NVENC session limits on RTX PRO, RTX Ada, RTX A1000 and A400: unrestricted, 8 or 12

NVENC session limit by card, from unrestricted to 8

How many NVENC sessions a card runs at once depends on the card, and NVIDIA publishes the figure per card in its Video Encode and Decode GPU Support Matrix. In the matrix of 16 October 2026 the RTX PRO Blackwell workstation cards and the RTX 5000, 4500 and 2000 Ada carry "Unrestricted" in the column for concurrent sessions, the RTX A1000 and RTX A400 carry 8, and the GeForce RTX desktop cards of the 50, 40 and 30 Series carry 12. The limit of three sessions that is still often quoted for GeForce cards is not in this document.

The figure matters to anyone who builds a system that encodes several video streams at once: a recorder for meeting rooms or lecture halls, a transcoding node, a control-room workstation that streams its screens, a video-analytics system that writes annotated streams back to disk. This page gives the engines and the session line of eleven of the NVIDIA professional cards we supply, sets the GeForce rows beside them, and separates two questions that are often mixed up: how many sessions a card admits, and how many encoder engines it has.

NVENC engines and sessions on the eleven professional cards

NVIDIA's Video Encode and Decode GPU Support Matrix of 16 October 2026 for encoders, sessions, AV1 and 4:2:2 encode; decoder counts and the two rows without a matrix line from NVIDIA's datasheets of each card. Each card name links to its spec page.

Two things in the table run against what the product class suggests. The two smallest professional cards have the lowest figure of all, 8 sessions on the RTX A1000 and RTX A400, below the 12 of a GeForce card. As a general sentence, "professional cards have unlimited NVENC sessions" is wrong. "Unrestricted", in turn, says nothing about capacity, since the RTX PRO 2000 and the RTX 2000 Ada are unrestricted with a single encoder engine.

For two of the eleven cards we have no session line to cite. The RTX 2000E Ada is not in the matrix at all; its datasheet gives one encoder and one decoder with AV1. For the NVIDIA RTX PRO 4500 Blackwell Server Edition we have three encoders and three decoders from its datasheet and no matrix row among those we read. The matrix row "NVIDIA RTX PRO 4500 Blackwell" names no edition and has two encoders; we print it as the Workstation Edition, whose datasheet gives two encoders and two decoders. NVIDIA's product page describes the Server Edition's ninth-generation encoder with 4:2:2 H.264 and HEVC encoding.

The text of the matrix we read contains no footnote that explains the limit. Its session column holds one of a few values: 12 in the GeForce table, 8 for a set of small professional cards, and "Unrestricted" for most RTX PRO, RTX Ada, RTX A and Quadro rows and for the server table.

For systems that encode video we supply all eleven cards of this page across the EU: name the number of streams, the codec and the chassis, and we reply within one business day with a written quote for the card that fits.

GeForce RTX desktop cards in the same matrix have 12 sessions

The same matrix lists the GeForce desktop cards, and the comparison is the usual reason for the search. Every GeForce RTX row in the table below carries 12 in the session column.

NVIDIA's Video Encode and Decode GPU Support Matrix of 16 October 2026, desktop rows of the cards named; maximum number of concurrent sessions 12 in each of these rows.

The GeForce rows show why the session figure alone is a weak reason to choose a card. By the matrix a GeForce RTX 5090 has three encoder engines and 12 sessions, an RTX PRO 2000 Blackwell one engine and no session limit. A design with more than 12 concurrent streams on one card is outside the GeForce rows; a design with fewer is decided by other things.

What NVIDIA documents for the professional cards beyond this column lies outside the encoder: memory with ECC on the Blackwell and Ada cards, the RTX Enterprise driver branch, and for a machine in a rack the driver licence. Section 2.8 of NVIDIA's driver licence agreement, as read on 10 November 2026, says GeForce or Titan software "is not licensed for datacenter deployment"; the clause concerns the driver licence, not the hardware. Our article on GeForce in a server and the driver licence sets out the text. We supply professional cards to companies and do not sell GeForce cards. Card-by-card comparisons by published figures are in RTX PRO 2000 Blackwell vs RTX 5060 Ti and 5070 and RTX A1000 and A400 vs RTX 3050.

Engines and sessions are two columns of the matrix

The matrix has one column for the total number of NVENC engines and another for the maximum number of concurrent sessions. The first counts hardware, the encoder engines on the card. The second caps how many encode sessions may be open at the same time. As we read it, a session is one stream being encoded: one camera being recorded, one rendition of a transcode, one screen being streamed.

For sizing, the two columns bound different things, and what follows is our reading, since the matrix explains neither. The session figure bounds the number of streams. A recorder for ten rooms cannot run on one RTX A1000 with its 8 sessions, whatever the resolution. The engine count bounds throughput. All sessions of a card share its engines, so a card with three engines has three times the encoding hardware of a card of the same generation with one, and the frame rate left for each stream falls as streams are added.

