diff --git a/.buildkite/README.md b/.buildkite/README.md
new file mode 100644
index 000000000..c38870a51
--- /dev/null
+++ b/.buildkite/README.md
@@ -0,0 +1,123 @@
+# Multi-Queue Buildkite Configuration
+
+This directory contains the configuration for running SciMLBenchmarks on multiple Buildkite queues, enabling GPU-intensive benchmarks to run on GPU-enabled compute resources while maintaining CPU-only benchmarks on the standard queue.
+
+## Files
+
+- `queue_config.yml`: Central configuration mapping benchmarks to queues
+- `generate_pipeline.jl`: Dynamic pipeline generator that creates appropriate Buildkite steps
+- `test_sciml.yml`: Main Buildkite pipeline configuration
+- `path_processors/project-coalescing`: Script to determine which benchmarks need rebuilding
+
+## How It Works
+
+1. **Queue Detection**: When a benchmark runs, the system determines which queue to use based on:
+   - Benchmark-specific `[buildkite]` section in `Project.toml` (highest priority)
+   - Central mapping in `queue_config.yml`
+   - Default queue (`juliaecosystem`) as fallback
+
+2. **Dynamic Pipeline Generation**: The `generate_pipeline.jl` script:
+   - Reads the queue configuration
+   - Processes changed files to determine which benchmarks to run
+   - Generates appropriate Buildkite steps with correct queue assignments
+   - Uploads the dynamic pipeline to Buildkite
+
+3. **Queue-Specific Configuration**: Each queue can have different:
+   - Architecture requirements (`arch`)
+   - Operating system (`os`)
+   - Environment variables (e.g., GPU-specific settings)
+   - Timeout values
+
+## Available Queues
+
+### `juliaecosystem` (Default)
+- **Purpose**: CPU-only compute queue
+- **Architecture**: x86_64 Linux
+- **Usage**: Standard Julia benchmarks, ODE solvers, linear algebra
+
+### `gpu`
+- **Purpose**: GPU-enabled compute queue  
+- **Architecture**: x86_64 Linux with GPU support
+- **Usage**: Neural PDE benchmarks, GPU-accelerated simulations
+- **Environment**: Includes CUDA-related environment variables
+
+## Configuring Benchmarks for Specific Queues
+
+### Method 1: Project.toml (Recommended)
+
+Add a `[buildkite]` section to your benchmark's `Project.toml`:
+
+```toml
+[deps]
+# ... your dependencies
+
+[buildkite]
+# Specify the queue this benchmark should use
+queue = "gpu"
+```
+
+### Method 2: Central Configuration
+
+Edit `.buildkite/queue_config.yml` and add your benchmark folder:
+
+```yaml
+benchmark_queues:
+  YourBenchmarkFolder: "gpu"
+```
+
+## Adding New Queues
+
+1. **Define the queue** in `queue_config.yml`:
+```yaml
+queues:
+  your_new_queue:
+    arch: "x86_64"
+    os: "linux" 
+    description: "Description of queue capabilities"
+```
+
+2. **Configure benchmarks** to use the new queue using either method above
+
+3. **Set up the Buildkite infrastructure** to have agents with the `your_new_queue` tag
+
+## Testing
+
+To test the pipeline generation locally:
+
+```bash
+# Test with specific files
+julia .buildkite/generate_pipeline.jl benchmarks/PINNOptimizers/poisson.jmd
+
+# Test with benchmark folders
+julia .buildkite/generate_pipeline.jl benchmarks/PINNOptimizers
+```
+
+## Backward Compatibility
+
+- Benchmarks without queue specifications continue to use `juliaecosystem`
+- Existing CI behavior is preserved for benchmarks that don't opt into GPU queues
+- The system gracefully handles missing configuration files or invalid queue names
+
+## Environment Variables
+
+For GPU benchmarks, the following environment variables are automatically set:
+
+- `JULIA_CUDA_USE_BINARYBUILDER=false`: Use system CUDA instead of artifacts
+- `JULIA_GPU_ALLOW_DEFAULT=true`: Allow default GPU selection
+
+## Troubleshooting
+
+### Pipeline Not Generated
+- Check that `YAML.jl` is available in the Julia environment
+- Verify the `queue_config.yml` syntax is valid YAML
+- Ensure changed files are properly detected
+
+### Wrong Queue Assignment
+- Check the `[buildkite]` section in your benchmark's `Project.toml`
+- Verify the benchmark name matches the folder name in `queue_config.yml`
+- Confirm the queue exists in the `queues` section
+
+### GPU Jobs Failing
+- Verify GPU agents are available and tagged with `queue: "gpu"`
+- Check that CUDA dependencies are properly configured on GPU nodes
+- Review environment variables for GPU-specific settings
\ No newline at end of file
diff --git a/.buildkite/generate_pipeline.jl b/.buildkite/generate_pipeline.jl
new file mode 100644
index 000000000..4cf297bba
--- /dev/null
+++ b/.buildkite/generate_pipeline.jl
@@ -0,0 +1,132 @@
+#!/usr/bin/env julia
+
+using YAML, Pkg
+
+"""
+Generate dynamic Buildkite pipeline based on queue configuration and changed files.
