Comparing processor: PassMark — CPU Comparison

Comparisons of Processor Speeds | Small Business

By John Papiewski

The speed of a processor strongly determines a computer’s performance, whether it is a desktop PC for business, a smartphone or Web server. Chip makers rate their processors according to the clock frequency in megahertz and gigahertz; higher frequencies indicate faster processors. By comparing the speeds of different processors, you can gauge the relative performance of the computers that use them.

Processor Families

  1. Comparisons of processor speeds are more meaningful if the chips all belong to the same general family. Chip makers sell hundreds of different kinds of processors; the ones powering PCs represent only one series. Though processors in smartphones and tablets approach the power of desktop machines, they belong to different families and comparing them becomes technically complicated. Historically, Intel’s x86 is the most popular processor series, used in business desktop and laptop computers going back to the original IBM PC in the 1980s. Although Intel developed the x86 design, several competitors sell compatible processors that let you run the same software.

Intel Core I3, I5 and I7

  1. Intel sells a line of processors called the Core i3, i5 and i7, designed for desktop computers. The i3 represents the economy end of the series, running at speeds starting at 1.6 GHz. The i7 processors are Intel’s high-performance models, the fastest of which runs at 3.8 GHz. Most of these processors have two or more cores, allowing several programs to run at the same time.

Intel Atom

  1. First introduced in 2008, Intel’s Atom processor has an energy-saving design to prolong battery life in netbook computers and other mobile devices. These processors run at speeds ranging from 600 MHz to 2.13 GHz. Relatively low clock speeds are common for mobile processors, because running the clock slower saves battery power.

VIA QuadCore

  1. The QuadCore processor by VIA is an x86-compatible processor intended for mobile computers. The introductory model has a clock speed of 1.2 GHz, a built-in cryptography engine for data security, 64KB of instruction and data caches on each core, and a 4MB level 2 data cache. A fast form of memory, cache improves efficiency by saving recently used data. The processor searches for data in the level 1 cache first and then searches its larger level 2 cache. Not all cache lookups are successful, but the arrangement greatly enhances processor speed.

AMD Athlon II

  1. AMD, one of Intel’s biggest competitors, offers the Athlon II series processors with speeds up to 3.4 GHz. This chip, like the Intel i7, is meant for desktop computers. The Athlon II comes in two series, the X3 and X4, which have three and four cores, respectively. It has a 64-bit design and virtualization technology that lets you run several operating systems at the same time.


  • Intel: 3rd Generation Intel Core i3 Processors

Writer Bio

Chicago native John Papiewski has a physics degree and has been writing since 1991. He has contributed to «Foresight Update,» a nanotechnology newsletter from the Foresight Institute. He also contributed to the book, «Nanotechnology: Molecular Speculations on Global Abundance.»

[PDF] Comparing CPU and GPU Implementations of a Simple Matrix Multiplication Algorithm

  • DOI:10.17706/IJCEE.2017.9.2.430-438
  • Corpus ID: 55110937
  title={Comparing CPU and GPU Implementations of a Simple Matrix Multiplication Algorithm},
  author={Tomaz Dobravec and Patricio Buli{\'c}},
  journal={International Journal of Computer and Electrical Engineering},
  • T. Dobravec, P. Bulić
  • Published 2017
  • Computer Science
  • International Journal of Computer and Electrical Engineering

In recent years, graphics processing units (GPU) have become a standard part of high-performance computing systems used for solving large scale computation problems. To relieve the main processor more and more time consumptive tasks are moved from CPU to GPU where algorithms run in parallel on a high number of GPU’s processors. In this paper we present both sequential and parallel implementations of a simple matrix multiplication algorithm and we compare the overall execution time. To further… 

View via Publisher

Scalable Multi-Node Fast Fourier Transform on Gpus

  • Manthan Verma, Soumyadeep Chatterjee, M. Verma
  • Physics, Computer Science

    SSRN Electronic Journal

  • 2022

The details of the multi-node GPU-FFT library, as well its scaling on Selene HPC system, and the timings of multicore FFT of 1536 grid with 196608 cores of Cray XC40 are presented.

GPU empowered pipelines for calculating genome-wide kinship matrices with ultra-high dimensional genetic variants and facilitating 1D and 2D GWAS

A new method and two pipelines are devised: KMC1D and KMC2D for kinship matrix calculation with high-dimensional genetic variants, respectively, facilitating 1D and 2D GWAS analyses and performance analyses show that the calculation speed can be accelerated by 100–400 times over the conventional CPU-based computing.

