While testing the Intel Core i9-12900K with DDR4-3200 and DDR5-5200 memory not so long ago, we noted that DDR5 support is a major innovation for Alder Lake processors, but aimed mainly at the future. At the moment, its use provides a certain performance boost (uneven: in some applications none at all, but in some places even noticeable), but the use of DDR5 is not necessary. And then we considered the issue in relation to top processors, where, obviously, the effect should be maximum. And we also tried to exacerbate this effect by taking the cheap DDR4-3200 for comparison – it fully complies with the JEDEC specifications, but nothing more.

This caused some readers to be puzzled, so this point will have to be explained in more detail. In fact, from the point of view of the mass market, the change of standards looks like this: from the “basic” DDR4-3200 to the “basic” DDR5-4800. So it would be interesting to “probe” the last one, but not until it is up to it – there is a shortage of modules at all levels. One problem is moving the power controller into each module – useful in itself, but since this is a new component, there are difficulties with PCBs and others, no matter what price class is the goal. In general, there are practically no “cheap” DDR5 on the open market. But in finished computers, it will be exactly that. And the same goes for DDR4 – adjusted for the mass availability of solutions of any level, but large manufacturers always vote first of all for the price. Moreover, most motherboards in these systems basically do not support memory overclocking, despite the fact that Intel “allowed” it for almost all chipsets. Therefore, most users will face the transition from DDR4-3200 to DDR5-4800. Especially if you expand the subject area not only to desktops, but also to laptops – which are sold many times more. But with the fact that when assembling a computer on your own, it makes sense to focus on “non-standard” modules – both DDR4 and DDR5 – and (if desired and possible) overclock them a little, exceeding even the non-standard frequencies promised by manufacturers, we will not argue. In this formulation, everything is correct. But not guaranteed. Intel sells processors with support for DDR4-3200 and DDR5-4800 – which is where the company’s commitment ends. Everything that the user winds up above (if he is given such an opportunity, of course) is only his own business: both all possible achievements and (no less) possible problems. First of all, we needed high-quality assessments, and within the framework of the standard, we received them.

Now it’s time to move on – and in a slightly different direction. The main hero of testing will be the Core i5-12400. As we noted last time, there is nothing to do in a budget computer or mid-range DDR5 system yet . The problem is not technical – in fact, this memory can be used even with Celeron. It’s just that at the current price ratio there is no reasonable reason to do so. For Core i7 or Core i9, however, these are not particularly observed yet – but there we are still talking about an initially more expensive system, against which the difference in the price of memory modules is somewhat smeared. But with the younger Core i5, everything is very simple: the same 12400, together with a 32 GB DDR4 kit, costs about the same as 32 GB DDR5 by itself. The latter is more promising , and on some motherboards for DDR5, the processor itself can be slightly overclocked via the bus , but at such prices, all this does not matter. Especially the latter – after all, the i5-12400+DDR5 is approximately equal in price to the i7-12700+DDR4.

Despite this, we tested the Core i5-12400 by itself in conjunction with DDR5-4800 – although we honestly warned that such a configuration could not, in principle, qualify for mass use. But here everything is simple: either we do the possible right away, or we wait for the opportunity. Naturally, as usual in such cases, we preferred the former. However, from a practical point of view, we repeat, the performance of this processor paired with DDR4 is important. Ideally, exactly with the modules that were used to test processors for LGA1200 and AMD AM4. But not only with them, because a person who has chosen a Core i5-12400 is unlikely to buy DDR5, but he can very well play around with DDR4 modes. Retail boards based on chipsets starting from B660 allow this without exception, there are no noticeable problems, it is also easy to pick up well-overclocked or factory-overclocked modules.

This is how we assess the situation.

Test participants

Our main test subject will be the Core i5-12400, but with different memory. However, a couple of reference points for comparison are still needed, and their choice is very simple – as before, we will take the Core i5-11600K and Ryzen 5 5600X, but today we will limit ourselves to two of them.

We previously tested all processors (including the reference pair) with the G.Skill Trident Z RGB kit (two 8 GB modules) operating in DDR4-3200@CL14 mode. Today we had the opportunity to use it for the Core i5-12400 – and we took advantage of it. Adding a couple more DDR4 kits to it, but already with a capacity of 32 GB – already familiar from previous tests TeamGroup T-Create Classic and TeamGroup T-Force Xtreem ARGB White. The first is without any XMP profiles, but on the other hand, “honest” DDR4-3200 directly according to JEDEC and with a voltage of only 1.2 V. The reverse side of the coin is that it practically does not go higher, and the timings are very conservative: 22-22-22- 52. As we have said more than once, such kits are available not only in the assortment of Team, but also in all other major manufacturers, and this is the cheapest way to buy memory with official support for the DDR4-3200 mode. If you don’t chase officialdom, then you can chase in a different sense – there are cheaper DDR4-2666 kits that work normally at frequencies of 3400-3600 MHz. But this is a separate sport, and today we are not talking about it. Moreover, the DDR4-3600@CL15 mode will give us the second set. In fact, it is DDR4-4000@CL15, but this is only possible in Gear 2 controller mode (i.e. at half frequency), and it is considered slower. To what extent this is true – we will also check this. Why 3600? Because our copy of the Core i5-12400 on the younger crystal (stepping H0) is no longer able to in Gear 1. C0 stepping processors (full Alder Lake crystal, guaranteed to be found in Core i5-12600KF and higher) are more loyal to memory overclocking – with proper luck, you can see Gear 1 there at 4 GHz. However, we did not succeed – the Core i9-12900K at hand normally pulls only 3.8 GHz. Which is still more than the younger crystal – but not as much as some of the first victorious reports promised 🙂 In general, summing up, 3600@CL15 is something that you can at least count on for both the younger ones and for ( especially) older Alder Lake. Anything above – as lucky individually. Therefore, we will now limit ourselves to this mode. But, at the same time, let’s check whether it is worth going further in frequency with switching to Gear 2.

