A generator for data center duty turns the hall's IT load and the UPS ride-through into a single set rating that the site can trust during a grid loss. Tesla Power measures that rating from the hall's metered load rather than from a nameplate guess, then states the ambient and the altitude basis so two supplier quotations can be compared step by step before an order is placed. The exercise settles three things at once: frame size, the duty class printed beside the rating, and the fuel autonomy the tank has to hold through the longest planned outage. Build it from a recording rather than from a walk round the plant, because a walk round misses the peaks that decide the frame. Three loads decide the frame before the average does, and none of them is the largest in kW terms. Ambient removes capacity before any load is connected, and that removal sits in the derating chart rather than on the plate. A standby profile needs one figure a prime profile does not: the outage length the site plans for, because a two hour ride-through and a three day ride-through size the tank and the cooling margin completely differently, and the steps below follow the recording rather than the estimate.
- Log the current on each circuit for a week, because a nameplate total overstates the steady load and hides the peak that decides the frame.
- Add the largest starting load to the running total, since that step decides the frame more often than the average day does.
- Correct the figure for the ambient at the set, which sits above the site average and moves the plate figure down.
- Keep the log beside the range of frames the factory builds, so a supplier quotes against a measured figure rather than an estimate.
What does a generator for data center need in the profile?
The profile decides three things at once: the frame size, the duty class printed beside the rating, and the fuel autonomy the tank has to hold. None of the three follows from a connected load total, because that total describes equipment at rest rather than equipment in use, and Tesla Power reads the recording instead. A hall with 800kW of installed IT may draw 420kW in the afternoon and 800kW for nine minutes after a chiller restart, and the second figure is the one that decides the frame. A generator for a data hall that is sized on the average runs at light load for most of the year.
The reason is that a diesel engine responds to a step rather than to an average. A frame chosen on the average runs at light load for most of the year, and a frame chosen on the nine minute peak carries capacity it touches twice a day. The profile resolves that argument, because it shows how long each figure lasts and how often it returns. Most generator manufacturers quote a single figure and leave the profile to the buyer, which is where two offers stop being comparable.
How is the IT load turned into a set rating?
Turn it from a recording rather than from a walk round the plant. A clamp meter read every fifteen minutes for a working week gives 672 samples, and four of them decide the answer: the highest fifteen minute figure, the highest one minute figure, the longest run above 70% of the peak, and the lowest sustained figure during a shift. A UPS sits between the hall and the set, and it hands over the load over a few seconds rather than instantly, which is why the step matters more than the steady draw.
Where a recording is not possible, the site log and the production schedule carry most of the same information on the duty. Shift patterns show when the large chillers run together, the maintenance book shows how often a compressor restarts, and a set in parallel operation shares the step differently. Yangzhou Tesla Power asks for both when a tender names a figure without a recording, because the two together close most of the gap that a nameplate total leaves open.
| Sample | What it shows | Why it changes the answer |
|---|---|---|
| Highest fifteen minutes | The steady ceiling the set holds | Sets the rating before any correction |
| Highest one minute | The step taken at transfer | Decides the duty class on the plate |
| Longest run above 70% | How hard the engine is worked | Drives the cooling and the service interval |
| Lowest sustained figure | How far the load falls | A run below 30% loads the exhaust |
The fourth line catches more halls than the other three together. A set that carries 800kW for an hour and 120kW for six leaves unburned fuel in the exhaust, and the deposit shows on the injector tips long before it shows on the hour meter, which is why the lowest sustained figure belongs in the profile.
Which loads decide the frame before the average does?
Three loads decide it, and none of them is the largest in kW terms. The first is the chiller that starts direct on line, because it draws about six times its running current for a few seconds. The second is any group of loads a control system restarts together after a transfer, since the group behaves as one step and the frame sees the total at once. The third is a large drive with a charging cycle, because the inrush lasts long enough to pull the voltage down on a set for standby duty.