Comparing Controlflow and Dataflow for Tensor Calculus: Speed, Power, Complexity, and MTBF

  • Miloš Kotlar, V. Milutinovic
  • Computer Science

    ISC Workshops

  • 2018

Ten different tensor operations, their generalizations, as well as their implementations for a dataflow paradigm, are introduced for addressing a number of big data problems in machine learning and computer vision.

Batch Sparse Recovery, or How to Leverage the Average Sparsity

  • Alexandr Andoni, Lior Kamma, Robert Krauthgamer, Eric Price
  • Computer Science


  • 2018

The question for p =1 up to polylogarithmic factors is resolved by presenting a randomized adaptive scheme that performs $\tilde{O}(km)$ measurements and with high probability has output satisfying $(1)$, for arbitrarily small $C > 1$.

Feasibility of Material Decomposition-based Photon-counting CT Thermometry

  • Nathan Wang, Petteri Haverinen
  • Physics


  • 2022

Over the past decades, thermal ablation procedures such as high intensity focused ultrasound (HIFU) have been developed vaporize cancerous tissues in a focal area. Thermal ablation has the potential…

Reduced Kernel Extreme Learning Machine for Traffic Sign Recognition

This paper proposes a TSR system based on a Reduced Kernel Extreme Learning machine (RK-ELM) which is efficiently implemented in a Graphic Processing Unit (GPU) because of the inherent simplicity of ELM-based models.


SORT BYRelevanceMost Influenced PapersRecency

GPU Computing

  • John Douglas Owens, M. Houston, D. Luebke, Simon Green, J. Stone, James C. Phillips
  • Computer Science

    Proceedings of the IEEE

  • 2008

The background, hardware, and programming model for GPU computing is described, the state of the art in tools and techniques are summarized, and four GPU computing successes in game physics and computational biophysics that deliver order-of-magnitude performance gains over optimized CPU applications are presented.

A performance study of general-purpose applications on graphics processors using CUDA

  • Shuai Che, Michael Boyer, Jiayuan Meng, D. Tarjan, J. Sheaffer, K. Skadron
  • Computer Science

    J. Parallel Distributed Comput.

  • 2008

GPGPU Processing in CUDA Architecture

  • Jayshree Ghorpade, Jitendra Parande, Madhura Kulkarni, Amit Bawaskar
  • Computer Science


  • 2012

It is shown how CUDA can fully utilize the tremendous power of these GPUs, and a comparison of CUDA C/C++ with other parallel programming languages like OpenCL and DirectCompute is taken through.

Graphics processing unit (GPU) programming strategies and trends in GPU computing

Generating Performance Analysis of GPU compared to Single-core and Multi-core CPU for Natural Language Applications

  • Shubham Gupta, M. Babu
  • Computer Science

  • 2011

This paper compares the performance of single-core, multi-core CPU and GPU using a Natural Language Processing application and shows that multi-cores on CPU need to be used explicitly to achieve high performance.

Parallel Complexity of Matrix Multiplication1

This paper determines the parallel complexity of multiplying two (not necessarily square) matrices on parallel distributed-memory machines and/or networks and provides an achievable parallel run-time that can not be beaten by any algorithm for solving this problem.

Faster Algorithms for Rectangular Matrix Multiplication

  • F. Gall
  • Computer Science, Mathematics

    2012 IEEE 53rd Annual Symposium on Foundations of Computer Science

  • 2012

A new algorithm for multiplying an n × n<sup>k</sup> matrix by an n–k × n matrix, which is better than all known algorithms for rectangular matrix multiplication and recovers exactly the complexity of the algorithm by Coppersmith and Winograd.


  • D. Coppersmiths
  • Mathematics

  • 2014

The number of essential multiplications required to multiply matrices of size N N and N N is bounded by CN log N. Key words, matrix multiplication, tensor rank, algebraic complexity Introduction. Let…

On the Asymptotic Complexity of Matrix Multiplication

  • D. Coppersmith, S. Winograd
  • Computer Science, Mathematics

    SIAM J. Comput.

  • 1982

A consequence of these results is that ω, the exponent for matrix multiplication, is a limit point, that is, cannot be realized by any single algorithm.

