Pixel streaming lets you run a heavy, photoreal Unreal Engine application on a cloud GPU and stream the live, interactive result straight to a web browser. The viewer clicks a link and explores a full-fidelity 3D scene on a phone, tablet or laptop – with no installation, no download, and no high-end graphics card on their device. 10ⁿ Tech designs, deploys and operates pixel-streaming platforms on Azure UAE North and other GCC clouds.
What is pixel streaming?
Traditional real-time 3D forces a choice. You either ship a multi-gigabyte desktop application that only runs on a gaming PC, or you fall back to pre-rendered video and 360 tours that look good but cannot be freely explored. Pixel streaming removes the trade-off: the Unreal Engine app runs on a server-side GPU, and only the video frames are streamed to the browser. User clicks, taps and camera moves travel back the other way in real time over WebRTC. The experience is identical to the desktop build – free navigation, real lighting, full quality – delivered by URL.
How pixel streaming works
Each Unreal Engine application runs on a GPU-enabled virtual machine in the cloud and streams to a web page served by a signaling server. The page embeds the stream and opens a real-time WebRTC channel between the browser and the engine instance. When a user opens the streaming URL, the signaling server hands back a page that establishes the session, finds an available GPU streamer, and connects the two.
The core building blocks are a browser portal, backend APIs, a real-time signaling path, a session agent on a Windows or Linux GPU host, the packaged Unreal Engine application, and a WebRTC media path. Persistent data and session metadata live in the platform backend; only the rendered frames cross the network. Where a direct peer connection is not possible, traffic is relayed through a STUN/TURN server for NAT traversal. Peer voice chat between session participants is available out of the box, and a WebRTC video-and-audio layer can be added for guided, two-way sessions.
Pixel streaming vs the alternatives
Pixel streaming is one of three ways to put real-time 3D in front of a remote audience. The right choice depends on fidelity, scale and infrastructure tolerance – and the channels combine well.
| Criteria | 3D web app (pre-rendered) | Pixel streaming | Local desktop build |
|---|---|---|---|
| Access | Any browser, no GPU server | Any browser, GPU streaming host required | Install required, gaming PC |
| Visual fidelity | Medium-high (pre-rendered assets, 360 tours) | Very high (Unreal rendered live) | Very high (Unreal, local) |
| Interactivity | Guided navigation, filters, floor plans, 360 tours | Full real-time Unreal interaction | Full real-time Unreal interaction |
| Infrastructure | Lowest – standard web server | Medium-high – GPU hosts + signaling stack | None server-side; heavy on the client |
| Best for | Mass reach, first impressions, mobile browsing | High-fidelity remote demos and guided tours at any scale-of-quality | On-site kiosks and controlled showrooms |
A common pattern is to lead with a lightweight 3D web app for reach, then switch high-intent viewers into a pixel-streamed session for the full-quality experience – both drawing from the same Unreal source.
Running pixel streaming on Azure and other clouds
Pixel streaming needs a GPU that exposes hardware video encoding (NVENC) to the virtual machine, with one Unreal Engine instance provisioned per concurrent stream. The platform is portable across any cloud with a suitable GPU SKU, networking and managed services; in the UAE we most often deploy on Azure UAE North.
Validated GPU options
| GPU SKU | VRAM | Streams per GPU (1080p60) | Notes |
|---|---|---|---|
| NVIDIA RTX PRO 6000 (Blackwell) | 96 GB | 4 to 8 | Highest density; cost-effective when consolidating many streams per node. |
| NVIDIA L40S | 48 GB | 2 to 4 | Sweet spot for data-center pixel streaming; strong NVENC throughput. |
| NVIDIA L4 | 24 GB | 1 to 2 | Cost-efficient single-stream option. |
| NVIDIA A10 | 24 GB | 1 to 2 | Established, widely available – the Azure NVads_A10_v5 family in UAE North. |
Sizing per GPU streaming node
- CPU – 8 to 16 modern x86 cores per GPU instance. The engine renders on the GPU, but game logic, audio mixing and WebRTC are CPU-bound.
- RAM – 32 GB per streaming instance; UE5 scenes regularly sit at 12 to 20 GB resident.
- Disk – 100 GB NVMe for OS, drivers, the Unreal build and shader cache; an optional scratch volume speeds cold starts.
- OS – Windows Server 2022 or Ubuntu 22.04 LTS, both validated for Unreal Engine 5 pixel streaming.
- Network – 25 to 50 Mbps sustained upstream per 1080p60 stream, with headroom for peaks.
