It is unlikely that anyone will argue with the fact that a sharp increase in the number of cores in mass desktop processors from four (which was a kind of “standard” for more than ten years) to eight in just a couple of years is the most significant event in this market segment. That is why in the first tests according to the new version of the methodology, we focused on eight-core models – as well as on the (rather successful) AMD initiative last year to endow the desktop platform with models with 12 and even 16 cores. Of course, they go beyond the mass segment primarily in terms of price, but they make you completely rethink the concept of HEDT. Well, and ideas about the “horizons” of mass platforms, of course, too. So far, Intel has nothing to answer such challenges directly, but the company has already sharply reduced prices for multi-core solutions for LGA2066 , increasing the competitiveness of this HEDT platform, and also plans to move from eight cores to ten in new mass platforms this year.

Against this background, the heroes of yesterday , such as six-core processors, somehow faded. They appeared ten years ago, but for a long time they were expensive, and then they became conditionally “slow” overnight. Quite an interesting process that deserves detailed consideration.

Summary of previous episodes

In principle, the first six-core x86 processors were presented by Intel at the end of 2008, but they did not leave noticeable traces in people’s memory. For objective reasons, these were Xeons for “multi-socket” systems based on the already outdated Core2 architecture and the “recharged” Socket604 platform (debuted back in the days of NetBurst), announced only a little earlier than the revolutionary first-generation Core at that time. And even though the latter had only four cores at the start, they “pulled” eight computation threads, and the integrated memory controller radically improved work with it. In addition, the abandonment of the archaic FSB itself allowed a significant increase in the speed of interprocessor data exchange in two-socket systems (which have always sold much better than a multisocket). In general, 2010 turned out to be more important – when Intel introduced six-core processors for LGA1366 and a new LGA1567 platform for “adult” solutions. Within the framework of the latter, not only six-, but also eight-core processors were already produced, later supplemented with ten-core models. But LGA1366 was the ancestor of the High-End Desktop, so not only Xeon, but also Core i7 were produced within its framework. At first, only the 980X Extreme Edition for $999, then the 990X began to “settle in” at this price level, and the company “recommended” selling the Core i7-980 for a little less than $600. At the end of 2011, we saw a “restyling” of the platform in the form of LGA2011 with second-generation Core processors, but the same number of cores for the same money, then its third-generation refresh, and then … Then 2014 happened and LGA2011-3 – the older (“extreme”) processor for which has already become eight-core, and the younger six-core was offered at a price of around $400. In general, for four years the entrance ticket has fallen in price by 2.5 times – but even this was not enough for mass popularity. Not because of any technical reasons – the price war between AMD and Intel in 2006-2007 radically brought down the prices of mass processors: 80% of buyers suddenly discovered that they already had enough devices from the $80-$200 price range, and not like before . Before him, it remained to reduce prices by a further two times – which, perhaps, by now would have happened in an evolutionary way. But the reality turned out to be more interesting – AMD returned to the market of processors with high performance.

This company introduced its six-core desktop computers in the same 2010 – and immediately at “humane” prices: initially the lineup included Phenom II X6 1090T for $289 and Phenom II X6 1055T for only $199. It is clear that this did not happen at all because of some kind of philanthropy – just about as many processors worked. Six-core (as in the case of Core2) was implemented at a relatively low price – a couple of cores were added to the four available in Phenom II. However, these processors were an evolutionary development of the legendary Athlon 64 of the beginning of the millennium, so by that time they were obsolete. In general, a temporary solution is in anticipation of the Bulldozer. Which was also much delayed at the start, so the FX sample of the second half of 2011 looked great against the background of the first (2008-2010), and not the second generation Core. In addition, the backlog in the development of technical processes and attempts to make a “big leap” forced the company to limit the number of cores in the FX to four. And so that such a retreat would not seem strange after Phenom II X6, verbal balancing act went into action: the cores were renamed into modules and the term “x86-core” was introduced – a part of the module capable of executing one stream of arithmetic-logical commands. The decoder is one per module (i.e., two such “cores”), the floating point unit is the same, the behavior is at the level of a regular Core core with Hyper-Threading – but cheap six- and eight-core processors. True, the latter competed only with the quad-core Core (and even then – so-so), except for the price, and the former had even worse. So the company only once updated the “productive” line in 2013 – and took up a radically new microarchitecture.

