macrostack
Browse

The AI stack

Categories

Local & Sovereign AINotes & KnowledgeObservability & MonitoringPassword ManagersWeb AnalyticsTeam ChatSmart HomeNetworking & RoutersVideo ConferencingCloud Storage & SyncPhotos & MediaAPI DevelopmentImage EditingWorkflow Automation & iPaaSDeveloper Tools & ContainersOffice & Productivity SuitesNo-Code DatabasesCode Hosting & Git ForgesProject ManagementEmail Marketing & NewslettersScheduling & BookingError Tracking & Exception MonitoringLog Management & SIEMVPN & PrivacyEmail & Secure MailVector Databases & AI SearchLLM & Agent FrameworksDomains & Web HostingData Removal & PrivacyAuthentication & IdentityHelp Desk & Customer SupportCloud & VPSKubernetes & Container PlatformsEmbedding ModelsPDF & DocumentsAI Coding AssistantsAI Voice & SpeechLLM Observability & EvaluationLLM Gateways & RoutingCloud GPU & AI ComputeCI/CD & build automationData & pipeline orchestrationModel serving & inferenceAI agent frameworksBackend as a serviceSecrets managementFeature flags & experimentationProduct analyticsSearch infrastructureUptime & status monitoringAffiliate & partner platformsVisitor identification & personalisationWikis & internal docsIdentity & access managementData warehouses & analytics enginesCustomer data platformsCRMObject storageBI & dashboardsE-signatureWhiteboards & diagrammingIn-memory data stores & cachingPlatform as a serviceTransactional & bulk emailHeadless CMSDesign & prototypingE-commerce platformsInternal tools & admin panelsManaged databasesForms & surveysFine-Tuning & Model TrainingRAG & Retrieval PlatformsLLM Evaluation & TestingAI Guardrails & Content SafetySpeech Recognition & TranscriptionExperiment Tracking & ML OpsDocument AI & OCR

About

How we rank & score

Layer 3 · self-hosting reality check

What it actually takes to self-host SGLang

The docs say 16 GB system. In practice you want 24 GB VRAM for useful production serving. Here is the honest version — real requirements, real monthly cost, what you will be maintaining, and the one thing that catches people out.

Usually reached from Replicate alternatives, where SGLang is one of the picks.

RAM — documented minimum16 GB system
RAM — what it really needs24 GB VRAM for useful production serving
CPU8 vCPU alongside the GPU
DiskModel weights
Monthly cost$300–900/mo for a rented A100 or L40S
Setup timeA day including CUDA
How you install itpip install plus a server command; exposes an OpenAI-compatible endpoint
Ongoing maintenanceModerate. CUDA and driver versions are the recurring pain, not SGLang itself.
Where it stops scalingThousands of concurrent requests per GPU, and further than vLLM on prefix-heavy workloads.

The thing that catches people out

Its advantage is RadixAttention, which reuses shared prompt prefixes across requests — so it is dramatically faster on agents, chat and structured extraction, and roughly ordinary on one-shot prompts that share nothing. Benchmark it on your real traffic shape. Teams adopt it expecting a universal speedup, measure independent single prompts, and conclude the published numbers were exaggerated.

When not to self-host SGLang

Your workload is one-shot prompts with no shared context. vLLM is the more mature choice there.

Every guide here carries this section. A site that only ever tells you to self-host is selling something — the useful answer is sometimes no.

Other Layer 3 self-hosting guides

Common questions

How much RAM does SGLang actually need?
24 GB VRAM for useful production serving in practice. The documented minimum is 16 GB system, which is the figure at which the process starts rather than the figure at which it works under real use. 8 vCPU alongside the GPU alongside it.
What does self-hosting SGLang cost per month?
$300–900/mo for a rented A100 or L40S This is commodity VPS pricing and excludes your time, which is the larger cost for most people — budget for moderate. CUDA and driver versions are the recurring pain, not SGLang itself.
How long does it take to set up SGLang?
A day including CUDA, via pip install plus a server command; exposes an OpenAI-compatible endpoint.
When should I NOT self-host SGLang?
Your workload is one-shot prompts with no shared context. vLLM is the more mature choice there.
What is the most common mistake when self-hosting SGLang?
Its advantage is RadixAttention, which reuses shared prompt prefixes across requests — so it is dramatically faster on agents, chat and structured extraction, and roughly ordinary on one-shot prompts that share nothing. Benchmark it on your real traffic shape. Teams adopt it expecting a universal speedup, measure independent single prompts, and conclude the published numbers were exaggerated.
The Macrostack brief

New swaps, worth your inbox.

A short, occasional email when we add a high-intent alternative or ship a new head-to-head. No spam, no selling your address — unsubscribe in one click.