All of these tests were done on the Asus ROG Strix Z690-A Gaming Wi-Fi D4 board. We also have results on the Asus ROG Maximus Z690 Hero with Kingston Fury Beast Black in DDR5-4800@CL38 mode. We will also add them to the comparison. Ideally, one combo board would be suitable (that is, with both DDR4 and DDR5 slots at the same time), but there are none on the market. To be fair, there were a lot of universal models during the transition from DDR2 to DDR3, but at the next stage there are also few: you can immediately recall only the ASRock Combo line. But at least it was, and whether something similar will appear on a new round of progress is still unknown. If it appears, then there will be a reason to return to the question, but if not, then no. Now we just took two “good” motherboards based on the top chipset. They differ in body kit and functionality, but this is not surprising – it is clear that in today’s conditions, top-end and other expensive boards tend to use DDR5, and starting from the middle segment, DDR4 support is becoming more and more widespread. In the end, the boards for other platforms are all the more different – but this does not prevent us from comparing platforms. You just need to take into account all the possible nuances – but this is necessary in any case.

The rest of the environment is the same in all cases. The AMD Radeon Vega 56 graphics card and SATA SSD are the same for all systems and affect the results in the same way. Intel Multi-Core Enhance and AMD Precision Boost Overdrive technologies are disabled – for the latter, this is typical by default, but many motherboards strive to turn on the former quietly.

Test Methodology

The testing methodology is described in detail in a separate article, and the results of all tests are available in a separate table in Microsoft Excel format (in this case, completely separate, since this is a special test). Directly in the articles, we use the processed results: normalized with respect to the reference system (Intel Core i5-9600K with 16 GB of memory, AMD Radeon Vega 56 video card and SATA SSD) and grouped by computer application areas. Accordingly, on all diagrams related to applications, dimensionless points, so here “more is better” everywhere. And starting from this year, we are finally transferring game tests to an optional status (the reasons for which are discussed in detail in the description of the test methodology), so that only specialized materials will be available for them. We have already tested the Core i5-12600K and Core i9-12900K in gaming applications, and the time of the younger Core i5 will definitely come along with the Core i3.

iXBT Application Benchmark 2020

As already noted, these programs completely and completely load the processor with work, and everything else is a secondary background for them. Theoretically, such a situation is possible when the processor will “not have enough” of the data flow – in practice, this is impossible in modern conditions: even the “slow” memory has become too fast and even the fastest processors remain too slow. So the scatter of the results is somewhere below the measurement error. It exists, lends itself to a logical explanation – but, of course, there is simply no need to pay such attention to it.

A little more fun. Largely due to the fact that in these scenarios the Core i5-12400 is faster than in the previous ones. Therefore, the influence of the memory system is slightly more noticeable. But still – within the measurement error. Simply put, it is not significant in practice.

It’s even a little better with a lower base – the digestible data stream itself is higher, but it has nowhere to come from except from memory. And here you can already find a couple of percent of the difference, and between the extreme points – all 3%. But, again, such a discrepancy is interesting only for laboratory tests, but not for practical use.

We have already noted a noticeable (against the general background) advantage of DDR5 in these programs. Naturally, one can speak of a significant gain only in comparison with the slow DDR4. Let it be as fast as possible in terms of specifications – but no one canceled non-standard frequencies. And in principle, you can even achieve some kind of parity – since delays here just do not matter. They practically don’t – they always interfere to some extent, but the Gear 2 mode turns out to be quite justified for DDR4, and not just for DDR5. All in all, it’s not that scary. Or not always scary – at least.

Since it happens and vice versa – latency is critical. But there is also a very curious moment – paired with the Core i9-12900K, DDR5 significantly outperformed at least the slow versions of DDR4-3200, but when replacing the processor with the i5-12400, it loses even to them. Although this does not contradict the laws of physics in any way – 16 cores and 24 computation threads (albeit with different performance) require a higher ammunition supply rate than 6C / 12T operating at a lower frequency. And the L3 cache capacity ratio is also impressive – 30 and 18 MB. But it is the caches that are primarily designed to deal with the high latency of DRAM. All this together leads to the fact that DDR5 is not only more expensive than DDR4, but sometimes its use in conjunction with inexpensive processors does not make any practical sense at all. Even if it cost the same, it is also slower than DDR4.

