Low-profile NVIDIA RTX workstation cards compared: RTX A400 to RTX PRO 2000 Blackwell

Five half-height NVIDIA RTX cards from 4 GB to 16 GB

This page compares five low-profile NVIDIA RTX cards that we supply. All are 2.7 in high, the half-height format usually called low profile: the NVIDIA RTX A400 with 4 GB, the NVIDIA RTX A1000 with 8 GB, and three cards with 16 GB, the NVIDIA RTX 2000E Ada Generation, the NVIDIA RTX 2000 Ada Generation and the NVIDIA RTX PRO 2000 Blackwell. The first three take one slot and 50 W, the last two take two slots and 70 W. All five have four Mini DisplayPort outputs and active cooling.

The page is written for IT departments and system builders who fit cards into small-form-factor (SFF) workstations, compact desktops and other systems with half-height slots. Each figure comes from the NVIDIA datasheet of the card unless another document is named.

NVIDIA datasheets: RTX A400 (3279761, May 2024), RTX A1000 (April 2024), RTX 2000E Ada Generation (4328932, October 2025), RTX 2000 Ada Generation (2999417, January 2024), RTX PRO 2000 Blackwell (5322351, June 2026). Every card is 2.7 in high with active cooling.

Converted, 2.7 in is about 69 mm, 6.4 in about 163 mm and 6.6 in about 168 mm. NVIDIA gives the sizes in inches only and publishes no weight for these cards.

Which of these cards NVIDIA itself calls low profile

"Low profile" is the common name for expansion cards of this height, and it is what buyers type: "RTX 2000 Ada low profile", "low profile workstation GPU", "RTX A400 SFF". NVIDIA's own documents use the word for three of the five. The datasheet of the RTX A1000 describes the card as a "low-profile GPU", NVIDIA's product page for the RTX A400 carries the heading "Low-Profile, Single-Slot Form Factor", and the datasheet of the RTX 2000E Ada of October 2025 calls that card "a powerful, low-profile GPU".

For the RTX 2000 Ada and the RTX PRO 2000 Blackwell, the NVIDIA documents read for this page give the height and the length and do not use the word. We describe these two by what their datasheets state, which is half-height at 2.7 in. The height is identical on all five. In a tender or a parts list the dimensions are the safer entry, because they are the figure NVIDIA prints for every one of them.

"SFF" is used in the same loose way. NVIDIA puts it into the name of one current Blackwell card, the RTX PRO 4000 Blackwell SFF Edition, which a later section covers. A search for "RTX PRO 2000 Blackwell SFF" or "RTX A1000 SFF" leads to the standard card, because NVIDIA's desktop line-up page of 8 September 2026 lists each of them in one desktop version.

One slot or two at the same height

Height is what the five share. Width separates them, and width is the figure most often mixed up.

• Single slot: RTX A400 and RTX A1000, both 6.4 in long, and the RTX 2000E Ada at 6.6 in.
• Dual slot: RTX 2000 Ada and RTX PRO 2000 Blackwell, both 6.6 in long.

The RTX 2000 Ada is the usual source of the confusion. It is often searched as a single-slot card, and its datasheet says "Dual Slot". The 16 GB card for one slot is the RTX 2000E Ada. It is a separate product: memory and CUDA core count match the RTX 2000 Ada, the board power is 20 W lower, and the single-precision figure in the datasheet drops from 12.0 to 8.9 TFLOPS. NVIDIA has no product page for the 2000E, so its figures rest on the datasheet of October 2025. Our comparison RTX 2000 Ada vs RTX 2000E Ada sets the two side by side, and the page on single-slot NVIDIA RTX cards extends the question to the full-height cards.

For the chassis this means two checks instead of one. Where the position next to the graphics slot holds a network or capture card, only the three single-slot cards go in. Where two slot widths are free, the two 70 W cards become possible, and with them the RTX PRO 2000 Blackwell, whose 17 TFLOPS in single precision are the highest datasheet figure of the five.

