You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: docs/en/academy/index.md
+42-28Lines changed: 42 additions & 28 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -4,48 +4,62 @@ title: Academy
4
4
5
5
# Academy
6
6
7
-
The academy is the ordered learning surface of this repository. Architecture gives the map. The academy gives the teaching sequence.
7
+
The academy is the ordered learning surface of this repository. Architecture gives the system map. The academy gives the teaching sequence — the order in which each kernel stage is explained, and why that order is non-negotiable.
8
8
9
-
## The rule of this section
9
+
## The structuring principle
10
10
11
-
Read kernels as a progression of bottleneck shifts:
11
+
Read kernels as a progression of bottleneck shifts, not as a list of tricks:
That order matters because every later page assumes the previous page already explained why its extra complexity is justified.
21
+
Each later page assumes the previous page already explained why its extra complexity is justified. Reading out of order makes the causal chain invisible.
20
22
21
23
## Academy map
22
24
23
25
| Track | Purpose | Start here |
24
26
|---|---|---|
25
-
| Orientation | Learn the route through the ladder |[Learning Path](./learning-path)|
27
+
| Orientation | Learn the route through the ladder before opening any kernel page |[Learning Path](./learning-path)|
7.[Diagnosis Loop](./diagnosis-loop) — turn measurements into decisions
42
+
8.[Optimization Playbook](./optimization-playbook) — structured tuning process
41
43
42
44
## Interview-ready framing
43
45
44
-
When you need to explain the project quickly:
46
+
When defending any kernel stage under review, use this four-part structure:
47
+
48
+
1.**Name the current bottleneck** — what resource is saturated or wastefully used?
49
+
2.**Name the specific structural change** — what does this kernel do differently at the hardware level?
50
+
3.**State the evidence requirement** — what measurement would confirm the change helped?
51
+
4.**State the constraint** — what assumption or shape condition limits this improvement?
52
+
53
+
That sequence keeps the discussion at the level of engineering reasoning rather than benchmark screenshots. The academy is designed to give you a defensible answer for each of the five stages.
54
+
55
+
## What the academy is not
56
+
57
+
The academy is not a reference manual for CUDA programming. For reference, use the [CUDA C++ Programming Guide](https://docs.nvidia.com/cuda/cuda-c-programming-guide/) and the [CUDA Memory Cheat Sheet](./cuda-memory-cheatsheet) in this section.
58
+
59
+
The academy is not a substitute for reading the source code. Each kernel page explains the architectural reasoning; the code itself contains the implementation. Both are necessary to give a complete account of any stage.
45
60
46
-
1. name the current bottleneck
47
-
2. name the specific structural change
48
-
3. say what evidence would prove that change helped
49
-
4. say what constraint still limits the design
61
+
## Related resources
50
62
51
-
That sequence keeps the discussion technical and keeps you out of vague “it got faster” claims.
63
+
-[Architecture Overview](../architecture/) — the system map that contextualizes the ladder
64
+
-[Validation Overview](../validation/) — the trust boundary for any number produced during academy study
65
+
-[Performance Model](../validation/performance-model) — analytical cost model behind each ladder stage
0 commit comments