NVIDIA RTX A4000
Four single-slot Ampere RTX A4000 GPUs per server, each with 16 GB of ECC GDDR6 and 448 GB/s of bandwidth, for 3D visualization, rendering previews, video decode and light inference.
- GPUs per server
- 4
- Memory per GPU
- 16 GB GDDR6 ECC
- FP32 per GPU
- 19.2 TFLOPS
- Power per GPU
- 140 W
Specifications
| GPU | |
|---|---|
| Architecture | NVIDIA Ampere |
| Memory | 16 GB GDDR6 with ECC, 256-bit |
| Memory bandwidth | 448 GB/s |
| CUDA cores | 6,144 |
| Tensor Cores | 192, third generation |
| RT Cores | 48, second generation |
| FP32 | 19.2 TFLOPS |
| RT Core performance | 37.4 TFLOPS |
| Tensor performance | 153.4 TFLOPS (with sparsity) |
| Video engines | 1 encode, 1 decode, AV1 decode |
| System interface | PCIe 4.0 x16 |
| Power per GPU | 140 W |
| Form factor | Single slot, 4.4 in x 9.5 in, active cooling |
| Server as delivered | |
|---|---|
| GPUs per server | 4 |
| GPU memory per server | 64 GB (4 x 16 GB) |
| GPU board power per server | 560 W (4 x 140 W) |
| CPU, system memory, storage and network | Configured to your requirements and confirmed in your proposal |
Typical workloads
- 3D visualization and design review
- Rendering previews and look development
- Video decode and analytics on many streams
- Light inference and computer vision
Rent dedicated 4-GPU servers with NVIDIA RTX A4000 cards for visualization, video decode and light AI work. RTX A4000 servers are available on request, configured to your requirements, on 36, 48 or 60 month contracts.
Who the RTX A4000 is for
The RTX A4000 is a single-slot, 140 W professional card on the Ampere architecture, with 16 GB of GDDR6 with ECC. It suits 3D visualization, design review and rendering previews, decode-heavy video work such as analytics on many camera or broadcast streams, and light inference.
We rent it as whole 4-GPU servers with 64 GB of GPU memory. Teams with steady, moderate graphics or video loads get dedicated hardware sized to the job rather than a share of a larger GPU. The ECC memory corrects single-bit errors, which matters for visualization sessions and analytics that run all day.
What it does well
- Memory: 16 GB of GDDR6 with ECC at 448 GB/s on a 256-bit bus, more bandwidth per GPU than the 360 GB/s of the RTX 4000 Ada.
- Graphics: 6,144 CUDA cores at 19.2 TFLOPS of FP32 and 48 second-generation RT Cores at 37.4 TFLOPS.
- AI: 192 third-generation Tensor Cores, rated at 153.4 TFLOPS with sparsity, for computer vision and small models.
- Video: one encoder and one decoder per GPU, with AV1 decode.
- Power: 140 W per card, actively cooled, 560 W of GPU board power per server.
Four RTX A4000 cards add up to 560 W of GPU board power, a fifth of the 2,800 W of eight L40S cards. That suits steady loads that need many modest GPUs rather than a few large ones.
Configuration as delivered
RTX A4000 servers are configured to your requirements and confirmed in your proposal: CPU, system memory, local NVMe and network. Tell us which applications you run and how many users or streams each server should carry, and we size the host around them.
Every server is burned in before handover, with the results in an acceptance report. Our burn-in and acceptance page explains why that replaces a trial.
How renting works
RTX A4000 servers are available on request. Tell us the number of servers, the term and the start date, and we confirm the configuration and delivery schedule in a written proposal.
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, with no egress fees. A fixed rate over 36 to 60 months suits steady loads such as camera analytics or visualization desks, where the work runs every day of the term. Our how to buy page covers each step, and 10 questions to ask your GPU infrastructure provider lists what to check in any contract.
Workloads and alternatives
The RTX A4000 fits 3D visualization, rendering previews, video decode and light inference. In video analytics, each GPU's decoder hands frames to the CUDA and Tensor Cores on the same card, so detection and tracking run next to the decode instead of on a separate system.
Our media and entertainment page covers studio and streaming pipelines, and ISV certification and GPU infrastructure explains why professional drivers matter for media software.
| RTX A4000 | RTX 4000 Ada | RTX A6000 | |
|---|---|---|---|
| Architecture | Ampere | Ada Lovelace | Ampere |
| Memory per GPU | 16 GB | 20 GB | 48 GB |
| Memory bandwidth | 448 GB/s | 360 GB/s | 768 GB/s |
| FP32 | 19.2 TFLOPS | 26.7 TFLOPS | 38.7 TFLOPS |
| AV1 | Decode | Encode and decode | Decode |
| GPUs per server | 4 | 4 | 8 |
For AV1 encoding, choose the RTX 4000 Ada. For 48 GB per GPU and NVLink bridges between pairs of cards, see the RTX A6000.
Questions buyers ask
RTX A4000 or RTX 4000 Ada: which should we rent?
The RTX 4000 Ada has 20 GB per GPU, two encoders with AV1 encoding and 26.7 TFLOPS of FP32. The RTX A4000 has 16 GB with more memory bandwidth, 448 GB/s against 360 GB/s, and suits visualization and decode-heavy work on Ampere.
Why are RTX A4000 servers rented with four GPUs?
We rent the smaller workstation cards as whole 4-GPU servers, with 64 GB of GPU memory per server. Larger workloads take more servers.
Does the RTX A4000 have ECC memory?
Yes. Each GPU has 16 GB of GDDR6 with error-correction code, according to NVIDIA.
Can the RTX A4000 encode AV1?
No. NVIDIA lists AV1 decode, not AV1 encode, for the RTX A4000. For AV1 encoding, choose the RTX 4000 Ada Generation.
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, with no egress fees.
Rent RTX A4000 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.