Once upon a time, we carried out all tests of computer systems and processors for them with the same discrete video cards. This approach did not cause problems – basically I had to work with modular desktop computers, which also provided the necessary flexibility in configuring them. Yes, and fixing the video card was required mainly for games – other applications at that time, as a rule, did not rely on their capabilities. And then games were part of the standard methodology – which is why it was required. Yes, and integrated graphics then took its first steps, and many of its implementations simply “slowed down” processors in general-purpose programs as well.

Later, the ability to maintain this approach disappeared as the number of “non-configurable” systems studied grew. However, the GPU still did not affect the work of the vast majority of programs, but the games had to be moved to the optional set. Outside of it, it remains possible to compare the performance of processors directly with each other – regardless of the specific GPU. At one time, most tests, even on desktop systems, could be carried out exclusively with a focus on the IGP. Simply because in the period from 2014 to 2017, most of the new products were supplied with it: AMD then developed only APUs, while Intel supplied all desktop processors with an integrated GPU. HEDT, as well as special game testing, did not fit into the main canvas, but both could be done using locally adapted methods.

But after 2017, I had to return to the original approach. To the extent that it was possible, of course, compact and / or portable systems cannot be flexibly configured, so you have to use as is. But already adjusted for the GPU – which many programs gradually began to use, shifting part of the work from the processor to it. However, as our last year’s testing of the AMD Ryzen 5 3400G and Intel Core i5-9600K with integrated graphics and two discrete video cards showed, this is not particularly difficult: for most test method programs, a specific video card model is not important: either it is not used at all, or it is used in the same way. Only Adobe Premiere Pro and Magix Movie Edit Pro work significantly differently: the versions we use already know how to use the Intel GPU, but still ignore all other solutions. Now the situation is gradually changing, but we will find out how and by how much within the framework of the new testing methodology, which will soon be ready. In the meantime, this moment should simply be taken into account, but it makes sense to return to the topic again, since both companies have updated their processors. This is Intel’s first microarchitecture change in five years, and AMD last year increased the number of cores in APUs from four to eight – and also transferred them to a new microarchitecture this year. It is clear that the “pure” processors of the company can have a larger number of cores – and if they are equal, they are cheaper, while maintaining an advantage in the amount of cache memory, PCIe 4.0 support, etc. Intel has a different problem – graphics are everywhere (just in terms of processors blocked – but without changing other characteristics), but its performance, to put it mildly, leaves much to be desired. And in models until last year, the same problem concerned even functionality. It would seem that the choice is obvious: if it is possible to install a video card, then it should be done. Even if there are no serious requirements for it – at least inexpensive. But exactly what it “seemed” – since there are practically no inexpensive ones on the modern market. There are, perhaps, budget “gags” that have not been updated for a long time – which, in some respects, are even worse than integrated devices (even Intel ones) – but in modern conditions they cost almost like real ones. And in such circumstances, focusing on integrated graphics (at least for a while – to wait for the normalization of prices) is fully justified. And just then the new Ryzen 5 5600G came to us, which (unlike the previous generation of APUs) the company plans to actively sell at retail, so it was decided to start working with it with just such material. For which it was necessary to test a few more AMD and Intel processors, but still they were going to do it for a long time.

Note that we will not touch on game tests today – recently we have already done this with respect to most of the subjects. With the exception of the 5600G, the GPU hasn’t changed much in the new line. Over time (no later than we collect the entire new collection) we will work on this issue in more detail. In the meantime – the processor part, since this is just the main change in both AMD Cezanne and Intel Rocket Lake.

Test participants

Last time we settled on the Core i5-9600K, however, as part of the transition from LGA1151 to LGA1200, the manufacturer first endowed this family with Hyper-Threading support (more precisely, stopped blocking it) without significantly changing anything in the crystal, and then Rocket replaced Comet Lake lake. In which the microarchitecture of the processor cores is already new, and the GPU is new, and the PCIe controller is already Gen4 … In general, everything is beautiful, except for one thing: all the splendor is based on the same 14-nanometer process, so the crystal turned out to be very large and voracious. Everything is like in a joke about a new comfortable airliner: now fasten your seat belts and we will try to take off with all this on board . We already know this well – since all these three processors have been tested more than once. But two of them – only with a discrete graphics card.

