop it . We can stop it with the next option
We recommend passing the test for approximately 10 minutes without AVX instructions , or better, 5 minutes with them and 5 minutes without them. To disable the use of AVX instructions in Prime95 we are going to edit the local.txt configuration file . We will add the following line at the end (the rest of the parameters do not have to be edited)
CpuSupportsAvx = 0
To re-activate the AVX instructions we can either delete the line , or write
CpuSupportsAvx = 1
It is not necessary to close and open prime for the change to take effect. Simply save the file, stop the test and start over.
Throughout the process we must have an eye on processor temperatures . Since these processors use thermal adhesive instead of welding , it is practically impossible to overclock without passing the Tcase no matter how good the cooling (we have taken it to the limit). We will try to reach the highest possible overclocking without reaching the throtling , ie, at most 90-95 ° C in these unreal tests , and no more than 60- 65ºC in normal use . HWInfo gives us direct information on whether any kernel is in throtling (should be “No” in all cases):
If some kernel is in throtling , it means that we are at the limit of what our processor allows with our cooling, and we should not keep going up . In fact, we will not gain performance on heavy loads, because in throtling it will go down to a safe frequency, let’s say the multiplier we put .
We also have information about the voltage and frequency of the processor in the upper display (it is normal that the numbers are not accurate, normally by the options of reduction of EMI interference), for example, for 4.7Ghz in our case we see 4699.4 MHz , we expect 4700.0, The difference is negligible. If anyone is bothered by this rounding, you can upload the BCLK to 100.1MHz, we have not bothered about it.
Here we take a look at an interesting point on this screen: Here we see which are the most vague colors, that is, the ones that require the most voltage according to the detection of the plate (in our case, core number 0 with 1.302 V), which would be the Candidates to lower the multiplier point if we want to improve consumption and temperatures. We also see which are the best colors, in our case # 2, # 6, # 7 and # 9, which will be the candidates to climb higher.
What if your computer is not stable?
The symptoms of unstable equipment are clear and indisputable . If the equipment fails in 10 minutes (or 1), no further testing is required. This failure can be seen in several ways, ordered from less significant (probably too little overclocking to be stable) to more serious (there is a long way to make the overclock stable):
- P rime95 does not hang nor the computer, but some test fails and is displayed in the Prime95 text boxes
- It hangs prime95 (most common) or some other program that is running in the background , but the operating system still works.
- It hangs the whole pc, either being frozen, with a blue screen, or with a sudden restart/shutdown.
In any of these cases, we will slightly increase the offset, with small steps, around 0.01V more each time, and re-test. We will stop climbing when temperatures rise too high (more than 90º in extreme tests) or when the voltage approaches dangerous levels. With air cooling, we should not go from 1.3V for all colors, 1.35 maximum with liquid. We can see the total value of voltage with HWInfo, since the offset is only what is added and not the final value.
What if your computer is stable?
In case our system is more or less stable, we will stop it after approximately 10 minutes with the option we have seen above. We say “more or less” since in 10 minutes we can not know for sure. After stopping the tests, we will see how the following, with all the workers (the blocks of work that run in each nucleus) finalizing correctly. We look at the part squared, all the tests must have finished with 0 errors/0 warnings. The number of tests that have been completed may vary, because the processor is doing other things while running prime95, and some colors may have had more free time than others.
This is the ideal case, since it means that we have multiplier and offset settings that we can test with a longer stability test, and that improve the performance of the processor. For the moment, if our temperatures are not high, we point them and keep raising the frequency, in the following section, to return to the last stable value when we reach a point where we can not go up.
We keep going up
In case of a fast test like the previous ones have been stable and our temperatures are in acceptable values, the logical thing is to continue raising the frequencies. For this, we increase another point the multiplier, up to 46 in our 7900X:
As the previous stability test has passed without raising the voltage (we remember that each processor is different, and may not be the case in your particular processor), we maintain the same offset. At this point we return to pass the stability tests. If it is not stable, we raise the offset slightly, from 0.01V to 0.01V (other steps can be used, but the smaller, the better we will adjust). When stable, we keep going up:
We pass the stability tests again. In our case we need an offset of + 0.010V for this test, being as follows:
After leaving it stable, we raise the multiplier again, to 48:
This time we needed an offset of + 0.025V to pass the stability test successfully.
This configuration has been the highest we could keep with our processor. In the next step, we have raised the multiplier to 49, but we have increased the offset, it was not stable. In our case we stopped at + 0.050V offset, as we were approaching dangerously 1.4V and almost 100ºC in the more vague colors, too much for it to make sense to keep going up, and more in an overclock thinking 24/7.
We take advantage of the fact that we have touched down our microprocessor to test with lower offset values for AVX instructions, down from 5 to 3. The final frequency for all colors is 4.8Ghz and 4.5Ghz in AVX, which means an increase of approximately 20% compared to the stock frequencies. The necessary offset, again in our unit, has been + 0.025V.
