Seven steps from a tank of compressed gas to the engine behind Starship. Then Raptor 1, 2 and 3 in 3D, and what SpaceX took off between them.
Felix Becker · September 2026
Raptor 1185 tfRaptor 2230 tfRaptor 3280 tf
The 3D models need WebGL, which this browser doesn't offer. Everything else on the page works without it.
My simplified 3D models of the three versions, lined up like SpaceX's side-by-side photo from 2024.
I read Brian Potter's How SpaceX Streamlined the Raptor Engine and got stuck at the fan-made schematics: dozens of lines, arrows in every direction, acronyms I didn't know. I only got it once I drew the engine myself, starting with the simplest rocket there is and adding one part per problem.
How to read the drawings
The dots show which way things flow. The drawing updates as you scroll.
1
Compressed gas and a nozzle
A rocket throws mass out of the back, and that pushes the rocket forward (Newton's third law). The faster the mass leaves, the more push you get per kilogram.
Simplest version: fill a tank with compressed gas and let it out through a nozzle. That's a cold gas thruster. Falcon 9 uses little nitrogen thrusters like this for steering, and NASA's Manned Maneuvering Unit, the jetpack astronauts flew in the 1980s, worked the same way.
Problem: the only energy you have is the pressure in the tank, and that's not much. Fine for steering, useless for getting to orbit.
Used on: Falcon 9 steering thrusters, the MMU jetpack
2
Burn something
Chemical bonds hold far more energy. Take a fuel (Raptor uses methane) and an oxidizer (liquid oxygen), push both into a chamber and burn them. The hot gas leaves the nozzle much faster than cold gas ever could.
The simplest way to push them in is tank pressure. That's a pressure-fed engine. The Apollo lunar module lifted off the Moon on one.
But propellant only flows from high pressure to low pressure. For propellant to flow into the chamber, the tanks have to be at higher pressure than the chamber. And you want the chamber at very high pressure, because that's how you get a small, powerful engine. Raptor runs at around 300 bar. Set a chamber pressure, then see what the tanks need to keep up:
Steel wall a 9-metre-wide tank needs at that pressure:
Thin-wall estimate: stainless steel at about 500 MPa, no safety margin. Starship's real tanks run at a few bar with walls a few millimetres thick.
Problem: high chamber pressure means thick, heavy tanks. At Raptor's pressure the walls would be about 30 cm of steel.
Used on: Apollo lunar module ascent engine
3
Add pumps
Fix: keep the tanks at low pressure, and put a pump in each line that squeezes the propellant right before it enters the chamber. Light tanks, high chamber pressure. Almost every rocket that reaches orbit does this.
The catch is power. The turbopump on the Saturn V's F-1 engine needed about 41 megawatts, roughly 55,000 horsepower, to move its propellant. Something has to spin it.
Rocket Lab's Electron uses batteries and electric motors. That works for a small engine. For a big one the batteries get far too heavy.
Problem: you need tens of megawatts from something light.
Used on: every big orbital rocket. Electron with batteries.
4
Spin the pumps with a turbine
Fix: the rocket is already carrying an enormous amount of energy in its propellant. Tap off a little fuel and oxygen, burn it in a small burner called a gas generator, and let the hot gas spin a turbine. The turbine sits on the same shaft as the pumps.
One catch: burn fuel and oxygen at the perfect mixture and you get over 3,000 °C, which melts any turbine blade. So the gas generator runs lopsided, with lots of fuel and just a little oxygen, which keeps the gas cool enough for the turbine. Try it:
all fuelall oxygen
Gas temperature:
The used gas gets dumped overboard through its own little pipe. On a Falcon 9 launch you can often spot it as darker, sooty exhaust next to the bright flames of the Merlin engines.
Problem: that dumped gas is propellant you paid for, and it barely adds thrust. More chamber pressure needs more pump power, which means dumping more. You hit diminishing returns fast.
Used on: Merlin (Falcon 9), F-1 (Saturn V)
5
Burn the turbine exhaust again
Fix: the turbine exhaust is still full of unburned propellant. Instead of throwing it away, pipe it into the main chamber and burn it there. The small burner is now called a preburner, and because the propellant burns in two stages, this is staged combustion.
Now nothing goes overboard, so you can push far more gas through the turbine, get far more pump power, and run at much higher chamber pressure. The Russian RD-180 works like this (the oxygen-rich version drawn here), and so did the Space Shuttle's main engine (a fuel-rich version).
Problem: one turbine, bathed in hot gas made mostly of one propellant, drives the pumps for both. Somewhere on that shared shaft, hot oxygen-rich gas sits right next to the fuel, kept apart only by seals. If a seal leaks, fuel meets hot oxygen inside the machinery.
Used on: RD-180 (Atlas V), RS-25 (Space Shuttle, SLS)
6
Full flow: give each side its own turbine
Fix: split it in two. One preburner gets all the methane and just a little oxygen. It makes hot, fuel-rich gas, which spins the methane pump. The other gets all the oxygen and a little methane. It makes hot, oxygen-rich gas, which spins the oxygen pump. Every drop of propellant passes through a turbine before it reaches the main chamber. That's what full flow means.
