Changing the standards of RAM in the personal computer market occurs regularly – and always in the same way. There is a conditional version of “N” – for many years of use, licked to a shine, inexpensively sold on every corner, but already quite tired of some advocates of novelty, and gradually becoming a brake on the implementation of something new. In the storerooms, the developers have a version of “N + 1” – formally improved, but it still needs to be learned how to produce in mass quantities. Therefore, no one is in a hurry with the introduction – and there is no need, it seems. But as soon as processor manufacturers are ready to switch to N + 1 and implement its support in their products, the uniform fever immediately begins. It turns out that production needs to be adjusted, there are some pitfalls in the process … in general, at first there is a shortage on the market and corresponding prices. Later it turns out that there is no particular reason for haste – since processors with N and N + 1 work approximately the same at first. The voices of enthusiasts are subsiding – supporters of reasonable economy come to the fore. Affirming that there was nothing to hurry. After some time, however, production problems are solved, new modules are produced in ever larger quantities, it turns out to reduce costs and prices to the level of the old ones, support for the outdated standard disappears – and everything calms down for several years. Until the next shift.

But are there alternatives? By and large, no. It makes no sense for memory manufacturers to get ahead of the father in hell – as long as the new standard is not supported by anyone, there is no need to start production. And processor manufacturers understand that at the first stages there will always be roughness – which is why they start the transition even before they completely “lock in”. Even if in the end the benefits of it turn out to be imperceptible – the solution of the chicken and egg problem in itself is important here. So that it does not become too serious, you usually have to support both N and N + 1 at least for some time with processors. In the retail market of components, motherboard manufacturers also come to the rescue, usually releasing products with two types of slots at once (although this is not recommended). In general, somehow everything is experienced. I experienced it before – both when switching from SDRAM to DDR SDRAM, and when introducing DDR2, and later when mastering DDR3, and then DDR4 …

Now it’s time for DDR5 – and everything will happen again. In the future, the new standard has significant advantages over the old one, since it provides a quarter increase in throughput at the same frequencies, and the frequencies are also much higher. Of course, we are talking about effective frequencies here – the DRAM crystals themselves have not changed significantly for a long time, so the real frequencies of each have grown over the years, perhaps from 100 to 200 MHz. But the number of banks is constantly increasing and prefetching schemes are improving, which allows you to “squeeze” more out of the same chips in terms of throughput. But the transition to new memory standards does not bring improvements in delays – it has long been customary to deal with them primarily by multi-level caching, etc. Usually, with each change in the standard, delays only increase – and then they decrease in the process of debugging production and switching to higher frequencies.

Potentially, DDR5 can be more economical due to the lower operating voltage of the microcircuits, and the transfer of the power controller directly to the module provides a higher quality of power supply. At the same time, this (along with doubling the number of bank groups) will increase the capacity of each module without any additional tricks (such as LRDIMM) from 64 to 256 GB. To work large volumes at high speeds, the parity system is expanded: if now 8 ECC bits fall on one 64-bit wide channel, then in the new standard – already on a 32-bit wide channel. At the same time, the width of the data bus for each module remains “old” – it’s just that each module becomes two-channel at the internal level. This approach does not change the peak memory bandwidth and remains “consistent” with the cache line length, but may eventually give some gain in a multitasking environment on its own.

In general, it can be seen that many of the advantages are primarily focused on the server market – which is not surprising, since these systems are most critical to speed and memory capacity. What about that ordinary user? Right now, almost nothing. With the exception of an increase in peak bandwidth by a factor of one and a half, but this is only if we compare official specifications: in fact, the frequency potential of DDR4 does not end at 3200 MHz. Actually, about 4 GHz has long been obtained without any problems – albeit unofficially . If there is a desire and an opportunity to overcome problems, then 5 GHz for the existing (and not some promising) DDR4 is also achievable for a long time (albeit limitedly, but in a PC these limitations are not critical). At the same time, the delays when operating at the same effective frequency for DDR5 are much higher, and at higher frequencies, they are also higher, i.e., to fully compensate for this effect, you will need to reach the DDR5-6000 level somewhere, which will happen massively and inexpensively not even tomorrow, but it’s good if the day after tomorrow. And today, the retail prices of the simplest and most inexpensive (relatively inexpensive) DDR5-4800 module are at the level of two of the same capacity from the same manufacturer, but already “elite” DDR4-4000. Shortage: First you need to establish production and supply. On top of that, the prime cost of DDR5 modules is objectively higher than that of DDR4 – a custom power controller has not only advantages. Over time, thanks to the effect of mass production, this problem will be weakened, but so far it is very noticeable.