A 500kVA frame at 0.8 power factor carries 400kW, so 500kVA × 0.8 = 400kW is the figure the profile has to stay under at that factor, and a 1000kVA frame carries 800kW on the same basis. An 80kW chiller motor started direct on line adds roughly 500kW of apparent demand for two seconds, which the alternator meets by dipping rather than by delivering. The 500kVA silent set with the ATS option carries that step on the same 0.8 factor.
- A direct on line start draws about six times running current for two to four seconds.
- A star delta start cuts that to about two and a half times, at the cost of a second step.
- A soft starter holds the figure near three times and removes the second step.
The dip recovers as the excitation responds, and the question is whether the hall tolerates it. The commissioning sheet sets out the tolerance for common loads, and Chinese generator suppliers usually quote a dip figure only when a buyer asks for one.
How does ambient change the figure on the plate?
Ambient removes capacity before any load is connected, and the removal is stated in the derating chart rather than on the plate. A set rated at 40 ℃ loses roughly 3% of its figure for every 10 ℃ above that point, so a 500kVA frame loses about 6% of its figure between 40 ℃ and 45 ℃. The hall then plans against the corrected value rather than the plate figure, which is the difference between four offers that can be compared and four prices that cannot.
Altitude compounds it. Air density falls with height, and a naturally aspirated engine loses about 4% of its rating per 1000m, while a turbocharged engine loses rather less because the compressor compensates for part of the drop. Tesla Power Equipment states both bases on the offer, because a correction applied in sequence gives a different answer from one added on top. The figure belongs on the offer sheet beside the rating, where a buyer can read it against the site conditions.
What does the set need for a long outage?
A standby profile needs one figure a prime profile does not: the length of the outage the site plans for. A set that carries the hall for two hours needs a tank and a cooling margin sized on two hours, and a set that carries it for three days needs both sized on three. The rating class follows the same reasoning, because a standby class allows a limited number of hours a year and no continuous overload.
The hours figure decides more than the tank. A hall that expects 40 hours a year reaches the calendar limit of a warranty schedule long before the hour cap, and a hall that expects 2000 hours closes the cap inside a season. Tesla Power writes the expected hours into the offer beside the class, so the two lines are read together rather than as a rating and a note, which keeps a generator for the data centre honest about its duty.
Where does the sizing go wrong on export sites?
The first error is a load list copied from an earlier project in a cooler country. The equipment matches, the ambient does not, and the correction is left off the figure on a genset for a data centre as readily as on a set for a mine site. The second is a proposal written from a standard sheet, which quotes a reference figure without the site basis and therefore reads higher than the site can use on the day the load steps. Both errors are invisible in a price comparison.
The third is a profile that stops at the main board and ignores the loads a tenant adds two years after handover. A chiller fitted later changes the profile, and the offer is then read as a promise rather than as a figure with a basis. Suppliers who ask for the load list before quoting close that gap, and the ones who quote from a catalogue leave it open, which is why a generator in a data centre should be quoted against a recording.
A tender that names the figure, the ambient, the altitude and the expected hours gives every bidder the same starting point. That is the difference between four offers that can be compared and four prices that cannot, and the difference shows at the transfer rather than at the invoice.
Which records prove the load was measured rather than guessed?
Three records carry the proof. The site recording, with the date and the instrument named, shows what the load actually did. The commissioning sheet shows the figure the set carried at handover and the ambient at the time. The annual load test shows the figure a year later, and the two together show whether the profile has drifted.
The value of the third record appears slowly. A hall that adds load over five years reaches the frame limit quietly, and the first sign is a trip on the largest motor rather than a rising bill. A load test at 80% of the rating for two hours each year catches the drift while the answer is still a setting rather than a new frame, which is the practice our own test bay follows for every export order.
How should a buyer compare two offers on the same profile?
Compare the basis before the price. Two offers on the same profile should state the same four figures: the rating, the class, the ambient and the altitude. Where one offer states a figure without a basis, the comparison stops there, because a higher number on a lower basis is not a larger machine. The ISO 8528-1 rating definitions set out the classes both offers have to name, and the NFPA 110 rule set for emergency supply adds the monthly test schedule a hospital site follows.
