NVIDIA RTX A1000 specs: 8 GB GDDR6, 2,304 CUDA cores, 50 W in a single slot

RTX A1000 specs in five figures

The NVIDIA RTX A1000 is a desktop workstation graphics card with 8 GB of GDDR6 memory, 2,304 CUDA cores and a total board power of 50 W. It occupies a single slot at 2.7 in of height and 6.4 in of length, has a memory bandwidth of 192 GB/s and four Mini DisplayPort 1.4a outputs. NVIDIA announced it on 16 April 2024 and builds it on the Ampere architecture.

Every value below comes from NVIDIA's datasheet of April 2024 unless the table names another document. Where NVIDIA publishes nothing, the row says so; we do not fill such rows from shop listings.

The complete RTX A1000 spec sheet

NVIDIA RTX A1000 datasheet (April 2024); NVIDIA blog of 16 April 2024; NVIDIA's line-up page "RTX PRO in Desktops" (8 September 2026); NVIDIA's Video Encode and Decode GPU Support Matrix of 16 October 2026; NVIDIA's "CUDA GPU Compute Capability" page (16 October 2026).

The datasheet footnotes the single-precision figure as a peak rate based on the GPU boost clock. The two Tensor figures are footnoted with sparsity, and FP16 TFLOPS and INT8 TOPS are different units that cannot be set against each other.

Power consumption and the connector question

NVIDIA's datasheet gives a maximum power consumption of 50 W and names no auxiliary power connector. It names no power-supply wattage either, so this page prints none. For the power budget of a compact chassis, 50 W is the number to plan with. As a reference point, NVIDIA's power guidelines for workstation products of July 2023, written before this card and without a row for it, put the typical rating of a PCIe slot at 75 W.

The host link is PCIe 4.0 with eight lanes. A footnote in the datasheet adds that the card uses a full-length PCIe Gen 4 x8 interface.

Dimensions, single slot and small-form-factor systems

NVIDIA gives the size in inches only: 2.7 high, 6.4 long, single slot. Converted, that is about 69 by 163 mm; a millimetre figure and a weight are not published. At 2.7 in the card is half-height, the format usually called low profile, and NVIDIA's datasheet itself describes the card as a "low-profile GPU".

The single slot separates the RTX A1000 from two of the three 16 GB cards of the same height: the NVIDIA RTX 2000 Ada Generation and the NVIDIA RTX PRO 2000 Blackwell are 6.6 in long and take two slots. The third, the NVIDIA RTX 2000E Ada Generation, is also 6.6 in long and stays in a single slot.

Dassault's SOLIDWORKS 2026 certification list, read 10 November 2026, has rows for the card in desktop workstations, among them the HP Z1 Tower G9 and Lenovo's ThinkStation P3 Tiny Gen 2, P3 Ultra Gen 2 and P3 Tower Gen 2. Each row stands for one computer model with one card and one driver, and the same list holds laptop rows under the name RTX A1000. We supply the RTX A1000 to companies across the EU; name the workstation models and the number of cards and you receive a written quote within one business day.

DisplayPort outputs, maximum resolution and Mosaic

The four outputs are Mini DisplayPort 1.4a, and the card drives four displays at most. The datasheet's lines for simultaneous displays are alternatives:

• four displays at 4096 × 2160 and 120 Hz;
• four displays at 5120 × 2880 and 60 Hz;
• two displays at 7680 × 4320 and 30 Hz.

The 8K line is the one to read twice, because it ends at 30 Hz. The card has no HDMI output, and HDCP 2.2 is listed.

The datasheet lists NVIDIA Mosaic technology with the footnote "Windows 10 and Linux". To synchronise Mosaic across two or more GPUs, NVIDIA's documentation calls for its sync card, and the RTX A1000 is not on NVIDIA's RTX PRO Sync list of 15 October 2026. Plan Mosaic with this card as one card with up to four displays; our article on four monitors with the NVIDIA RTX A400 and the RTX A1000 covers the setup. One line of the datasheet has aged: it names RTX Desktop Manager, and the release notes of driver 621.14 state that Release 580 was the last branch to support it.

One encoder and two decoders for video

The datasheet's video row reads "1x encode, 2x decode (+AV1 decode)". NVIDIA's blog of 16 April 2024 names the engines as seventh-generation NVENC and fifth-generation NVDEC. Three limits follow from NVIDIA's support matrix of 16 October 2026: the card encodes H.264 and HEVC but not AV1, 4:2:2 encoding is not listed, and at most 8 NVENC sessions run at once. The RTX Ada and RTX PRO Blackwell cards in the same matrix are listed as unrestricted, a difference our article on NVENC session limits works through.

Drivers, CUDA version and operating systems

The card runs on the NVIDIA RTX Enterprise Production Branch driver and is listed in the newest release of each current branch.

NVIDIA RTX Enterprise release notes 621.14, 597.30 and 582.99 and NVIDIA's Data Center GPU Driver release notes for the same releases, as read on 10 November 2026.

The notes name Windows 11 from 22H2 to 25H2 and Windows 10 22H2, and Release 620 is the last branch for Windows 10. The Linux distributions in the notes include, among others, the Ubuntu LTS releases 22.04, 24.04 and 26.04, RHEL and Rocky from 8.y to 10.x, SLES 15 SP7 and 16, and Debian 12 and 13.