In general, the first buyers will have to pay “goodies” for subsequent ones. Or not to pay – there is only one line of desktop processors with DDR5 support on the market now, but it can also work with DDR4. And it is unlikely that it will work much worse with DDR4 – it is clear that Intel tried to make the new memory better than the old one, but not everything depends only on Intel. With such a difference in prices, you can put more DDR4 into the system, which will make a difference in many “serious” tasks, and you can also go beyond the DDR4-3200 mode. Moreover, it is “allowed” to overclock memory not only on top-end motherboards with Z-series chipsets – everything is suitable, except for the H510 for LGA1200 and H610 for LGA1700, but their customers, it seems to us, are not at all up to overclocking anything, and not up to “elite” modules in principle, so the difference in the price of memory in their case will reach up to three or four times at all – which in itself is more than enough.

Test participants

For today’s testing, we decided to limit ourselves to the standard DDR4-3200. Moreover, they even tried to play along with DDR5 a little, at least to the extent of current capabilities. For the first tests of the LGA1700, we used the Kingston Fury Beast Black kit, which supports two modes: DDR5-4800 CL38 and DDR5-5200 CL40 – we will use both today (at the same time comparing them with each other). By the way, this is one of the most inexpensive DDR5 kits, which you can relatively freely go and buy at any time without pre-orders and waiting – 31,999 rubles for 32 GB. In general, it’s easy to count – a ruble per megabyte 🙂

And let’s take a simple and inexpensive DDR4 – TeamGroup T-Create Classic without any XMP profiles, but DDR4-3200 CL22 directly from JEDEC. Such 32 GB kits are in the assortment of many manufacturers, they cost an average of about 10 thousand rubles, that is, three times cheaper than Kingston Fury Beast Black. In general, the problem of saving is solved immediately. And from the point of view of the mass buyer, it seems to us, this comparison can be completed – the difference in the price of these two kits is slightly more than the cost of the Core i5-12400. Or (if you look from the other side) allows you to purchase a Core i7-12700K instead of a Core i5-12600K. In such a situation, it is simply ridiculous to evaluate the difference in the performance of the memory itself. Over time, maybe. But not now.

Maybe even gamers will come in handy? It’s also unlikely – because even in our fun times, without any difficulties, for 25-30 thousand you can buy, for example, a video card based on the GeForce GTX 1650 or Radeon RX 6500 XT. Neither the first nor the second is far from the dream of a demanding buyer – but in a budget gaming computer, they will more or less fit. And integrated graphics – definitely not.

In general, there are no rational reasons to pay extra for DDR5 at the moment – regardless of how well it is able to work. Well, except that the buyer is generally doing well : he can easily buy (and buy!) a top-end motherboard, a top-end processor, a powerful video card, one or more of the very best SSDs of sufficient capacity. And in general, everything that can be useful to him in the foreseeable future is bought immediately, so there is nothing to improve – and there is still money left. And I want to invest them immediately in the prospects , since there is such an opportunity.

Will such a hypothetical buyer actually get anything even compared to inexpensive and “simple” DDR4? This is what we decided to test by taking the Core i9-12900K and a couple of motherboards: Asus ROG Maximus Z690 and Asus TUF Gaming Z690-Plus WiFi D4. For maximum correctness, a combo board would be suitable, since different models may behave differently, but so far nothing suitable has been observed on the market. Yes, and it usually “works” in one direction: motherboards with DDR5 support are usually more expensive and more “wound”, but DDR4 gravitates towards the budget segment. Another reason why you may have to buy DDR5 right now is if there are serious requirements for the board, so you can’t find anything suitable among inexpensive models. Although this in itself separates the potential areas of application of DDR4 and DDR5 even further apart, since most of the sales of boards fall on inexpensive models. And there is no point in releasing an inexpensive board for expensive memory – therefore, for the majority of buyers, the choice of the memory itself will generally be purely theoretical for the time being. Except for the very top of the top segment.