In this section we will test the possibilities of overclocking by core, keeping the Turbo Boost 3.0 technology active and trying to scratch 100-200mhz in the two best colors without increasing the voltage. We say advanced overclocking because we multiply the possible tests, and there is much more trial and error time. These steps are not essential, and at best only provide improvements in applications that use few colors.
We will not discuss the voltage increase in other parameters relative to the memory controller or the BCLK, since usually the limiting will be the temperatures before reaching frequencies that make it necessary to touch nothing more, and the competition overclocking with extreme cooling is outside Of the Scope of this guide. In addition, the overclocker der8auer mentioned, the phases of a medium/high-end board of this socket may be insufficient for the consumption of a 7900x i9 (or even its younger siblings) raised well above its stock frequency.
First of all, it is interesting to comment on one of the advantages of this boost 3.0 technology, because the board detects the best cores automatically , that is, the ones that need less voltage and apparently will be able to increase their frequency. We note that this detection may be correct or not, and that on our board we can force the use of other colors, and choose the voltage for each. In our processor the board tells us, as we anticipated when viewing information from HWInfo, that the best colors are # 2, # 6, # 7 and # 9.
We can corroborate this choice in the program Intel Turbo Boost Max Technology 3.0 Application, which will be installed automatically through windows update, and is minimized in the taskbar, since these colors will be the first, and will be the ones that are They will send The tasks that are not paralelized when possible.
In our case it seems logical to try to raise the two best colors to 4.9Ghz in the first place , 100mhz more than all colors. To do this, we changed the CPU Core Ratio option from XMP to By Core Usage. Then the values appear Turbo Ratio Limit # , which allow us to choose the multiplier for the fastest core (0 for faster, 1 for the second fastest, etc.), and the option Turbo Ratio Colors # , we Allows you to choose which core You want to upload, or leave it in Auto, in such a way that the board will use the detection that we have seen in the previous step to determine which are the fastest colors
For this we set the values of Turbo Ratio Limit 0/1 to 49, which will make the two colors faster to 4.9Ghz. The remaining Turbo Ratio values are left at 48, since we know that all other colors work well at 4.8Ghz.
The way to test stability is the same, although now we must be careful to throw only 1 or 2 test threads, because if we put more the processor will work at the usual turbo frequency. For this we only choose one thread in the screen that we already know about Prime95 :
It is convenient to check in the task manager that the work is being assigned to the correct colors (we count 2 graphs per core, since with hyperthreading every 2 threads are a physical core , and in Windows they are ordered together), as well as The frequency Is what we expect in HWInfo64 . Below we can see the # 6 kernel at full load, and how the frequency is at 5Ghz.
Personally I have not had much success using the previous method , even with a little extra voltage , although each processor is different and may be different for another person. The result that is seen in the previous capture was achieved using the manual option, with which we have been able to upload a pair of colors up to 5Ghz . With this mode we can choose the voltage and multiplier for each core, so we can give a high voltage, around 1.35V, to the higher nuclei, without unduly worsening the TDP or uncontrol our temperatures. Go for it:
First we choose the By Specific Core option
We must open it. In this new screen, putting all values of Core-N Max Ratio to 48 with the rest in Auto would leave us in the previous steps, at 4.8Ghz all colors. We will do that, except in two of the best cores (7 and 9, marked with * on the plate, and two of the four that we had identified as better), that we will test with 50 51, but this value Did not work correctly)
As a suggestion, although the voltage in Manual Mode is faster to adjust to the value that we want , it would be more correct to do the same with Offset, testing until obtaining the desired VID.
The gain in tasks that only use a kernel is notorious. As a quick example, we have passed the popular Super Pi 2M benchmark, getting a 4% improvement in test time, which is expected to increase in frequency (5/4.8 * 100 = 4.16%) .
Once we have found a configuration that convinces us, it is time to test it thoroughly, since it should not only seem stable 10 minutes, it should be over several hours . In general this configuration will be the one immediately preceding the one we were touching, but in some processors we will have to lower 100mhz more if we do not get it to be stable. Our candidate is 4.8Ghz at + 0.025V Offset .
The process to follow is the same as in the tests of stability that we have done, but now we must leave it several hours. From here we recommend about 8 hours of Prime95 to consider a stable overclock . Although I personally have not noticed temperature problems in the phases of the Asus X299-E Gaming board , it is advisable to make small breaks of about 5 minutes every hour , so that the components can be cooled.
If we can measure the temperature of the phases, we can skip this step. In our case we see That, after 1 hour of prime, the sink is About 51 ° C. If we do not have an infrared thermometer, we can gently touch the top heatsink of the motherboard. The maximum temperature that can be endured without removing the hand by the hairs, is of about 55-60ºC for a normal person. So if the heatsink burns but can stand, we’re on the right margins.
The screen we want to see is the same as before, all the workers stopping, with 0 warnings and 0 errors. In our case we had an error after 1 hour of testing, so we raised the offset slightly, up to + 0.03V, which is the minimum that allowed us to finish the test correctly.
What did you think of our overclocking guide for the LGA 2066 socket and X299 motherboards? What have you been overclocking with this platform? We want to know your opinion