The "CUDA 11.6" in the datasheet is the version at the card's launch and no ceiling. The CUDA version follows the driver: CUDA 13.5 pairs with R620, 13.2 with R595 and 13.0 with R580, and CUDA 13.x applications need a driver of at least 580. The Shader Model version in the table carries the date of the datasheet in the same way.

NVIDIA's "CUDA GPU Compute Capability" page of 16 October 2026 gives 8.6 for the Ampere cards RTX A2000, A4000, A5000 and A6000 and does not name the RTX A1000, so we print no figure for it. NVIDIA's product page names DLSS among the card's AI features without stating a version.

What NVIDIA does not list for the RTX A1000

• ECC. NVIDIA's line-up page writes "GDDR6 with ECC" for the RTX Ada cards and plain "8 GB GDDR6" for this card.
• AV1 encode and 4:2:2 encode. Not listed in the support matrix.
• MIG and vGPU. The card is in neither NVIDIA's MIG guide (4 November 2026) nor its vGPU list (2 October 2026), and nothing is documented on SR-IOV or passthrough.
• Sync board. Not on the RTX PRO Sync list.
• NVLink. Not mentioned in the datasheet.
• FP8 and FP4. NVIDIA publishes Tensor figures for FP16 and INT8 only and mentions neither precision for this card.
• A wattage for the power supply, a weight and millimetre dimensions.

RTX A1000 with 4 GB or 6 GB is the laptop GPU

Searches for "RTX A1000 4GB" and "RTX A1000 6GB" lead to different products, because NVIDIA uses the name for two laptop GPUs as well.

NVIDIA RTX A1000 datasheet (April 2024); NVIDIA's laptop line card of February 2024.

Only the first row is a PCIe card. The 6 GB laptop GPU has more CUDA cores than the desktop card and less memory, so a benchmark or a certification row has to be read with the full product name; driver lists write the laptop parts as "RTX A1000 Laptop GPU" and "RTX A1000 6GB Laptop GPU". A fourth product, an embedded module called RTX A1000 on NVIDIA's page on RTX PRO embedded GPU solutions (31 July 2026), has 4 GB and 60 W.

Performance figures NVIDIA and others have published

NVIDIA measures the card against the T1000 8GB, which its datasheet of November 2021 lists with 896 CUDA cores, no RT or Tensor Cores, 8 GB and the same 50 W in a single slot; our article on replacing Quadro P1000, T400 and T1000 cards covers that step. NVIDIA's product page states up to 3.6 times the generative AI performance of that card; the test note names Stable Diffusion webUI v1.7.0 with TensorRT at 512 × 512, a Core i9-12900K and a pre-release build.

Independent numbers exist for one system. StorageReview's review of the Lenovo ThinkStation P3 Tiny Gen 2 (20 November 2025, modified 14 August 2026; Core Ultra 9 285) reports 39.379, 27.734 and 24.323 tokens per second for the RTX A1000 with the Phi, Mistral and Llama3 models of UL Procyon's AI text generation test, 11.196 s per image with Stable Diffusion 1.5 in FP16, and 581.75, 363.47 and 323.37 samples per minute in the Monster, Junkshop and Classroom scenes of the Blender benchmark. Blender Open Data, read 10 November 2026, shows a median of 1,248.25 samples per minute over 6 submissions with Blender 4.5.0; the database is crowd-sourced.

For AI models the memory decides first. By our sizing rule an 8 GB card has a budget of 4.2 GiB for weights plus KV cache: 0.9 times the memory the driver reports, less 3 GiB, with the reported memory scaled from the 95.6 GiB the driver shows on the 96 GB card. The scaling is an assumption, and the driver on the delivered card decides. Granite 4.2 3b in FP8, a 3.89 GiB file on Hugging Face as read on 10 November 2026, fits with 0.3 GiB to spare. For speech recognition, OpenAI states about 6 GB of VRAM for Whisper turbo and about 10 GB for Whisper large-v3, so the first lies within the card's 8 GB and the second above it. Our article on AI on the RTX A1000 and RTX A400 lists what else fits.

Where the RTX A1000 sits in the line-up

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).

Below the card, the RTX A400 keeps the 50 W and the single slot with half the memory, one decoder instead of two and no 8K line: RTX A400 specs, buy the RTX A400, and the comparison RTX A1000 vs RTX A400. Above it are three 16 GB cards of the same height. The RTX 2000E Ada stays in a single slot at 50 W: RTX 2000E Ada specs, buy the RTX 2000E Ada. The RTX 2000 Ada and the RTX PRO 2000 Blackwell take two slots and 70 W. The RTX 2000 Ada is described in its specs and on the page to buy the RTX 2000 Ada, the RTX PRO 2000 Blackwell in its specs and on the page to buy the RTX PRO 2000 Blackwell; RTX A1000 vs RTX 2000 Ada and RTX PRO 2000 compares the RTX A1000 with these two.

The step to 16 GB is due when a software maker names more than 8 GB. On the makers' pages read on 10 November 2026, the card meets the 8 GB memory figures, such as the 8 GB in Autodesk's recommended line for AutoCAD 2027, and it lies below the 10, 12 and 16 GB tiers that Lumion, Twinmotion and Autodesk's table for large datasets name. Our guide to the GPU for CAD workstations goes further.

NVIDIA names no successor for the RTX A1000. The card is on NVIDIA's desktop line-up page of 8 September 2026 and in the R620 driver branch, and ordering is described on our page to buy the RTX A1000. If you equip a number of seats, write to us with the applications and the chassis, and we quote the RTX A1000 or the 16 GB card that fits.

FAQ