A gas engine vs gas turbine decision for a data center turns on block size, duty and site. Where the requirement is a few tens of megawatts, phased, or run at part load, a block of reciprocating gas engines usually fits better. Each unit is small, a reserve unit adds little installed capacity, and efficiency holds up as load falls.
Where one aeroderivative package can carry the load near full output, or plot area is the constraint, the turbine usually wins. The engine shortlist here is the mtu 20V4000 GS, INNIO Jenbacher J624, Wärtsilä 20V34SG and Cat G3520H or G3520K. The turbine shortlist is the GE Vernova LM2500XPRESS and LM6000. Compare them model by model, with rating type, heating-value basis and frequency beside every figure.
Key Takeaways
- Block size and duty decide the choice. Output per unit, efficiency basis, start, part load, emissions path and footprint are the six axes, and neither technology wins all six.
- The four engine platforms span 2.5–9.4 MW per unit. At 60 Hz, from the Cat G3520H's 2,490 ekW continuous to the Wärtsilä 20V34SG's 9,388 kW, rating type unstated.
- One aero package carries 34–57 MW ISO net. LM2500XPRESS+G4: 34.4 MW at 60 Hz; LM6000 PF+ Sprint: 56.9 MW; both simple cycle, per GE Vernova factsheets.
- A spare unit costs engines far less. With one spare, mtu 20V4000 blocks install 111% of a 25 MW requirement and 103% of 100 MW; LM2500XPRESS, 275% and 138% (calculated).
- Part load favors engine blocks. At half load, EPA puts a typical lean-burn engine 8–10% below full-load efficiency and turbines 15–25% below, so deep turndown favors many units.
The six axes that decide a gas engine vs gas turbine choice#
This comparison runs on six axes: output per unit and per block, efficiency and its basis, start and ramp, part-load behavior, the NOx control route, and footprint and delivery. The table shows where each axis tends to point, using the model figures below.
Decision axes: where each points and what to ask the supplier for#
| Axis | Leans to an engine block when | Leans to an aero package when | Ask for |
|---|---|---|---|
| Output per unit and block | The requirement is phased or a few tens of MW, so 2.5–9.4 MW units divide it finely | One 34–57 MW package covers the load at steady output | Rated output with rating type, frequency and power factor |
| Efficiency basis | Simple-cycle electrical efficiency matters most | Combined cycle is planned (LM6000 PF+ Sprint 2x1: 55.0% LHV) | LHV or HHV named, reference conditions, included losses |
| Start and ramp | Full load is needed in under 5 minutes | A 5-minute start meets the requirement | Published start-to-full-load time and the conditions behind it |
| Part load | The plant will run well below full output for long periods | The package will run near full output most hours | Fuel consumption at 100%, 75% and 50% load |
| NOx route | Limits can be met by lean-burn engine-out levels or SCR | DLE meets the limit, or SCR is planned anyway | The NOx level of the offered build and the aftertreatment scope |
| Footprint and delivery | Land is ample and capacity arrives in stages | Plot area per MW is the binding constraint, or one fast-installed package covers the load | Package dimensions, installation scope and the contractual delivery and operating dates |
Reciprocating engines, typically used for backup and standby power, were "becoming increasingly popular for larger utility-scale power generation applications", an EIA analysis said on 19 February 2019. As of November 2018 the average engine generator had 4 MW of capacity, against 56 MW for natural gas combustion turbines.
Compare whole plants on matching boundaries. Neither side should arrive as a bare generator set against a complete plant scope, and an offered used unit should be distinguished from a new-build reference. The data-center power procurement roadmap places this choice inside the wider connection program.
Reciprocating gas engines: the four models that matter here#
A reciprocating gas engine burns gas in cylinders whose pistons turn a crankshaft and generator. The four platforms below are gas engine generator sets, and each model has a SecondWatt dossier. The gas-engine category holds the wider range.
