Render farm The theoretical part of the selection of iron

The profession of 3D visualizer appeared relatively recently. It is especially in demand in architectural design and interior design. The main tools of the visualizer are a computer, 3D modeling and visualization programs, and work with raster and vector graphics. Almost all applications are resource-intensive, and the faster the “hardware” on which the professional works, the more money he can earn.



This article is devoted to the selection of the "iron" part.

How visualizers work

The work of the visualizer consists of two parts: the creation of a 3D scene and visualization. Working with a 3D scene usually occurs behind a workstation and can be divided into several workers. The process of visualization (rendering) is well parallel to several computers. As practice shows, the software for working with the scene is not particularly demanding on the “hardware” of the workstation, and the emphasis is mainly on the amount of RAM and the video card. Rendering also requires the maximum amount of computing resources. Since the visualization process can take several hours, the error in the final stages can be costly.



Usually, visualizers try to use maximum configurations in workstations, since any rendering assistance is only beneficial. However, in conditions of a limited budget, everyone is trying to squeeze the maximum performance at minimum cost.

')

CPU or GPU?

Now in visualization programs such as mental ray , V-Ray , Brazil r / s , Renderman, the main scenes are calculated on the CPU, with occasional patches of help in the form of a GPU. However, the situation is changing for the better and programs that can use idle graphics card power in 3D visualization calculations have already begun to appear. These include iray .



The video cards have appeared a long time ago, but only now at least some gestures have begun towards using the GPU in resource-intensive calculations. This was facilitated by the promotion of manufacturers of such technologies as CUDA and OpenCL. Perhaps in a couple of years, the load balance will shift towards the GPU, but at this point in time it is the birthplace of the CPU.

The main problems of the render farm

There are only two of them: heat and money, and they are closely linked. There is no perfect configuration that solves all problems at once. Therefore, you need to find a compromise between the cost of the system and its performance.

CPU

The choice of processor as the main “worker” of the farm is the most difficult task. In order to dot the i's, consider a few indicators that we have to take into account.

CPU frequency

Everyone knows the simple truth that the frequency of the processor is directly proportional to the speed of its work. It is not always so. Different processors at the same frequency can show different speeds. This is influenced by such factors as:

• processor architecture
• Number of Cores
• cache size
• CPU temperature

The fresher the architecture, the more productive the processor. History knows only a couple of examples where a more recent architecture at one frequency was slower than the previous one.



The number of cores and the amount of cache memory have a positive effect on the processing speed. Although one and the second a lot happens.



CPU temperature, we consider a little later.

Tdp

TDP is the amount of heat that needs to be removed from the processor, because almost all the electricity it consumes is converted into thermal energy. Intel and AMD differently calculate this figure, and if it is simplistic, then for Intel this is a typical processor heat release, and for AMD it is the maximum. This is very important, since in selecting the cooling system for systems on Intel processors, you need to take at least a little, but with a margin.



The processors of both companies, although they have energy-saving technologies, however, they almost do not work in conditions of 100% of the load that arises during calculations.

Overclocking

With overclocking you need to be very careful. Yes, overclocking gives more performance for the same money, but there are also limiting factors.



The greater the frequency of the processor, the more heat it emits. When overclocked, CPUs easily fly out of the heat pack and their heat dissipation grows exponentially. Since the processors can be quite a long time under load, you need to carefully monitor the performance of temperature sensors. If the critical temperature is reached, the system will be automatically turned off and all work will go on a smack. Also, an additional mechanism is implemented to reset the frequency when the critical temperature is reached, and it may turn out that the processor due to a weak cooling system will never work at the frequency to which you overclocked it.



The performance of the system is not directly proportional to overclocking, when overclocking by 50% you get only about 25% of the performance increase.



Attention should be paid to the fact that not all of the processors on the market can be overclocked. Intel marks its models with the suffix K, and AMD adds the Black Edition. The rest will either have a weak overclocking potential or overclocking will be impossible in principle.

Turbo Boost and Turbo Core

These are two technologies that allow you to temporarily increase the frequency of individual processor cores to speed up work. Neither one nor the second work at 100% loading of all cores. In other words, forget that they exist at all.

Hyper threading

This is Intel's technology, which allows two threads to be executed on one core. For an operating system, a processor with 4 cores and HT technology turned on looks like an 8 core processor. However, do not roll out the lip, virtual kernels give about 20-30% increase in performance. But they are not superfluous.

Cooling system

The cooling system is the key to stability and overclocking efficiency. Currently, several types of cooling systems are used. It can be active or passive, air or water, with or without external conditioning.



Passive cooling is not recommended for high loads. An exception may be server solutions using special enclosures and air-conditioned containment zones. However, the price will be appropriate.



Air cooling has now caught up with water efficiency, and this is at a much lower cost.

Heat pipes

Heat pipes allow you to efficiently and quickly remove heat from the CPU heat distributor cap. Their number is of course important, but you should not chase the radiators with the maximum number of heat pipes, you should always check the effectiveness of CO by practical reviews.