A profile for a data hall adds a fifth figure, which is the step the uninterruptible supply applies when it hands over. A profile for a mine site adds a different fifth figure, which is the restart group after a power cut. Both belong on the enquiry, and both change the class rather than the frame, which is why a supplier should answer the four items inside the quotation.
Ask for the corrected figure in writing, and ask which document the supplier will issue with the set. A supplier who answers with a schedule can be compared with a second supplier; one who answers with a number can only be compared on price. Tesla Power answers the four items inside the quotation, and the reply arrives as a schedule for that reason.
What does the set look like on a container build?
A container build carries the same profile as an open frame, and the enclosure changes the correction rather than the load. The intake sits lower, the air is warmer inside the box, and the louvre free area sets the volume that reaches the radiator. The 1000kW container build for standby duty carries its derating chart beside the plate for that reason, because a generator for a hospital wing shares the same enclosure logic as a data hall.
The container adds one more figure to the profile, which is the ambient inside the box rather than outside it. A site at 40 ℃ outside may see 47 ℃ at the intake, and the correction follows the higher figure. generator for a site with a container build needs that number on the enquiry, because the answer changes by about 2% per degree at that point, and the practice came from a coastal site where the box ran 8 ℃ above the yard for most of August.
Our own test bay measures the intake figure with the doors closed, which is the condition the set meets in service rather than the condition it meets on a bench, and the test bay at the works records that figure for every order. We record the figure with the load bank at 80% of the rating and the ambient noted beside it, and the record travels with the set so the buyer can read it against the site.
Which paperwork proves the correction was applied?
The offer sheet carries the corrected figure, the data sheet carries the basis, and the nameplate carries the class. Three documents, and a buyer who reads only the third sees a number without a reason. conformity assessment for the data centre then checks that the plate matches the file rather than the quotation, which is the step a customs inspection performs first.
Ask for the correction as a line on the offer rather than as a paragraph in a covering letter, because a line survives a change of salesperson and a paragraph does not. Tesla Power Equipment prints the ambient and the altitude beside the rating for that reason, and the practice has settled more than one disagreement at handover when two offers were read against the same recording.
The plate itself carries the class, the frequency and the power factor, and each field is checked at the port against the file. A plate that omits the class leaves the rating open, which is the gap a customs inspection finds first. Send the load list and the site sheet to our engineering support desk, and the reply comes back as a schedule with the correction already applied and the quotation built on the measured figure.
Frequently asked questions about a generator for data center
A working week covers most halls, because it captures the weekday peak, the weekend trough and at least one restart cycle. A single day misses the shift pattern that produces the peak. Where a week is impossible, record the two hours around the largest restart and the two hours after a transfer, which still catches the step the frame is sized on.
It should, and 20% covers most steady halls. The margin is a decision rather than a rule, because a set that runs at 45% for years meets its own failure mode. Take the margin from the profile: a hall with a large chiller start needs more than a hall with a resistive load, and the margin should sit in the recorded figure rather than on the plate.
It is the share of a connected group that runs at the same moment, and it is applied group by group rather than to the whole hall. Lighting runs at about 0.9, socket outlets at 0.4 and cooling sets at 0.3 on a busy shift. A total built without factors overstates the load by a wide margin and oversizes the frame that follows.
It does, because the alternator is rated in kVA and the load is often quoted in kW. A 400kW load at 0.8 power factor needs 500kVA, and the same load at 0.95 needs less capacity on the same alternator. The factor is measured rather than assumed on a hall with drives, and it is written into the offer beside the rating.
They can, and the arrangement gives redundancy that a single frame cannot. The cost sits in the switchgear and in the load sharing controller, and the maintenance work doubles. Two sets also change the step behaviour, because the second frame starts against a load the first already carries, which is the trade a buyer should weigh before quoting.
The current recorded at the rating point during the commissioning run, taken with the ambient noted beside it. A figure without the ambient cannot be compared with the factory test. Tesla Power issues that record with the set, and it is the document a warranty question is read against when the load has drifted two years after handover.