Let’s note an interesting point – according to the results of work on the Zen project, it is clearly seen that the company tried first of all to make a good … quad-core processor, that is, it acted, in general, in the spirit of Intel. But at the same time, it was decided to fix the long-standing shortcomings of AMD platforms, namely the lack of a “good” interprocessor interface. Therefore, by 2017, the company had both a “base module” (CCX) of four processor cores with memory controllers and PCIe, and an Infinity Fabric bus ready. With its help, it was possible to connect several blocks for different purposes (for example, CCX with a GPU to build an APU), as well as several CCXs in one chip, several chips on one substrate, and even several sockets in one computer system. The potential possibilities of this interface are endless – and still not fully realized, which allows the company not to slow down the pace of release of new products. In practice, to restore its position in the market, AMD for the first time limited itself to one “homogeneous” crystal from two CCXs, which made it possible to release relatively inexpensive eight-core desktop processors. It is clear that desktops have long been niche solutions, but on the basis of such semiconductor devices it was possible to make multi-chip assemblies, bringing the number of cores in one socket to 32, which was more than that of Intel (albeit in a monolithic crystal). And up to two sockets, i.e. up to 64 cores in a two-socket system – Intel “kept” up to 192 cores per system, but when using eight sockets. Well, as for laptops, other compact systems or just budget PCs, another crystal was planned for them – with CCX + GPU. Actually, that’s why Ryzen-based APUs for a long time remained fundamentally a maximum of quad-core, but today we are not talking about them.

Desktop processors without a graphics core initially received eight cores – and at the beginning of 2017, Ryzen 7 occupied an intermediate position between mass Intel processors for LGA1151 (up to four cores plus graphics) and HEDT solutions for LGA2011-3 (six to ten cores – and also without graphics). They turned out to be “intermediate” in terms of the number of PCIe lanes – the controller of the first crystals supported 32 PCIe 3.0 lanes, of which 24:16 for video cards (by default), 4 for communication with the chipset, and 4 for one fast solid state drive. Buyers of systems based on Intel processors received at their disposal either 16 available “processor” lines, or already from 28. The memory controller was similar to the mainstream Intel platform – only two channels, which limited both bandwidth and maximum capacity. Therefore, despite the high performance, Ryzen 7 had to be positioned in the $300-$500 price segment – competition in the HEDT market was postponed until the fall. But this, we recall, is eight cores – obtained from crystals of “ideal” quality: where everything works. At the same time, at the first stage, marriage also happened – with one or even two faulty nuclei in the crystal. Nobody was going to “throw away” them, so deliveries of Ryzen 5 began in the summer: if no more than one core per CCX turned out to be non-working, then six-core 1600/1600X were obtained, and if two, then quad-core 1400/1500X (with the help of the first they were also disposed of chips with bad L3 cache). Over time, the yield of suitable crystals increased, so that the working cores were already turned off – to saturate the market. But, if the younger Ryzen 7 cost a little over $ 300, then the six-core Ryzen 5 should, by definition, be sold cheaper. And so it was – in fact, this pair of models competed in price with the Core i5, but the “sevens” – with the Core i7 and higher. But in all cases, with a significant head start in the number of cores: Intel at that time offered only four of them at prices up to about $ 350 and six – above.

The “correction” came in the fall of 2017 – when the second version of LGA1151 and six-core processors for it entered the market. Given that the process of manufacturing processor chips itself takes about six months, two conclusions can be drawn. Firstly, this is really an “answer” to the appearance of Ryzen 7 in the spring. Secondly, it didn’t have to be specially developed – in fact, the company already had a ready-made six-core design for mass production. Which goes well with the rumors of 2014 – the older Skylake endowed with so many cores (then just being developed). Now it is already difficult to judge how events would develop if Intel released such models in 2015 or at the beginning of 2017 (when Skylake was “refreshed” in the form of Kaby Lake without major changes). Perhaps the whole story would have gone in a completely different direction. In fact, new processors had to be brought to the market in a hurry, and even a little bit to change the platform, and for some time, budget motherboards and processors were only for the “old version” of LGA1151, and high-performance ones for the new one. But the old and new Core i5 and i7 cost the same, and more powerful cores allowed the six-core Core i7 to compete directly with the eight-core Ryzen 7, often overtaking it. Here, AMD had to lower prices – “cramming” all Ryzen 7 at $ 300. Ryzen 5 also fell in price, becoming quite a “folk” in six-core versions. However, the prices were still comparable to the Core i5 – but with the ability to perform 12 computation threads versus six.

