NVIDIA RTX PRO 6000 Blackwell Max-Q power: two, four and eight cards on 230 V

How much power the Max-Q draws

The NVIDIA RTX PRO 6000 Blackwell Max-Q draws up to 300 W per card, so the cards alone take 600 W in a two-card workstation, 1,200 W with four and 2,400 W with eight. By our estimate the whole system needs about 1.1 kW at the wall with two cards, 1.8 to 2 kW with four and 3.6 to 3.9 kW with eight; on 230 V mains the first two run on one 16 A circuit, while eight cards need a 32 A or three-phase feed.

NVIDIA's datasheet for the Max-Q (5349650, June 2026) gives a total board power of 300 W, one PCIe CEM5 16-pin power connector and an active cooler, and its product page states that the card "enables up to four GPUs in a single system". Every card needs its own 16-pin cable from the power supply. The rest of the system adds the processor, memory, drives, fans and the losses of the power supply, which is why the wall figure is well above the sum of the cards. Most of what is written about four-card builds comes from the United States and assumes 120 V circuits; in Europe the same machine sits on 230 V, which changes the arithmetic in its favour.

Two, four and eight cards: wall power by our estimate

The table adds the rated power of the cards and processors and then allows 15 per cent for the rest of the system. These are planning figures; the only published reading we found for a multi-card Max-Q system is quoted below the table.

Our estimate from NVIDIA's board power figures; workstations with one 350 W processor, the rack server with two 280 W AMD EPYC 9355, the eight-card servers with two 360 W AMD EPYC 9555 or, for the Max-Q row's upper figure, two 500 W Xeon 6900; plus 15 per cent; current = wall power ÷ 230 V.

The workstation rows follow the method of our guide to four RTX PRO 6000 Max-Q in one workstation: 1,200 W for the cards plus a 350 W processor, the rated power of both a Threadripper PRO 9995WX and a Xeon 698X, gives 1,550 W before memory, drives, fans and conversion losses, and roughly 1.8 to 2 kW at the wall. The two-card figure applies the same allowance to 950 W. For the eight-card rows we use two 360 W EPYC 9555, as in our other eight-card guides; Supermicro's SYS-422GA-NRT takes two Xeon 6900 rated up to 500 W, which moves the figure to about 3.9 kW. The last row, eight 600 W Server Edition cards, shows why the Max-Q matters at this size: the same 768 GB of GPU memory at about 2.7 kW less with the same processors. A Level1Techs build log with four Max-Q gives one published reading: about 1,650 to 1,760 W on its UPS under full GPU load, with the processor capped at roughly 200 W, which is in line with our four-card estimate.

A build log on the Level1Techs forum (July 2026) with four Max-Q also shows what a lower power limit does. Its author ran the cards at 250 W and at 300 W; at 300 W gpu-burn reached 17 to 22 per cent more dense throughput, while two models served with SGLang gained about 4 to 8 and 8 to 12 per cent, depending on the load. Four cards at 250 W save 200 W. How to set and keep such a limit on Linux is in our guide to the RTX PRO 6000 power limit.

230 V against 120 V: why US build advice does not carry over

A US 15 A circuit at 120 V carries 1,800 W; a European 16 A circuit at 230 V carries 3,680 W. The same Level1Techs build log reports a tripped 15 A breaker and a processor capped to keep the machine under about 1.8 kW, a problem that a four-card tower on a 230 V circuit of its own does not have: 2 kW is about 54 per cent of a 16 A circuit.

The makers' own figures show the difference. Lenovo's power configurator for the ThinkStation PX gives the current of its 1,850 W supply as 12 A at 200 to 240 V and 20 A at 115 to 127 V, and for two supplies in team mode, 2,350 W, 15 A against 25 A. At 100 to 110 V a single supply drops to a power-limited mode of 1,400 W. Advice to install a 20 A or 30 A circuit, common in US build guides, answers a 120 V problem; on 230 V the questions are which outlet the power supply needs and what else shares the breaker.

Power supplies: full rating only at 200 V or more

Workstation and server power supplies are rated by output, and many reach the full figure only on high-line input. The table shows what the makers state for the systems that take four or more Max-Q.

HP Z8 Fury G6i specifications; Lenovo PSREF ThinkStation PX (9 October 2026) and ThinkStation PX power configurator v1.8; Supermicro product pages; all read on 24 October 2026.

Two consequences follow. First, a power supply's rating is its output: a 1,700 W unit at full load at 92 per cent efficiency, the figure HP and Lenovo give, takes about 1,850 W from the wall by our arithmetic. Second, two supplies give redundancy only if one of them carries the whole load. HP's aggregate mode and Lenovo's team mode add the two supplies together, so losing one stops a four-card machine at full load. For a tower that must keep running, ask the maker whether the configured load fits on one supply; for four Max-Q at about 1.55 kW of rated components, a single 1,700 W supply runs close to its limit, and a dual supply in aggregate mode gives headroom without redundancy.

