With the advent of the seventh generation Intel Core processors, it became clear to many that the tick-tock strategy that Intel had been following all this time had failed. The promise to reduce the process technology from 14nm to 10nm remained a promise, the long era of "Taka" Skylake began, during which Kaby Lake (seventh generation) happened, the sudden Coffee Lake (eighth) with a slight change in process technology from 14nm to 14nm + and even Coffee Lake Refresh (ninth). Looks like Intel really needed a little coffee break. As a result, we have several processors of different generations, which are based on the same Skylake microarchitecture, on the one hand. And Intel's assurances that each new processor is better than the previous one, on the other. True, it is not very clear what exactly ...
So let's go back to our generations. And let's see how they differ.
Kaby Lake
The appearance of processors in retail took place in early 2017. What is new in this family relative to its predecessor? First of all, this is a new graphics core - Intel UHD 630. Plus support for Intel Optane memory technology (3D Xpoint), as well as a new 200 series chipset (the 6th generation worked with the 100 series). And that's all of the really interesting innovations.
Coffee Lake
The 8th generation, codenamed Coffee Lake, was released in late 2017. The processors of this generation added cores and proportionally to the third level cache, raised Turbo Boost by 200 MHz, added support for DDR4-2666 (before that it was DDR4-2400), but cut off support for DDR3. The graphics core remained the same, but it was given 50 MHz. For all the frequency increases, I had to pay with an increase in the heat pack to 95 watts. And, of course, the new 300-series chipset. The latter was not at all necessary, since soon enough the specialists were able to launch this family on 100-series chipsets, although Intel representatives said that this was impossible due to the peculiarities of building power circuits. Later, however, Intel officially admitted that it was wrong. So what's new in the 8th family? In fact, it looks more like a regular refresh with the addition of cores and frequencies.
Coffee Lake Refresh
Ha! And here is our refresh! In the fourth quarter of 2018, 9th generation Coffee Lake processors were released, equipped with hardware protection against some Meltdown/Spectre vulnerabilities. Hardware changes made to the new chips protect against Meltdown V3 and L1 Terminal Fault (L1TF Foreshadow). Software and microcode changes protect against Specter V2, Meltdown V3a and V4 attacks. Protection against Specter V1 will continue to be patched at the operating system level. The appearance of fixes at the chip level should reduce the impact of software patches on processor performance. But Intel realized all this joy with protections only in processors for the mass market segment: i5-9600k, i7-9700k, i9-9900k. Everything else, including server solutions, did not receive hardware protection. For the first time in Intel's consumer processor history, Coffee Lake Refresh processors support up to 128GB of RAM. And that's it, no more changes.
What do we have in the dry residue? Two years of refreshes, games with cores and frequencies, plus a set of minor improvements. I really wanted to objectively evaluate and compare the performance of the main representatives of these families. Therefore, when I had a kit from the seventh to the ninth generation at hand - fresh i7-7700, i7-7700k and i7-8700k were recently added to our i7-9700 and i9-9900k, I took advantage of the situation and forced five different Intel Core processors to show what they are capable of.
The test is
Five Intel processors participate in testing: i7-7700, i7-7700k, i7-8700, i7-9700k, i9-9900k.
Tactical and technical characteristics of platforms
Intel i7-8700, i7-9700k and i9-9900k processors have the same basic configuration:
- Motherboard: Asus PRIME H310T (BIOS 1405),
- RAM: 16GB DDR4-2400 MT/s Kingston 2 pieces, total 32GB.
- SSD: 240 GB Patriot Burst 2 in RAID 1 (a habit developed over the years).
The Intel i7-7700 and i7-7700k processors also run on the same platform:
- Motherboard: Asus H110T (BIOS 3805),
- RAM: 8GB DDR4-2400MT/s Kingston 2 pieces, total 16GB.
- SSD: 240 GB Patriot Burst 2 in RAID 1.
We use custom-made chassis with a height of 1,5 units. They house four platforms.
Software part: OS CentOS Linux 7 x86_64 (7.6.1810).
Π―Π΄ΡΠΎ: 3.10.0-957.1.3.el7.x86_64
Made optimizations relative to the standard installation: added kernel launch options elevator=noop selinux=0.