This balance was maintained in 2018. AMD updated Ryzen – but the new models weren’t too different from the old ones. Intel has mastered eight cores – which could change the situation in the market, since the older Core i9s have already become faster than Ryzen 7 (still able to compete only with the Core i7), but they are a little cheaper. The problem is that Intel had to increase the number of cores in both desktop and laptop processors. While maintaining the same process technology, this led to an increase in the area of u200bu200bthe crystal – and a lower yield of finished products from each silicon wafer. But it was not possible to increase production volumes: since at the same time the demand for server solutions also grew (which turned out to be higher than predicted by Intel), and the company got involved in the LTE modem business … Therefore, the situation has not fundamentally changed.

Unlike last year – when Ryzen were significantly redesigned. The new chiplet layout made it possible to separate processor cores and cache from other bindings, transferring them to a more subtle production process, and even “shove” processors with 12-16 cores into the same dimensions. And the cores themselves are no worse than Core (as we have already seen), so Ryzen 9 is not needed to compete with Core i9 – now Ryzen 7 is enough for this. But new Ryzen 5 have also appeared – with a price like in Core i5, but the performance of Core i7. We already know this too, because we tested the Ryzen 5 3600 using the old method. Now it’s time to see how it and some other processors work in more recent programs.

Test participants

There is one funny fact: for Intel processors, the formula “six cores twelve threads” was strategically important for four whole years (since the company’s top models were like that), and for the same amount of time it was simply important , but now there is practically no such relevant one in the company’s assortment. desktop models. Notebook Core i7 of the “tenth” and “ninth” generations meet such a formula, and on the desktop it ended along with the “eighth”. Officially. So, these models are still on sale, the platform has not changed since 2017, and retail prices in the corresponding pairs (for example, i7-8700K and i7-9700K) are almost the same, so the choice of a specific model is a practical issue. We just took the best (albeit rare) Coffee Lake – the limited edition Core i7-8086K, which is comparable to the i5-9600K and i7-9700K. And in the future, its results will be useful to us, because within the framework of the soon-to-be-expected LGA1200 platform, Core i5 will receive such a nuclear formula, so it can become “popular” when applied to Intel processors (10 years after the appearance – but better late than nobody ).

And AMD is no stranger to releasing affordable six-cores. And over the past few months, the company has expanded its understanding of those by releasing the Ryzen 5 3500 and 3500X, which are more similar to the Core i5, since the cores in them are “single-threaded”. But this is understandable: before, the company needed a head start in the number of cores, or at least computation threads, since Core was architecturally better than Ryzen, but now there is no need for this. As a result, models with a “full” configuration already compete with Core i7 in terms of performance, while “simplified” models fit perfectly between them and quad-core APUs (which still retain the old architecture). Today we will limit ourselves to the first, of which there are two – but fundamentally they differ from each other only a little more than the Ryzen 7 3700X and Ryzen 7 3800X. The latter, as we remember, behaved almost identically in the tests – so let’s see if this will be repeated in the lower pair.

As for the rest of the environment, only the motherboards were different: ASRock X570 Phantom Gaming X based on the AMD X570 chipset and Asus ROG Maximus X Hero based on the Intel Z370 chipset. And also the frequency of RAM – DDR4-2666 for Intel and DDR4-3200 for AMD. It is clear that the same frequencies could be set on these boards, but since Intel still does not officially support high-frequency memory, this is its choice, and it is worse for it. Moreover, the efficiency of memory controllers is also different, so the question of what to consider equal conditions is completely separate. Therefore, we test it this way, in strict accordance with the specifications and recommendations – we also use boards based on the X570 chipset for the Ryzen 3000. Here, too, the situation is ambiguous: on the one hand, AMD actively advertises the compatibility of the new processor family with old motherboards, and on the other hand, it still recommends using new motherboards. Which, generally speaking, is not very logical for Ryzen 5: the processors themselves are inexpensive, motherboards based on the X570 are expensive, and new chipsets for “inexpensive” are still not ready. Therefore, in the near future we will be engaged in a practical study of the need for the X570 chipset – but for now, for maximum correctness, we will test the Ryzen 5 3600/3600X in this (not too typical for them in practice) environment.

