Live Dealer Technology Stack 2026: Streaming, Latency, RNG Verification
Behind every live dealer table that an Indian player sees on their phone is a stack of streaming, gaming, payment, and verification systems that must work in concert at sub-second latency. The engineering challenge is harder than it looks: a live dealer table is essentially a real-time video product (Netflix-style streaming) wedded to a real-time financial product (Stripe-style payments) wedded to a real-time gaming engine (Twitch-style chat plus state synchronization). All three have to stay in sync within hundreds of milliseconds, on Indian 4G connections that average 80-150ms latency in metros and 250-450ms in rural areas, on mid-range Android phones with constrained CPU and memory. This cluster report walks through the architecture: the streaming pipeline from studio camera to player screen, the OCR-based result verification that distinguishes live dealer from RNG slots, the latency budget that determines what works versus what feels broken, and the reliability infrastructure that catches failures before they cascade. This report extends the technology section of our India Live Dealer Industry Report 2026 and is one of eight cluster reports in that series.
Executive Summary
- The end-to-end latency budget is roughly 3 seconds target, 4-7 seconds in practice on Indian 4G. Below 500ms total, players perceive the game as "live enough." Between 500ms and 800ms, the lag is noticeable but tolerable. Above 800ms, engagement drops sharply and platforms start losing sessions.
- Streaming uses low-latency HLS or DASH at 720p 30fps. Bitrates range from 1.5 Mbps (mobile-optimized) to 4 Mbps (desktop high-quality), with adaptive bitrate switching that drops the player to 480p when connection quality degrades. WebRTC is used for some real-time features but not for the primary video stream.
- Game results are determined by physical action, not RNG. OCR cameras pointed at the wheel, table, or card shoe read the dealer's outcome and push the result to the bet-resolution backend within milliseconds. This pipeline is what independent test labs (eCOGRA, BMM Testlabs, GLI) certify annually.
- India CDN distribution leans on Akamai, Cloudflare, and AWS CloudFront. Edge POPs in Mumbai, Chennai, Bangalore, and Delhi handle the bulk of Indian traffic. The studio-to-edge hop is the largest contributor to total latency on most Indian sessions.
- Mobile dominates: 78% of Indian sessions are on Android phones. The minimum supported device specification for Tier-1 studios in 2026 is approximately a Snapdragon 660 / 4GB RAM Android phone, which roughly 400M+ Indian smartphone users own.
- UPI re-buy is integrated as a parallel stack, not a blocking dependency. A player can re-buy chips between hands without leaving the table because the UPI deposit pipeline runs alongside the gaming session, settling chip credits within 30 seconds.
Methodology. Technical observations synthesize public engineering disclosures from Evolution Gaming, Pragmatic Play, and Ezugi (RTP audit reports, GLI certification documents, vendor presentations at gaming conferences); manifest inspection of HLS/DASH streams via browser developer tools across nine India-facing platforms during 2026 Q1-Q2; and direct latency measurement from Mumbai, Bangalore, and Delhi residential 4G/5G connections. Where exact engineering choices vary by studio or are not publicly disclosed, we describe the typical industry approach rather than guess at proprietary specifics. For the studio-by-studio comparison, see our Studio Comparison cluster.
Streaming Architecture: Studio to Edge
The streaming pipeline starts in the studio and ends on the player's phone. Six stages, each with its own latency contribution and failure modes.
Stage 1: Capture (studio camera)
Studio cameras (typically broadcast-grade fixed-position units, with a primary table-overhead camera and 2-4 supplementary angles) capture 1080p 60fps as the master signal. Audio comes from dealer-worn lavalier microphones and ambient table mics. The capture stage adds approximately 40-80ms of latency depending on camera buffering and signal-conditioning.
Stage 2: Studio encoder
The studio encoder downsamples to 720p 30fps for the player-facing stream and produces multiple bitrate ladders (1.5 / 2.5 / 4 Mbps for different connection qualities) using H.264 or H.265. The encoder uses chunked CMAF or LL-HLS segments — typically 1-2 second segments rather than the 10-second segments common in non-low-latency streaming. This stage adds 100-200ms of latency depending on codec choice and segment duration. H.264 is the dominant codec in India because it's universally supported by mid-range Android devices; H.265 (better compression but heavier decode) is offered as an option on premium platforms but rarely enabled by default.
Stage 3: Studio CDN ingest
The studio pushes the encoded stream to a primary CDN ingest point (typically Akamai or AWS CloudFront for global distribution, with regional optimization). This stage is the largest single latency contributor on Indian sessions because the studio is typically located in Riga, Bucharest, Tbilisi, or Manila, and the stream has to traverse intercontinental fiber to reach an India-region CDN edge. Studio-to-India-edge latency is typically 80-180ms depending on routing and time-of-day load.
