RTX PRO 6000 Server Edition compatible servers, and adding the card to a server you already run

Which servers support the RTX PRO 6000 Server Edition

The RTX PRO 6000 Blackwell Server Edition is a passively cooled, dual-slot, full-height, full-length PCIe Gen5 card with a power limit configurable from 300 to 600 W (NVIDIA product brief SP-12355-001_v02, 27 June 2025), so it runs only in a server whose manufacturer has qualified it for that model, slot, power cable and fan set. As of 9 October 2026, the compatible servers named in the manufacturers’ own documents or by NVIDIA are the Dell PowerEdge XE7745 and R7725, the HPE ProLiant Compute DL380a Gen12, ProLiant Compute DL380 Gen12 and ProLiant DL385 Gen11, the Lenovo ThinkSystem SR650a V4 and SR675 V3, the Cisco UCS C845A M8 and the Supermicro SYS-422GL-NR. The maximum ranges from two to eight cards per server, depending on the model. For any server, the first check is the manufacturer’s GPU support list for the exact model, before the empty slot.

NVIDIA’s family comparison lists the card with passive cooling, 96 GB of GDDR7 with ECC and a PCIe Gen 5 x16 interface, and it takes its air from the server’s fans. The product brief gives two power modes, 600 W and 450 W maximum, each with a 300 W minimum. NVIDIA also lists a liquid-cooled, single-slot version for its 6U MGX design; this article covers the air-cooled card. How it differs from the Workstation and Max-Q editions is set out in our comparison of the three RTX PRO 6000 editions.

Servers that list the RTX PRO 6000 Server Edition

Each model below is named for the card by its manufacturer, or by NVIDIA, on a page we read on 9 October 2026.

Read on 9 October 2026: Dell PowerEdge Server GPU Matrix (August 2026), XE7745 (July 2025) and R7725 (December 2025) specification sheets; NVIDIA press release of 11 August 2025; NVIDIA Accelerators for HPE QuickSpecs (version 83, 8 September 2026) and the DL380a Gen12 and DL385 Gen11 QuickSpecs; Lenovo Press LP2263 (28 July 2026); Cisco UCS C845A M8 data sheet (24 August 2026); Supermicro PCIe GPU systems and SYS-422GL-NR pages. Counts are maximums; the configurator decides per configuration.

Dell’s PowerEdge Server GPU Matrix, dated August 2026, lists the card as “RTX Pro 6000 BSE” in its seventeenth-generation table and in neither the fifteenth nor the sixteenth-generation table, so a sixteenth-generation server such as an R760xa is not on Dell’s list. The XE7745 specification sheet names the card as an example of the “up to 8 double-wide 600W PCIe AI accelerators” the server takes. NVIDIA’s announcement of 11 August 2025 names “Dell PowerEdge R7725 2U servers featuring two RTX PRO 6000 GPUs”, while Dell’s R7725 specification sheet of December 2025 gives the GPU options as “Up to 2 x 450W DW”. Confirm the power per slot with Dell before you plan on 600 W.

HPE’s NVIDIA Accelerators QuickSpecs list the card as SKU S6A73C for the ProLiant Compute DL380a Gen12, the ProLiant Compute DL380 Gen12, the ProLiant DL385 Gen11 and two edge systems, and refer to the platform QuickSpecs “for configuration rules including enablement kits”. The DL380a Gen12 QuickSpecs state “RTX Pro 6000 GPU limited to max quantity of 8”, although the same document allows most other double-wide GPUs in quantities of 2, 4, 8 and 10. They also exclude the card from the 8DW NIC FIO configuration, P75011-B21. The DL385 Gen11 QuickSpecs list “4x RTX PRO 6000 BSE” among their newly supported configurations.

Lenovo’s product guide LP2263, updated on 28 July 2026, offers the card under two feature codes, one for 600 W and one “Slot Capped at 450W”. The SR650a V4 takes two cards at 600 W or four of the capped version. The SR675 V3 takes eight, with a footnote limiting support to 5th Gen AMD EPYC processors, and every other current ThinkSystem server in the table is marked as not supported. Another footnote says “RTX PRO 6000 GPU is only supported on SR650a V4”, followed by its machine types, so the SR650 V4 without the “a” is excluded.

Cisco’s data sheet for the UCS C845A M8, updated on 24 August 2026, describes a 4RU server with two AMD Turin processors and “Two, four, six, or eight” PCIe GPUs, the RTX PRO 6000 at 600 W among them. Supermicro’s PCIe GPU systems page shows a 4U RTX PRO Server, the SYS-422GL-NR, with up to eight cards, and its product page lists the card. The same page shows a 2U RTX PRO Server for up to two cards, but the product page it links to names neither the card nor a GPU count, so confirm that model with Supermicro. NVIDIA’s RTX PRO Server page names other manufacturers as well.

We supply the RTX PRO 6000 Server Edition for servers you already run, after a check of the platform, power and cooling. Send us the server model and its configuration, and we say in advance if the card will not work.

NVIDIA RTX PRO Servers and NVIDIA-Certified Systems

NVIDIA calls an RTX PRO Server a “Universal data center platform for enterprise AI and industrial AI”, built by system partners on its MGX reference designs. Its RTX PRO Server page, updated on 11 June 2026, lists “8x RTX PRO 6000 Blackwell Server Edition PCIe GPUs” in the quick specifications. It shows two cards in a 2U design and eight in the 4U and 6U designs, both with “NVIDIA ConnectX-8 SuperNICs with PCIe Gen 6 switch”; the 6U design uses the liquid-cooled card. The page states that “RTX PRO Servers are NVIDIA-Certified”.

