Strengthening science and technology capabilities is considered to be the most important issue for Japan to survive in the world in the future, and this is due to the increase in the budget request for science and technology in the next fiscal year (Reiwa 4) announced the other day. You can see the high need.
Science and technology are divided into various fields, but there are eight major fields that are considered to be the most important in Japan, and the emphasis of science and technology in "life science, information and communication, environment, nanotechnology / materials". There are 8 promotion fields and 4 promotion fields of science and technology of "energy, manufacturing technology, social infrastructure, frontier".Among these, there are important R & D issues 4 and strategic priority science and technology issues 273, most of which are related to our lives.
Main research topics (Cabinet Office)
By the way, in order to solve these problems of science and technology, it is important that the environment (soil) where research and development is actively carried out is in place.In addition, one of the tools that support these scientific and technological capabilities may be the need for a "computer," which can be called a brain.
This time, I would like to introduce the points for selecting a PC for science and technology from among the PCs for various purposes that contact Tegsys, our research and industrial PC manufacturing and sales service.
table of contents
- Before choosing a computer
- How to choose a computer?
- What is computer design?
Before choosing a computer
Computers range from the fastest supercomputers in the world (hereinafter referred to as supercomputers) to standard desktops, laptops, workstations, or tablets.
Speaking of supercomputers, the supercomputer "Tomitake", which has been recognized as having the best computing power in the world for three consecutive terms, was crowded with people, but the website of the Institute of Science explains the features of Tomitake with the expression "ordinary and high performance". increase.This "ordinary and high performance" can be said to be proof that it is not ordinary.If you want to achieve specialized performance only for a specific application, you may continue to pursue it, but it means that you will continue to produce the best performance results for anything. It's quite difficult.Tomitake is a flexible all-round supercomputer, that is, a strong man who can be expected to be used in a wide range of fields.
Just as Tomitake's design aimed at "ordinary and high performance", what do you want to do (what kind of result you want to obtain and how much) when designing the computer required for research and development, what kind of thing You need to be careful and flexible in organizing complex issues, such as what you want to do or how much R & D you have to spend.
How to choose a computer?
For example, specialized software has the necessary conditions for operation (* memory capacity, installation of a high-end video card, etc.), but in research and development, if you want to produce results with the software alone, or if you want to use another program Since complex factors such as whether it is necessary to run are intertwined, it is not always the case that only one software operation requirement needs to be cleared.
First, it is a good idea to organize information on the following items.
Information needed for computer design
* We are good at providing the most suitable computer for the operation without the trouble of searching for a computer that meets these conditions.
What is computer design?
In terms of processing speed (speed), it is necessary to focus on some parts such as CPU, memory, storage, and video card.However, each performance is involved in a complex manner, and consideration must be given to the installation location and temperature, so it is important to consider the overall balance rather than focusing on individual parts.Here are some examples.
It is necessary to pay attention to the clock, the number of cores, etc. from the viewpoint of software specifications, or based on manufacturers such as intel and AMD.
For example, "AMD Threadripper PRO 3995WX" and "Ryzen Threadripper 3990X" have the same number of CPU cores, 64 cores, but 3990X is selected for the purpose of giving top priority to the operating clock.However, the 3995WX has a higher memory channel (memory bandwidth) and upper limit of maximum memory, so which CPU is selected depends on whether memory is important or not.
|Spec.||AMD Threadripper PRO 3995WX||Ryzen Threadripper 3990X|
|Number of CPU cores||64||64|
|Number of threads||128||128|
|Maximum boost clock||4.2GHz||4.3GHz|
|Memory channel (memory bandwidth)||8ch / DDR4 / Up To 3200MHz||4ch / DDR4 / Up To 3200MHz|
Also, depending on "how many cores are secured" when selecting a CPU, the operating speed of a single core (single core) may be faster if the number of cores is not increased.
For reference, if you take the figure on the right (benchmark of software called Adobe Lightroom) as an example, you can see that the benchmark is higher for 64 cores than for 24 cores even with the same architecture.Ryzen 9 etc., which has a fast single core (single core), are under 64 cores, so it also shows that the number of cores is effective to some extent.
* For details, please refer to the benchmark trends of Adobe Lightroom (for free trial lessons).
Number of CPU cores
When consulting with computers for bioinformatics analysis, thermo-fluid analysis, geographic information system analysis, CFD analysis, deep learning, and deep learning, we may receive requests such as "I want to increase the number of CPU cores." ..
When emphasizing the number of cores, the basic idea is to improve the processing speed by parallelization.As mentioned above, CPUs with a large number of cores tend to have a low clock speed, and as a result of prioritizing the number of cores, the processing speed may rather slow down.Therefore, we are making proposals that take into consideration the balance, such as adopting the CPU with the highest number of clocks from the CPUs that are close to the desired number of cores.
Suggestions for inquiries about PCs for CFD analysis:
In general, multi-core CPUs are often effective for CFD analysis, so select a CPU with a larger number of cores.
For example, in response to a request to "adopt Xeon and increase the number of cores as much as possible within the budget", the number of cores that secures the performance and clock of 1 CPU as high as possible, considering the operation effect due to the memory bandwidth difference of the CPU, etc. We also consider optimization of memory capacity that may affect the calculation, future expandability, and power consumption that is a concern due to the increase in the number of cores.Depending on the specifications, some configuration patterns are presented for comparison.
Please be assured that the design will focus on the clock, such as when calculating a single unit.
Depending on the analysis software, there may be a limit on the number of cores depending on the license type.If you are considering a multi-core CPU, please let us know the software information you plan to use.
In addition, if you have any questions regarding the selection of the number of cores, please ask.
