The de-rating issue of electronic load
How to procure authentic specifications? Don’t waste your procurement budget!
Keywords：Artificial Intelligence Server (AI Server), Modular Hardware System-Common Redundant Power Supply (M-CRPS)
Foreword: AI computing power drives the demand for electricity to increase significantly
AlphaGo defeated Lee Sedol in 2016; AlphaGo Master defeated Ke Jie in 2017; and AlphaGo Zero defeated AlphaGo Master 89:11 at the end of 2017, establishing a milestone for artificial intelligence to defeat the human brain. Open AI released ChatGPT in 2022, allowing people to see that generative AI can be implemented in daily applications. Since then, the computing power demand of artificial intelligence (AI) has led to the explosive growth of AI servers. Computing power must rely on the support of electricity, which is the biggest difference between the chess king and AlphaGo. AlphaGo consumed megawatts of power to play against the chess king, while the power consumption of the chess king against AlphaGo was only 20 watts (one fifty thousandth of that of AlphaGo).
Modular Hardware System -Common Redundant Power Supply
The Open Compute Project is a non-profit organization founded in 2011 by Facebook, Intel, Rackspace and Goldman Sachs. The current main contributors are AMD, Dell, Google, HP, Intel, Meta, and Microsoft that mainly aim to create open data center hardware architecture. At the time of writing this article, the version of the Modular Hardware System -Common Redundant Power Supply M-CRPS in OCP is Version 1.04 RC3 February 22nd, 2024. At the OCP Summit held in Lisbon, Portugal on April 24-25, 2024, Meta proposed the outlook of ORv3 High Power Rack (HPR) Ecosystem Solution, the following is an excerpt comparing ORv3 with ORv3 HPR.
|
|
ORv3 |
ORv3 HPR |
|
Rack Depth |
42” |
48” |
|
Busbar Capacity |
18kW+ |
92kW+ |
|
PSU Shelf (Power Supply Unit) |
18kW (6*3kW PSUs) |
33kW (6*5.5kW PSUs) |
|
BBU Shelf (Battery Backup Unit) |
18kW (6*3kW BBUs, 90sec) |
33kW (6*5.5kW BBUs, 90 sec) |
|
PSU/BBU Shelves/Rack |
2/Rack |
3/Rack + |
Table 1: Comparison between ORv3 and ORv3 HPR
From Table 1, it can be seen that the single server power supply specification is increased from 3kW to 5.5kW (single module). One shelf has 6 modules, and one rack ORv3 has two shelves (one shelf 3kW*6 =18kW); ORv3 HPR is three shelves (one shelf 5.5kW*6=33kW), and in the test of power supply, dynamic load is an important test item. From this specification, it can be seen that the electronic load of 4kW and 6kW is the basic requirements for single module testing for ORv3 and ORv3 HPR, but most engineers do not understand the truth of electronic load specifications. The following will explain the tricks of electronic load specifications for you.
Understanding the tricks of electronic load specifications
Electronic loads are like flight simulators. Flight simulators provide various flight scenarios to train pilots. Flight simulators can simulate normal takeoffs, landings and abnormal situations through route selection. Similarly, the main function of the electronic load is to simulate various load scenarios that the power supply will encounter to confirm the design goals and adaptability of the power supply. Electronic loads, through the setting of the current profile, establish basic operating modes such as constant current (C.C), constant voltage (C.V), constant resistance (C.R) in the specifications, and sink the output power of the power supply.
Electronic loads are widely used. In addition to testing power supplies, battery chargers, batteries, solar panels and other power devices, they can also test components that are driven by current and carry current, such as various types of current protection relays, switches, relays, fuses, cables, etc. The detailed test items of the power supply include the power supply load regulation rate, efficiency and temperature rise under different loads, etc., therefore, where there is a power supply, a load is required for testing.
The specifications of electronic loads usually boast the best state. The best state is based on specific excellent working conditions [warm-up 30 minutes to 60 minutes; power laboratory temperature requirement (23.0 ± 2.0) °C]. The specifications of electronic loads will be affected by factors such as operating voltage and operating temperature, resulting in de-rating.
The following is the derating curve in the brand C 632XA series specifications. The left picture in Figure 1 shows that after the power laboratory environment exceeds 25°C, when the operating temperature is 40°C, the rated power capacity drops by 17% (only 83 % are available), and when the working voltage is 1200V, the rated power capacity drops by 50%.
Figure 1: The curves on the brand C catalog, the left is the temperature derating curve; the right is the voltage derating curve
Figure 2 shows the derating curve of the 150V/600V model of GW Instek PEL-5000G series. The 4kW and 5kW models can provide rated power at 40°C, while the power rating of the 6kW model at 40°C decreases by 14% (86% available, which is 3% higher than the same category models).
Figure 2: Derating curve of PEL-5000G 150V/600V models
Figure 3 shows the derating curve of the PEL-5000G 1200V model. The rated capacity dropped by 40% (60% remained, which is 10% higher than the same category). From these derating specifications, if you are applying for a 1200V application, the power of the brand C is only 90% of that of GW Instek products, so you can have a quantitative benchmark in your mind by how much it should be cheaper than that of GW Instek products. You may be curious. Why GW Instek products can achieve rated power at 4kW/5kW? The reason is that there is usually one set of heat sinks for every 1kW of electronic loads.
Generally, manufacturers only use 4 or 5 sets of heat sinks for 4kW/5kW models. However, GW Instek utilizes a 6kW heat dissipation design for 4kW/5kW, using 6 sets of heat sinks. Therefore, the heat dissipation capacity at 4kW/5kW is better, so it can maintain rated power. In addition, the single-set heat dissipation design is also better than that of the brand C, so the derating of 6kW is still better than that of the same category.
Figure 3: Derating curve of PEL-5000G 1200V model
How to meet the instantaneous transient load demand at a lower cost?
In applications such as testing the dynamic load specifications of power supplies or testing fuses and circuit breakers, there are often short-term, high-current application requirements. Of course, you can purchase higher-power electronic loads to cope with these demands. However, this method requires additional budget, and the extra budget spent is of no use in general applications. The 1.5-fold turbo mode of GW Instek PEL-5000G allows users to meet these testing needs with a reasonable budget. The 400A model can be upgraded to 600A; 500A model can be upgraded to 750A; and 600A model can be upgraded to 900A. Why does GW Instek electronic loads have turbo mode function? The reason is that we use 20% more MOSFETs than our competitors to provide this 1.5-fod (instantaneous 2 seconds) capability.
Conclusion: Understand the specifications to get the most out of your investment
GW Instek's dedication and materials make the PEL-5000G series electronic loads the most effective for customers. Either full power or derating, our products allow you to obtain authentic power within your budget, and furthermore, the turbo mode can meet instantaneous applications. Only by understanding these specifications can prevent you from overspending your budget when purchasing or feeling the pinch during applications.
Figure 4: Illustration of GW Instek Turbo
The recommended equipment for verifying server power supply is as follows：
Figure 5: PEL-5000G electronic load with turbo mode
Figure 6: ASR-6000 series AC/DC power supply required for M-CRPS testing
Contact Us:
Diana
Digital Service Specialist
E-mail: diana@goodwill.com.tw