ASUS ESC8000 for the NVIDIA RTX PRO 6000 Blackwell Server Edition: models, specs, power

Five ASUS ESC8000 models for the RTX PRO 6000 Blackwell Server Edition

Five 4U servers of the ASUS ESC8000 series carry the NVIDIA RTX PRO 6000 Blackwell Server Edition on their pages at ASUS's server site, read 12 November 2026: the ESC8000-E12 and ESC8000-E12P with two Intel Xeon 6 processors, and the ESC8000A-E13, ESC8000A-E13P and ESC8000A-E13X with two AMD EPYC. Each has eight slots for dual-slot GPUs of up to 600 W. The ESC8000A-E13X is the one ASUS calls an NVIDIA RTX PRO Server, and its page is the only one that puts the number eight directly in front of this card.

This page deals with the ESC8000 series only. Why a passive card of up to 600 W runs only in a chassis whose maker lists it, and which servers of other makers do, is in our article on RTX PRO 6000 Server Edition compatible servers; the card itself is described in the RTX PRO 6000 Blackwell specs. We supply ASUS, Supermicro, Gigabyte, Dell and HPE servers per project; tell us which ESC8000 model you have in mind and how many cards it should take, and you receive a configuration and a written quote within one business day.

What E12, E13, A, P and X mean in an ESC8000 model name

The name carries the platform. Models without an A are Intel: the E12 generation takes two Xeon 6500/6700-series processors in LGA 4710 sockets. An A marks AMD: the E13 generation takes two EPYC 9005 or 9004 in SP5 sockets, and the E13X the EPYC 9005 only. The suffix tells how the eight GPU slots are connected.

ASUS product pages for the five models and the overview pages of the ESC8000-E12P and ESC8000A-E13P (undated), read 12 November 2026. The E13 datasheet adds that the actual TDP support varies by configuration.

The P models add slots for the network: ASUS speaks of five PCIe 5.0 slots for NICs and a DPU, against two in the models without a suffix. The X model "supports eight 400Gb/s InfiniBand/Ethernet QSFP ports and PCIe 6.0 connectivity for GPU communication". What that board does is described in ConnectX-8 SuperNIC in an AI server, and the reference designs behind the P and X models in NVIDIA MGX servers explained.

Two older servers have names one letter away from the current ones. The ESC8000A-E12 is the earlier AMD model for EPYC 9004, and the ESC8000-E11P the earlier Intel model for 5th Gen Xeon Scalable, both with four 3,000 W supplies. Neither page names the RTX PRO 6000 Blackwell Server Edition. An order for an "ESC8000 E12" should therefore say whether the Intel ESC8000-E12 or the older AMD ESC8000A-E12 is meant.

What ASUS states about GPU support

On the pages of the ESC8000-E12, E12P, ESC8000A-E13 and E13P the card stands in a list. The E13 page, for example, speaks of "eight dual-slot NVIDIA H200 GPUs or NVIDIA RTX PRO 6000 Blackwell Server Edition", with up to 600 watts each. Read strictly, the count belongs to the H200. The E13X page says "eight dual-slot NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs", "supporting up to 600 watts each", and its datasheet reads "Support up to eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs".

Three details help when reading these documents:

• "H200" on the pages is the PCIe card H200 NVL. The datasheet file names carry "RTX_PRO_6000_H200_NVL", and the E13 datasheet is headed "ESC8000A-E13 NVIDIA H200 NVL, RTX PRO 6000 BSE".
• "BSE" is ASUS's short form for the Blackwell Server Edition, in datasheets and in its GPU list.
• The NVLink bridges ASUS lists for the E12, E12P and E13 ("2-Way or 4-Way") concern the H200 NVL. The RTX PRO 6000 has no NVLink, as RTX PRO 6000 without NVLink explains.

The same four pages name the NVIDIA RTX PRO 4500 Blackwell Server Edition, the single-slot card with 32 GB and 165 W; its figures are in the RTX PRO 4500 Server Edition specs, and the page to buy the RTX PRO 4500 Server Edition covers the order.

ASUS's GPU compatibility list on its server support site has product filters for the ESC8000-E12, E12P, ESC8000A-E13 and E13P and GPU filters named "NVIDIA RTX PRO 6000 BSE" and "NVIDIA RTX PRO 6000 Max-Q". The rows of that list are loaded by script and were not readable, so the quantity per model is not confirmed from it here.

None of the ASUS pages read says NVIDIA-Certified; the P and X models are described as "Fully compatible with NVIDIA MGX architecture". ASUS appears among the system partners on NVIDIA's RTX PRO Server page (11 June 2026), which says "RTX PRO Servers are NVIDIA-Certified". In NVIDIA's list of NVIDIA-Certified Systems, as updated on 4 October 2026 and read on 12 November 2026, the ESC8000A-E13P stands among the eight-GPU PCIe reference configurations with the L40S as the tested GPU. NVIDIA-Certified Systems: what the label covers explains why the row for the exact model and card is the one that counts.

ESC8000 specs from the ASUS datasheets

ASUS datasheets for the ESC8000-E12 (V1), ESC8000-E12P (V1), ESC8000A-E13 (V2) and ESC8000A-E13X (V1), all undated, read 12 November 2026. The E12 and E12P weights exclude processors, memory and drives.

All four have eight 2.5-inch hot-swap bays at the front. How many of them take NVMe drives depends on the model and, in the E12 and E13, on the RAID card. These two models have no network port on board apart from the management port: NICs go into the free slots, one x16 slot for a NIC or a BlueField-3 DPU and one x8 slot in each. Management is the ASMB12-iKVM module with an AST2600 controller and ASUS Control Center; the E12P page also names a web interface, IPMI and the Redfish API.

