NVIDIA RTX A1000 vs RTX A400: the difference between the 8 GB and the 4 GB card

RTX A1000 vs RTX A400 in one table

The NVIDIA RTX A1000 and the NVIDIA RTX A400 are boards of the same size with the same 50 W and the same four outputs, and they differ in what is on the board: the RTX A1000 has 8 GB of memory and 2,304 CUDA cores, the RTX A400 has 4 GB and 768. The RTX A400 is the card for a desk that mainly needs displays. The RTX A1000 is the card as soon as 3D models, several video streams or a software maker's 8 GB figure come into play.

NVIDIA RTX A400 datasheet 3279761 (May 2024); NVIDIA RTX A1000 datasheet (April 2024); NVIDIA's laptop line card of February 2024. The last row is our reading.

What is identical on both cards

Every published value that concerns the chassis is the same for the two cards, and so are the generation, the encoder and the drivers.

• Board. 2.7 in high, 6.4 in long, single slot, active cooling; converted, about 69 by 163 mm. NVIDIA's datasheet describes the RTX A1000 as a "low-profile GPU", and its product page for the RTX A400 carries the heading "Low-Profile, Single-Slot Form Factor".
• Power. For each card, NVIDIA's datasheet gives a maximum power consumption of 50 W and names no auxiliary power connector. A power-supply wattage is not published.
• Host interface. PCIe 4.0 x8; a footnote in each datasheet adds that the card uses a full-length PCIe Gen 4 x8 interface.
• Outputs. Four Mini DisplayPort 1.4a outputs with HDCP 2.2 and no HDMI.
• Architecture. NVIDIA Ampere with third-generation Tensor Cores and second-generation RT Cores. NVIDIA announced both cards on 16 April 2024.
• Encoder. One seventh-generation NVENC engine, at most 8 concurrent sessions and no AV1 encoding in NVIDIA's support matrix of 16 October 2026.
• Drivers. Both are listed in the newest releases of the R580, R595 and R620 branches of the RTX Enterprise driver.
• Not listed for either card. ECC, a compute capability on NVIDIA's CUDA page, MIG, vGPU and the RTX PRO Sync board.

For a fleet this means that the slot, the power budget, the display cables and the driver branch stay the same whichever of the two is ordered. A mixed order raises no new question in the chassis.

Three times the cores, twice the memory and bandwidth

NVIDIA RTX A400 datasheet 3279761 (May 2024) and product page; NVIDIA RTX A1000 datasheet (April 2024). The Tensor figures are effective figures with sparsity.

The core counts stand in a ratio of exactly three to one, for CUDA, Tensor and RT Cores alike. Memory and bandwidth double. NVIDIA's performance figures grow by about 2.5 times, by our arithmetic from the datasheets: 6.7 against 2.7 TFLOPS in single precision and 53.8 against 21.7 TFLOPS for FP16 on the Tensor Cores. One value exists twice. NVIDIA's datasheet gives 5.4 TFLOPS for the RT Cores of the RTX A400, its product page 5.3.

Independent measurements are thin. Blender Open Data, read 10 November 2026, shows 1,248.25 samples per minute for the RTX A1000 (median of 6 submissions) and 419.90 for the RTX A400 (median of 14) with Blender 4.5.0, a ratio of about 3.0 by our arithmetic. With Blender 5.2.0, whose scores are not comparable with those of 4.5.0, the medians are 1,202.49 (5 submissions) and 445.30 (4). The database is crowd-sourced, it pools OptiX and CUDA runs, and the systems around the cards differ. For the RTX A1000 there is also StorageReview's review of a Lenovo workstation, quoted on our RTX A1000 specs page. For the RTX A400 we know of no independent review, and we found no SPEC result for either card.

NVIDIA's own reference cards differ for the two. It measures the RTX A1000 against the T1000 8GB and the RTX A400 against the T400 4GB; our guide to replacing Quadro P1000, T400 and T1000 cards follows those steps.

Displays and video with 8K and a second decoder on the RTX A1000

Each card drives four displays, one per output. The datasheets give the same two lines for both: four displays at 4096 × 2160 and 120 Hz, or four at 5120 × 2880 and 60 Hz. Only the RTX A1000 has a third line, two displays at 7680 × 4320 and 30 Hz. The RTX A400 datasheet has no 8K line. An 8K panel is therefore a reason for the RTX A1000, and the 30 Hz in that line is a reason to check the 16 GB cards for anything that moves on the screen.

Both datasheets list NVIDIA Mosaic technology with the footnote "Windows 10 and Linux", and neither card is on NVIDIA's RTX PRO Sync list of 15 October 2026. Plan either of them as one card with up to four displays; our article on four monitors with the RTX A400 and RTX A1000 goes into the setup.

For video the datasheet rows read "1x encode, 2x decode (+AV1 decode)" for the RTX A1000 and "1x encode, 1x decode (+AV1 decode)" for the RTX A400. The second decoder matters at a desk that shows many camera or video streams at once. On the encoding side the cards are equal, with one encoder for H.264 and HEVC and the limit of 8 sessions.

