# Render decouples memory from CPU with new compute plans up to 96GB RAM Render has added more than 15 memory-heavy compute options that let services buy additional RAM without proportionally buying CPU, plus a new 12-CPU tier up to 96GB for web services, private services and background workers. Memory-bound agents, retrieval systems and stateful services can now choose 2-, 4-, 8- and 12-CPU Render plans with much wider RAM ratios. Existing plan prices and legacy IDs stay compatible; the new choices change the cost trade-off for workloads that previously had to overbuy CPU to get enough memory. - Status: Active - Published: 2026-08-28T19:05:23+12:00 - Updated: 2026-08-28T19:05:23+12:00 - Categories: Cloud & Infrastructure, Hosting - Tags: cloud hosting, Cloud pricing, memory-optimized compute, Render - Canonical HTML: https://beyondthe.news/dossiers/render-memory-optimized-compute-plans-12cpu-96gb ## What changed On August 26–27, 2026, Render expanded its service compute catalog with more than 15 new CPU/RAM combinations. At every tier with two or more CPUs, customers can choose higher-memory options independently of CPU count; for example, four CPUs can now be paired with 8GB, 16GB or 32GB of RAM. Render also added 12-CPU web-service, private-service and background-worker plans with 24GB, 48GB or 96GB of RAM. At the same time it replaced descriptive plan names with spec-based IDs such as `4c-32g`, while keeping legacy IDs backward compatible across the API, CLI, SDK, Blueprints and Terraform integrations. ## Why it matters Memory-heavy application economics do not always scale with CPU. Agent harnesses can hold long-lived session state, retrieval systems keep indexes or embeddings resident, and databases or workers may need large heaps without saturating processors. Previously, getting more memory could force teams into a higher CPU tier they did not need. Render’s new catalog makes RAM a more independent cost/sizing decision and raises the ceiling for stateful application workloads without requiring a move to another hosting model. ## RAM can now scale independently at the same CPU tier Render’s expanded catalog adds multiple memory ratios at 2, 4 and 8 CPUs, including 2c-8g, 2c-16g, 4c-32g, 8c-16g and 8c-64g options for supported service types. That lets teams tune for memory pressure without automatically stepping up processor count. ## The top long-running service tier reaches 12 CPUs and 96GB Web services, private services and background workers can use new 12-CPU plans with 24GB, 48GB or 96GB RAM. Render explicitly cites agent platforms and other memory-intensive workloads as a motivation. Cron jobs also receive additional memory-heavy combinations, although their plan matrix differs from continuously running services. ## Existing deployments do not need a naming migration Render renamed 'instance types' to 'compute plans' and introduced spec-based IDs. Existing legacy identifiers such as `standard`, `pro` and `pro_ultra` remain valid in tooling and map to identical existing specifications and prices. Teams can adopt the new IDs gradually; the entirely new memory configurations have only the new spec-based identifiers. ## The economic benefit depends on the new-plan prices Render says existing plans keep their prices unchanged. The significance of the new choices is that a workload can select a different resource ratio rather than overprovisioning CPU solely for RAM. Builders should compare the new plan’s actual monthly rate with their current tier and with horizontal scaling; more memory on one instance may reduce CPU waste but can also increase failure concentration. ## Key details - Render announced the expanded compute catalog August 26–27, 2026. - More than 15 new memory-optimized plan options were added. - At 2+ CPU tiers, users can choose multiple RAM quantities independently of CPU count. - Four-CPU web services can now choose 8GB, 16GB or 32GB RAM. - New 12-CPU tiers for web services, private services and background workers offer up to 96GB RAM. - New plan IDs encode specifications, such as `4c-32g` and `12c-96g`. - Legacy plan IDs remain supported across Render tooling. - Render says existing plans receive no pricing changes. ## Builder takeaways - Profile actual peak and sustained memory before moving to a higher-RAM plan; choose the smallest CPU tier that satisfies real compute demand rather than treating RAM and CPU as one scaling unit. - For agent and retrieval workloads, measure memory growth over long sessions and concurrent jobs, not only a short synthetic request. - Compare a single large-memory instance with multiple smaller instances on both cost and failure-domain risk. - Update Render CLI, Terraform provider and other integrations before using new spec-based plan IDs, even though legacy identifiers remain backward compatible. - Record the plan ID and specs explicitly in infrastructure configuration so future plan-catalog expansion does not make an old descriptive name ambiguous. ## What to watch - Pricing and adoption of the new high-memory plans relative to competing hosting providers. - Whether Render extends independent CPU/RAM sizing to smaller sub-2-CPU tiers or offers even larger memory ratios. - Whether autoscaling gains memory-aware policies that make the expanded plan matrix easier to use dynamically. - Operational evidence from long-running agent, vector-search and database workloads on the 64GB and 96GB tiers. ## Uncertainties - Render’s claim that the new ratios reduce overprovisioning is workload-dependent; CPU, memory bandwidth, I/O and horizontal scaling can still dominate performance. - Existing plan prices are unchanged, but the new plan options have their own rates and should be compared against alternatives rather than assumed cheaper. - A larger single instance can reduce resource waste while increasing the impact of an instance-level failure. ## Sources - [New compute plans for memory-intensive applications](https://render.com/blog/new-compute-plans-for-memory-intensive-applications) — Render · primary/vendor · 2026-08-27T00:00:00+12:00. Primary explanation of the memory-heavy use cases and expanded plan catalog. - [New compute plans, and new IDs for existing plans](https://render.com/changelog/new-compute-plans-and-new-ids-for-existing-plans) — Render · primary/vendor · 2026-08-26T00:00:00+12:00. Primary changelog for new CPU/RAM tiers, 12-CPU plans, spec-based IDs and backward compatibility. - [Render Compute Plans](https://render.com/docs/compute-plans) — Render Docs · primary_documentation. Current plan matrix and legacy-ID mapping. - [Updates to Render Compute Plans](https://render.com/docs/compute-plans-update) — Render Docs · primary_documentation · 2026-08-26T00:00:00+12:00. Migration details and confirmation that existing plan pricing is unchanged.