Testing is done with all patches against Specter, Meltdown and Foreshadow attacks backported to this core. It is possible that the results of testing on newer and more recent Linux kernels may differ from those obtained, but the performance will be better. But, firstly, I personally prefer CentOS 7, and, secondly, RedHat is actively backporting innovations related to hardware support from new kernels to its own, LTS. I hope so π
Tests used for research
- sysbench
- Geekbench
- Phoronix Test Suite
Sysbench test
Sysbench is a package of tests (or benchmarks) for evaluating the performance of various computer subsystems: processor, RAM, data storage devices. The test is multi-threaded, for all cores. In this test, I measured two indicators:
- CPU speed events per second - the number of operations performed by the processor per second: the higher the value, the more productive the system.
- General statistics total number of events - the total number of completed events. The higher the score, the better.
Geekbench test
A suite of tests conducted in single-threaded and multi-threaded mode. The result is a performance index for both modes. Below are links to test results. In this test, we will look at two main indicators:
- Single-Core Score - single-threaded tests.
- Multi-Core Score - multi-threaded tests.
Units of measurement: abstract "parrots". The more "parrots", the better.
Phoronix Test Suite
Phoronix Test Suite is a very rich test suite. Despite the fact that all the tests from the pts/cpu package were carried out, I will give the results of only those that I personally found particularly interesting, especially since the results of the missed tests only reinforce the general trend.
Almost all the tests presented here are multi-threaded. The only exceptions are two of them: single-threaded Himeno and LAME MP3 Encoding tests.
In these tests, the higher the score, the better.
- John the Ripper multi-threaded password guessing test. Let's take Blowfish cryptalgorithm. Measures the number of operations per second.
- The Himeno test is a linear Poisson pressure solver using the Jacobi point method.
- 7-Zip Compression - A 7-Zip test using p7zip with an integrated benchmark feature.
- OpenSSL is a set of tools that implement the SSL (Secure Sockets Layer) and TLS (Transport Layer Security) protocols. Measures the performance of RSA 4096-bit OpenSSL.
- Apache Benchmark - The test measures how many requests per second a given system can handle when executing 1 requests, with 000 requests running concurrently.
And in these, if less, then better.
- C-Ray tests CPU performance on floating point calculations. This test is multi-threaded (16 threads per core), will shoot 8 rays from each pixel for anti-aliasing, and generate a 1600x1200 image. The test execution time is measured.
- Parallel BZIP2 Compression - The test measures the time required to compress a file (Linux kernel source .tar package) using BZIP2 compression.
- Encoding of audio and video data. The LAME MP3 Encoding test runs in a single thread, while the ffmpeg x264 test is multi-threaded. The test time is measured.
As you can see, the test set consists of purely synthetic tests that allow you to show the difference between processors when performing certain tasks, for example, clicking passwords, encoding media content, cryptography.
A synthetic test, unlike a test that is carried out under conditions close to reality, is able to provide a certain purity of the experiment. Actually, that's why the choice fell on synthetics.
It is possible that when solving particular tasks in combat conditions, you will be able to get extremely interesting and unexpected results, but still the βgeneral temperature in the hospitalβ will be as close as possible to what I got from the test results. It's also possible that by disabling Specter/Meltdown protection when testing 9th generation processors, I could get better results. But, looking ahead, I will say that they have shown themselves so well.
Spoiler: cores, threads and frequencies will rule the ball.
Even before testing, I carefully studied the architecture of the families of these processors, so I expected that there would be no significant differences between the test subjects. Moreover, not so much significant as extraordinary: why wait for interesting results in tests if you are measuring on processors built, in fact, on one core. My expectations were met, but something still turned out not quite as I thought ...
And now, in fact, the test results.
The result is quite natural: whoever has more streams and higher frequency gets points. Accordingly, the i7-8700 and i9-9900k are ahead. The gap between i7-7700 and i7-7700k is 10% in single-threaded and multi-threaded test. The i7-7700 lags behind the i7-8700 by 38% and from the i9-9900k by 49%, that is, almost 2 times, but at the same time the lag behind the i7-9700k is only 15%.
Links to test results:
Test results from The Phoronix Test Suite
In the John The Ripper test, the difference between the twin brothers i7-7700 and i7-7700k is 10% in favor of "k", due to the difference in Turboboost. The i7-8700 and i7-9700k processors have a very slight difference. The i9-9900k outperforms everyone with more threads and higher clock speeds. Twins almost 2 times.
The result of the C-Ray test seems to me the most interesting. The presence of Hyper-Treading technology in the i9-9900k in this multi-threaded test gives only a slight increase relative to the i7-9700k. But the twins lagged behind the leader by almost 2 times.