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 . 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 – this article is also directly involved in today’s article) and grouped by areas of application of the computer. Accordingly, all diagrams related to applications have dimensionless scores – so more is always better. 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. In the main lineup there are only a couple of “processor-dependent” games in low resolution and medium quality – synthetic, of course, but conditions close to reality for testing processors are not suitable, since nothing depends on them in such conditions.

iXBT Application Benchmark 2020

It’s clear that the 8/8 formula is better than the 6/12, so the upgrade to the Core i7 is welcome. But the fact that the older Core i7 is a little slower than Ryzen 5 is impossible. It is also clearly seen that at present it is not worth paying attention to clock frequencies. This used to be a major difference between models, but it ended as the spread in the number of cores increased, the presence / absence of SMT support, cache and memory systems, etc. This is more serious. And the frequency has since begun to change dynamically – depending on the load and power consumption. Therefore, last time we observed close results for Ryzen 7 3700X and Ryzen 7 3800X, since in real software and their “working” frequencies turned out to be close, and now we see the same unanimity in a pair of Ryzen 5 3600 and Ryzen 5 3600X. And, again, they are both faster than any Core i7 – although they compete with the Core i5 in price.

Renderers can squeeze out a little more, so the gap between Ryzen 5 3600 and Ryzen 5 3600X has increased a bit. But against the backdrop of the difference between Ryzen 5 and Ryzen 7, these are trifles. In any case, Intel’s solutions are already behind, no matter how you think – at least taking into account prices, at least abstracting from them and paying attention only to the “nuclear formula”.

As for working with video, there is parity between Ryzen 5 and Core i7. However, this is an average – if you look at the results of specific applications, you can see that everything is not so clear . First of all, due to the fact that programmers are mastering new features of hardware with a long lag – and the Core microarchitecture has not changed for five years now: since Skylake itself, the number of cores has been growing, but not their “quality”. Ryzen came later, and the Ryzen 3000 family did last year. So over time, they will probably “squeeze” more out of it. This always happens – Intel has proven since 2011: at the time of release, each generation overtook the previous one by the notorious 5% (about which they regularly scoffed in forums, etc.), and then they grew every year … So it will be and with AMD. And, by and large, it has only an academic value – technically it is already parity, but until Intel can reduce the prices of the Core i7 to the level of Ryzen 5, it is more logical for the buyer to “vote” for the latter with his labor ruble. Or for Ryzen 7 – where they give a little more for the same money .

And the fact that the cores “leveled up” played a bad joke on Intel in these programs, where their quantity is not so important, like “quality”, work with memory, etc. Previously, under such conditions, Core consistently won – but last year’s AMD’s big leap was so impressive that the 3000 family smashes both them and the “old” Ryzen to smithereens. And already in almost any form – even the “stripped down” Ryzen 5 3500 is only slightly behind the Core i7-9700K, noticeably outperforming all Core i5 and old Core i7 (however, we have already run far ahead by announcing the content of one of the future materials 🙂 ).

Simple integer code, so here “virtual” computation threads are comparable in efficiency to physical cores – as before, updating the Core i7 only leads to a decrease in performance. This does not help intercompany competition in any way – just the opposite. Especially if you take into account the “cache-loving” program – and the “giant” L3 of the new Ryzen (thanks to the 7 nm process technology in many respects – it is clear that the microarchitecture is always linked to an affordable production process).

