EZHARDYNAMICS/ezhar-logic-kernel-twin

1import streamlit as st
2import time
3import pandas as pd
4import numpy as np
5import plotly.graph_objects as go
6from utils import inject_industrial_css, verify_session
7# Import the kernel simulator explicitly
8from kernel_simulator import fake_cuda_kernel_matrix_mul, simulated_gpu_latency
9
10st.set_page_config(page_title="SIM CORE", layout="wide")
11inject_industrial_css()
12verify_session()
13
14# --- HARDWARE SPECS ---
15GPU_SPECS = {
16    "NVIDIA H100 NVL": {"arch": "Hopper", "mem": "188 GB HBM3", "bw": "7.8 TB/s", "tdp": "700 W", "fp8": "3,958 TFLOPS"},
17    "NVIDIA A100 SXM4": {"arch": "Ampere", "mem": "80 GB HBM2e", "bw": "2.0 TB/s", "tdp": "400 W", "fp8": "624 TFLOPS"},
18    "NVIDIA L40S": {"arch": "Ada Lovelace", "mem": "48 GB GDDR6", "bw": "864 GB/s", "tdp": "350 W", "fp8": "733 TFLOPS"}
19}
20
21st.title("Module 01 — Simulation Core")
22st.markdown("""
23<div style='background-color:#111; padding:15px; border-left:3px solid #76b900; margin-bottom:20px'>
24    <strong style='color:#76b900'>MISSION TARGET:</strong> Configure compute topology. <br>
25    <span style='font-size:0.8em; color:#888'>NOTE: System is running on CPU. CUDA Kernels will be emulated for latency projection.</span>
26</div>
27""", unsafe_allow_html=True)
28
29# --- CONFIGURATION ---
30col_config, col_specs = st.columns([1, 1])
31
32with col_config:
33    st.subheader("1. CLUSTER TOPOLOGY")
34    with st.container(border=True):
35        gpu_model = st.selectbox("GPU ARCHITECTURE", list(GPU_SPECS.keys()), index=0)
36        c1, c2 = st.columns(2)
37        with c1: node_count = st.number_input("NODE COUNT", value=32, min_value=1, step=8)
38        with c2: matrix_size = st.selectbox("MATRIX SIZE (N)", [512, 1024, 2048, 4096], index=1)
39        
40        # Add a visual selector for "Compute Backend" to make CUDA visible
41        backend = st.selectbox("COMPUTE KERNEL", ["CUDA (Tensor Cores)", "OpenCL (Legacy)"], index=0)
42        pue = st.slider("PUE RATING", 1.0, 2.0, 1.2, 0.01)
43
44current_spec = GPU_SPECS[gpu_model]
45
46with col_specs:
47    st.subheader("2. HARDWARE TELEMETRY")
48    with st.container(border=True):
49        st.markdown(f"**TARGET SKU:** `{gpu_model}`")
50        s1, s2 = st.columns(2)
51        s1.metric("VRAM", current_spec['mem'])
52        s2.metric("BANDWIDTH", current_spec['bw'])
53        s3, s4 = st.columns(2)
54        s3.metric("TDP", current_spec['tdp'])
55        s4.metric("FLOPS", current_spec['fp8'])
56        
57        if backend == "CUDA (Tensor Cores)":
58            st.info("⚡ CUDA DRIVER: v12.2 DETECTED (EMULATION MODE)")
59
60st.markdown("---")
61
62# --- EXECUTION ---
63exe_col, status_col = st.columns([1, 3])
64with exe_col: 
65    run_btn = st.button("EXECUTE KERNEL", type="primary", use_container_width=True)
66
67if run_btn:
68    # 1. VISUALIZATION: Fake CUDA Init
69    progress_bar = status_col.progress(0, text="INITIALIZING RUNTIME...")
70    
71    # Cinematic steps to show "CUDA" is doing work
72    steps = [
73        (10, "LOADING CUDA DRIVERS (nvcc)..."),
74        (25, f"ALLOCATING VRAM ON {gpu_model}..."),
75        (40, "TRANSFERRING TENSORS TO GPU..."),
76        (60, "EXECUTING GEMM KERNEL (FP32)..."),
77        (85, "RETRIEVING RESULTS FROM DEVICE...")
78    ]
79    
80    for pct, text in steps:
81        time.sleep(0.12) # Just for effect
82        progress_bar.progress(pct, text=text)
83
84    # 2. REAL LOGIC: CPU Stress Test (Emulating the workload)
85    # This runs the code from kernel_simulator.py
86    cpu_ms = fake_cuda_kernel_matrix_mul(matrix_size, repeat=1)
87    
88    # 3. PROJECTION: Calculate H100 Speed
89    gpu_ms = simulated_gpu_latency(cpu_ms, target_factor=40.0)
90    
91    progress_bar.progress(100, text="COMPUTATION COMPLETE")
92    
93    st.success("KERNEL EXECUTION SUCCESSFUL")
94    
95    # 4. RESULTS
96    m1, m2, m3 = st.columns(3)
97    m1.metric("CPU LATENCY (ACTUAL)", f"{cpu_ms:.1f} ms", delta="HIGH LATENCY", delta_color="inverse")
98    m2.metric("H100 LATENCY (SIMULATED)", f"{gpu_ms:.2f} ms", delta="-97%", delta_color="normal")
99    m3.metric("ACCELERATION FACTOR", f"{cpu_ms/gpu_ms:.1f}x")
100    
101    # Save State
102    st.session_state['sim_result'] = {
103        "matrix_size": matrix_size, "cpu_ms": cpu_ms, "sim_gpu_ms": gpu_ms,
104        "nodes": node_count, "pue": pue, "model": gpu_model
105    }
106    
107    # Chart
108    fig = go.Figure()
109    fig.add_trace(go.Bar(x=['Latency'], y=[cpu_ms], name='Current CPU', marker_color='#444'))
110    fig.add_trace(go.Bar(x=['Latency'], y=[gpu_ms], name=f'Target {gpu_model}', marker_color='#76b900'))
111    fig.update_layout(
112        title="PERFORMANCE GAP ANALYSIS", 
113        paper_bgcolor='rgba(0,0,0,0)', 
114        plot_bgcolor='rgba(0,0,0,0)', 
115        font_color='#ccc', 
116        height=300
117    )
118    st.plotly_chart(fig, use_container_width=True)
119else:
120    status_col.info("Waiting for execution command...")