On requestAda Lovelace

NVIDIA RTX 4000 Ada Generation

Four single-slot RTX 4000 Ada GPUs per server, each with 20 GB of ECC GDDR6 and two encode and two decode engines with AV1, for live and on-demand transcoding, streaming and light rendering.

GPUs per server
4
Memory per GPU
20 GB GDDR6 ECC
Encode engines per GPU
2, with AV1
Power per GPU
130 W

Specifications

Contracts run 36 to 60 months. Every price is given in a written proposal.

GPU
GPU
ArchitectureNVIDIA Ada Lovelace
Memory20 GB GDDR6 with ECC, 160-bit
Memory bandwidth360 GB/s
CUDA cores6,144
Tensor Cores192, fourth generation
RT Cores48, third generation
FP3226.7 TFLOPS
RT Core performance61.8 TFLOPS
Video engines2 encode, 2 decode, AV1 encode and decode (eighth-generation NVENC)
System interfacePCIe 4.0 x16
Power per GPU130 W
Form factorSingle slot, 4.4 in x 9.5 in, active cooling
Server as delivered
Server as delivered
GPUs per server4
GPU memory per server80 GB (4 x 20 GB)
Video engines per server8 encode, 8 decode (4 x 2)
GPU board power per server520 W (4 x 130 W)
CPU, system memory, storage and networkConfigured to your requirements and confirmed in your proposal

Typical workloads

  • Live transcoding and adaptive bitrate ladders
  • Video on demand transcoding with AV1
  • Streaming and broadcast workflows
  • Light rendering and visualization

Figures from NVIDIA product documentation: NVIDIA RTX 4000 Ada Generation, NVIDIA RTX 4000 Ada Generation datasheet (September 2023).

Rent dedicated 4-GPU servers with NVIDIA RTX 4000 Ada Generation cards for video transcoding, streaming and light rendering. RTX 4000 Ada servers are available on request, configured to your requirements, on 36, 48 or 60 month contracts.

Who the RTX 4000 Ada is for

The RTX 4000 Ada is for media teams that transcode and stream: live channels, video on demand libraries and broadcast workflows that need AV1. It is a single-slot, 130 W professional card on the Ada Lovelace architecture, with 20 GB of GDDR6 with ECC.

We rent it as whole 4-GPU servers with 80 GB of GPU memory and eight hardware encoders each. Pipelines that grow add servers, and each server stays dedicated to you.

Encoding on the GPU's dedicated engines leaves the host CPU free for packaging and the other steps of the pipeline. The ECC memory corrects single-bit errors, which matters for channels that run for months without a restart.

What it does well

Encode and decode

Two encode and two decode engines per GPU, with AV1 encoding on NVIDIA's eighth-generation NVENC and AV1 decoding: eight encoders and eight decoders per server.

Memory with ECC

20 GB of GDDR6 with error-correcting code per GPU, at 360 GB/s on a 160-bit bus.

Graphics and compute

6,144 CUDA cores at 26.7 TFLOPS of FP32, 48 third-generation RT Cores at 61.8 TFLOPS and 192 fourth-generation Tensor Cores.

Low power

130 W per GPU in an actively cooled, single-slot card, 520 W of GPU board power per server.

The Tensor Cores and 20 GB per GPU also leave room to run small vision models next to the transcode, for tasks such as scene detection or thumbnail selection, without a separate set of servers.

Configuration as delivered

RTX 4000 Ada servers are configured to your requirements and confirmed in your proposal: CPU, system memory, local NVMe and network. Transcoding leans on the host as well as the GPU, so tell us your input formats, output ladders and stream counts, and we size CPU and network around them.

Every server is burned in before handover, with the results in an acceptance report. Our burn-in and acceptance page explains the process.

How renting works

RTX 4000 Ada servers are available on request. Tell us the number of servers, the term and the start date, and we reply with a written proposal that sets out the configuration and delivery schedule.

Contracts run 36, 48 or 60 months, and longer terms carry lower rates. The company that will run the servers completes three onboarding forms before signing. You pay a down payment at signing, then one fixed monthly rate for the full term. There are no egress fees, so the volume you stream out is not billed per gigabyte. Live channels usually run around the clock, and a fixed rate for 36 to 60 months means the cost of each channel is known for the whole term. Our how to buy page covers each step.

Workloads and alternatives

Stream density depends on codec, resolution, frame rate and preset, so we size the number of servers with you rather than publish a stream count. Our media and entertainment page covers transcoding and streaming pipelines. Moving your transcoding solution to the cloud with the RTX 4000 Ada covers migration, and 10 questions that reveal the real cost of your cloud GPU setup covers egress and utilization.

RTX 4000 AdaRTX A4000L40S
Memory per GPU20 GB GDDR6, ECC16 GB GDDR6, ECC48 GB GDDR6, ECC
Encode and decode engines2 and 21 and 13 and 3
AV1Encode and decodeDecode onlyEncode and decode
Power per GPU130 W140 W350 W
GPUs per server448

For decode-heavy work without AV1 encoding, see the RTX A4000. For more engines and 48 GB per GPU, see the L40S.

Questions buyers ask

Why are RTX 4000 Ada servers rented with four GPUs?

We rent the smaller workstation cards as whole 4-GPU servers. Each server gives you 80 GB of GPU memory and eight hardware encoders; larger pipelines take more servers.

How many streams can one GPU transcode?

It depends on codec, resolution, frame rate and preset, so we do not publish a stream count. NVIDIA rates each card with two encode and two decode engines, and we size the number of servers with you before the proposal.

Does the RTX 4000 Ada support AV1?

Yes. NVIDIA lists AV1 encode and decode for the RTX 4000 Ada Generation, with encoding on its eighth-generation NVENC.

Are there egress fees on the streams we send out?

No. Every 1Legion contract bills one fixed monthly rate for the full term, with no egress fees.

What contract terms are available?

36, 48 or 60 months, with lower rates for longer terms. You pay a down payment at signing, then one fixed monthly rate for the full term.

Rent RTX 4000 Ada servers on 36 to 60 month terms.

Share the hardware, the number of servers, the term and the start date. You receive a written proposal built on our delivery specification.