Initially, it was decided to limit ourselves to six-core processors, since the 5600G (and today it is the main character, since it is the only completely new one) is just that, however, on reflection, we added last year’s older APU to the set, even though it has eight cores. But the old ones – actually the year before last, and in the desktop processors of the 3000 line (which are also Zen2) and in 2019 there were up to 16 cores. But there it happened thanks to the chiplet layout. And all APUs are more like Intel processors – they use one monolithic crystal. The 7-nanometer process technology made it possible to “shove” eight cores along with the GPU there, but more is still difficult. And without that, I had to greatly limit the capacity of the cache memory, and they still do not have support for PCIe 4.0 APUs. When switching from Renoir to Cezanne, nothing has changed significantly – for this, the company also needs new technical processes. But the cores are now Zen3, not Zen2 – and in the form of a single block: without splitting into quad-core CCX, as before. There is also a single cache of the third level, and its capacity has doubled. Despite the limited transistor budget, it was necessary to go for it – the low capacitance of L3 was perhaps the weakest point of Renoir. In chipsets of “clean” processors of both lines, we recall, it is generally 32 MiB, so the gap from them has only been reduced, but not eliminated. But this should also have a beneficial effect – as well as the new microarchitecture itself.

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.

iXBT Application Benchmark 2020

The task is primarily for calculations, and multi-threaded calculations – therefore, the number of cores and threads is important. Can we increase their number? This is more efficient than improving the quality of the kernels. That’s why the 4750G remains the leader (since it’s the only one with the 8C/16T formula), and Intel processors benefited more from the return of Hyper-Threading than from a deep upgrade. But in general, the latter is clearly visible in both companies – and the efficiency is almost the same. And the performance is at the same level: if the 4600G/4650G were direct competitors to the Core i5-10600K, then the 5600G is already head to head with the i5-11600K.

Nothing changed. Although it couldn’t. Theoretically, software optimization could allow the Core i5-11600K to speed up somewhere (thanks to AVX512 support), but for now, parity. More precisely, its preservation is the same as last year. It was worse the year before last – then AMD APUs contained only four cores of the “old” architecture, much less efficient than Skylake. Therefore, the Ryzen 5 CPUs were at least as good as the Core i5 (in fact, at one time they competed well with the more expensive Core i7), and the Ryzen 5 APUs were incapable of such feats. Now, in pairs (2020 or 2021 edition) are the same, that is, when choosing, first of all, you need to look at other circumstances.

Pogrom 🙂 The reasons for which are indicated above – in two out of five programs in Intel processors, the GPU took over the lion’s share of the work, but the versions used still cannot work in the same way with discrete or integrated solutions from AMD and Nvidia. It happens so – we warned more than once. However, it is clear that this is nothing more than a temporary state – Intel generally took care of video encoding and decoding issues before others, and the company’s market share has always been too large to ignore, and AMD has just yet to “convince” all programmers it is more responsible to treat (at least) APU. But, by the way, the effect of the upgrade of Intel processor cores turned out to be blurry – AMD has a larger increase. And it is clearly seen that not everything is decided by quantity here: the new Ryzen 5 even managed to outperform last year’s Ryzen 7 even a little.

In this case, single-threaded performance generally comes first. And it is clear that Intel is doing better with this at times – depending on other conditions, of course. But the difference of the order of 5% can not be taken into account – you still won’t notice it without instruments .

And again we return to the exact equality. But only in the 11600K-5600G pair – last year’s products differed slightly not in favor of AMD. Remembering the “cache-loving” nature of this program – and the fact that Cezanne L3 has doubled, the mechanism for solving the problem is clear.

But here doubling the cache is clearly not enough. Inside AMD’s own lineup, it (together with the improvement of the cores) worked perfectly – the new Ryzen 5 managed to overtake the Ryzen 7 from the previous collection. But Intel processors are still noticeably faster. Note that “pure” AMD processors do not have such a problem – but there is L3 32 MiB, and not 16 and not, especially 8. Moreover, archivers themselves are no longer very relevant for many users today (and when they are needed – in practice, weaker processors do just fine), but this can also manifest itself in other classes of software. For example, the behavior of games has much in common with just these programs. And how things are going there – we will definitely check in a configuration with a discrete graphics card.