Performance analysis of GPU compared to single-core and multi-core CPU for natural language application

  • International Journal of Advanced Com puter Science and Applications,

  • 2011

is the best Ryzen for 3D gaming

> AMD Ryzen 7 5800X3D


AMD Ryzen 7 5800X3D is the world’s fastest gaming processor! Beats the 12-core Ryzen 9 5900X and the cool Intel Core i9-12900K! Uses 3D editing technology for the first time and equipped with 96 MB L3 cache — 3 times more than the «blue» competitor! All this is very . .. very loud statements by Lisa Su — the head of AMD!

But, who spoke of strong heat? And what will happen if you fight him with «undervolting». And what is the dependence of the novelty on memory and how does it look against Intel in reality? This will be a full exploration of the new Ryzen 7 5800X3D gaming processor!

What is «3D» in the name of the processor?

So, what is the difference between our good friend 8-core Ryzen 7 5800X, which received the prefix «3D»? Well, in short — this is the first «two-story processor»! The fact is that before this CPU did not grow up. Cores, cache, controllers — it was all one layer. AMD, on the other hand, has made a revolution — it has increased an additional floor to the main one. This addition turned out to be a 64 MB SRAM die. nine0003

Moreover, it is not connected crookedly on the bus or somehow with a significant decrease in throughput. No, those 64MB are literally dotted to the main 32MB L3 cache on the main die. Thus, AMD has a large 16-channel cache, which all 16 processor threads have access to. And it sounds so easy. AMD, together with TSMC, has done a titanic job to build an additional die … even without physically increasing the dimensions of the processor — it looks completely identical to the regular Ryzen 7 5800X. It has the same die height and cooling compatibility. In general, without exaggeration, AMD has made a revolution — this is the first step towards the mass introduction of multilayer processors. But since this is only the beginning of the journey, there are also nuances that we will show in comparison with competitors. nine0003

Who are we comparing the Ryzen 7 5800X3D to?

The new gaming processor was made based on the 8-core 16-thread AMD Ryzen 7 5800X, 2020 release. It operates at frequencies from 3.8 to 4.7 GHz. The L3 cache is also not small — 32 MB. Also supported is an interface for graphics cards version 4 and DDR4 memory with a guaranteed frequency of up to 3200 MHz. Prices, as usual, we will discuss at the end, against the backdrop of test results.

Now about the review hero Ryzen 7 5800X3D. It has the same configuration of cores and threads, but at a noticeably lower frequency — from 3.4 to 4.5 GHz. This is done to reduce heat generation. Of course, the «sandwich» of crystals is hotter. And the AMD Ryzen 7 5800X3D has overclocking blocked for the same reason. However, look at this beauty — unreal 96 MB cache in the third level.

Intel Core i7-12700K has 12 cores, but, as we remember, only 8 of them are high-performance, the remaining 4 are energy efficient. The latter do not parallel, so there are 16+4 threads in Hyper-Threading — or a total of 20. Yes, the blue ones also get out as best they can. This processor supports PCI Express up to version 5 and memory, both DDR4-3600 and DDR5-4800. But it’s official. We will take a faster kit with a frequency of 5600 MHz and compare turbo memory against turbo cache. nine0003

But first, let’s look at the components of the system, because in this test it is more important than ever to understand what worked and how.

System Components

We needed three motherboards for these tests. And one of them is for the Intel bundle with DDR5:

  • Ryzen 7 5800X and Ryzen 7 5800X3D — ASUS ROG Strix B550-F Gaming (Wi-Fi)
  • Core i7-12700K + DDR4 — ASUS PRIME Z690-P D4
  • Core i7-12700K + DDR5 — ASUS PRIME Z690-A

And, when we have already touched on the memory, we note that all Ryzen tests and half of the Intel tests passed with the «popular» PATRIOT VIPER Steel DDR4-3600 dies. In stock, they worked with timings of 18-22-22-42.

And the second set, just to confront the memory against the cache. Here Intel has already been transplanted into the second board and Kingston FURY Beast DDR5-5600 has been installed. Namely, 32 GB with two slats. Imagine, the next Ryzen will also support this. nine0003

Finally, the high point of Cooler Master MasterLiquid PL240 FLUX dropsy has come. This time, she will have to work on her conscience. It generates an airflow of 2 cubic meters per minute or 72 CFM. Plus, it boasts an increased area of ​​\u200b\u200bthe copper base of the cool two-chamber pump.