We supply both cards across the EU; send us the number of seats and what each seat displays and we reply with a written quote within one business day.

4 GB and 8 GB against the software makers' figures

The makers of CAD, video and visualisation software publish memory figures, and these separate the two cards more clearly than any benchmark we could cite. The table compares memory only. Autodesk's AutoCAD and Inventor lines pair the memory figure with a "Bandwidth" value that Autodesk does not define, and Lumion and Twinmotion also set a benchmark score that we did not check. A cell therefore says which memory figure a card reaches and nothing beyond that.

the makers' requirement pages as read on 10 November 2026: Autodesk for AutoCAD 2027 (27 October 2026), AutoCAD LT 2027 (25 March 2026), Revit 2027 (7 April 2026) and Inventor 2027 (11 May 2026); Adobe Premiere (10 November 2026); the DaVinci Resolve Studio 21.2 readme (4 November 2026); Enscape (19 March 2026); Twinmotion 2026.1 and Lumion 2026 (undated). Memory only.

Three patterns show. With 4 GB the RTX A400 sits at the minimum or entry figure of most applications and reaches a recommended figure only for AutoCAD LT. It is below the 6 GB minimum of Twinmotion and Lumion. With 8 GB the RTX A1000 reaches every 8 GB figure in the table and remains below the 10, 12 and 16 GB tiers that Lumion, Twinmotion and DaVinci Resolve name.

Certification is a separate matter from memory. Dassault's SOLIDWORKS 2026 certification list, read 10 November 2026, has rows with desktop workstations for both cards, each row for one computer model with one card and one driver. One practical note on that list: its filter entry "RTX A400" also returns rows of the RTX A4000, so the card name has to be read in each row.

Our guides to the NVIDIA GPU for AutoCAD, Revit and Inventor and to the GPU for Enscape, Twinmotion, D5 Render and Lumion go through the makers' pages, and the guide to the GPU for CAD workstations covers certification lists.

AI workloads on 4 GB and 8 GB

NVIDIA's product page speaks of "light AI workloads" for the RTX A400. Our sizing rule puts numbers on that. An 8 GB card has a budget of 4.2 GiB for model weights plus KV cache, and a 4 GB card has 0.6 GiB. The rule takes 0.9 times the memory the driver reports and subtracts 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.

For speech recognition OpenAI states about 2 GB of VRAM for Whisper small and about 6 GB for Whisper turbo, which places the first within the RTX A400 and the second within the RTX A1000 only. NVIDIA publishes Tensor figures for FP16 and INT8 for both cards and mentions neither FP8 nor FP4. The article on AI on the RTX A1000 and RTX A400 lists the model files that fit.

RTX A1000 laptop GPU, RTX A4000 and RTX A2000 are other products

Three other products turn up in searches for these two cards.

• RTX A1000 laptop GPUs. NVIDIA uses the name for two laptop GPUs with 4 GB and 6 GB, which have 2,048 and 2,560 CUDA cores at 35 to 95 W. A listing with 4 GB or 6 GB is not the desktop card. NVIDIA lists no laptop GPU called RTX A400; the RTX A500 is a laptop-only GPU.
• RTX A4000. One digit more and a different class: 16 GB of GDDR6 with ECC, 6,144 CUDA cores and 140 W with a 6-pin connector, in a single slot of 4.4 × 9.5 in.
• RTX A2000. The card above the RTX A1000 in the same generation has 6 GB or 12 GB of GDDR6 with ECC at 288 GB/s, 3,328 CUDA cores and 70 W, in two slots at 2.7 × 6.6 in, and the support matrix lists it with unrestricted NVENC sessions.

RTX A400 or RTX A1000 by type of seat

• Trading, dispatch, control-room and office desks with up to four displays. The RTX A400. Its outputs and its two resolution lines are those of the RTX A1000.
• 2D CAD with AutoCAD LT. The RTX A400 reaches the recommended memory figure of 4 GB.
• 3D CAD with AutoCAD, Revit or Inventor. The RTX A1000, for the 8 GB figures of all three.
• Desks that decode many video streams. The RTX A1000, for its second decoder.
• Real-time visualisation and GPU rendering. 4 GB is below the minimum of Twinmotion and Lumion, and 8 GB is below the next tiers they name, 10 GB for Lumion and 12 GB for Twinmotion. Here a 16 GB card is the sounder order: the NVIDIA RTX 2000 Ada Generation (specs, buy) or the NVIDIA RTX PRO 2000 Blackwell (specs, buy), both 2.7 in high in two slots at 70 W. Our comparison RTX A1000 vs RTX 2000 Ada vs RTX PRO 2000 covers that step.
• A fleet with mixed seats. Because the two boards are identical from the outside, the choice can be made seat by seat. If it has to be one card for all seats, the 8 GB of the RTX A1000 reach more of the makers' figures.

The orders are described on the pages to buy the RTX A400 and to buy the RTX A1000, and the full sheet of the smaller card is in our RTX A400 specs. If you are equipping a number of seats, write to us with the applications and the chassis model, and we quote the RTX A400, the RTX A1000 or both, seat by seat.

FAQ