+"""
+
+# Read queue configuration
+config_path = joinpath(@__DIR__, "queue_config.yml")
+if !isfile(config_path)
+    error("Queue configuration file not found: $config_path")
+end
+
+config = YAML.load_file(config_path)
+default_queue = get(config, "default_queue", "juliaecosystem")
+queues = get(config, "queues", Dict())
+benchmark_queues = get(config, "benchmark_queues", Dict())
+
+# Get changed files from environment or arguments
+changed_files = if haskey(ENV, "BUILDKITE_CHANGED_FILES")
+    split(ENV["BUILDKITE_CHANGED_FILES"], "\n")
+elseif length(ARGS) > 0
+    ARGS
+else
+    # Default to testing folder for demonstration
+    ["benchmarks/Testing/test.jmd"]
+end
+
+# Process changed files using the existing project-coalescing logic
+coalescing_script = joinpath(@__DIR__, "path_processors", "project-coalescing")
+if isfile(coalescing_script) && haskey(ENV, "BUILDKITE_AGENT_ACCESS_TOKEN")
+    # Only use project-coalescing if we're in a real Buildkite environment
+    try
+        build_targets = strip(read(`$coalescing_script $changed_files`, String))
+        targets = split(build_targets)
+    catch e
+        @warn "project-coalescing failed, using fallback" exception=e
+        targets = changed_files
+    end
+else
+    # Fallback: use the provided targets directly or extract from file paths
+    targets = if any(startswith(f, "benchmarks/") for f in changed_files)
+        unique([startswith(f, "benchmarks/") ? f : "benchmarks/$f" for f in changed_files])
+    else
+        changed_files
+    end
+end
+
+# Generate pipeline steps
+steps = []
+
+for target in targets
+    if isfile(target)
+        # Single file target
+        folder = dirname(target)
+        benchmark_name = basename(folder)
+    else
+        # Folder target  
+        benchmark_name = basename(target)
+    end
+    
+    # Determine queue for this benchmark
+    # Priority: 1. Project.toml metadata, 2. queue_config.yml, 3. default
+    queue_name = default_queue
+    
+    # Check if benchmark has its own Project.toml with queue specification
+    project_path = if isfile(target)
+        joinpath(dirname(target), "Project.toml")
+    else
+        joinpath(target, "Project.toml")
+    end
+    
+    if isfile(project_path)
+        try
+            project_content = read(project_path, String)
+            # Simple regex to extract queue from [buildkite] section
+            queue_match = match(r"\[buildkite\].*?queue\s*=\s*\"([^\"]+)\""s, project_content)
+            if queue_match !== nothing
+                queue_name = queue_match.captures[1]
+            end
+        catch e
+            @warn "Could not parse Project.toml for queue info: $(project_path)" exception=e
+        end
+    end
+    
+    # Fallback to configuration file
+    if queue_name == default_queue
+        queue_name = get(benchmark_queues, benchmark_name, default_queue)
+    end
+    
+    queue_info = get(queues, queue_name, Dict("arch" => "x86_64", "os" => "linux"))
+    
+    # Create step configuration
+    step = Dict(
+        "label" => ":julia: $(benchmark_name) on $(queue_name)",
+        "command" => "julia benchmark.jl $(target)",
+        "plugins" => [
+            Dict("JuliaCI/julia#v1" => Dict("version" => "1.10")),
+            Dict("JuliaCI/julia-test#v1" => nothing)
+        ],
+        "timeout_in_minutes" => 60,  # Increased timeout for potential GPU jobs
+        "artifact_paths" => [
+            "html/$(benchmark_name)/*.html",