On the other hand, the already well-known Palit GeForce RTX 3080 Ti GameRock OC video card warmed up the system and at the same time provided the maximum expansion of the framework from the GPU. nine0003

Regarding the disk system, stability was provided by the 1TB M.2 PCIe SSD PATRIOT P300. Games, benchmarks, everything was read very quickly and flew into faster memory — efficiency and caches of different levels.

There were no comments from the side of food either. The platinum Seasonic Prime PX-850 gave out stable equal voltage on all lines and sufficient power. In particular, up to 390 W for video cards and from 100 W for processors. And other components are not offended. nine0003

And here is the second processor cooling component that needs to be fixed in memory — the be quiet! PURE BASE 500DX Black. In addition to the fact that it is beautiful, roomy and generally charged, it is also well ventilated by three large complete turntables 140mm Pure Wings 2.

Plus, we improved this base with three be quiet! LIGHT WINGS 140mm PWM. See what great cooling conditions the processors were in? Go ahead.

Competitive comparison

First, let’s compare Ryzen 7 5800X3D turbo cache vs Core i7-12700K turbo RAM in workloads. Will this second floor give us anything outside of games?

And immediately the 3DMark Time Spy test. Even with the same DDR4 memory at 3600 MHz, the Core i7 overtakes the review hero by 47%, according to our main CPU Score indicator. And when switching to DDR5, blue looks 55% better. A comparison of the Ryzen 7 5800X3D with its brother showed no difference at all. The 3DMark test does not need an additional cache. nine0003

7-ZIP Benchmark throws the Ryzen processor a lifeline. With the same memory, the Ryzen 7 5800X3D was able to beat the competitor by 6% in archive packing speed. But with DDR5, blue is tearing everyone to pieces, increasing its advantage by 35% in this indicator. In unpacking, its lead is not so great — 6-8%

In CineBench R23C, Intel processors are again ahead of the new Ryzen. What is curious here is the frequency and type of RAM do not decide anything for blue. It is better by 32-64%, both with DDR4 and DDR5. Although Intel outperforms even a relative due to the higher frequency of cores — by 10-11%. nine0003

If we take the results of all the benchmarks we made, then for work tasks the Ryzen 7 5800X3D is the least suitable. It is 2.5% faster than the base Ryzen 7 5800X, 21% faster than the Core i7-12700K with DDR4-3600 memory, and 37% faster than the same processor with DDR5-5600.

In Cyberpunk, DLSS at maximum performance puts extra load on the CPU. So, are you ready to check out the fps boost from adding 64MB of L3 cache over the 5800X? Given that the frequency of the 3D novelty is 200-400 MHz lower? nine0003

There are 31% higher average and 45% very rare events! These are not AVX instructions for you!

So, yes, the chip really works — it gives a noticeable increase. True, the temperature of the new one is something not at all happy. But we will talk about this separately in due course.

In the meantime, let’s look at the progress compared to Intel: in terms of average frequency, the hero of the review is the best — by 25% bypasses Core i7 with DDR4 and by 18% — with DDR5. But Intel loses in terms of 0.1% Low — 8 and 10%, respectively. nine0003

When tracing is enabled, the balance of forces is preserved. On average, the Ryzen 7 5800X3D has the highest numbers on the counter. And for rare and very rare events — a better Core i7. And friends, pay attention to how the cache boosts AMD better than Intel RAM. In the reds, their chip gave a 27% acceleration to the 2020 processor, and 2000 MHz Intel RAM — only 11%.

On the contrary, the slightest increase came out in Assassin’s Creed Valhalla at ultra-settings and 50% render scale. Compared to the Ryzen 7 5800X3D, it is better only in the middle frequency — by 9%. And it also outperformed the Core i7 with DDR4 memory by 6% in very rare events and compared with a bunch of DDR5 in this indicator.

And now our CPU killer Total War Saga: Troy with maximum grass and army size. She literally smokes Ryzen 7 5800X3D — 90°C. There is a hot crystal sandwich. Nevertheless, we get 23-32% better performance than the younger brother, and 18-23% worse than the Core i7-12700K with DDR4 memory.

If you equip an Intel 8+4-core with DDR5 modules, it becomes an unattainable leader at a distance of 54-62%.