Low Level Tests

For a quick evaluation, we used the benchmark in AIDA64. DDR5 bandwidth is impressive – especially in DDR5-5200 mode. Latency in it is also reduced – but only when compared with the DDR5-4800 mode, which we used earlier in most tests. DDR4 even with such conservative timings is still ahead. But this is far from the limit – we have seen less than 55 ns on also generally inexpensive kits . Someday we will see it on DDR5 – but definitely not today (and the money will not help). And how it affects in real applications – now we will evaluate.

Test Methodology

The testing methodology is described in detail in a separate article. The only change we made for some of the materials was the use of Windows 11. Accordingly, we did not add the results of all tests to the general table – they are available in a separate one (as usual, in Microsoft Excel format) . 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, all diagrams related to applications show dimensionless scores, so here “bigger 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.

iXBT Application Benchmark 2020

As expected, the processor is completely loaded with work – the memory’s task is only to supply shells . And the time when the dual-channel DDR4-3200 will no longer cope with this will not come very soon in such tasks.

Rendering is even more so. Yes – the memory system in such tasks works with a high load. However, this is not yet the case when desktop processors can have faster memory become vital.

But an example of a completely different kind – a weighty (against the general background) 10%. Not that it can justify the current price difference. But in the future, it is clear who should look at the new memory in the first place. Although if you look at the detailed results, you can see that the lion’s share of the acceleration falls on Vegas: other applications are more stable in terms of memory speed. But this or that effect is everywhere – just more or less modest.

The acceleration is even more noticeable – and again uneven: Lightroom is accelerated by one and a half times (almost linear growth), and the other two programs – by only 10%. It is understandable – multi-threaded “development” and RAW processing requires quick access to RAM: in which the whole process takes place. Actually, Photoshop also keeps all the images in memory in an “unpacked” form, so only the sequential mode of operation in this test prevents it from speeding up significantly (unlike Lightroom, the batch mode of this program processes photos one by one, not in parallel). In any case, we repeat, there is a performance increase in all photo programs – and it was higher in all video programs. So for the production of multimedia content, the new standard is useful: although the main load still falls on the processor and sometimes on the GPU, they need to be able to quickly provide data for work. And here it is just the increase in memory bandwidth that is more important than delays. On the other hand, in these tasks, the amount of memory also matters, so with today’s price difference for the same money , a system with DDR4 can be faster than one equipped with DDR5 – this must also be taken into account in practice.

Something in between the first and second pairs – there is an increase, but small. However, there is no fall either – enough already. In theory. In practice, we repeat, with the current price difference, this can not be taken into account at all, and with its reduction or disappearance, there will be many arguments in favor of DDR5 without performance.

A slightly unexpected result – after all, archivers usually react more strongly to delays than to bandwidth. On the other hand, it is obvious that Intel did not need to “squeeze” everything possible out of DDR4 – for Alder Lake, this is, in fact, a compatibility mode, that is, it works fine . On the contrary, there was an incentive to slow down such work a little by default. Which we exacerbated with an inexpensive DDR4 kit, and without manual settings. As it turned out, even in this case, paired with a top-end DDR4 processor, it is difficult to lose a lot. So to consider it objectively obsolete does not work.

Once again, we are convinced that 10% is still a good result. In this group, for example, only two out of four applications accelerated by this amount, while the rest simply did not notice the memory replacement.

Averaging “good” and “bad” results gives an overall difference of 8%. Like it or not, but a little. And part of the loss of DDR4 can be “winned back” even on these modules by working with manual settings. Considering the peculiarities of the current pricing, this is not the only possible method 🙂 In general, DDR5 performs no worse than DDR4 on average – this can be said for sure, but nothing more. In fact, everything depends very much on the specific nature of the loads – some are generally very “resistant” to memory, while some have shown a noticeable increase. Although it is mostly noticeable only against the background of the average, we also saw an acceleration of almost one and a half times on our Lightroom test. That is, we repeat, it is too early to “chase” a new memory just for the sake of speed. But the reasons for its purchase may already be now. And it definitely won’t get worse from this – in some cases it can turn out even better, albeit disproportionately to the costs.