That solves the seal problem from step 5, and you get two more things on top:
No dangerous seals. Each turbopump only ever touches one propellant. Methane-rich gas spins the methane pump, oxygen-rich gas spins the oxygen pump.
Cooler turbines. A turbine's power is roughly how much gas flows through it times how much energy each kilogram gives up. Push all the propellant through and each kilogram can give up less, so the gas can be cooler for the same power. Cooler turbines last longer, which matters when you want to reuse the engine.
Gas meets gas. Both propellants reach the main chamber already hot and gaseous, so they mix and burn fast and completely. That helps run at very high pressure.
Energy each kilogram has to deliver:
Toy model with the pump power held fixed. Real turbines also depend on pressure drop, gas type and a lot more.
The price: hot oxygen-rich gas attacks metal like a cutting torch, so SpaceX had to develop its own alloy for that side of the engine. And two preburners that feed each other are hard to balance and hard to start. Before Raptor, only two full-flow designs had ever been built: the Soviet RD-270 in the 1960s, and the American Integrated Powerhead Demonstrator in the 2000s, a test rig for the front half of such an engine. Neither flew. Raptor first flew in July 2019, on Starhopper's 20-metre hop.
Used on: Raptor (Starship)
7
Now add the plumbing
The core engine is done. Most of the lines on the real schematics belong to the systems around it. Keep scrolling and they switch on one by one, or tap them yourself:
With all four on, this is roughly the fan-made schematic from Brian's article. Almost every extra line there belongs to one of these systems.
The support systems up close
Step 7 showed the support systems as lines on a drawing. Three of them are worth a closer look.
Cooling the chamber with fuel
The gas in the chamber is over 3,000 °C. Every metal you could build the chamber from melts well below that. So the wall is built as a sandwich: a thin inner wall that conducts heat very well (typically a copper alloy), full of small channels, with a steel jacket around it. Cold methane from the pump runs through those channels on its way to the preburner. It carries the heat away before the wall can melt, and the heat isn't wasted: it warms the fuel up for the preburner.
The hardest spot is the throat, the narrowest point. The gas there is dense and fast, so it hammers the wall with more heat than anywhere else. Turn the coolant down and watch where the wall fails first:
Hot face of the wall:
Toy model with illustrative numbers. The copper alloys used for chamber walls melt at around 1,000 to 1,085 °C and lose strength long before that.
Film cooling is the old trick for the throat: squirt a little fuel along the wall so a cooler layer of gas protects it. It works, but that fuel barely burns, so you lose some efficiency. According to Musk, Raptor 2 got rid of the film cooling at the throat. And on Raptor 3, SpaceX went further and added regenerative cooling to exposed parts outside the chamber, which is part of why it no longer needs a heat shield.
Keeping the tanks pressurized
The pumps in step 3 let the tanks stay at low pressure. But not at no pressure. As the engine drinks, the liquid level drops and the empty space above it grows. Nothing fills that space, so the pressure falls. That starves the pump: the pressure at its inlet drops so low that the methane starts to boil right inside it. Bubbles form and collapse against the blades, the pump loses its grip on the liquid, and the engine has to shut down. This is called cavitation.
So something has to keep pushing gas into the tank as it empties. Try the three options:
Tank pressure:
Toy model: the same burn each time, with the tank pressure shown relative to what the pump needs.
Falcon 9 carries helium in high-pressure bottles for this. Helium is expensive, the bottles are heavy, and problems around them cost SpaceX two Falcon 9s (2015, 2016). Starship uses its own propellant instead: it taps a little methane and oxygen from the engines, warms it into gas and pipes it back up to the tanks. No helium to buy or carry, and on Mars you'd only need to make methane and oxygen. The fan schematics show these lines on Raptor 1 and on Raptor 3.
Starting the engine
A running full-flow engine feeds itself: the preburners make the gas that spins the pumps that feed the preburners. At the start, though, nothing is spinning and there's no hot gas. Getting from a cold standstill to full thrust takes a carefully timed sequence, and a lot can go wrong in it. In July 2026 the first launch attempt of Starship's thirteenth flight aborted because several Raptor 3s failed to start.
Chill-down. Cold propellant runs through the pumps and pipes and gets vented overboard. Warm metal would make the liquid boil the moment it arrived. This is part of the venting you see around a rocket before launch.
Purge. Inert nitrogen blows through the preburners and the chamber to clear out anything that shouldn't burn yet.
Spin-up. Gas from the ship spins the turbines so the pumps start pushing propellant before there is any fire. According to the fan schematics, early Raptors used helium for this and Raptor 3 uses nitrogen.
Light the preburners. Igniters light both preburners. Hot gas takes over from the start gas and the turbines speed up on their own. Both sides have to come up in step: if one runs ahead, the mix in the main chamber goes wrong.
Main chamber ignition. The fuel-rich and oxygen-rich gases meet and ignite. Raptor 1 had spark igniters here; Raptor 2 deleted them.