So, what does the average come out. According to the results of all gaming benchmarks, the Ryzen 7 5800X3D outperforms the regular Ryzen 7 5800X by 20-25%. Assassin’s Creed Valhalla certainly likes a higher frequency, and all other games have accelerated very well.

In comparing the hero of the review with a bunch of Core i7-12700K + DDR4, we already have mixed success. Ryzen 7 5800X3D is 6% better in average fps, but loses a little in rare and very rare events. nine0003

The combination of the Intel processor with DDR5 can be called the «first» in quotation marks. Yes, we have a 7% to 24% advantage over the Ryzen 7 5800X3D. But here the scales were strongly tilted to the whole side of the Total War Saga: Troy benchmark. Now, if we selected benchmarks like Lisa Su…

And now let’s calculate the number of fps in terms of 0.1% low for all benchmarks:

  1. Core i7-12700K + DDR5 — 359 fps
  2. Core i7-12700K + DDR4 — 336 fps
  3. Ryzen 7 5800X3D — 323 fps
  4. Ryzen 7 5800X — 263 fps

The hero of the review takes an honorable third place with a noticeable advantage only over the domestic competitor. It seems that Lisa was a little hasty in this comparison. Although the data obtained will still have to be revised, taking into account the prices, both for the CPUs themselves and for the bundles of the motherboard and memory. And we will definitely do this after three stops with interesting information.

Overheating and undervolting Ryzen 7 5800X3D

Now let’s talk about something that may have annoyed you when comparing processors — namely, the temperature regime of the Ryzen 7 5800X3D.

Let’s go back to our favorite Total War Saga: Troy. Competitors have about 75°C, and the hero of the review has 90°C! And this is its critical temperature for a minute. And now let’s think about our cool case, six 140mm fans and a charged liquid cooling system. This is the case — a real silicone lottery. For example, the Ryzen 7 5800X3D is only a couple of degrees hotter than its younger brother, and we got a black mark out of the bag. nine0003

Can you imagine what will happen to him in an ordinary building, in a cave of an ordinary Ukrainian gamer? This is the fee for participation in the project «multilevel CPU — in every home!».

But what temperature we got when we started the stress test — more than 93 ° C, that is, above the critical maximum. There was no throttling.

It’s not for nothing that AMD banned this processor from overclocking. In this regard, we decided to explore the other side of this coin — undervolting, that is, reducing the CPU supply voltage. How many millivolts does it need to drop to get an acceptable temperature and how will this affect power? nine0003

We reduced the voltage in two stages — by 0.05 V and 0.1 V or minus 50 and 100 mV, respectively.

The first stage of the experiment had almost no effect on the temperature during the stress test — it is still in the red zone.

But in the gaming benchmark, this figure fell to the level of competitors — around 75 ° C. And, pleasantly, it had almost no effect on the final result.

Reset 100 mV cooled down the «3D skin» by 9°C in the stress test and another 2-3°C in the game.

Again, only a few percent of the productivity paid for this.

Therefore, we recommend future owners of this processor to use undervolting at least for the summer heat.

How is the Ryzen 7 5800X3D affected by RAM frequency?

Now let’s answer another important question. «And if I already have a Socket AM4 system with not the fastest memory, and the Ryzen 7 5800X3D gets into it, how is the maximum upgrade of this platform?» How will the speed change in games with 2400, 3200 and 3600 MHz memory? Will a large cache smooth out this kind of RAM slowdown? nine0003

Of course, the AIDA64 memory test depends more on the memory frequency, but it is not particularly representative when you need to estimate the real performance in real programs.

In turn, other benchmarks have enough L3 cache to avoid using slower RAM. Because with a frequency of 2400, which is with 3600 MHz — the difference is within the margin of error.

But there is almost no cache for games, because it is used for almost everything — weather cycle changes, vehicle physics calculation, NPC display, etc. And when there is enough cache, we see that the Ryzen 7 5800X3D keeps a high average frequency with all three memory values. And as soon as it overflows a little, it turns to slower RAM and sags by 24%. We get a very rare event, which against the background of 175 fps can be felt as a slight twitch. nine0003

If the graphics are Ultra-Ultra and with the addition of ray tracing, the focus shifts more towards the video card. The processor prepares fewer frames and the impact of the RAM frequency is minimized. In this case, 2400 MHz loses 5-10% to 3600 MHz.