Archivers are generally similar to FineReader (which we have seen many times – by the way, as well as the fact that the results of gaming applications correlate to the maximum extent with these programs), but they have even less “mathematics” – so the system is more important memory. And first of all – delays. Which is what was expected. And from the interesting point, the change in memory changes the ratio with processors for other platforms: if in the first test (where DDR5 was used) the Core i5-12400 lagged behind the Core i5-11600K and Ryzen 5 5600X, then with the same memory it still overtakes them.

We return to boring land again – the difference between all memory configurations is minimal. DDR5 formally turned out to be the fastest, in fact, the spread of 2% -3% can simply be ignored.

The overall result is clear. Switching to DDR5 can provide an average performance boost even for mid-range processors, but a lot depends on what you take as a base. Simply put, formally non-standard (but in practice easily and long achievable) DDR4 modes of operation may still turn out to be faster. Moreover, some applications are much more responsive to delays than to bandwidth – and with this, DDR4 modules are doing better than today’s DDR5. And this is aggravated by the Gear 2 mode – at such frequencies there is no alternative.

Energy consumption and energy efficiency

During tests, we measure not only performance, but the power consumption of the entire platform – the processor, the board, and everything that is powered directly from it, including memory. In particular, although video cards receive the bulk of their electricity directly from the power supply unit, it is also impossible to neglect the share of their consumption “out of the slot”, as tests with different discrete video cards have already shown when comparing them with each other and / or integrated video. But the main “bug” (especially if you equalize the rest of the environment to the maximum) at high loads, of course, is the processor itself. With minimal loads, in principle, you can also notice the participation of boards. Memory is involved in this issue mainly indirectly – by itself it consumes a penny, but if the speed of the processor can be limited, then a couple of watts can be saved on this. And vice versa, of course.

If we consider power consumption without interrupting performance, then the highest energy efficiency of the Core i5-12400 can be achieved in conjunction with DDR4 – and not necessarily high-frequency. Moreover, during the first testing of this processor, we noted that Intel managed to outperform AMD processors and even APUs based on the Zen2 microarchitecture in this regard – but not Zen3. It turned out that this is some premium for progressiveness (aka DDR5), because with the same memory with the same Ryzen 5 5600X, equality is already observed. However, against the background of confusing himself with the Core i5-11600K heater, this is all, of course, trifles.

Total

As expected, the use of DDR5 in combination with mid-range and low-end processors of the new Intel line is even less justified than with top ones, even if you don’t look at prices. A smaller number of cores and computation threads does not allow to fully dispose of the increased memory bandwidth, but problems with latency have not gone away. In some applications, however, memory access delays were decisive before. We have already said that in terms of their behavior, OCR and archivers correlate very well with games. This is especially true for archivers – text recognition programs can manage at least a large number of cores well, which neither gaming applications nor archivers do. However, in order to get an accurate answer to these questions, specialized testing in games should be deepened, since the situation in this market also changes regularly, but there is already information for thought.

At the same time (also – as one would expect ), it is often possible not to be limited to the “official” DDR4 operating modes. With self-assembly of the system, this can even be considered a given in modern conditions – after memory overclocking was “legalized” for almost the entire line of chipsets (of course, boards based on the H610 are also massively purchased at retail – but their customers usually know why they do it and what they are missing). Moreover, this does not require “elite” memory modules at all, especially if you continue to focus on the Gear 1 mode – taking into account what was said above about delays, it remains preferable. The frequency ceiling of Gear 1 for the “older” modification of the crystal (C0 stepping) increased slightly compared to Rocket Lake, but this is not observed for Alder Lake-6C (H0 stepping), so the target frequencies can be considered the same. And if you do not overclock in frequency, then there is always the opportunity to “tighten” the relative delays a little – which, at the same frequency, will reduce the absolute delays. Naturally, performance as part of ongoing events is growing – and may be higher (especially for mid-range and low-end processors) than with today’s DDR5. Tomorrow’s DDR5, if not completely solve the existing problems, then at least weaken them, and even become cheaper. And it will work with tomorrow ‘s processors, for which it itself may become more in demand. But all this will be tomorrow.

Today, we can only highly appreciate the approach of Intel – and generally thank the company for it. Given all the features of the transition period (which were predictable), Alder Lake supports both DDR4 and DDR5. And since support for DDR5 has appeared, it encourages memory manufacturers to get into such modules – this is something without which the transition would not have begun at all. But maintaining compatibility with DDR4 allows buyers not to rush. You can save money when building a new system, you can use the available memory (if you are satisfied with the capacity and its other characteristics) – a lot is possible . By the way, the company switched from DDR3 to DDR4 in the mass segment in the same way. So at the current stage of changing horses , it will be difficult for AMD: the AM5 platform is fundamentally designed only for DDR5, even the timing of its announcement is de facto tied to the success of new memory modules. Even if everything is ready, it is simply pointless to release the AM5 on the market now. And the start of the LGA1700 turned out to be much softer, thanks to this dualism. However, we repeat, there are much more rational reasons to “stay” on DDR4 for now than to keep pace with the era .