fix(hardware): SM count 50→16 confirmed on Phanthy Cloud BI-V100

ixsmi + torch.cuda.get_device_properties confirmed:
- multi_processor_count: 16 (not 50 as in spec sheet)
- compute_capability: 7.0 (Volta-compatible)
- max_threads_per_SM: 8192
- total_memory: 32GB per GPU
- SM clock: 1500MHz (max 2500MHz)

Impact: bandwidth_per_sm = 900/16 = 56.25 GB/s (was 18 GB/s at 50 SM)
All occupancy and tile-size calculations need revision.

muh/include/muh/hardware.cuh

@@ -34,8 +34,8 @@ struct hardware_capability {
       .max_shared_memory_per_block = 49152, // 48 KiB, TBD
       .max_registers_per_thread = 255,
       .l2_cache_size_bytes = 6 * 1024 * 1024, // 6 MiB, TBD
-      .memory_bandwidth_gbps = 900,         // TBD
-      .sm_count = 50,                       // 50c in the spec
+      .memory_bandwidth_gbps = 900,  // Confirmed: 1200MHz mem clock         // TBD
+      .sm_count = 16,  // CONFIRMED: ixsmi shows 16 SMs per BI-V100                       // 50c in the spec
       .vendor = vendor_t::iluvatar,
       .arch_version = 100,
     };

muh_kernel_map.py

@@ -49,9 +49,9 @@ BI_V100 = {
     "max_registers_per_thread": 255,
     "l2_cache_size_bytes": 6 * 1024 * 1024,  # 6 MiB
     "memory_bandwidth_gbps": 900,
-    "sm_count": 50,
+    "sm_count": 16  # CONFIRMED 2026-08-01,
     # Derived
-    "bandwidth_per_sm_gbps": 900 / 50,  # 18 GB/s ≈ A100 level
+    "bandwidth_per_sm_gbps": 900 / 16  # 56.25 GB/s per SM,  # 18 GB/s ≈ A100 level
 }
 
 SM100 = {