Approximately the same picture with Total War Saga: Troy — 2-7% difference. In general, as you can see, the influence of the RAM frequency depends on many factors: somewhere around 24%, somewhere around 2-7%. Therefore, in order to always have maximum comfort with the 5800X3D on average, it is advisable to keep the frequency not lower than the recommended one, but preferably 3600 MHz. nine0003

Test in difficult games

And our usual gameplay test. Now let’s see how the Ryzen 7 5800X3D handles real heavy games — without undervolting and in FullHD. By the way, we will also pay attention to the temperature.

The first thing I want to launch is the long-suffering 128-player Battlefield 2042 mode. With ultra-graphics, but without ray tracing. Has this day really come, after a bunch of tests of retro processors and budget AMD CPUs, when the 0.1% low indicator has finally exceeded 60 fps! nine0003

God of War and ultra graphics in DX11. The video card is loaded to the brim — it consumes as much as 400 watts and heats up to 83 ° C. And the processor does not lag behind it in terms of temperature. It is not visible on the recording, but there were still small twitches in the gameplay.

Red Dead Redemption 2 and maximum graphics on the Vulkan API. The video card is loaded almost to the ceiling, and the processor is cooled both literally and figuratively. In the FPS line, you can see at least 60 fps, with an average of 98 fps.

In general, all the most difficult games are excellent. Here, the capabilities of the system are limited precisely by the graphics subsystem, even the RTX 3080 Ti does not pull!! In terms of temperatures, the Ryzen 7 5800X3D got a headroom of 10-15°C to a critical mark of 90°C, given that in many projects it was only 70-80% loaded.


So, it’s time to look at the prices together and compare them with the capabilities of all test participants. It is clear that now they are not stable, but the gradation between the processors can still be determined. nine0003

Prices for processors with a boxed cooler. hotlines. End of July:

  • Ryzen 7 5800X3D from UAH 15,169
  • Core i7-12700K from 14 742 UAH
  • Ryzen 7 5800X from 11 097 UAH

If we consider only the CPUs themselves, then the distribution is as follows. The Ryzen 7 5800X3D is even slightly more expensive than the Core i7-12700K.

Prices for motherboards. hotlines. End of July:

  • LGA1700 under DDR5 | Core i7-12700K | from 6 770 UAH
  • nine0057 LGA1700 for DDR4 | Core i7-12700K | from 2 562 UAH

  • Socket AM4 both Ryzen | from 1 568 UAH

Regarding motherboards. It is unlikely that anyone will look towards the most budget options for such processors, without heatsinks on the VRM. But again, we appreciate the trend. Here, whatever one may say, DDR5 support costs horse money.

CPU+MB+RAM communication prices. hotlines. End of July:

  • Core i7-12700K | from 6770 UAH MB| + from 4 407 UAH RAM | = from 25 919 UAH
  • Core i7-12700K |+ from UAH 2562 MB|+ from UAH 2145 RAM| = from 19 449 UAH
  • Ryzen 7 5800X3D |+ from UAH 1568 MB | + from UAH 2 145 RAM | = from 18 882 UAH
  • Ryzen 7 5800X |+ from UAH 1568 MB | + from UAH 2 145 RAM | = from 14 810 UAH

Therefore, we add the price of the most affordable DDR4-3600 and DDR5-4800 kits.

As a result, the Ryzen 7 5800X3D turned out to be 27% more expensive than its brother without the «3D» prefix. At the same time, it is a couple of percent weaker in synthetics, but 20-25% better in games. On the one hand, everything is fair — how much you paid extra, so much you got. But given the increased heating of the 3D-shki, it is more attractive, especially for a non-pure gaming PC, that the good old Ryzen 7 5800X looks like. nine0003

Compared to the Core i7-12700K+DDR4 bundle, the review hero is 3% cheaper. But he is 15% worse than her in work tasks and loses by 5-8% in drawdowns in games. Plus, Intel is a little colder, and support for the PCIe 5.0 interface may come in handy sometime. Therefore, in this case, the Ryzen 7 5800X3D flies by.

And finally, the Intel bundle with DDR5 memory is already at least 37% more expensive than the Ryzen 7 5800X3D. At the same time, the same 37% speed is provided in work tasks and from 7 to 24% in games. nine0003

Therefore, in our opinion, here you can either save a little and take the standard Ryzen 7 5800X or pay a little extra for the Core i7-12700K with DDR4.

