oracle/graal · critical · GraalError

No backend available for specified GPU architecture \"%s\"

Error message

No backend available for specified GPU architecture \"%s\"

What it means

After the host backend is created, HotSpotGraalRuntime iterates the JVMCI runtimes's non-host JVMCIBackends (e.g. GPU architectures registered by external JVMCI providers) and needs a HotSpotBackendFactory for each; a missing factory throws GraalError 'No backend available for specified GPU architecture "arch"' during initialization.

Source

Thrown at compiler/src/jdk.graal.compiler/src/jdk/graal/compiler/hotspot/HotSpotGraalRuntime.java:189

            HotSpotBackendFactory factory = backendMap.getBackendFactory(hostArchitecture);
            if (factory == null) {
                throw new GraalError("No backend available for host architecture \"%s\"", hostArchitecture);
            }
            if (replayCompilationSupport != null) {
                factory = replayCompilationSupport.decorateBackendFactory(factory, jvmciRuntime);
            }
            hostBackend = registerBackend(factory.createBackend(this, compilerConfiguration, jvmciRuntime, null));
        }

        for (JVMCIBackend jvmciBackend : jvmciRuntime.getJVMCIBackends().values()) {
            if (jvmciBackend == hostJvmciBackend) {
                continue;
            }

            Architecture gpuArchitecture = jvmciBackend.getTarget().arch;
            HotSpotBackendFactory factory = backendMap.getBackendFactory(gpuArchitecture);
            if (factory == null) {
                throw new GraalError("No backend available for specified GPU architecture \"%s\"", gpuArchitecture);
            }
            try (InitTimer t = timer("create backend:", gpuArchitecture)) {
                registerBackend(factory.createBackend(this, compilerConfiguration, null, hostBackend));
            }
        }

        // Complete initialization of backends
        try (InitTimer st = timer(hostBackend.getTarget().arch.getName(), ".completeInitialization")) {
            hostBackend.completeInitialization(jvmciRuntime, options);
        }
        for (HotSpotBackend backend : backends.getValues()) {
            if (backend != hostBackend) {
                try (InitTimer st = timer(backend.getTarget().arch.getName(), ".completeInitialization")) {
                    backend.completeInitialization(jvmciRuntime, options);
                }
            }
        }

View on GitHub (pinned to a66e9ccd1d)

Solutions

Add/enable the Graal backend module matching the registered GPU architecture, or upgrade the provider+Graal pair so they align
Remove/disable the JVMCI provider registering the GPU backend you do not intend to use
Select a compiler configuration whose BackendMap includes the GPU architecture

Defensive patterns

Strategy: validation

Validate before calling

for (JVMCIBackend b : jvmciRuntime.getJVMCIBackends().values()) {
    if (b != host && backendMap.getBackendFactory(b.getTarget().arch) == null) { /* unregister or refuse before init */ }
}

Prevention

Match GPU provider versions to the Graal build
Remove unused GPU JVMCI providers from the module path
Test startup with the exact module graph you will deploy

When it happens

Trigger: A JVMCI runtime that registers a secondary (non-host) backend — typically a GPU arch via an external JVMCI/Graal-GPU provider — while the selected compiler configuration's BackendMap has no factory for that architecture; thrown inside the backend-registration loop of the HotSpotGraalRuntime constructor.

Common situations: Installing a GPU-accelerated JVMCI provider without the matching Graal GPU backend module; version skew between the provider and the Graal build; configurations that deliberately restrict backends while optional GPU providers are present.

Related errors

AI-assisted analysis of oracle/graal@a66e9ccd1d (2026-08-14). Data as JSON: /api/errors/c088b9303f232d90. Report an issue: GitHub.