diff --git a/PerfectNumbers.Core/Cpu/MersenneNumberDivisorByDivisorCpuTester.cs b/PerfectNumbers.Core/Cpu/MersenneNumberDivisorByDivisorCpuTester.cs
index 4ee09a7..70cc274 100644
--- a/PerfectNumbers.Core/Cpu/MersenneNumberDivisorByDivisorCpuTester.cs
+++ b/PerfectNumbers.Core/Cpu/MersenneNumberDivisorByDivisorCpuTester.cs
@@ -18,8 +18,6 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
 
     private ulong _configuredMaxPrime;
     private ByDivisorDeltasDevice _deltasDevice = ByDivisorDeltasDevice.Cpu;
-    private MersenneNumberDivisorByDivisorGpuTester? _gpuDeltasTester;
-    private bool _gpuTesterConfigured;
 
 
     public int BatchSize
@@ -29,14 +27,6 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
         {
             int sanitized = Math.Max(1, value);
             _batchSize = sanitized;
-
-            lock (_sync)
-            {
-                if (_gpuDeltasTester is not null)
-                {
-                    _gpuDeltasTester.GpuBatchSize = sanitized;
-                }
-            }
         }
     }
 
@@ -60,7 +50,6 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
             _divisorLimit = ComputeDivisorLimitFromMaxPrime(maxPrime);
             _lastStatusDivisor = 0UL;
             _isConfigured = true;
-            _gpuTesterConfigured = false;
         }
     }
 
@@ -167,31 +156,6 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
         }
     }
 
-    internal IMersenneNumberDivisorByDivisorTester.IDivisorScanSession CreateGpuDeltasSession()
-    {
-        MersenneNumberDivisorByDivisorGpuTester tester;
-
-        lock (_sync)
-        {
-            if (!_isConfigured)
-            {
-                throw new InvalidOperationException("ConfigureFromMaxPrime must be called before using the tester.");
-            }
-
-            tester = _gpuDeltasTester ??= new MersenneNumberDivisorByDivisorGpuTester();
-
-            if (!_gpuTesterConfigured)
-            {
-                tester.ConfigureFromMaxPrime(_configuredMaxPrime);
-                _gpuTesterConfigured = true;
-            }
-
-            tester.GpuBatchSize = _batchSize;
-        }
-
-        return tester.CreateDivisorSession();
-    }
-
     public IMersenneNumberDivisorByDivisorTester.IDivisorScanSession CreateDivisorSession()
     {
         lock (_sync)
@@ -514,7 +478,7 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
         private ulong[]? _cycleRemainders;
         private ulong[]? _montgomeryDeltas;
         private int _bufferCapacity;
-        private IMersenneNumberDivisorByDivisorTester.IDivisorScanSession? _gpuSession;
+        private MersenneNumberDivisorResidueGpuEvaluator? _gpuResidueEvaluator;
 
         internal DivisorScanSession(MersenneNumberDivisorByDivisorCpuTester owner)
         {
@@ -572,13 +536,6 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
                 }
             }
 
-            if (_owner._deltasDevice == ByDivisorDeltasDevice.Gpu)
-            {
-                EnsureGpuSession();
-                _gpuSession!.CheckDivisor(divisor, cachedData, divisorCycle, primes, hits);
-                return;
-            }
-
             EnsureCapacity(length);
 
             Span<ulong> deltaSpan = _primeDeltas!.AsSpan(0, length);
@@ -587,11 +544,27 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
             Span<ulong> remainderSpan = _cycleRemainders!.AsSpan(0, length);
             ComputeCycleRemainders(divisorCycle, primes, deltaSpan, remainderSpan);
 
+            ulong montgomeryOne = cachedData.MontgomeryOne;
+
+            if (_owner._deltasDevice == ByDivisorDeltasDevice.Gpu)
+            {
+                EnsureGpuEvaluator();
+                Span<ulong> residueSpan = _montgomeryDeltas!.AsSpan(0, length);
+                // Reuse the Montgomery delta buffer to hold the final residues returned by the GPU evaluator.
+                _gpuResidueEvaluator!.ComputeResidues(primes, cachedData, residueSpan);
+
+                for (int i = 0; i < length; i++)
+                {
+                    hits[i] = remainderSpan[i] == 0UL && residueSpan[i] == montgomeryOne ? (byte)1 : (byte)0;
+                }
+
+                return;
+            }
+
             Span<ulong> montgomerySpan = _montgomeryDeltas!.AsSpan(0, length);
             ComputeMontgomeryDeltas(cachedData, primes, deltaSpan, montgomerySpan);
 
             ulong nPrime = cachedData.NPrime;
-            ulong montgomeryOne = cachedData.MontgomeryOne;
             ulong currentMontgomery = montgomerySpan[0];
 
             hits[0] = remainderSpan[0] == 0UL && currentMontgomery == montgomeryOne ? (byte)1 : (byte)0;
@@ -603,23 +576,23 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
             }
         }
 
-        private void EnsureGpuSession()
+        private void EnsureGpuEvaluator()
         {
-            if (_gpuSession is null)
+            if (_gpuResidueEvaluator is null)
             {
-                _gpuSession = _owner.CreateGpuDeltasSession();
+                _gpuResidueEvaluator = new MersenneNumberDivisorResidueGpuEvaluator(_owner._batchSize);
             }
         }
 
-        private void ReleaseGpuSession()
+        private void ReleaseGpuEvaluator()
         {
-            if (_gpuSession is null)
+            if (_gpuResidueEvaluator is null)
             {
                 return;
             }
 