And we are back to normal again, even with a slight preponderance of APU. It used to be, and here Intel even “won back” a little backlog – but such a difference can hardly be considered significant.

Here’s one that won’t work. However, it is due to literally two groups of applications – archivers and video editors. The first problem is “hardware” – it cannot be solved, at least until it is possible to speed up the memory system and, in particular, increase caches. It just has to be taken into account. And the second is purely software. And, in the process of updating the software, it can resolve itself, since this is purely a matter of compatibility. What is in newer versions of programs – as already mentioned, we will try to check soon. For now, let’s just say that the possibility of such a development of events cannot be ruled out. Under normal conditions, we can talk about parity between Core i5 and APU Ryzen 5.

As for the processor part, as far as GPU performance is concerned, there is no “parity”, as we well know. Rocket Lake has only learned how to overtake Athlon – but at least Ryzen 3 (including the very first three-year-old models) is still far away. And this is not the only problem.

Energy consumption and energy efficiency

Sometimes you can hear that for a desktop processor, they say, power consumption is not important . In principle, if we are talking about a gaming PC with a top-end video card, then against its background … Maybe not very important. But if a simple (perhaps even compact) system is being assembled, then I would like it to be smaller .

But the old 14-nanometer process technology “smaller” does not allow. In general, the models of three years ago did not look very good in this parameter, the LGA1200 was developed in many respects to feed the voracious updated models – but against the backdrop of Rocket Lake, this all pales. AMD APUs, on the other hand, have about twice as modest appetites (on average), so you can assemble a compact computer on AM4 and not suffer with all sorts of supercoolers. With Intel’s new products, it’s better not to be smart. With younger models and tightening the power consumption limits, if only – but in this case, performance will decrease.

Simply because the “old” 14 nm does not allow you to be both rich and healthy at the same time. You can make a low-power processor, or you can make a high-performance one – but these will be two completely different processors. Although once this technical process was the best in the industry, and until recently it allowed Intel to somehow get out. But it’s good that this horse skeleton can no longer be spurred on. Alas, not yet in the desktop segment. But AMD does not have such a problem – exactly the same crystals that are designed for the mobile market go to desktop APUs. Much more power efficient than existing Intel desktop solutions.

Total

If we ignore individual vzbryka, then the conclusion is simple – in terms of processor performance, the Ryzen 5 APUs are equivalent to the Core i5 of the corresponding year of release. Over the past time, both companies have increased performance – but just in approximately equal proportions. So, most likely, the same can be extended to the competition of Ryzen 7 with Core i7. But in the budget segment, AMD already has the new Ryzen 3, while Intel remains with last year’s Core i3. With all the consequences. However, Alder Lake and LGA1700 will be released soon, which can greatly change the alignment. But all this will start, as usual, with top-end processors and motherboards, then prices will drop for several more months … In general, it is unlikely that many will assemble a computer on a new platform by the New Year. And without a discrete graphics card – almost no one is for sure.

In such conditions, first of all, you have to pay attention to other parameters. For example, prices. Or prohibitive (for this segment) power consumption of Rocket Lake. Or significantly more powerful graphics in the APU – at the level of junior discrete video cards, which, in general, allows you to play many games. Although, for some, the greater functionality of Intel platforms may be more important – for example, support for PCIe 4.0 (AMD still has to choose between integrated graphics and a new interface – they do not live together) or USB3 Gen2 × 2. In general, it seems to us that AMD AM4 looks much better than the Intel LGA1200 as a platform for a home non-gaming (but with the ability to run some games) computer in terms of the totality of characteristics at the moment. In the case of the latter, the use of integrated graphics is rather a temporary measure: when a gaming system is planned, but there is no money for a gaming video card yet. And the APU is a balanced all-in-one solution: fast, inexpensive, economical.

Another interesting point: although both companies have updated their processor lines this year, old solutions still should not be discounted yet. Yes, they are slower – but also cheaper. At the same time, Intel’s “tenth” line is much more economical. And AMD’s IGP once again hasn’t changed much, so if you focus on it, then new items are not really needed. On the other hand, the 4000th line did not enter the mass retail market, but there will be no problems with the acquisition of the 5000th in ordinary retail stores. And this is good – many people have been waiting for it there for a long time.