Four gas-engine platforms at a glance: electrical output with rating type, efficiency basis, start time, NOx path (OEM datasheets, revision noted)#
| Model | Electrical output, rating type, frequency | Efficiency or heat rate, basis | Start to full load | NOx path | Genset L × W | Document |
|---|---|---|---|---|---|---|
| mtu 20V4000 GS L64FNER | 2,520 kWel, 60 Hz, pf 1.0; rating type not stated | 43.6% electrical; references ISO 3046/I-2002, heating value not stated on the sheet | Not on the sheet | NOx < 500 mg/Nm³ build | 7,250 × 2,000 mm | mtu 3237291, Edition 01/24 |
| Jenbacher J624 | Up to 4,477 kW, 60 Hz, natural gas, at generator terminals; rating type not stated | Up to 46.8%; energy input references DIN ISO 3046 / DIN 6271, heating value not stated | Not stated for the J624 | Data based on 250 and 500 mg/Nm³ NOx at 5% O₂ | 570 × 100 in | J624 page |
| Wärtsilä 20V34SG | 9,388 kW, 60 Hz, pf 1.0, at generator terminals, 25 °C; rating type not stated | 48.8%; heat rate 7,374 kJ/kWh, basis not named | Under 2 min (fast start); under 5 min (regular) | Lean burn; catalyst designed per project | Not listed | 34SG page |
| Cat G3520H | 2,490 ekW continuous, 60 Hz, 0.8 pf (High Efficiency, no pumps) | 8.05 MJ/ekW-hr at 100% load (1.0 g NOx build), basis not named | Not on the sheet | 0.5 or 1.0 g/bhp-hr NOx builds | 7,672 × 2,173 mm | Cat LEXE1552-03 (06/23) |
| Cat G3520K | 2,552 ekW continuous, 60 Hz, 1.0 pf (High Efficiency, no pumps) | 46.3% genset electrical (1.0 g NOx build), basis not named | Not on the sheet; the High Response build (G3520K HR) can "start, take on full load, and sync to the grid in 4.5 minutes", in Caterpillar's own release, Dec 2024 | 0.5 or 1.0 g/bhp-hr NOx builds | 7,375 × 1,830 mm | Cat LEHE22852-01 (11/25) |
Rating types differ across the table. The Cat sheets carry continuous ratings, while the mtu, Jenbacher and Wärtsilä figures carry no rating type, so ask each OEM which duty its number supports.
The heating value behind each engine efficiency goes unnamed in all five documents; the Cummins and GE Vernova sheets below state LHV. The Jenbacher figures are also "up to" values, calculated across "various gas quality levels and ambient conditions".
Part-load operation runs "down to 35%" on mtu's Series 4000 gas units, and their fast-start option is listed at "< 120 sec", end point unstated. NOx of 250 or 500 mg/Nm³ at 5% O₂ is reached "without external aftertreatment", according to mtu's Series 4000 gas page.
Aeroderivative turbines: LM2500 and LM6000 class#
The combustion turbine chapter of EPA's March 2015 CHP catalog says aeroderivative gas turbines for stationary power "are adapted from their jet and turboshaft aircraft engine counterparts". For readers arriving turbine-first, the aeroderivative and industrial turbine comparison covers the larger frame and industrial classes.
The figures below are ISO net simple-cycle ratings on natural gas from GE Vernova's LM2500XPRESS factsheet and LM6000 factsheet. They include inlet and exhaust losses and exclude balance of plant. Package dimensions are from the LM2500 and LM6000 product pages.
Aeroderivative reference figures: ISO output, heat rate, start time (OEM datasheets)#
| Model | ISO simple-cycle net output, frequency | Net heat rate and efficiency, LHV | Fast start | NOx path | Package L × W | Document |
|---|---|---|---|---|---|---|
| LM2500XPRESS+G4 DLE UPT | 34.4 MW ISO net at 60 Hz and 34.5 MW at 50 Hz on the factsheet; the product page shows 34.5 MW | 8,563 Btu/kWh, 39.8% at 60 Hz and 8,669 Btu/kWh, 39.4% at 50 Hz on the factsheet; the product page shows 8,669 and 39.4% | 5 min | DLE; +G5 variants listed at 25 and 15 ppm NOx | 28.5 × 3.04 m | GEA35746 (09/2025) |
| LM6000 PF+ Sprint | 56.9 MW ISO net, same figure at 50 and 60 Hz | 8,328 Btu/kWh, 41.0% | 5 min | Not stated on the factsheet | 20.43 × 4.1 m | GEA35743 (09/2025) |
| LM6000 PC Sprint | 51.1 MW ISO net, same figure at 50 and 60 Hz | 8,585 Btu/kWh on the factsheet, whose 9,068 kJ/kWh converts to 8,595 (calculated), the product-page figure; 39.7% | 5 min | Not stated on the factsheet | 20.43 × 4.1 m | GEA35743 (09/2025) |
Aeros lead on output per unit and on plot area. On its listed 28.5 × 3.04 m package, an LM2500XPRESS takes about 2.5 m² per MW (calculated).