In Himeno's single-threaded benchmark, the difference isn't that big. A noticeable gap between the 8th and 9th generation from the twins: the i9-9900k overtakes them by 18% and 15%, respectively. The difference between the i7-8700 and i7-9700k is within the margin of error.
The twins pass the 7zip compression test 44-48% worse than the leader i9-9900k. Due to the greater number of threads, the i7-8700 outperforms the i7-9700k by 9%. But this is not enough to overtake the i9-9900k, so we see a lag of almost 18%.
The BZIP2 compression time test shows similar results: streams win.
MP3 encoding is a "ladder" with a maximum margin of 19,5%. But in the ffmpeg test, the i9-9900k loses to the i7-8700 and i7-9700k, but bypasses the twins. I redid this test several times for the i9-9900k, but the result is always the same. This is already unexpected π In the multi-threaded test, the most multi-threaded of the tested processors showed such a low result, lower than that of the 9700k and 8700. There are no clear explanations for this phenomenon, and I donβt want to make assumptions.
The openssl test shows a "ladder" with a gap between the second and third rungs. The difference between the twins and the leader i9-9900k is from 42% to 47%. The gap between the i7-8700 and i9-9900k is 14%. The main thing is flows and frequencies.
In the Apache i7-9700k test, it outperformed everyone, including the i9-9900k (6%). But in general terms, the difference is not significant, although there is a 7% gap between the worst result of the i7700-7 and the best result of the i9700-24k.
In general, the i9-9900k is the leader in most tests, only ffmpeg fails. If you are going to work with video, take the i7-9700k or i7-8700 better. The i7-9700k is in second place in the overall standings, slightly behind the leader, and even ahead in the ffmpeg and apache tests. So I boldly advise him and the i9-9900k to those who regularly experience large influxes of users to the site. Processors should not fail. I have already mentioned the video.
The i7-8700 scores well in Sysbench, 7zip and ffmpeg.
In all tests i7-7700k is better than i7-7700 from 2% to 14%, in ffmpeg test 16%.
Let me remind you that I didnβt do any optimizations other than those indicated at the beginning, which means that when you install a clean system on a newly purchased Dedicated drive from us, you will get exactly the same results.
Cores, threads, frequencies - our everything
In general, the results were predictable and expected. In almost all tests, a βladder to heavenβ appears, demonstrating the dependence of performance on the number of cores, threads and frequencies: more of this, better results.
Since all the test subjects are actually refreshes of the same core on the same process technology and do not have any fundamental architectural differences, we were unable to obtain βoverwhelmingβ evidence that the processors are qualitatively different from each other.
The difference between the i7-9700k and i9-9900k processors in all tests, except for Sysbench, tends to zero, since in fact they differ only in the presence of Hyper-Threading technology and a hundred additional megahertz in Turbo Boost mode for the i9-9900k. In the test, Sysbench is just the opposite: it decides not the number of cores, but the number of threads.
There is a very large gap in multi-threaded tests between the i7-7700(k) and i9-9900k, in some places as much as twice. There is also a difference between the i7-7700 and the i7-7700k - the extra 300 MHz adds agility to the latter.
Also, I can't talk about the qualitative impact of the cache memory size on the test results - we have what we have. Moreover, the included protection of the Specter/Meltdown family should greatly reduce the impact of its volume on the test results, but this is not certain. If a dear reader demands "bread and circuses" from our marketing department, I will gladly download you testing with disabled protection.
Actually, if they asked me: what processor would you choose yourself? - I would first count the money in my pocket, and choose the one that is enough. In short, you can also get from point A to point B in a Zhiguli, but a Mercedes is still faster and more pleasant. Processors based on the same architecture will one way or another cope with the same range of tasks - someone is just good, and someone is excellent. Yes, as testing has shown, there are no global differences between them. But the gap between the i7 and i9 hasn't gone away.
When choosing a processor for some particular highly specialized tasks, such as working with mp3, compiling from sources or rendering 7D scenes with light processing, it makes sense to focus on the performance of the relevant tests. For example, designers can immediately look at the i9700-9k and i9900-7k, and take a processor with Hyper-Threading technology for complex calculations, that is, any other than the i9700-XNUMXk. Streams run here.
So I advise you to choose what you can afford, taking into account the specification, and you will be happy.
In testing, servers based on i7-7700, i7-7700k, i7-8700k, i7-9700k and i9-9900k processors with . Any of them can be ordered with a 5% discount for 3 months - contact with the code phrase "I'm from Habr". If you pay for the year minus another 10%.
All evening in the arena , FirstDEDIC System Administrator
Source: habr.com