Therefore, the Ryzen 7 5800X3D at the current price is more of a pioneer for geeks and enthusiasts, charting a new and bright path for AMD. Probably the future belongs to multilayer 3D processors. And it is they who improved the processor in 2020. And imagine this technology in service with the cooler 5nm Ryzen 7000? 16 cores, 32 threads, a hundred megabytes of L3 and DDR5 — we would look at this! nine0373 Intel Core i5-12400 vs. Core i5-11400, Core i5-12600K and Ryzen 5 5600X: price/features

results and impressions from the domestic processor — Technique on vc.


Now more than ever it is important to have a plan «B», so we took Elbrus-8SV to study. The text contains the results and conclusions about the potential of such a solution for our data centers. With photo!


Hello! My name is Maxim, I work as a hardware tester at Selectel Lab. Recently I took the domestic Elbrus-8SV processor for a test. It came with the 2E8SV-EATX motherboard manufactured by MCST. nine0003

As part of the task, I assembled a test server of arbitrary configuration, installed the operating system, configured the network, and ran a benchmark test. I’ll tell you more about the experience in numbers and nuances below. And here I share the main conclusion.

So far, it is not easy to use domestic hardware: there are too many obstacles.

To make work on Elbrus convenient for the end user, you need to invest a lot of effort and time. For example, figure out how to automate processes, recompile necessary programs, decide how to get around existing restrictions. Without a catastrophic need, these investments will not be justified. nine0003

Here the processor is not yet covered by a heatsink.

I will make a reservation that Elbrus 32-C is already more realistic for use in data centers. Most likely, the software will also be “finished”. But will they be able to produce it on a 7nm process? This is a question of the future. MCST was founded in 1992, and Intel in 1968. Perhaps the domestic company will still be able to catch up with competitors.

The following is a basic overview. Perhaps it is not as detailed and scrupulous as it could be. Our main goal was to quickly evaluate the potential of the Elbrus server in Selectel data centers. nine0003

Review navigation:

→ Assembly of test configuration

→ Installation of the operating system

→ Network Settings

→ Review of E2K

→ Programs for OS «Elbrus»

→ Preparation for tests

2222 → Benchmark test results

Building test configuration

Only a board with a processor came to our laboratory (the processor is soldered into the board, without the possibility of upgrading the motherboard) without heatsinks, fans and a case. Therefore, we assembled a custom server on Elbrus. nine0003

As a result, the test config looked like this:

  • Chassis: 2U NVMe

  • Processor: Elbrus-8SV
  • Motherboard: «2E8SV-EATX»
  • RAM: 8 pcs. DDR4 32GB 2933MHz ECC Reg DIMM (MTA36ASF4G72PZ-2G9J3) Micron
  • SSD: 1 pc. Micron 5300 Pro 1TB
  • HDD: 1 pc. ST4000NM0035 Seagate 4TB
  • nine0057 SSD: Samsung 480 GB MZ7Lh580HAHQ

  • SSD: Intel 240 GB S4510

Heatsinks Intel BXSTS100A with active cooling are suitable for heat dissipation. The case was taken for two units from Supermicro 825TQ-R720/R740, as they initially wanted to add GPU and NVMe to the config. True, the plan did not succeed.

Motherboard from MCST.

HDD and SSD drives were connected to test IOPS. The ST4000NM0035 Seagate 4 TB and Micron 5300 Pro 1 TB came to hand. These drives were tested on test cases, so they were selected for comparison of performance with other test configurations. nine0003

In the hope that we can test NVMe SSDs, we connected PCIe NVMe Host Bus Adapter AOC-SLG3-4E4T, but the system did not see NVMe drives. The same thing happened with the Samsung 970 EVO Plus NVMe M.2 SSD. This is about the Elbrus operating system. How the assembly works on other operating systems, I will describe below.

Operating system installation

First, we decided to install the native Elbrus OS on the e2k architecture. The system is installed both from USB-drive and from DVD. It is also possible to install over the network (PXE). nine0003

There is no BIOS in the system, but there is its analogue — «Loader», or boot. In fact, this is a shell that accepts commands and options through the command line.

This is a minus for working in a data center. Usually, the initial configuration of the platform is done in the BIOS, including the Baseboard Management Controller (BMC). The operating system is installed later. Roughly speaking, when ordering a dedicated server.