-            _gpuSession.Dispose();
-            _gpuSession = null;
+            _gpuResidueEvaluator.Dispose();
+            _gpuResidueEvaluator = null;
         }
 
         private void EnsureCapacity(int requiredLength)
@@ -767,7 +740,7 @@ public sealed class MersenneNumberDivisorByDivisorCpuTester : IMersenneNumberDiv
             }
 
             _disposed = true;
-            ReleaseGpuSession();
+            ReleaseGpuEvaluator();
             _owner.ReturnSession(this);
         }
     }
diff --git a/PerfectNumbers.Core/Gpu/MersenneNumberDivisorResidueGpuEvaluator.cs b/PerfectNumbers.Core/Gpu/MersenneNumberDivisorResidueGpuEvaluator.cs
new file mode 100644
index 0000000..8fdd163
--- /dev/null
+++ b/PerfectNumbers.Core/Gpu/MersenneNumberDivisorResidueGpuEvaluator.cs
@@ -0,0 +1,114 @@
+using System;
+using System.Buffers;
+using System.Runtime.InteropServices;
+using PerfectNumbers.Core;
+using ILGPU;
+using ILGPU.Runtime;
+
+namespace PerfectNumbers.Core.Gpu;
+
+internal sealed class MersenneNumberDivisorResidueGpuEvaluator : IDisposable
+{
+    private readonly GpuContextPool.GpuContextLease _lease;
+    private readonly Accelerator _accelerator;
+    private readonly Action<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<ulong>> _kernel;
+    private MemoryBuffer1D<ulong, Stride1D.Dense> _exponentBuffer;
+    private MemoryBuffer1D<ulong, Stride1D.Dense> _resultBuffer;
+    private ulong[] _hostBuffer;
+    private int _capacity;
+    private bool _disposed;
+
+    internal MersenneNumberDivisorResidueGpuEvaluator(int batchSize)
+    {
+        _lease = GpuContextPool.RentPreferred(preferCpu: false);
+        _accelerator = _lease.Accelerator;
+        _kernel = _accelerator.LoadAutoGroupedStreamKernel<Index1D, MontgomeryDivisorData, ArrayView<ulong>, ArrayView<ulong>>(ComputeResiduesKernel);
+        _capacity = Math.Max(1, batchSize);
+        _exponentBuffer = _accelerator.Allocate1D<ulong>(_capacity);
+        _resultBuffer = _accelerator.Allocate1D<ulong>(_capacity);
+        _hostBuffer = ArrayPool<ulong>.Shared.Rent(_capacity);
+    }
+
+    public void Dispose()
+    {
+        if (_disposed)
+        {
+            return;
+        }
+
+        _disposed = true;
+        _exponentBuffer.Dispose();
+        _resultBuffer.Dispose();
+        ArrayPool<ulong>.Shared.Return(_hostBuffer, clearArray: false);
+        _lease.Dispose();
+    }
+
+    internal void ComputeResidues(ReadOnlySpan<ulong> exponents, in MontgomeryDivisorData divisorData, Span<ulong> destination)
+    {
+        if (_disposed)
+        {
+            throw new ObjectDisposedException(nameof(MersenneNumberDivisorResidueGpuEvaluator));
+        }
+
+        int length = exponents.Length;
+        if (length == 0)
+        {
+            return;
+        }
+
+        EnsureCapacity(length);
+
+        Span<ulong> hostSpan = _hostBuffer.AsSpan(0, length);
+        exponents.CopyTo(hostSpan);
+
+        ArrayView1D<ulong, Stride1D.Dense> exponentView = _exponentBuffer.View.SubView(0, length);
+        exponentView.CopyFromCPU(ref MemoryMarshal.GetReference(hostSpan), length);
+
+        ArrayView1D<ulong, Stride1D.Dense> resultView = _resultBuffer.View.SubView(0, length);
+        _kernel(length, divisorData, exponentView, resultView);
+        resultView.CopyToCPU(ref MemoryMarshal.GetReference(hostSpan), length);
+
+        hostSpan.CopyTo(destination);
+    }
+
+    private void EnsureCapacity(int required)
+    {
+        if (required <= _capacity)
+        {
+            return;
+        }
+
+        int newCapacity = _capacity;
+        if (newCapacity == 0)
+        {
+            newCapacity = 1;
+        }
+
+        while (newCapacity < required)
+        {
+            newCapacity <<= 1;
+        }
+
+        _exponentBuffer.Dispose();
+        _resultBuffer.Dispose();
+        ArrayPool<ulong>.Shared.Return(_hostBuffer, clearArray: false);
+
+        _capacity = newCapacity;
+        _exponentBuffer = _accelerator.Allocate1D<ulong>(_capacity);
+        _resultBuffer = _accelerator.Allocate1D<ulong>(_capacity);
+        _hostBuffer = ArrayPool<ulong>.Shared.Rent(_capacity);
+    }
+
+    private static void ComputeResiduesKernel(Index1D index, MontgomeryDivisorData divisorData, ArrayView<ulong> exponents, ArrayView<ulong> results)
+    {
+        int idx = index;
+        int length = (int)exponents.Length;
+        if (idx >= length)
+        {
+            return;
+        }
+
+        ulong exponent = exponents[idx];
+        results[idx] = exponent.Pow2MontgomeryModWindowedGpu(divisorData, keepMontgomery: true);
+    }
+}