The matrix gives no frames per second, and the NVIDIA documents behind this page contain no encode throughput for any of the eleven cards. How many streams a card carries at a given resolution, codec and preset is therefore a figure to measure on the delivered system, with the software that will run on it.

Decoders are counted separately. For camera systems they are usually the first limit, and NVIDIA's decode rates per engine for data-centre cards are in our guide to GPU servers for video analytics. Between the two RTX A cards the decoder count is the one difference in the video engines: two on the RTX A1000, one on the RTX A400, as our comparison RTX A1000 vs RTX A400 shows.

AV1 encode and 4:2:2 encode by generation

AV1 encode. The matrix lists AV1 encode for every RTX PRO Blackwell and RTX Ada row of the table above and for the GeForce RTX 50 and 40 Series. It lists none for the RTX A1000, the RTX A400 and the GeForce RTX 30 Series. The datasheets of the two RTX A cards say the same in their own way: the video line names AV1 for decode only. For AV1 output from a card 2.7 in high the candidates are the RTX 2000 Ada, which NVIDIA's blog of 12 February 2024 introduced with an eighth-generation encoder with AV1 support, the RTX 2000E Ada, whose datasheet names AV1 for its encoder and decoder, and the RTX PRO 2000 Blackwell.

4:2:2 encode. Among the rows on this page, H.264 and HEVC encoding in 4:2:2 is in the matrix for the RTX PRO Blackwell workstation cards and the GeForce RTX 50 Series. The Blackwell datasheets name a ninth-generation NVENC and a sixth-generation NVDEC, and NVIDIA's product pages describe the pair as adding 4:2:2 H.264 and HEVC encoding and decoding. No Ada row has it; for the RTX 5000 Ada the matrix lists 4:4:4 and no 4:2:2. Whether an editing application uses the hardware path is the application maker's matter: Adobe's Premiere pages, read 10 November 2026, name hardware encoding and decoding for H.264 and HEVC and mention neither NVENC nor AV1 nor NVIDIA 4:2:2. Our guide to NVIDIA cards for Premiere Pro and DaVinci Resolve covers the applications.

H.264 and HEVC. Every matrix row cited on this page lists both. For the RTX A1000 and RTX A400 the matrix lists H.264 in 4:2:0, 4:4:4 and lossless, and HEVC up to 8K with 10-bit, 4:4:4 and B-frames.

RTX A1000 and RTX A400 with 8 sessions and one encoder

The two 50 W cards are the ones where the session line can decide a design. Both have one seventh-generation encoder; the RTX A1000 has two fifth-generation decoders, the RTX A400 one. The matrix gives both at most 8 concurrent sessions and no AV1 encode.

For the desks these cards are bought for, that is rarely a constraint. A workstation that records its screen and sends one stream into a video conference opens two sessions, by the reading above. The line starts to matter in three cases: a recorder or encoder box with more than eight channels, a delivery format in AV1, and footage or output in 4:2:2. The next cards with a documented unrestricted line are the RTX 2000 Ada and the RTX PRO 2000 Blackwell, both 2.7 in high and dual slot, with 70 W and no power cable.

Older small cards carry the same line. The matrix lists the T1000, T600 and T400 and the Quadro P400, P620 and P1000 with a sixth-generation encoder and 8 sessions, while the Quadro P2000 and P2200 and the RTX A2000 are unrestricted. "Unlimited NVENC on Quadro" does not describe the whole range either. Moving such a fleet to current cards is the subject of our article on upgrading from the Quadro P1000, T400 and T1000 to the RTX A400 and A1000.

Cards for recording and transcoding systems by stream count

• Up to 8 streams in H.264 or HEVC in a small chassis. The RTX A1000 and the RTX A400 stay within their documented line, with one encoder each.
• More than 8 streams, or AV1, at 70 W. The RTX 2000 Ada and the RTX PRO 2000 Blackwell are unrestricted with one encoder; the RTX PRO 2000 adds 4:2:2 encode.
• One slot and 50 W. The RTX 2000E Ada has one encoder and one decoder with AV1 by its datasheet. It is not in the matrix, so a design that depends on a session count should not assume one for it. Our article on the RTX 2000E Ada for edge AI covers the card.
• Two encoders. The RTX PRO 4000 Blackwell has two encoders and two decoders in a single slot at 145 W, with AV1 and 4:2:2. The RTX PRO 4500 Workstation Edition, the RTX 4500 Ada and the RTX 5000 Ada have two encoders as well, the Ada cards without 4:2:2.
• Three encoders in a workstation. The RTX PRO 5000 Blackwell with 48 GB is the card of the eleven with three encoders, three decoders and a session line in the matrix.
• Rack server. The RTX PRO 4500 Server Edition has three encoders and three decoders at 165 W on a passive single-slot card. Check its row in NVIDIA's current matrix before you size on a session count.

If the stream count is fixed and the card is not, describe the recording or transcoding system, with streams, resolution and codec, and the written quote names the card that fits.

FAQ