On September 2006, 9 (15 years ago), we delivered an e-mail newsletter called Tegara News, and the article "Custom PC of the Week" at that time introduced the AMD Opteron Dual CPU server.Regarding CPU selection at this time, we are proposing a Dual CPU server using AMD Opteron in order to respond to the customer's request "I just want a fast server".
Around that time, the adoption of Opteron was increasing even at major manufacturers, and Opteron's market share was increasing considerably.It was architecturally designed for SMP specifications, the Opteron core policy was a specification that disliked the look-ahead penalty, and overall speed. The point of selection was that it was designed to be hard to fall off.
In terms of speed, the latest Intel Core architecture at that time is also capable of competing sufficiently, but due to the stability problem of FB-DIMM and the judgment that the server needs a dead part as much as possible, it became AMD Opteron. I did.
For some time, we have been designing machines by comprehensively judging customer requests and evaluation of parts.
As a trend in recent years, there are increasing cases where "larger memory capacity" is required, and we have a track record of proposing machines equipped with 2TB and 3TB of memory for research and development.Computers for which large capacity is desired are widely used for finite element method simulation (Comsol) calculation, Synopsys Rsoft FullWave parallel calculation machine, CAE (multiphysics) machine, image processing / image analysis, deep learning, etc. ..
■ Example of Comsol
Comsol is a simulation tool for modeling designs, processes and devices that accelerate scientific and technological innovation.Finding solutions to a wide range of physical phenomena such as structural mechanics, acoustics, fluid mechanics, electromagnetics, heat transfer, and chemical engineering based on the latest numerical analysis methods, which is the most important for quick calculation. The spec is memory.
What hardware do you recommend for COMSOL Multiphysics
When selecting memory, it is necessary to consider the selection of CPU and motherboard (the number of memory that can be physically installed), but the configuration may consider the memory channel of the installed CPU, or select the CPU from the desired memory capacity. Or, we will decide the specifications according to the customer's conditions, such as a proposal with a configuration (Massive Core EX manufactured by our company) that is superior in terms of the number of cores.
■Which is faster, RADIC 2 CPU or TEGSTAR Massive Core EX? (For COMSOL)
However, the situation will change when the budget is 100 million yen. RADIC is not the fastest in terms of memory channels depending on the configuration, and does not take advantage of each memory bus, which is the advantage of 2 CPUs.When the memory channel is maximized, physical memory cannot be added later, so memory replacement is required to increase the capacity.
In this way, we will consider memory bandwidth and future expandability, so please feel free to contact us.
As another example, some solvers used for analysis have a high memory load, so if there is a limit such as "memory capacity determines the calculation size" in the absolute condition when selecting a computer, use it. The memory capacity will be selected based on the basic premise.
It is also recommended to secure more main memory capacity than the video memory capacity for deep learning applications.
Since it is an area for storing data, select the capacity according to the amount of data to be handled, taking into consideration the case where writing and reading speeds are required.
For example, in the case of NGS (next generation sequencer), handling of a large amount of data is a prerequisite, so the storage capacity may inevitably be required to be in tera units, which is extremely large.Since there may be physical restrictions on the amount of storage that can be installed, it is important to estimate the size of the data to be handled in advance in the design.
■ Examples of ArcGIS Pro and Metashape
The required specifications of ArcGISPro and Metashape are similar. The amount of data handled, such as photos taken with UAVs, is large, and data analysis is also required, so storage speed is also important.In such a case, M.2 SSD (NVMe specification) is adopted in consideration of both speed and capacity.
As another example, even if the image size itself is not large, random performance and write resistance are also important when handling a large number of images (frequent daily access), so NVMe support for U.2 connection We have selected products.We also propose NVMe-compatible SSDs for improving read and write speeds during multitasking.
About video card
Due to the dramatic improvement in the performance of video cards, the emphasis on video cards is increasing.It is required for inference processing, AI, simulation, deep learning, image processing, research on interaction functions, etc.
Quadro and NVIDIA RTX series are also well-established.In addition, NVIDIA's TESLA and A series are also very popular in AI and HPC.
When selecting a video card, the number of cards that can be physically mounted differs depending on the type of video card, there are differences in exhaust (internal exhaust and external exhaust models), and if there is power, the power consumption will also increase accordingly. There are some points to consider, such as when it is necessary.
In addition, due to the recent shortage of semiconductors, the availability of video cards is unstable, and depending on the model, long-term supply may be difficult (GeForce is a standing position for gaming, so distribution is limited. There is).If you are uncertain about the selection, please feel free to contact us.
■ Lightning Chart JS example
Since GPU acceleration is used to draw the chart, it is possible that the dependency on the video card will increase.The following models are equipped with the NVIDIA RTX series, which can consistently perform from chart development to operation tests including GPU acceleration.
As another example, when using the distribution system Zoom or Teams, it is possible that a certain amount of GPU load will occur due to GPU acceleration, so consider the performance to some extent and use the video card. Has been selected.
[Feature Article] The future of the international community opened up by STI for SDGs However, as I introduced, in the future society, science and technology innovation will be the driving force for economic revitalization, and I think that science and technology will change and expand the future society better.And the computer situation required for research and development will continue to evolve as if to adapt to changes in society.
So far, I have given some points about the selection of parts as an example, but I sincerely hope that the optimal computer designed in this way will accelerate research and development.
In addition, we are a custom-made PC production service "TegusisNot only "", but also an overseas product procurement and consulting service for R & D "UniposWe provide a service called ".Only because of Tegara, "R & DPurchasing overseas products for customers and manufacturing custom PCs optimized for their useWe are also developing a one-stop service called ".
We are constantly refining our computer design and manufacturing technology to help accelerate our research and development.If you have any problems in selecting a computer, please feel free to contact us.