In the case of a server on Elbrus, in order to configure the BMC, you first need to install the native OS or another system that fits the e2k architecture. nine0003

And now imagine: the data center received 100 servers on Elbrus. How will we configure the BMC? That’s right, first we will install the OS from a USB flash drive on each server and only then configure the BMC. It will take, to put it mildly, a lot of time. Of course, there is an option with a PXE server, which will make the task a little easier, but the BMC will still have to be configured “by hand” through the OS.

Points adds that there are no manuals for setting up the BMC. And then the «dancing with a tambourine» begins. nine0003

As I wrote above, the setting is carried out only after the OS is installed. This is due to the fact that the system does not have a BIOS and this board does not implement a “bridge” between the BMC interface and the “Loader” (analogous to BIOS). Connecting a laptop directly to the BMC Ethernet port did not work. I take out the second tambourine.

After consulting with the MCST specialists, it was possible to set up the BMC. To do this, I connected the BMC and eth0 ports with a patch cord jumper, raised the DHCP server on the eth0 interface and specified a pool of one IP address we needed. After that, the BMC received the address we needed, and we had access to the BMC web interface. nine0003

A data center specialist can configure the BMC only from a laptop directly in the server room, this is not very convenient. We try to automate the work of system engineers with servers as much as possible — this way we prepare servers much faster and eliminate possible human errors.

Network setting

Network setup is similar to other Linux systems.

Overview of operating systems for e2k

I’ll digress and talk about suitable operating systems. There are currently four operating systems that support the e2k architecture. We tried to install everything. nine0003

OS «Elbrus». Installed from USB, 1,256 software packages, unfriendly GUI. There is no network repository, only local. There is no test version of the OS — it does not come with the motherboard and processor.

ALT Linux. The installation was carried out from a USB drive, the interface is nicer than in the Elbrus OS, the number of packages is 17,211 pcs. It is also limited by its repository, but already has a much larger number of development packages — for example, Java, which is not available on the Elbrus OS. Another advantage is the presence of a network repository. To write to USB, a script is included with the OS. nine0003

Astra Linux 8.1, Leningrad release. Installation failed. We clarify the reasons with the OS developers.

ZOSRV «Neutrino». We were not provided with the OS distribution kit, explaining that the Neutrino RTOS and the Neutrino-E RTOS are rarely used on servers. Unlike, for example, on-board industrial equipment that solves problems in real time. According to the manufacturer, this OS supports the e2k architecture.

Programs for OS «Elbrus»

Back to our configuration. After setting up the platform, we got to the Elbrus OS itself. Here we immediately encountered the fact that there are no network repositories, and the stack of compiled programs for the e2k architecture is limited to one DVD that comes with the operating system. In our set there were 1,256 programs, the list can be found at the link. Everything that is required beyond this list will have to be “rebuilt” or written by ourselves.

To solve the problem of compatibility with software for the x86 platform, MCST developed a proprietary binary translator. It works in two modes, somewhat reminiscent of a hypervisor. The first is implemented in Lintel, the second — in «RTC». The latter is running an already running operating system. nine0003

In the attached list, we did not see Docker, Prometheus, Java, and many other packages necessary for the normal operation of the server in the data center. At the moment, only one DBMS is implemented on e2k — PostgreSQL, which will work outside the binary translation mode. The rest of the DBMS and software that is not included with the OS can only be run through a binary translator. It would take a whole team to recompile the missing tools for Elbrus.

nine0002 On the plus side: there are necessary packages for software development, Python, Git, libraries and compilers from MCST.

Preparation for tests

We used the RTC binary application translator and the Ubuntu 22.04 distribution for testing.

Significant disadvantage of «RTC»: performance, which is already small, will be limited. According to tests from different sources, it drops to 30% depending on the software.

I also had to suffer with the distribution kit. After deploying Ubuntu-server 20.04 x86 there was a problem with the system time. The system time lags behind the real time. For server hardware, this is critical. There is an assumption that this is due to the binary translation system. nine0003

Of the connected 32 RAM strips, only 3 are used. According to information from the MCST, only one KPI is used in the binary translation mode (you can see it on the diagram).

Accordingly, all peripherals connected to KPI_1 will not work in this mode.

Benchmark test results

We ran a Geekbench5 benchmark in binary translation mode and got the following results:

It can be seen that while Elbrus in the binary translation mode is trying to catch up with AMD Ryzen 5 2400G in terms of performance. But this is a desktop processor, which was released in February 2018.