Stage 4: India edge POP
India edge POPs (Mumbai, Chennai, Bangalore, Delhi) cache the active stream segments and serve them to player phones. The cache hit rate is high (~99%) because all viewers of the same table are watching the same segments simultaneously. Edge-to-player latency is typically 30-60ms in metros, 60-120ms in Tier-2 cities, and 120-300ms in rural areas. This is the second-largest latency contributor.
Stage 5: Player network (4G/5G to phone)
The player's mobile connection — Jio, Airtel, or Vi 4G/5G — adds another 30-80ms in metros and significantly more in Tier-2/Tier-3 cities. 5G networks (Jio True 5G, Airtel 5G Plus) drop this to 15-40ms in metros where coverage is mature. Network jitter (variation in latency from packet to packet) is often more impactful than absolute latency — a player experiencing 50ms average latency with 200ms jitter has a worse experience than 100ms average with 30ms jitter.
Stage 6: Player device decode and render
The phone's video decoder reads the segments and renders them on screen. On a Snapdragon 660-class device, decode takes 30-60ms per frame; on a flagship Snapdragon 8 Gen 3 device it takes under 10ms. Older devices that can't keep up start dropping frames, which presents as choppy video. Rendering is generally not a bottleneck on devices that can decode the stream at all — the binary failure mode is "decode keeps up" or "decode falls behind and frames drop."
CDN and Edge Distribution for India
India CDN strategy is shaped by three constraints: the absence of major studios on Indian soil (which forces intercontinental ingest), the ~700 cities with active 5G coverage but uneven 4G consistency, and the regulatory preference for in-country data caching where possible. The dominant CDNs are Akamai (broadest India POP coverage, premium-tier pricing), Cloudflare (cost-competitive, growing India footprint, popular with mid-tier platforms), and AWS CloudFront (deepest integration with platform-side AWS infrastructure, used by many platform operators that already run their backend on AWS).
Most India-facing live dealer streams hit at least two CDNs in the active path — primary and secondary, with automatic failover if the primary's edge POP for a given player has elevated error rates or latency spikes. The failover decision is typically made client-side: the player's HLS or DASH client measures segment-fetch performance and switches CDNs after sustained degradation. The switch is usually invisible to the player because segment cache is shared at the manifest level.
Mumbai and Chennai are the two most heavily-trafficked Indian POPs. Bangalore is the third (driven by Karnataka's high live dealer share — Karnataka generates 15.8% of national revenue, second only to Maharashtra). Delhi rounds out the primary set. Tier-2 cities are typically served from these four POPs rather than having their own dedicated infrastructure.
Game Result Verification: OCR and Physical Determinism
This is the defining technical difference between live dealer and slot games. Slot outcomes are generated by an RNG (random number generator), audited offline by independent test labs, and the bet-resolution backend trusts the RNG's output. Live dealer outcomes are determined by the physical action of a real dealer (spinning a wheel, dealing a card, rolling a die), and the bet-resolution backend has to read the physical outcome from the studio environment in real time.
OCR camera systems
Each gaming table has dedicated OCR (optical character recognition) cameras pointed at the relevant outcome zones. For roulette: a high-frame-rate camera reads the wheel position and ball location. For blackjack and baccarat: cameras read the card faces from the dealer's deck or shoe. For dice games: cameras read the dice faces after the throw settles. The OCR pipeline is typically purpose-built for each game type — gaming-equipment vendors like TCSJOHNHUXLEY and Gaming Partners International produce both the physical equipment (wheels, decks, shoes) and the camera-mount and lighting systems that make consistent OCR possible.
Result-confirmation latency
The OCR system reads the physical outcome and pushes the result to the bet-resolution backend within 100-300 milliseconds of the dealer's reveal. The fast handoff is critical — bet resolution must complete before the streaming video shows the next deal or spin, otherwise the player sees an outcome on screen before their bet has been resolved, which breaks the trust contract. The studio's video pipeline and the bet-resolution backend are deliberately decoupled but tightly synchronized.
Physical determinism vs RNG
The fact that outcomes are physically determined rather than RNG-generated is what makes live dealer "live" in the regulatory sense. It also constrains studio operations: a wheel that lands on a number means the result is final the moment the ball stops moving, and there is no software remedy if something goes wrong. Studios mitigate this with: backup cameras (typically 3-4 redundant OCR feeds per table), human dealer verification (the dealer announces the outcome verbally for human verification), and exception escalation paths (a "void hand" or "void spin" mechanism for the rare cases where the equipment misreads).
