Question Drawing millions of triangles (single mesh vs instancing)

I've recently been working on a project for which I need to draw millions of meshes (single triangles) at positions from a compute buffer.

The method I'm using is DrawPrimitives, but I've noticed using instancing seems to almost half performance at larger numbers of triangles (32 million), compared to when I just pass everything through as one instance, and I'm trying to work out if there's a reason.

Examples below:

Method 1 (~30fps) - As instances, indexing by vertexID

Code (CSharp):

1. // C# code
2. Graphics.RenderPrimitives(rp, MeshTopology.Triangles, 3, numBoids);

Code (Shader):

1. // Shader code
2. v2f vert(uint vertexID : SV_VertexID, uint instanceID : SV_INSTANCEID) {
3. Boid boid = boids[instanceID];
4. v2f o;
5. float3 pos = _Positions[vertexID];
6. rotate2D(pos.xy, boid.vel);
7. o.vertex = UnityObjectToClipPos((pos * _Scale) + float4(boid.pos.xy, 0, 0));
8. return o;
9. }

Method 2 (~70 fps) - As one large mesh, indexing by (vertexID % vertexCount)

Code (CSharp):

1. // C# code
2. Graphics.RenderPrimitives(rp, MeshTopology.Triangles, numBoids * 3);

Code (Shader):

1. // Shader code
2. v2f vert(uint vertexID : SV_VertexID) {
3. uint instanceID = vertexID / 3;
4. Boid boid = boids[instanceID];
5. v2f o;
6. float3 pos = _Positions[vertexID - instanceID * 3];
7. rotate2D(pos.xy, boid.vel);
8. o.vertex = UnityObjectToClipPos((pos * _Scale) + float4(boid.pos.xy, 0, 0));
9. return o;
10. }

I'm still quite new to shader programming in general so any info on why this performance difference exists would be greatly appreciated.