Not surprisingly, the picture is similar here. With the exception of a radical reduction in the difference between Ryzen 5 and Ryzen7 – simply because the bottleneck from the point of view of archivers quickly turns out to be the memory system. Ryzen 9 breaks away from Ryzen 7 more significantly – but thanks to doubling the capacity of L3. But Ryzen 5 and Ryzen 7 have the same caches – with the corresponding result. Well, the times when Core were ahead in such tasks are also already behind (such a slightly sad pun). Even (once again) without taking into account the price.

We return to the situation when physical cores are better than virtual threads. As a result, Intel processors line up in a neat ladder. AMD too – but the “steps” are shorter, since Ryzen 5 and Ryzen 7 differ just in the number of cores with the same caches and memory system. On the other hand, the first ones are already enough to compete with any Core i7 – despite the fact that they are cheaper.

The main conclusion is that the time for throwing cores and threads is over: even under equal conditions, Ryzen with cores is no worse than Core. Even better. So we can conclude that just adding Hyper-Threading support to Core i3 / i5 is not enough. This will give a good performance boost – but not enough to catch up with AMD. It is necessary to refine the microarchitecture, which Intel promises to do. So let’s see how well it will turn out. And the current round remains entirely with AMD – in almost all segments.

Energy consumption and energy efficiency

Once again, we are convinced that Coffee Lake was more economical than Coffee Lake Refresh, so if it weren’t for the need for eight cores … As a result, in the “ninth” generation, the loss to the “three thousandth” Ryzen also “drew”. With the exception of the minimum load mode, which we use file operations – so here the “hot” chipset can also interfere (but we will thoroughly understand this issue in one of the following materials).

The most interesting thing is that at the moment neither AMD nor Intel can find a direct correlation between the number of cores and power consumption. No – eight, of course, ceteris paribus, “eat” a little more than six – but just what a little . Well, you should not pay attention to TDP for a long time if you are interested in consumption – they are not connected in any way. However, this is not news – it has always been so.

With comparable power consumption, the one who works faster is more efficient – but this number of cores is affected. With AMD, everything is the same – for Intel, the crystals today are not exactly the same, so once again it can be noted that with the Coffee Lake refresh, energy efficiency has fallen. And, most likely, most of the “ninth” generation six-core mobile Core i7 are made on “eighth” crystals.

Games

As already mentioned in the description of the methodology, it makes no sense to keep the “classic approach” to testing gaming performance – since video cards have long been determining not only it, but also significantly affect the cost of the system, you need to “dance” exclusively from them. And from the games themselves – too: in modern conditions, fixing a game set for a long time does not make sense, since literally everything can change with the next update. But we will carry out a brief test in (albeit) relatively synthetic conditions – using a couple of games in the “processor-dependent” mode.

But there is nothing interesting here – except that the Core i5 lagged behind everyone in the “formula” greedy for threads (and not just cores), and the “old” Core i7 turned out to be worse than the others in non- greedy “tanks”. But Ryzen is smoother – and at the level of Core i7. Therefore, the Core i9-9900K can continue to be considered the best gaming processor – and gaming computers, nevertheless, are assembled in practice based on the Ryzen 5 3600: fortunately, inexpensively and paired with any video card, it will rest on the video card. So it goes.

Total

In principle, we have tested almost all of these processors before, so we did not expect any discoveries. This is the third testing of the new methodology, necessary to replenish the database of results – and the final verification of the methodology itself. Now we can move on: a new AMD HEDT platform is waiting for us, and in the short term, a new mainstream Intel platform is already looming. Yes, and information on old solutions needs to be updated, since the difference between them and “new solutions” usually increases with software updates.

So far, the main conclusion is this: if not immediately, but AMD managed to bring Ryzen to the same level as the latest Core. It is clear that Intel “played along” in this, which has not changed the architecture since 2015, when there were no Ryzens yet, but in practice the result is important. And it is such that now (right now) AMD’s mass solutions do not need any head start to compete – they are better under equal conditions. And Intel can change this situation either by a serious price reduction (very serious – since there is not even parity now), or by an equally serious revision of the Core. Better yet, both at the same time. Roughly speaking, the new Core i5 should cost the same as the old ones, but work faster than any six-core Core i7. The maximum task is not to be slower than the eight-core Core i9. What happens in practice – we will soon find out. But this round, we repeat, is for AMD.