Throttle up. The engine climbs to full power, and warm gas starts flowing back to the tanks to keep them pressurized.
SpaceX hasn't published Raptor's start sequence. This is the general idea for a full-flow engine, not SpaceX's timing.
How hard these engines squeeze
Chamber pressure isn't everything, but it's a decent measure of how hard an engine works: more pressure means more thrust from a smaller, lighter engine. Here's what each step bought:
F-1 gas generator70 bar
Merlin 1D gas generator97 bar
RS-25 staged, fuel-rich206 bar
RD-180 staged, oxygen-rich267 bar
Raptor 2 full flow300 bar
Raptor 3 full flow350 bar*
*Reached in a 2023 test; SpaceX has announced up to about 330 bar in operation. In February 2019 an early Raptor hit 268.9 bar and took the record from the RD-180. Sources at the bottom.
Now in 3D
The same full-flow layout as a 3D model. The oxygen turbopump sits on top of the engine, the methane turbopump hangs off the side, just like in the fan schematics. The shapes are simplified; SpaceX doesn't publish drawings.
Drag to rotate. Pinch or ctrl + scroll to zoom. Tap the numbers.
The 3D model needs WebGL, which this browser doesn't offer.
Raptor 1, 2 and 3
The part of Brian's article that stuck with me: between Raptor 1 and Raptor 3 the architecture didn't change. Same cycle, same two turbopumps in the same places, same cooling. SpaceX got its gains from everything around that core.
Raptor 1Raptor 2Raptor 3
Thrust185 tf230 tf280 tf
Engine mass2,080 kg1,630 kg1,525 kg
Engine + vehicle-side hardware3,630 kg2,875 kg1,720 kg
Thrust per tonne installed5180163
Raptor 3 pushes about 50% harder than Raptor 1 while the engine itself got a quarter lighter. Count the hardware the ship has to carry around each engine too, and every tonne you install now makes more than three times the thrust. You can flip between the versions in the 3D model above.
These are the figures SpaceX posted in August2024. For flight, SpaceX currently rates the sea-level Raptor 3 at 250 tf (May 2026), so the 280 tf is what the engine has shown, not what it flies at today.
R1 → R3Sensors came off or moved inside. Raptor 1 was largely a development engine, covered in temperature and pressure sensors so the team could see what it was doing. Once you understand an engine, most of those can go. On Raptor 3 the remaining sensors and controllers sit inside the engine's own thermal protection.
R1 → R2The main-chamber spark igniters were deleted. The gases from the two preburners are hot enough to ignite each other when they meet in the chamber.
R2Parts got merged. Many valves were combined into valve plates, and controllers scattered over the engine moved into boxes.
R1 → R3Bolted flanges became welds. Lighter, fewer joints that can leak. The price is that you can't unbolt it to fix it.
R3Plumbing moved inside the metal. Many secondary lines are now channels inside 3D-printed parts instead of pipes on the outside.
R3The heat shield disappeared. With nothing fragile left on the outside, and exposed parts cooled by propellant, Raptor 3 needs no heat shield and no fire suppression system. That's where most of the saved mass came from: the hardware around each engine went from about 1,550 kg to 195 kg.
R3Helium went away (according to the fan schematics). Early versions used helium to spin up the turbines and work valves; Raptor 3 is drawn with nitrogen instead, and the small lines are bundled into one umbilical to the ship.
“The amount of work required to simplify the Raptor engine, internalize secondary flow paths and add regenerative cooling for exposed components was staggering.”
Build a lot of them. SpaceX fired its first small development Raptor in 2016 and the first full-size one in February 2019. By late 2022 it was building more than one engine a day, and by October 2025 it had built more than 600. When engines are that cheap to make, you can afford to test them until they break, change something, and test again. Raptor 2 cost roughly half as much as Raptor 1.
Make it work first, make it simple later. Raptor 1 was built to learn: sensors everywhere, bolts so you could open it up. Only once the team knew what every part was doing could they start taking parts away.
Delete, then simplify. Musk's much-quoted engineering “algorithm”, which he walked through on a 2021 Starbase tour, goes: question every requirement, delete every part and process you can, only then simplify and optimize what's left, then speed it up, then automate it. His line is that “the best part is no part.” Raptor 3 is what years of step two look like.
Simpler outside, harder inside. The clean look is a bit misleading. A lot of the complexity moved into printed metal and software. And it's not finished: Raptor 3 first flew in May 2026 on Starship's twelfth flight, where both the booster and the ship lost an engine, and in July the first launch attempt of flight 13 aborted at T-0 because several Raptor 3s didn't start.
Raptor 1 as a 1:10 print kit by hisdirt (Thingiverse, CC BY). 51 parts with the real external plumbing. The only open model I found that has it.
Raptor 3 by VoitAa (Sketchfab, CC BY 4.0). Exterior only, but faithful to the 2024 photos.
Raptor by MartianDays (Sketchfab, CC BY 4.0), the artist behind the renders in Everyday Astronaut's 2019 video.
None of them shows the inside of the engine: preburners, turbines, injector, cooling channels. That part you only get as diagrams, like the ones on this page.