+            "markdown/$(benchmark_name)/*.md", 
+            "notebook/$(benchmark_name)/*.ipynb",
+            "pdf/$(benchmark_name)/*.pdf",
+            "script/$(benchmark_name)/*.jl"
+        ],
+        "agents" => Dict(
+            "queue" => queue_name,
+            "arch" => get(queue_info, "arch", "x86_64"),
+            "os" => get(queue_info, "os", "linux")
+        )
+    )
+    
+    # Add GPU-specific environment variables if needed
+    if queue_name == "gpu"
+        step["env"] = Dict(
+            "JULIA_CUDA_USE_BINARYBUILDER" => "false",
+            "JULIA_GPU_ALLOW_DEFAULT" => "true"
+        )
+    end
+    
+    push!(steps, step)
+end
+
+# Generate final pipeline
+pipeline = Dict("steps" => steps)
+
+# Output as YAML
+println(YAML.write(pipeline))
\ No newline at end of file
diff --git a/.buildkite/queue_config.yml b/.buildkite/queue_config.yml
new file mode 100644
index 000000000..0fa53ba95
--- /dev/null
+++ b/.buildkite/queue_config.yml
@@ -0,0 +1,34 @@
+# Queue Configuration for SciMLBenchmarks
+# This file maps benchmark folders to their preferred Buildkite queues
+# If a benchmark is not listed here, it will use the default queue
+
+# Default queue for all benchmarks (backward compatibility)
+default_queue: "juliaecosystem"
+
+# Available queues and their characteristics
+queues:
+  juliaecosystem:
+    arch: "x86_64"
+    os: "linux"
+    description: "CPU-only compute queue"
+  
+  gpu:
+    arch: "x86_64" 
+    os: "linux"
+    description: "GPU-enabled compute queue"
+    # Add other GPU-specific requirements as needed
+
+# Benchmark-specific queue assignments
+benchmark_queues:
+  # GPU-intensive benchmarks
+  PINNOptimizers: "gpu"
+  PINNErrorsVsTime: "gpu"
+  NBodySimulator: "gpu"  # Can benefit from GPU acceleration
+  
+  # CPU-only benchmarks (explicitly set for clarity)
+  NonStiffODE: "juliaecosystem"
+  StiffODE: "juliaecosystem"
+  LinearSolve: "juliaecosystem"
+  
+  # Add more mappings as needed
+  # BenchmarkFolder: "queue_name"
\ No newline at end of file
diff --git a/.buildkite/test_sciml.yml b/.buildkite/test_sciml.yml
index 99f9d6734..c16609fa7 100644
--- a/.buildkite/test_sciml.yml
+++ b/.buildkite/test_sciml.yml
@@ -1,21 +1,11 @@
 steps:
-  - label: ":julia: Run tests on LTS"
-    plugins:
-      - JuliaCI/julia#v1:
-          version: '1.10'
-      - JuliaCI/julia-test#v1:
-    timeout_in_minutes: 20
-    artifact_paths:
-       # Upload .html
-      - "html/Testing/*.html"
-      # Upload markdown
-      - "markdown/Testing/*.md"
-      # Upload notebook
-      - "notebook/Testing/*.ipynb"
-      # Upload .pdf files
-      - "pdf/Testing/*.pdf"
-      # Upload Julia script
-      - "script/Testing/*.jl"
+  - label: ":gear: Generate Dynamic Pipeline"
+    command: |
+      # Install required Julia packages
+      julia -e 'using Pkg; Pkg.add("YAML")'
+      
+      # Generate dynamic pipeline based on changed files and queue configuration
+      julia .buildkite/generate_pipeline.jl | buildkite-agent pipeline upload
     agents:
       queue: "juliaecosystem"
       arch: "x86_64"
diff --git a/benchmarks/PINNOptimizers/Project.toml b/benchmarks/PINNOptimizers/Project.toml
index 7cd4bce2a..86e307866 100644
--- a/benchmarks/PINNOptimizers/Project.toml
+++ b/benchmarks/PINNOptimizers/Project.toml
@@ -17,3 +17,7 @@ OptimizationOptimJL = "0.4"
 OptimizationOptimisers = "0.3"
 Plots = "1"
 SciMLBenchmarks = "0.1"
+
+[buildkite]
+# Specify that this benchmark should run on the GPU queue
+queue = "gpu"