Energy consumption and energy efficiency

Theoretically, DDR5 is more energy efficient, but in practice, the memory power consumption itself is small compared to other system components. Especially when it comes to the Core i9-12900K – glorious for its outstanding appetite. And especially his appetite is noticeable in those cases when he manages to work faster – that is, just with DDR5. What else is added some difference between motherboards, etc. In general, we got + 8% performance – and + 11% average power consumption. Although, we repeat, these results should be treated very carefully – it is true for specific assemblies on a specific pair of boards. The use of one board with two types of slots would give greater accuracy, but it is not known whether such will appear (there were some during the transition from DDR3 to DDR4 , but since then a lot of water has flowed under the bridge, so “ordinary” boards have become much more complicated and expensive) . From a practical point of view, this means that such nuances can be ignored. In portable computers, the introduction of DDR5 can bring some benefit, the development of LPDDR5 even more so, but the desktop is not the kind of technology where you should pay attention to memory consumption. Against the general background, it is difficult to notice.

Total

As mentioned at the beginning, current DDR5 prices may cause the memory selection process to end before it even starts. On the other hand, the situation was similar with the introduction of DDR4, and earlier – DDR3, and even earlier – DDR2 … “Lucky” on the Intel platform was only the very first DDR SDRAM, and even then only because before that you had to choose between conventional SDRAM ( already too slow) and RDRAM (which ended up not taking off). And DDR was actually “run in” at the expense of AMD Athlon buyers – it was already submitted to the Pentium 4 ready-made. Later, it was AMD that tried not to rush to support new types of memory, so it was for her that the process went more smoothly. The same will happen in the coming year, when the LGA1700 is already collecting all the transitional bumps, and the AM5 is just getting ready to enter the market.

But someone has to start the process because of the chicken and egg problem. By and large, the introduction of DDR5 support could have begun earlier. Initially, the industry was preparing for this, focusing that way on 2017-2018. The dates (quite suddenly!) are similar to the expected dates for the development of the 10-nanometer process technology by Intel. But this process, to put it mildly, dragged on, so it was decided not to hurry too much. On the other hand, the delay was beneficial: three years ago we would have seen exactly the same thing, only in an even more grotesque form. But, of course, if everything was as it should be, then the situation on the processor market would be more unambiguous, and some external effects (such as pandemics or another mining boom) would not interfere. In any case, the history of the subjunctive mood does not tolerate.

The facts for today are simple. The 12th Gen Core (i.e. Alder Lake desktop line) supports both DDR4 and DDR5 SDRAM. As tests show, they can run faster with DDR5, but for all processors (even top models) it is not necessary to use DDR5. Therefore, there are boards for LGA1700 on the market with slots of both types – but not at the same time. Top boards are mainly focused on DDR5, but they always sell worse than mid-range models and/or budget products, which is exacerbated by the very large difference in the prices of the memory modules themselves. In such conditions, it is difficult to focus on DDR5, and, from a practical point of view, it is not necessary. In any case, the cost of DDR5 is not a reason to refuse to buy Alder Lake, if it was supposed to be. And not even a reason to strongly postpone the purchase. And if there is a desire and an opportunity to join the ranks of the first users of the new memory, this can be done right now. In some applications, the performance will be higher, sometimes noticeably higher, but there were and still are even more loads that did not respond to the memory system at all. Therefore, the evidence base can be adjusted to any conclusions – it is enough to “correctly” select applications and conditions for comparison. If you work well with the latter, you can even prove that DDR4 is better. But objectively, it is cheaper in any case, and much more so far. Therefore, we repeat, there is nothing to do in a budget computer or a mid-range DDR5 system yet – unlike the LGA1700 itself. For a top-end system, the prospects are usually attractive, you just shouldn’t expect the costs to be recouped with this approach.