1.5 m² per MW — the LM6000 PF+ Sprint package on its listed 20.43 × 4.1 m outline, against 5.3–8.2 m² per MW for the engine gensets above (calculated).
The boundaries differ, since GE Vernova lists package dimensions and the engine OEMs list genset dimensions. Both figures exclude the spacing between units and the balance of plant.
Aeros also extend into combined cycle, where the LM6000 PF+ Sprint 2x1 reaches 55.0% LHV; the simple-cycle and combined-cycle guide sets out the added plant.
Block architecture: N units plus reserve#
In block architecture, N running units carry the requirement and at least one more is held in reserve, all feeding shared switchgear. Where the generator voltage is below the site's distribution voltage, a step-up transformer follows, and any emissions control sits on the exhausts.
Large engine plants follow the same pattern. Pearsall Power Plant near San Antonio, Texas, came online in 2010 with 24 engines of 8.4 MW capacity each. EIA's February 2019 analysis says it "can provide 25% of total generating capacity within two minutes" and reaches full output within eight minutes.
The arithmetic shows what that buys. Below, each unit's datasheet rating is used as a screening input; site derates, auxiliary loads and rating-basis differences are ignored, so these are calculations, not site ratings.
Holding 25 MW and 100 MW with one unit out: datasheet ratings at 60 Hz, before site derate (calculated)#
| Unit (rating used) | 25 MW: N | 25 MW: installed with N+1 | 25 MW: installed ÷ requirement | 100 MW: N | 100 MW: installed with N+1 | 100 MW: installed ÷ requirement |
|---|---|---|---|---|---|---|
| mtu 20V4000 GS L64FNER (2.520 MW) | 10 | 27.72 MW | 1.11 | 40 | 103.32 MW | 1.03 |
| Cat G3520K (2.552 MW continuous, 1.0 pf) | 10 | 28.07 MW | 1.12 | 40 | 104.63 MW | 1.05 |
| Jenbacher J624 (up to 4.477 MW rated output) | 6 | 31.34 MW | 1.25 | 23 | 107.45 MW | 1.07 |
| Wärtsilä 20V34SG (9.388 MW) | 3 | 37.55 MW | 1.50 | 11 | 112.66 MW | 1.13 |
| LM2500XPRESS+G4 (34.4 MW ISO net) | 1 | 68.8 MW | 2.75 | 3 | 137.6 MW | 1.38 |
| LM6000 PF+ Sprint (56.9 MW ISO net) | 1 | 113.8 MW | 4.55 | 2 | 170.7 MW | 1.71 |
10% — the share of a 25 MW requirement that one spare mtu 20V4000 GS or Cat G3520K unit adds, against 138% for one spare LM2500XPRESS (calculated).
At 100 MW, installed capacity with one spare is 103–113% of the requirement for the engines against 138–171% for the aeros. The same granularity lets a block grow with demand, one unit at a time. A documented example is the mtu 20V4000 25 MW gas block, on the power-block hub.
The ratings in the screen sit on different bases. The aero ratings use ISO conditions of sea level and 59 °F (15 °C), while the Wärtsilä rating assumes 25 °C. The mtu, Cat G3520K and Wärtsilä ratings are at power factor 1.0, and the GE Vernova factsheets state none. Rerun the screen on site-rated output at the design ambient before comparing blocks.
Unit count feeds an availability study and cannot replace one. Shared fuel supply, switchgear, cooling and controls can take out several units at once. The screen also cannot show whether the remaining units pick up the load when one trips. The turbine sizing and redundancy guide works the same question for turbine plants.
Part load, phased demand and fast start#
Part load separates the two classes. For a typical lean-burn gas engine, efficiency at 50% load is "approximately 8 to 10 percent less than full load efficiency". Gas turbines "typically experience efficiency decreases of 15 to 25 percent at half load". The same EPA chapter, on reciprocating engines in the March 2015 CHP catalog, adds that "multiple engines may be preferable to a single large unit" where deep load reductions recur.