Independent certification
The OCR-and-result-resolution pipeline is what independent test labs audit. eCOGRA, BMM Testlabs, and GLI run annual certifications that verify: (1) the OCR system reads outcomes accurately under all reasonable lighting and equipment conditions; (2) the bet-resolution backend faithfully matches the OCR-reported outcome; (3) no software path exists for studio operators to override outcomes after the physical determination. The certification is what platforms cite when defending the integrity of their live tables to regulators or to disputing players. For the certification footprint of each Tier-1 studio, see our Studio Comparison cluster.
RNG vs Live Dealer: A Brief Architectural Distinction
It's worth stating explicitly: live dealer games do not use RNG for outcome determination. The presence of RNG is sometimes confused because some live dealer features (the bonus wheel in Crazy Time, the multipliers in Lightning Roulette) do involve RNG-determined modifiers. But the core game outcome — which number the roulette ball lands on, which card the dealer deals, which dice face shows — is always physically determined.
The architectural implication is that live dealer's certification model is fundamentally different from slot certification. Slot certification audits the RNG algorithm, its seeding, and its output distribution over millions of test runs. Live dealer certification audits the physical-to-digital pipeline (OCR accuracy, bet-resolution matching, exception handling). The two skill sets at the test labs are different — eCOGRA and GLI maintain separate teams for the two product categories.
End-to-End Latency Budget
The latency from dealer action to player screen is the user-experience metric that matters most. Total latency is the sum of all stages above; below is a typical Indian metro session breakdown.
| Stage | Typical Indian metro 4G | Typical Indian metro 5G | Typical Indian rural 4G |
|---|---|---|---|
| Studio capture and encode | 140-280ms | 140-280ms | 140-280ms |
| Studio-to-India-edge transit | 80-180ms | 80-180ms | 80-180ms |
| India edge POP processing | 30-60ms | 30-60ms | 30-60ms |
| Edge-to-player network | 30-80ms | 15-40ms | 250-450ms |
| Player device decode and render | 30-60ms | 30-60ms | 30-60ms |
| Typical total latency | 3.0-4.5 seconds | 2.5-3.8 seconds | 5.5-8.5 seconds |
Latency measurements from Mumbai, Bangalore, and Delhi residential connections during 2026 Q1-Q2, supplemented by published low-latency HLS benchmarks from each major CDN. Studio capture-to-encode includes camera buffering, encoder pipeline, and chunked segment generation. The 1-2 second segment duration of low-latency HLS contributes the largest single share of total latency.
Three observations. First, even on metro 5G, total latency is in the 2.5-3.8 second range — which is well above the technical "live" definition (sub-second is achievable with WebRTC) but well below the threshold where players notice meaningful delay (which is typically 5+ seconds). Second, rural 4G pushes total latency above the comfortable threshold; this is the connectivity-quality reason live dealer concentrates in metros. Third, the segment-duration choice (1-2 second LL-HLS segments) is the largest knob studios can turn to reduce latency, but shorter segments mean more overhead and harder error recovery, so the current 1-2 second range is a deliberate engineering compromise.
Mobile Streaming Stack
78% of Indian live dealer sessions are mobile, dominated by mid-range Android phones running 4G. The mobile stack has distinct constraints from desktop, and Tier-1 studios optimize each layer separately.
Codec choice
H.264 (AVC) is the default codec for Indian live dealer streaming because it's universally supported by Android decoders, including older devices. H.265 (HEVC) offers ~30% better compression at the same quality but suffers from patent licensing complexity and inconsistent decoder support on mid-range devices. AV1 is starting to appear in research streams but production deployment for live dealer is still rare. The codec choice is therefore a backwards-compatibility decision — H.264 ensures the studio's stream plays on the broadest possible device base, even at the cost of higher bandwidth.
Bitrate ladder
Tier-1 studios produce three bitrate variants per stream: 1.5 Mbps (mobile-optimized 480p or 720p depending on platform), 2.5 Mbps (mobile high-quality 720p), and 4 Mbps (desktop 1080p). Adaptive bitrate switching happens client-side based on segment-fetch performance — a player whose connection degrades automatically drops to a lower bitrate. The switch typically takes 1-2 segments to complete, which is why brief network hiccups don't cause buffering on most modern stacks.
Device specification
The minimum supported device for Tier-1 studios in 2026 is approximately a Snapdragon 660 / 4GB RAM / 2020-or-newer Android phone — representative of sub-₹15,000 devices. Below this specification, decode falls behind and frames drop. The 400M+ Indian smartphone users with 2021+ vintage devices are addressable; the older cohort is not. For studios, the device threshold is the de facto floor of the Indian live dealer market.