Sometimes manufacturers "cunning", bringing the heat pipes to the bottom of the radiator, they say there is no heat loss from soldering. It should be remembered that it is important not to have direct contact with heat pipes or “extra” soldering, but the area of ​​contact of the radiator with the heat distributor cap, the larger it is, the more efficiently the heat will be removed.

Fans

The size of the fan and their number also does not play a role, since the rate of heat transfer is important for cooling. In other words, the greater the airflow, the better. The same result can be achieved with both a slowly rotating 140 mm fan and four speed 70 mm. Large fans do in order to reduce the noise from the cooling system. You should always monitor the condition of the fans and periodically clean them from dust. Dust greatly reduces the efficiency of the cooling system.

Noise level

If the noise level is not critical for you, the task is simplified. You can safely take the middling from the available on the market, they will still cope at the expense of higher fan speeds. For low-noise systems, selecting the right solution will almost always be a blow to the wallet.



It is worth noting that the higher revolutions of the fans wear them faster, causing an even greater level of noise. Stopping the fans due to sintering of the lubricant in the bearings results in at least a loss of part of the work. It is best to replace the fans in the first case of extraneous noise in the work.

Thermal interface

Good thermal grease can help lower the CPU temperature by an additional 5 degrees (compared to the cheapest on the market). And if the question arises whether it is worth buying something expensive, the answer is always positive. For more efficient operation of the cooling system, it is often necessary to remove it for cleaning. Thermal grease should be replaced with each dismantling of the radiator, and periodically it should be replaced, since constant high temperatures lead to its sintering, which, in turn, significantly reduces its thermal conductivity.

Housing

If you are using a rack, then the choice is obvious - the server case. For all other cases, you need to rather carefully approach the choice of "box". Definitely not a fully enclosed case. After 10 minutes, the system will go off because of the temperature inside. Heat, after all, needs to go somewhere! But a completely “leaky” case is also not an option - dust will get inside faster.



The key to success is a stream of colder air. The body, which allows you to create such a movement of air masses and will be the most effective. The fewer barriers to flow, the better. It should be understood that the ideal solution does not exist, the effectiveness of solutions can be easily reduced due to the banal location of the computer in the room.

Motherboard

The requirements for the motherboard are not many.

VRM

If you overclock the processor, it is best to take motherboards that are at least minimal, but designed for this very overclocking. In overclocking, the CPU consumes much more energy (currents up to 10 Amps can “walk!”) Than during normal operation, so the more stable the power system, the better. Chasing the most expensive thing on the market makes no sense, for a simple number of crushers with a processor, even mini-ITX motherboards can be enough for a simple overclocking. Manufacturers love to boast the number of phases of power. But, as in the case of heat pipes, you need to understand that the banal use of the cheapest components can level the quality and stability of the power supply.

Number of memory slots

A lot of memory does not happen. However, in many entry level motherboards only two slots are installed, which seriously limits the available total amount of installed memory. Now, 4GB modules are optimal for the price, and you can put only 8GB in two slots. For some scenes this may not be enough.



The maximum amount of memory for modern non-server solutions is 32GB. However, to achieve this attractive figure, you will have to use very expensive 8GB modules, which will also have to be searched. Perhaps a cheaper solution would be the Intel platform with a 1366 socket for the first generation i7 processors. These CPUs use a three-channel memory controller, which allows you to install 6 modules with a total capacity of up to 24GB.

RAM

Here the rules are also simple.



There are two main indicators of memory performance: operation frequency and data access latency. For rendering, these figures play almost no role. If you still want to squeeze an extra one percent acceleration for a disproportionately overpriced price, then take the highest speed (= expensive) that is on sale, you can’t go wrong. The best choice at this time is DDR3-1333 (PC3-10600) standard memory.



It is also worth remembering that modules with parity function (ECC) are not suitable for desktop solutions.

Video card

As mentioned above, at this point in time, the GPU helps little in the process of visualization. Relying on the fact that the most popular programs for working with 3D are able to use video chip resources while working with the scene, we conclude that for a workstation a powerful video card will be in place. For primitive threshing machines, the video card is not particularly needed, and you can easily get by with an integrated solution (built into the CPU or the motherboard chipset)



If we assume that tomorrow there will be visualization programs that actively use the "bad power" of the GPU, then it is not a fact that the change in the computing paradigm will be beneficial.



Modern high-performance video chips are very energy-hungry and emit much more heat than processors. Motherboards for gamers, though not cheap, but allow you to install up to 4 video cards at the same time. Now the main question is where to put extra kilowatts of heat in summer? Yes, calculations will be performed faster, but at what cost?

Power Supply

The power of this device should not be taken from the ceiling and not according to the formula “the most powerful or expensive of what is on the market”, but according to the formula “the maximum system consumption multiplied by approximately one and a half times”. Many power supplies can not provide a stable voltage under insufficient load, which is fraught with failures. Additionally, you should pay attention to the efficiency of the power supply. The higher it is, the better.



If you choose devices with passive or active cooling, then preference should be given to active cooling. Although the fan in the power supply and will be an additional source of noise and consumables, but it will provide a more stable operation in conditions of prolonged maximum loads.

Conclusion

I hope this article will help you at least a little to understand the marketing jungle and choose devices that will be optimal in terms of price / quality ratio.



Fast renders to you! Admin