A published load-point table shows what to ask every supplier for, as on Cummins' April 2019 datasheet D-6409e for the C2000N6CD continuous rating.
Cummins C2000N6CD load points: 2,000 kW continuous, 60 Hz, LHV (datasheet D-6409e)#
| Load point | Electrical output | Fuel input, LHV | Electrical efficiency, LHV |
|---|---|---|---|
| 100% load | 2,000 kW continuous | 4,599 kW (15.71 MMBtu/h) | 43.5% |
| 50% load | 1,000 kW (calculated) | 2,495 kW (8.52 MMBtu/h) | 40.1% |
Those figures assume 25 °C air inlet, 1,013 mbar, no engine-driven coolant pumps, power factor 1.0 and 97% alternator efficiency. The conditions travel with them into any site calculation.
Named models can sit outside EPA's typical range. The C2000N6CD loses 7.8% of its efficiency at half load. The G3520K High Efficiency 1.0 g build's fuel use rises from 7.88 to 8.99 MJ/ekW-hr, a 12.3% efficiency drop (both calculated).
EPA's combustion turbine chapter notes that turbine output is reduced by lowering turbine inlet temperature, which "also reduces efficiency", and that "emissions are generally increased at part load". For phased demand, test the first stage as hard as the last; the part-load and phased-demand guide does this for turbines.
Start claims need reading side by side. Fast start is 5 minutes for both aeros on GE Vernova's factsheets, against the engine start times in the model table.
Each OEM's own comparison favors its product. Engines start in "45 sec – 5 min" against "5 – 10 min" for turbines in Jenbacher's comparison. Aeros ramp at "around 50 MW/min" against "about 5 MW/min" for medium-speed engines on GE Vernova's LM6000 page. Load acceptance and black start are covered in the black-start and load-acceptance guide.
Fuel, emissions and the permit path, briefly#
Engine ratings carry a minimum methane number for the gas they burn. The mtu 20V4000 GS L64FNER needs a methane number of 72 or more, and 80 for its hot-ambient version; the C2000N6CD lists a minimum methane index of 70. The C2000N6CD also takes low-pressure gas: 0.15–0.45 bar at its fuel shutoff-valve inlet.
Many aeroderivatives run compression ratios near 30:1, "requiring a high-pressure external fuel gas compressor", EPA's combustion turbine chapter notes. Gas engines need "< 500mbar – 12bar" against "20bar – 50bar+" for gas turbines, in Jenbacher's own comparison. Size that compressor load into the turbine's site-net figure. The turbine fuel requirements guide covers gas quality limits.
On NOx, EPA's 2015 combustion turbine chapter says most turbine manufacturers "typically offer a range of 15-25 ppm DLN/DLE combustion systems" on natural gas. SCR, the main post-combustion control, removes "80 to 90 percent" of turbine NOx. Which route a site needs depends on its limit and permit; the emissions permitting guide owns that question.
Operating-cost basis from heat rate#
Fuel cost per MWh is the heat rate multiplied by the gas price, both on the same heating-value basis. In MMBtu per MWh (LHV), the LM6000 PF+ Sprint is 8.33 at ISO full load. The C2000N6CD is 7.86 at full load and 8.52 at half load (calculated from its MMBtu/h figures).
8.08 MMBtu/MWh — the C2000N6CD's energy-weighted heat rate (LHV) when half its hours run at full load and half at 50% (calculated).
Sum fuel and output across the load profile; never average efficiencies. In that example, 4,380 hours at each point give 13,140 MWh from 106,127 MMBtu. A simple average of 43.5% and 40.1% would imply 8.16 MMBtu/MWh and overstate fuel use by about 1%.
Basis errors are larger. The gap shows in EPA's catalog, where large lean-burn engines reach "over 46 percent LHV (42 percent HHV)". Pricing gas on one basis and quoting heat rate on the other would put the result out by about 10% on those figures (calculated).
Rating type sets the hours behind the arithmetic. The Cat G3520 ratings are continuous, defined as "output available with non-varying load for an unlimited time". The G3520K sheet adds that "Average power output is 70-100% of the continuous power rating".
Standby and prime power generator ratings carry different hour and load terms, set out in the rating guide. The fuel cost per MWh guide completes the method.
What changes the answer#
Five site facts can flip a close comparison.