NVIDIA’s reference architecture for the RTX PRO AI Factory, updated on 18 May 2026, gives the server components behind that design. It specifies up to eight cards, four BlueField-3 SuperNICs per server at up to 400 Gbps and one BlueField-3 DPU for north-south traffic. Per GPU, it asks for at least one Gen5 x16 link, seven physical CPU cores and 128 GB of system memory. Our article on how many GPUs fit in one server explains the lane, switch and airflow limits behind these counts and compares inlet temperature limits.

NVIDIA says NVIDIA-Certified Systems “undergo a rigorous suite of functional, security, and performance tests”, from single-GPU to multi-node, and that “Hardware support for NVIDIA-Certified Systems is available from the system vendors”. Its page of 3 September 2026 points to the Qualified System Catalog for certified partner systems; check the exact model there.

For a new server, we build AI servers to order with 2 to 8 RTX PRO 6000 Server Edition cards and check the rack, power and airflow before we quote. Describe the workload and the rack position in the form below.

Adding the card to a server you already run

A server already in the rack can take the card only if its exact model and configuration are on the manufacturer’s list.

Compiled from the manufacturer documents named under the first table, read on 9 October 2026.

The same order of checks applies to the H200 NVL, with that card’s own cable and bridge rules, in our H200 NVL checklist for an existing server, which also covers the BIOS settings for large memory windows.

Power cables, 450 W slots and power supplies

Lenovo states that “The RTX PRO 6000 Blackwell GPU option part number does not ship with auxiliary power cables” and that “Cables are server-specific due to length requirements”. For a field upgrade, in Lenovo’s words, “cables will need to be ordered separately”. Lenovo names a 320 mm 16-pin (2x6+4) cable kit, 4X97B04860, for the SR650a V4 and a 235 mm 16-pin cable, 4X97A84510, for the SR675 V3.

The card reads the power class its cable declares through the sense pins and caps itself to match. On a cable for 450 W it runs at 450 W, and the comparison of the three editions shows how that looks in nvidia-smi. Some servers set the cap on purpose: Lenovo’s second feature code is “Slot Capped at 450W” and is the version that allows four cards in the SR650a V4.

The power supplies are the next limit. HPE’s DL380a Gen12 QuickSpecs state “2400W or 3200W PSUs must be selected if H100 NVL, H200 NVL, or RTX Pro 6000 GPUs are configured”, with five power supplies for two or four double-wide GPUs, eight for eight or ten, and C19 or C19-C20 power cords only. Supermicro fits four 3,200 W supplies in a 3+1 configuration in the SYS-422GL-NR. Four cards at 600 W draw 2.4 kW and eight draw 4.8 kW before processors and fans, so the rack feed has to be planned with the server, as our guide to GPU rack power and cooling explains.

Fans, airflow and inlet temperature

A passive 600 W card depends on the server’s fan set, and the manufacturers name the fan kits that go with it. HPE’s DL385 Gen11 QuickSpecs state that the “Maximum Performance Fan kit is only supported when 600W GPUs are in the configuration”, naming the RTX PRO 6000 BSE among them. In the DL380a Gen12, eight RTX PRO 6000 cards require HPE’s Front Chassis Fan Module Kit, P79656-B21, and its Front Panel Kit, P79660-B21.

Inlet temperature limits are set per server and GPU configuration in each manufacturer’s thermal documentation, so measure the inlet temperature at the rack position before ordering.

Operating system, driver and firmware

Lenovo lists supported operating systems per server; for the SR650a V4 they include Red Hat Enterprise Linux 9.7 and 10.1, SUSE Linux Enterprise Server 15 SP6, 15 SP7 and 16, and Ubuntu 22.04.5 LTS or later and 24.04 LTS or later, both with the HWE kernel.

Our guide to NVIDIA driver branches and CUDA versions sets out the software minimums for the card from NVIDIA’s documents: the open kernel modules, which NVIDIA requires for Blackwell, driver R575 and, on x86, CUDA 12.9 or later from the Server Edition’s product brief, and vGPU from release 19.0. Bring the server’s BIOS and management controller firmware to the level the manufacturer lists for the GPU before the card goes in. Once the card runs, the host is hardened like any other GPU server, as described in securing a GPU server.

Support and warranty for a card bought separately

The manufacturers state their position for their own option part numbers. Lenovo’s guide LP2263, updated on 28 July 2026, gives the card a “One year limited warranty” and adds that “When installed in a Lenovo server, the GPU assumes the server’s base warranty and any warranty upgrades.” NVIDIA states that hardware support for NVIDIA-Certified Systems comes from the system vendors.

On the pages we read on 9 October 2026, Dell, HPE, Cisco and Supermicro do not state support terms for a card bought outside their own option part numbers. Ask the server manufacturer before you order, and keep the cable, fan kit and drive configuration within its list.

What we supply

We supply the NVIDIA RTX PRO 6000 Blackwell Server Edition across the EU, as a card for servers you already run or in AI servers built to order, on one EU contract and invoice with manufacturer warranty. For a server you own, we check the platform, power and cooling first and say in advance if something will not work. Our reference configurations include two cards with one AMD EPYC processor and 256 GB of DDR5 ECC memory, and four cards with two AMD EPYC processors and 512 GB, and we check the rack, power and airflow before we quote. The professional NVIDIA GPUs we supply include the other RTX PRO, L40S, L4 and H200 NVL cards, and NVIDIA AI Enterprise and vGPU licences come on the same invoice.

FAQ