The ESC8000A-E13P is missing from the table because its datasheet could not be opened: the link on ASUS's own product page is broken. The product and overview pages give the EPYC 9005/9004 platform with up to 500 W per socket, eight slots for GPUs of up to 600 W, "up to 24 DIMM", the line "Backplane supports up to 8 x NVMe" and a 3+1 layout of 3,200 W supplies. Dimensions, weight, memory maximum and on-board LAN of the E13P are not confirmed.

Power supplies and feeds for eight 600 W cards

The datasheets of the E12, E12P, E13 and E13X print the same row: four 3,200 W 80 PLUS Titanium supplies, 3+1 redundant, with the rating "220-240 Vac, 16A (x4) 50/60Hz". The E12 and E12P add "(following different configuration by region)". We read the row as four inputs of 16 A each; it names neither a connector nor a lower voltage range, so the power supply configuration for a delivery in the EU belongs in the quote.

Three of the four supplies deliver 9.6 kW by our arithmetic. By our estimate a server with eight cards at 600 W and two AMD EPYC 9555 processors draws about 6.3 kW at the wall. A 230 V 32 A feed carries 5.89 kW when planned at 80 per cent, so an eight-card ESC8000 belongs on three-phase power; the figures are worked out in GPU server power on 16 A, 32 A and three-phase feeds. With 3+1 the server keeps running when one supply fails. Dividing the supplies over two feeds A and B, so that either feed carries the server alone, requires an N+N layout, which this is not; PSU redundancy in GPU servers sets out the difference.

Cooling, depth and weight

Cooling is the same in the four datasheets: five 8080 fan modules for the GPUs and five for the system, hot-swap and removable without tools. The operating temperature is 10 to 35 °C, with the footnote "For operating temperatures above 30°C, please contact your local sales representative". A plan that puts eight 600 W cards into a room of 35 °C therefore needs ASUS's answer first. The figure that counts is the air temperature in front of the server, and GPU server inlet temperature shows how to take it.

The chassis measures 800 × 439.5 × 175 mm in all four datasheets. The E12 and E12P weigh 42 kg net and 44.23 kg gross without processors, memory and drives; the E13 and E13X datasheets give 45 kg net and 56 kg gross. ASUS publishes no weight for a server populated with eight cards. Before we quote an ESC8000 we check rack, power and airflow; send us the depth of the rack, the feeds at the position and the room temperature.

Four cards in an ASUS server and the Max-Q datasheets of the ESC4000

ASUS's 2U servers for four dual-slot cards are the ESC4000A-E12, with one AMD EPYC 9005/9004 of up to 400 W, 12 DIMM slots and two 2,600 W Titanium supplies in 1+1, and the ESC4000-E11, with two 5th Gen Intel Xeon Scalable processors, 16 DIMM slots and two 2,600 W Titanium supplies. The ESC4000A-E12 page speaks of "Up to four dual-slot active or passive GPUs". The datasheets, however, are for another card. The ESC4000A-E12 datasheet is titled "ESC4000A-E12 NVIDIA PRO 6000 Max-Q", and the file name of the ESC4000-E11 datasheet carries "PRO_6000MaxQ". Neither document names the RTX PRO 6000 Blackwell Server Edition.

The Max-Q is the 300 W blower edition for workstations. It has the same 96 GB, but NVIDIA's vGPU list does not include it, and in a rack it belongs only in a chassis whose maker lists it; RTX PRO 6000 Max-Q in a rack server and our comparison of the three editions explain the differences, and we supply all three editions through our RTX PRO 6000 Blackwell page. An ESC4000 documented for the Max-Q is not thereby documented for the Server Edition.

ASUS's GPU server listing, as read, shows no four-card ESC4000 of the current generation, and the product filter of its GPU list has no ESC4000-E12 or ESC4000A-E13. Four Server Edition cards in an ASUS chassis would mean an ESC8000 with four of its eight slots filled, which ASUS's documents do not describe as a configuration either. What four cards run is in four RTX PRO 6000 Server Edition cards in one server.

Points to settle with ASUS before the order

ASUS's documents leave six points open for an eight-card order.

1. Cards per model. The quantity and the slot order for the RTX PRO 6000 Blackwell Server Edition in the E12, E12P, E13 and E13P come from ASUS's GPU list, whose rows could not be read.
2. The ESC8000A-E13P datasheet. Dimensions, weight, memory maximum and LAN stay unconfirmed until ASUS supplies the document.
3. Power supplies by region. The connector and the configuration for the EU.
4. Operation above 30 °C. ASUS asks to be contacted for it.
5. NVMe count. It depends on the SKU and the RAID card in the E12 and E13.
6. Warranty period. ASUS's server warranty page states "A Limited Warranty is offered on Server products." and that the period "may vary depending on your region", without a number of years. For a server ordered from us, the manufacturer warranty runs through us; GPU server warranty in the EU covers the terms the makers publish.

Eight cards hold 768 GB of GPU memory, of which 664.3 GiB carry model weights and KV cache by our users-per-card rule; which LLMs fit eight RTX PRO 6000 Server Edition cards works through the models, and our page to buy a server with eight RTX PRO 6000 Server Edition cards through a complete build. Across makers, see Supermicro vs Gigabyte vs ASUS and the Gigabyte servers for the same card. Describe the project and the points above that apply to it, or send them to request@eurokommerz.at; the answer is a written quote with price, lead time and warranty terms.

FAQ