- Ambient and altitude. In EPA's example for the Solar Taurus 70, an industrial turbine, output drops to 80% at 100 °F against the 59 °F ISO point. Ask GE Vernova for the LM ambient curve. The Wärtsilä and Cummins engine figures are quoted at 25 °C, and the G3520K High Efficiency 1.0 g build lists 935 m altitude capability at 25 °C.
- Gas pressure. A low-pressure distribution supply adds compression to the turbine scope and its auxiliary load.
- Phasing. A first stage far below the final load favors units that can be added one at a time.
- Frequency. Some ratings change between 50 and 60 Hz; the 50 Hz and 60 Hz guide covers conversion.
- Emissions zone. A tight NOx limit can add SCR to either class, so compare the aftertreatment scope in both offers.
Used equipment adds one more: operating hours, starts and repair history change the work needed before installation.
Like-for-like scope to request from both technologies#
| Scope area | Items to identify in each offer |
|---|---|
| Generating equipment | Units, generators, reserve unit and installed configuration |
| Auxiliaries | Starting, cooling, lubrication, fuel compression and treatment |
| Electrical | Switchgear, step-up transformer, protection and controls |
| Emissions | NOx build, SCR or catalyst scope and monitoring |
| Service | Maintenance plan, parts and overhaul approach |
| Acceptance | Defined site-net output, heat rate and test boundary |
FAQ: engines vs turbines#
Can gas engines and gas turbines run together on one site?#
They can share a site bus if the switchgear and controls are designed for both. Paralleling switchgear, paralleling controls and a load-share module appear among the G3520H options on Cat LEXE1552-03. Ask both suppliers how their controls will share load with the other technology.
Which class handles a 50 Hz site more easily?#
The models here change little between frequencies. The Wärtsilä 20V34SG's rated output is 9,795 kW at 50 Hz against 9,388 kW at 60 Hz, 4.3% more (calculated). The mtu 20V4000 GS L64FNER gives 2,539 against 2,520 kWel, according to mtu's spec sheet 3234731, Edition 01/24.
The LM6000 PF+ Sprint figure is the same at both frequencies on GE Vernova's product page. The mtu 60 Hz sheet pairs a 1,500 rpm engine with a 1,500–1,800 rpm gearbox, so compare each model at the site frequency.
Does a gas engine need SCR?#
It depends on the limit. Lean-burn engines "produce the lowest NOx emissions directly from the engine", but must add SCR to meet extremely low limits such as California's CARB 2007 standard of 0.07 lb/MWh. Rich-burn engines can use three-way catalysts instead, according to EPA's reciprocating engine chapter. An engine-out figure such as mtu's, quoted without aftertreatment in the engine section, is the starting point to compare with the permit.
How do heat rate and efficiency relate?#
Efficiency equals 3,600 divided by the heat rate in kJ/kWh, or 3,412 divided by Btu/kWh. The Wärtsilä 20V34SG's 7,374 kJ/kWh gives 48.8%, matching its published efficiency (calculated). The conversion carries the heating-value basis with it, so an unnamed basis stays unnamed.
How is a whole-plant output promise tested?#
ASME PTC 46-2015 (reaffirmed 2025) covers overall plant performance and "applies to any plant size", according to ASME's PTC 46 page. Its "explicit methods and procedures" are for combined cycle and most Rankine cycle plants. Other heat-cycle plant types can use it "providing the explicit procedures can be met".
Component tests such as gas turbines under PTC 22 are outside its scope, as are emissions, operational demonstration and reliability tests. For an engine block or simple-cycle plant, the contract should name the test code and the boundary.
Which option can operate sooner?#
Both OEMs claim speed. An LM2500XPRESS "can be installed in about 14 days by 20 people", in GE Vernova's factsheet claim. Commercial operation takes "several weeks" for engines and "several months" for turbines in Jenbacher's comparison. Fuel, grid connection, permits and acceptance can set the date, so compare documented milestones in each offer.
Choosing between a block and a package#
SecondWatt publishes equipment research and helps buyers source power equipment; it earns a fee when a sale it arranges closes. Weighing an engine block against an aero package? Send the load profile, phasing, gas pressure and site conditions through a request for quote, and say whether both technologies are acceptable. For model-by-model background, start with the gas-engine dossiers.