50 W and 70 W through the slot

None of the five datasheets has a row for a power connector, and what the other documents add differs from card to card. For the two 70 W cards NVIDIA is explicit: its quick start guides for the RTX Ada and the RTX PRO Blackwell cards say that the RTX 2000 Ada and the RTX PRO 2000 Blackwell do not use a power cable. For the RTX A1000 and the RTX A400, NVIDIA's datasheet gives a maximum power consumption of 50 W and names no auxiliary power connector. The datasheet of the RTX 2000E Ada gives 50 W of total board power as well, and here it is Lenovo's product guide that states the card does not require an auxiliary power cable.

NVIDIA publishes no power-supply wattage for any of the five. Its power guidelines for workstation products, version 15 of 27 July 2023, put the typical rating of a PCIe slot at 75 W; that version predates NVIDIA's announcements of the RTX 2000 Ada, the RTX A1000, the RTX A400 and the RTX PRO 2000 Blackwell and is a reference point only. What a given slot delivers is decided by the maker of the system, so the chassis documentation is the place to confirm that the half-height slot is rated for the 50 W or 70 W of the card. The article on NVIDIA RTX power connectors goes through the same question for the full-height cards as well.

All five use eight PCIe lanes: PCIe 4.0 x8 on the two A-series and the two Ada cards, PCIe 5.0 x8 on the RTX PRO 2000 Blackwell. The datasheets add in a footnote that the cards use a full-length PCIe interface, and NVIDIA's quick start guides for the RTX 2000 Ada and the RTX PRO 2000 Blackwell ask for a PCI Express x16 slot.

We supply all five cards across the EU; send us the chassis model, the slot height and the quantity and we reply with a written quote within one business day.

Four Mini DisplayPort outputs and the resolutions behind them

Every card of the five has four Mini DisplayPort outputs and no HDMI output, and each drives four displays at most. The resolution lines in the datasheets are alternatives and do not add up.

the "Max Simultaneous Displays" rows of the datasheets named under the first table; for the RTX PRO 2000 Blackwell the multi-monitor footnote of datasheet 5322351.

The 8K line of the RTX A1000 ends at 30 Hz, and the RTX A400 has no 8K line at all. For a desk with four 4K monitors all five are sufficient by their datasheets; for 8K at 60 Hz or more the choice starts at the 16 GB cards.

The datasheets of the RTX A400, the RTX A1000, the RTX 2000E Ada and the RTX 2000 Ada list NVIDIA Mosaic, the RTX 2000 Ada with the footnote that this configuration offers no frame-lock synchronisation and no display overlap. None of the five is listed for NVIDIA's sync board; the rule in NVIDIA's RTX PRO Sync user guide is "4000-class and above". Mosaic on these cards therefore means one card with up to four displays.

4 GB, 8 GB or 16 GB for the seat

Once the slot is settled, memory decides. The table adds the rows that differ most between the five. CUDA core counts compare only inside one generation: Ampere for the two A-series cards, Ada Lovelace for the two Ada cards, Blackwell for the RTX PRO 2000.

the datasheets named under the first table; AV1 and 4:2:2 encode rows from NVIDIA's Video Encode and Decode GPU Support Matrix of 16 October 2026, which does not list the RTX 2000E Ada.

4 GB, RTX A400. The card for display desks, 2D work and light 3D. On the makers' pages read on 10 November 2026 its 4 GB lie below the 6 GB minimum that Twinmotion and Lumion name and below every 8 GB recommendation. See the RTX A400 specs and the page to buy the RTX A400.

8 GB, RTX A1000. Three times the CUDA cores of the RTX A400, twice the memory and twice the bandwidth in the same single slot at the same 50 W. It meets the 8 GB memory figures in the makers' pages, such as the 8 GB in Autodesk's recommended line for AutoCAD 2027, and it stays below the 10, 12 and 16 GB tiers. See the RTX A1000 specs and the page to buy the RTX A1000. NVIDIA's support matrix gives both A-series cards at most 8 concurrent NVENC sessions and no AV1 encode.