Rack servers follow the same logic at a larger scale. Puget Systems' 2U server with up to four Max-Q has two 2,600 W supplies in a 1+1 arrangement and states that 240 V is required for power-supply redundancy and that the server needs two power connections on separate circuits. Supermicro's SYS-422GA-NRT, which names the Max-Q in its GPU list and takes up to eight double-width cards, uses four 3,200 W supplies in a 3+1 arrangement; its datasheet gives no Max-Q count and no input voltage. Many server makers do not list the Max-Q for their GPU servers at all, and for continuous operation in a rack we recommend the RTX PRO 6000 Server Edition. Which rack chassis list the card at all is the subject of our page on buying a Max-Q rack server.

We supply the Max-Q and build two- and four-card workstations and rack servers with it to order. Send us your circuit rating and the room, and we size the power supply, cords and UPS with the quote.

Cords and outlets: C13, C19 and industrial sockets

The appliance connectors on power supplies follow IEC 60320. NetApp's summary of the standard, last modified on 9 June 2026, gives C13 and C14 a rating of 10 A at 250 V and C19 and C20 16 A at 250 V, and notes that a device or circuit may be rated lower. At 230 V a C13 connector carries at most 2.3 kW and a C19 connector 3.68 kW.

NetApp knowledge base on IEC 60320 connector types (9 June 2026); HPE DL380a Gen12 QuickSpecs; our guide to a GPU server in an office for IEC 60309.

A supply that draws 12 A, as Lenovo gives for its 1,850 W unit at 200 to 240 V, is above the 10 A rating of a C13 connector, so check the inlet on the machine and the cord the maker ships before planning the outlet. Server supplies of 2,400 and 3,200 W take C19 cords; HPE, for example, allows only C19 or C19-C20 cords for them in its DL380a Gen12 QuickSpecs. A rack PDU with only C13 outlets cannot feed such a server. For 32 A single-phase or three-phase power, plug and socket-outlet follow IEC 60309, and an electrician installs the outlet, as our guide to a GPU server in an office explains.

One 16 A circuit, two feeds or three-phase

We plan feeds at 80 per cent of their rating, as in our guide to GPU rack power and cooling: 2.94 kW for a 230 V 16 A circuit, 5.89 kW for 32 A single-phase and 8.87 kW for a 400 V three-phase feed of 16 A.

• Two Max-Q. About 1.1 kW fits a 16 A circuit with room for a monitor and a small switch, but not for a printer room or a kitchen on the same breaker. Ask the electrician which outlets share it.
• Four Max-Q in a tower. About 2 kW needs a 16 A circuit of its own. Two such towers, 4 kW together, do not fit one circuit. Where Lenovo recommends a dedicated circuit for each supply, a PX with two supplies takes two circuits.
• Four Max-Q in a rack server. About 2 kW fits one 230 V 16 A feed; with A and B feeds, each 16 A side carries the whole server.
• Eight Max-Q. 3.6 to 3.9 kW is above the 2.94 kW we plan for one 16 A circuit. Spread over two 16 A circuits, each carries about half in normal operation, but if one fails the other has to carry the whole load, about 16 to 17 A. With A and B feeds, plan 32 A single-phase per side or a three-phase feed.

For eight Server Edition cards at 600 W, about 6.3 kW, even 32 A single-phase is above our 5.89 kW planning figure, so three-phase power is the answer. Our guide to GPU server power on 16 A, 32 A and three-phase covers the feeds for server builds in full.

UPS sizing for a Max-Q workstation or server

Size the UPS from the full-load wall power, not from the power-supply rating, and check both figures on its label: the rating in watts and the rating in VA. For two Max-Q that means a unit comfortably above 1.1 kW, for four above 2 kW, and for eight about 4 kW or more, which at 230 V means an input of more than 16 A. Decide first what the UPS is for: bridging short interruptions, or carrying the machine through an orderly shutdown of the operating system and the running jobs. The second needs the UPS's management interface connected to the host.

Where the power supplies sit on separate circuits, each UPS has to carry the load that falls on it when the other side fails, which with redundant supplies is the whole machine. The UPS's own losses add to the heat in the room. Our guide to UPS sizing for a GPU server works through runtime and redundancy for two, four and eight GPUs.

Heat in the room: kW and BTU/h

Practically all the power a computer takes from the mains becomes heat. With NIST's factor of 3.412 BTU/h per watt, two Max-Q at about 1.1 kW put roughly 3,750 BTU/h into the room, four at 2 kW about 6,800 BTU/h and eight at 3.9 kW about 13,300 BTU/h. An air conditioner labelled 12,000 BTU/h has about 3.5 kW of cooling capacity, so a four-card tower takes more than half of it on its own.

The Max-Q's blowers exhaust through the card brackets at the back of the machine, so keep the rear clear and point it away from people and from the cooling unit's sensor. Lenovo and Supermicro rate the PX, the SYS-551A-T and the SYS-422GA-NRT for 10 to 35 °C operating temperature, and ASHRAE recommends 18 to 27 °C at the server inlet; our page on GPU server inlet temperature explains the limits. Cooling has to run whenever the machine runs, including nights, weekends and winter.

We build Max-Q workstations and servers for company offices, labs and server rooms across the EU. Write to us with the circuit, the room size and its cooling, and we tell you which card count and power setting fits before you order.

FAQ