Controlling the cost of GPU compute
GPU hours are the main cost of pixel streaming, so scheduling matters more than anything else. A business-hours schedule (roughly 187 hours a month per VM, auto-starting before the working day and deallocating in the evening) drops compute spend to about 26% of a 24/7 baseline. Where always-on capacity is required, an Azure Capacity Reservation guarantees a slot, and a one or three-year Reserved Instance offsets the cost by 36 to 56% versus on-demand. Because UAE North is a finite regional capacity pool, we monitor allocation success and keep validated fallback SKUs ready so a morning start never fails silently.
Indicative monthly cost
The figures below are rough Azure UAE North list prices to size a deployment – your actual bill is metered by Microsoft on real consumption and varies with SKU, resolution, hours and traffic. Model your own configuration in the Azure pricing calculator.
| Component | Indicative cost | Notes |
|---|---|---|
| GPU VM per stream – business hours (~187 hr/mo) | ~$200 to $350 / month | A10 class (NVads_A10_v5). The single biggest line, and the one scheduling controls. |
| GPU VM per stream – 24/7 | ~$750 to $1,100 / month | Cut 36 to 56% with a 1 or 3-year Reserved Instance. |
| Managed disk (Premium SSD, 256 GB) | ~$40 / month | One per GPU node. |
| Static public IP | ~$4 / month | Per endpoint. |
| Egress (outbound streaming) | ~$180 / TB | The variable to watch – budget ~25 to 50 Mbps per 1080p60 stream. |
| Storage, monitoring, logs | Tens of dollars / month | Blob for the UE build, Azure Monitor ingestion. |
As a rule of thumb, a single business-hours-scheduled A10 stream lands in the low hundreds of dollars a month all-in, and a small four to five-stream cluster in the low thousands. GPU hours and egress dominate; everything else is marginal.
Where pixel streaming is used
- Remote executive and sales demos – show a project at full 4K Unreal quality over a web link, with no install on the viewer’s device.
- Guided property and project tours – a presenter drives a live, shared session for buyers or stakeholders anywhere.
- Digital twins – stream a heavy city, plant or building twin that no laptop could run locally.
- Design and engineering review – walk a photoreal model together before anything is built.
- Training and simulation – deliver interactive, GPU-rendered scenarios to a browser at scale.
For off-plan property sales, pixel streaming is the engine behind our remote interactive sales tool – a sales agent drives a full-quality, synchronized 3D session for buyers anywhere, with no app to install.
Why 10ⁿ Tech
We build the Unreal Engine application, package it for streaming, stand up the GPU and signaling infrastructure, and validate it end to end against a written specification – NVENC exposure, latency, failover and stream quality. You can run it yourself once it is handed over, or we operate it as a managed service and own the GPU scheduling, monitoring and capacity strategy for you. Our pixel-streaming work underpins the digital-twin and immersive experiences we deliver across the GCC.
Frequently asked questions
What is pixel streaming?
Pixel streaming runs a real-time 3D application – typically Unreal Engine – on a cloud GPU and streams the rendered video to a web browser, sending the user’s clicks and camera moves back over WebRTC. The viewer gets a full-quality, interactive 3D experience by URL, with nothing to install.
Do users need a powerful device or a download?
No. All the heavy rendering happens on the server GPU. Any modern phone, tablet or laptop with a browser and a stable connection can run a pixel-streamed experience – there is no app to download.
What GPU do you need for pixel streaming?
An NVIDIA data-center GPU that exposes hardware encoding (NVENC) to the VM – A10, L4, L40S or RTX PRO 6000 depending on how many streams you consolidate per node. On Azure UAE North we typically use the NVads_A10_v5 family.
How much bandwidth and what latency should I expect?
Budget 25 to 50 Mbps of upstream per 1080p60 stream from the host. End-to-end latency on a well-provisioned regional deployment feels interactive – close to a local application – because rendering and encoding happen on the GPU and only frames are sent.
Can it run on Azure, or other clouds?
Both. The platform is provider-agnostic and runs on any cloud with a validated GPU SKU, the right networking and managed services. In the UAE we most often deploy on Azure UAE North for data residency; AWS Middle East and local UAE providers are also viable.
How is pixel streaming different from a pre-rendered 3D web tour?
A pre-rendered web tour is lightweight and scales cheaply, but navigation is fixed to predefined paths and 360 viewpoints. Pixel streaming delivers the full, freely explorable Unreal Engine experience at much higher fidelity, at the cost of a GPU host. Many projects use both – the web tour for reach, pixel streaming for high-intent, high-fidelity sessions.
Put your Unreal project in the browser
If you have a heavy 3D or Unreal Engine experience that is stuck on a desktop or a kiosk, pixel streaming opens it to any browser. Talk to us or explore the full solutions portfolio.