16 GB in one slot, RTX 2000E Ada. The only card of the five that combines 16 GB and ECC with a single slot and 50 W: RTX 2000E Ada specs, buy the RTX 2000E Ada.

16 GB in two slots, RTX 2000 Ada or RTX PRO 2000 Blackwell. Both draw 70 W. The Blackwell card has 17 TFLOPS in single precision against 12.0, 288 GB/s against 224 GB/s, GDDR7 instead of GDDR6, PCIe 5.0 instead of 4.0 and 4:2:2 encoding; its Tensor Cores add FP4, where the Ada generation stops at FP8. Puget Systems' professional GPU roundup of 18 December 2025, run on a Ryzen 9 9950X3D, measured the RTX PRO 2000 Blackwell 31 per cent ahead of the RTX 2000 Ada in Blender Cycles and 44 per cent ahead in V-Ray GPU. The Ada card remains the sound order where a fleet and its images are validated on it. The pages: RTX 2000 Ada specs, buy the RTX 2000 Ada, RTX PRO 2000 Blackwell specs, buy the RTX PRO 2000 Blackwell, and the existing comparison RTX PRO 2000 Blackwell vs RTX 2000 Ada.

For local AI models the same three steps apply. By our sizing rule a card has 0.6, 4.2 or 11.3 GiB for model weights plus KV cache at 4, 8 or 16 GB: 0.9 times the memory the driver reports, less 3 GiB. We scale the reported memory from the 95.6 GiB the driver shows on the 96 GB card, which is an assumption, and the driver on the delivered card decides. The step from 8 GB to 16 GB in this format is worked through in RTX A1000 vs RTX 2000 Ada vs RTX PRO 2000. If the choice between two of the five is open, describe the application, the monitors and the chassis, and we quote the card that fits, or both.

Half-height cards beyond the five, RTX PRO 4000 SFF and L4

Three further cards belong in the picture: two half-height cards with display outputs and the L4, a low-profile card built for servers.

NVIDIA datasheet 4016700 (August 2025) for the RTX PRO 4000 Blackwell SFF Edition; NVIDIA's product page and line card of March 2025 for the RTX 4000 SFF Ada Generation; L4 figures as in our comparison of the 70 W cards.

The RTX PRO 4000 Blackwell SFF Edition has the most memory of the half-height cards with display outputs on this page: 24 GB at 432 GB/s, 8,960 CUDA cores, 70 W without a power cable, 2.7 by 6.6 in, dual slot. Its datasheet uses the words "half height". The L4 is a different kind of card. It is passive, has no display outputs and belongs in servers whose airflow cools it, which is why it is no substitute for the five cards above in a desktop. Both are compared in RTX PRO 4000 SFF vs RTX PRO 2000 vs L4, and the server side is covered in our article on moving from the NVIDIA T4 to the L4, the RTX PRO 2000 or the RTX PRO 4000 SFF.

Replacing an older half-height card in the same slot

Where a half-height card takes the place of one that is already in the machine, the old card tells which of the five fits.

A single-slot NVIDIA T400, T600 or T1000 or a Quadro P400, P620 or P1000 leaves room for one slot width. Their datasheets give 30 W, 40 W and 50 W for the three T cards and 30 W, 40 W and 47 W for the three Quadro cards, so the RTX A400 and the RTX A1000 at 50 W can draw more than the card whose slot they take; the power budget of the chassis is worth a look. Our article on replacing Quadro P1000, T400 and T1000 cards goes through the pairs.

An RTX A2000, with 6 GB or 12 GB, is a dual-slot card of 2.7 by 6.6 in at 70 W. The RTX 2000 Ada and the RTX PRO 2000 Blackwell have the same height, length, slot width and power and bring 16 GB.

The whole range from 4 GB to 48 GB, including the full-height cards, is mapped in our guide to NVIDIA RTX workstation GPUs. For a refresh of many seats, tell us which cards are installed today and in which systems, and the written quote names a card for each of them.

FAQ