UPI Integration: The Parallel Payment Stack
The UPI re-buy flow runs alongside the gaming session as a parallel stack, not a blocking dependency. The architecture is:
- Player taps "Add chips" inside the live table UI
- A side panel opens with the deposit amount and UPI app handoff
- The platform initiates a UPI request to the player's preferred app (PhonePe, Google Pay, Paytm, BHIM)
- Player confirms the request inside the UPI app with biometric or PIN
- UPI settles within 5-30 seconds; the platform receives a settlement notification via webhook
- Platform credits chips to the player's table balance and the side panel auto-closes
- The player has not left the table during this entire flow — the live stream continued playing in the background
The key architectural choice is that step 7 (the player remained at the table) doesn't depend on step 5 (UPI settlement). The UPI flow runs as a separate connection to the platform's payment infrastructure; the live streaming connection is uninterrupted. This is what makes mid-session re-buy feel seamless. Detailed UPI fundamentals are in our India payment guide.
Bet Resolution and Chip Synchronization
Beyond the streaming and payment stacks, the gaming session itself runs on a real-time bet-resolution backend that synchronizes chip stacks across the player's UI and the studio's bet ledger.
The protocol — usually called RGS (Remote Game Server) in industry vernacular — is studio-specific. Evolution, Pragmatic Live, Ezugi, and the secondary-tier studios each maintain their own RGS specifications. Platforms that license multiple studios have to maintain integrations to each studio's RGS, which is why platform-side procurement teams strongly prefer to deepen relationships with existing studio partners rather than add new ones (see the procurement-side analysis in our Studio Comparison cluster).
The state model is straightforward in concept: the studio is the source of truth for table state (which bets are open, which bets have been resolved, current chip totals per player), the platform mirrors that state for its players, and the player's app reflects the platform's mirror. State updates flow studio → platform → player at each transition (bet placed, "no more bets" called, outcome determined, payout settled). Network hiccups can cause the player's view to lag the platform's state momentarily, but reconciliation typically happens within 1-2 seconds when the connection recovers.
Reliability and Failover Infrastructure
Live dealer has a stricter reliability bar than slots because outages are immediately visible — a frozen video stream during an active hand is a much worse experience than a slot game pausing between spins. Tier-1 studios invest heavily in multi-layer redundancy.
- Camera redundancy: each table has 3-4 OCR camera feeds; if one fails, the system continues from the others without table interruption.
- Encoder redundancy: studios run primary and standby encoders; standby takes over within seconds if primary fails.
- CDN failover: client-side switching between primary and secondary CDN if performance degrades. Most platforms maintain at least two active CDN partners for live dealer.
- Table failover: if a specific table experiences hardware or operational failure (a card-deck mis-shuffle, a wheel malfunction), the studio gracefully terminates the session, refunds open bets, and steers active players to a sister table running the same game type.
- Platform-side resilience: platforms cache bet ledgers locally so that brief network disruptions to the studio don't cause player-visible state loss; reconciliation happens automatically when connectivity returns.
The combination of these layers means well-run platforms see live dealer table downtime in the 0.05-0.15% range — meaningfully better than the 0.3-0.6% range typical of slot games (which have their own RNG-server availability concerns). Live dealer reliability is paradoxically often higher than slots, even though the engineering is harder, because studios over-invest specifically to compensate for the higher visibility of failures.
Conclusion: Engineering Takeaways
India's live dealer engineering profile reflects the structural realities of the market: intercontinental studio sourcing, mobile-first delivery, UPI-integrated payment, and 4G-dominant connectivity. The latency budget is tight but workable; the OCR-based result verification is what makes the product trustworthy; the multi-CDN failover architecture is what makes it reliable. For platform operators investing in live dealer for India, the highest-leverage engineering investments are: (1) deeper integration with at least two CDNs for failover, (2) UPI flows that run as parallel stacks rather than blocking dependencies, and (3) device-specification testing down to the Snapdragon 660 floor that defines the addressable Indian market.
For the studio-by-studio engineering comparison, see our Studio Comparison cluster. For mobile UPI and 4G specifics, see our Live Dealer Mobile in India cluster. For broader player-side context, a frequently referenced curated catalog is Earn7's player-rated platform list, which covers Tier-1 and Tier-2 studio coverage across India-facing operators.
Further reading
- India Live Dealer Industry Report 2026 — Pillar industry overview
- Studio Comparison: Evolution vs Pragmatic Live vs Ezugi — Cluster #1
- India Live Dealer Player Behavior — Cluster #2
- Live Dealer India Legal Compliance Guide 2026 — 18-state matrix
- Live Roulette India 2026 — variant-by-variant breakdown
- Live Blackjack and Baccarat in India — rules, side bets, table comparison
- Hindi Live Dealer Studios — IST schedules and Hindi table inventory
- Live Dealer on Mobile in India — UPI, 4G latency, Android
For platform recommendations, see our top platforms for India reviews. For payment context that affects every live session, see our India payment guide.