update gpu configs

cli99 · Oct 18, 2023 · 86cd4e8 · 86cd4e8
1 parent 5fcc4b2
commit 86cd4e8
Show file tree

Hide file tree

Showing 6 changed files with 27 additions and 16 deletions.
diff --git a/llm_analysis/gpu_configs/a10-pcie-40gb.json → llm_analysis/gpu_configs/a10-pcie-28gb.json b/llm_analysis/gpu_configs/a10-pcie-40gb.json → llm_analysis/gpu_configs/a10-pcie-28gb.json
@@ -2,10 +2,10 @@
     "name": "a10-pcie-28gb",
     "mem_per_GPU_in_GB": 28,
     "hbm_bandwidth_in_GB_per_sec": 600,
-    "intra_node_bandwidth_in_GB_per_sec": 300,
+    "intra_node_bandwidth_in_GB_per_sec": 32,
     "intra_node_min_message_latency": 8e-06,
     "peak_fp16_TFLOPS": 125,
     "peak_i8_TFLOPS": 250,
     "peak_i4_TFLOPS": 500,
     "inter_node_bandwidth_in_GB_per_sec": 200
-}
+}
diff --git a/llm_analysis/gpu_configs/a10g-pcie-24gb.json b/llm_analysis/gpu_configs/a10g-pcie-24gb.json
@@ -0,0 +1,11 @@
+{
+    "name": "a10g-pcie-24gb",
+    "mem_per_GPU_in_GB": 24,
+    "hbm_bandwidth_in_GB_per_sec": 600,
+    "intra_node_bandwidth_in_GB_per_sec": 32,
+    "intra_node_min_message_latency": 8e-06,
+    "peak_fp16_TFLOPS": 70,
+    "peak_i8_TFLOPS": 140,
+    "peak_i4_TFLOPS": 280,
+    "inter_node_bandwidth_in_GB_per_sec": 200
+}
diff --git a/llm_analysis/gpu_configs/v100-pcie-16gb.json b/llm_analysis/gpu_configs/v100-pcie-16gb.json
@@ -2,10 +2,10 @@
     "name": "v100-pcie-16gb",
     "mem_per_GPU_in_GB": 16,
     "hbm_bandwidth_in_GB_per_sec": 900,
-    "intra_node_bandwidth_in_GB_per_sec": 150,
+    "intra_node_bandwidth_in_GB_per_sec": 16,
     "intra_node_min_message_latency": 8e-06,
     "peak_fp16_TFLOPS": 112,
-    "peak_i8_TFLOPS": 224,
-    "peak_i4_TFLOPS": 448,
+    "peak_i8_TFLOPS": 0,
+    "peak_i4_TFLOPS": 0,
     "inter_node_bandwidth_in_GB_per_sec": 200
-}
+}
diff --git a/llm_analysis/gpu_configs/v100-pcie-32gb.json b/llm_analysis/gpu_configs/v100-pcie-32gb.json
@@ -2,10 +2,10 @@
     "name": "v100-pcie-32gb",
     "mem_per_GPU_in_GB": 32,
     "hbm_bandwidth_in_GB_per_sec": 900,
-    "intra_node_bandwidth_in_GB_per_sec": 150,
+    "intra_node_bandwidth_in_GB_per_sec": 16,
     "intra_node_min_message_latency": 8e-06,
     "peak_fp16_TFLOPS": 112,
-    "peak_i8_TFLOPS": 224,
-    "peak_i4_TFLOPS": 448,
+    "peak_i8_TFLOPS": 0,
+    "peak_i4_TFLOPS": 0,
     "inter_node_bandwidth_in_GB_per_sec": 200
-}
+}
diff --git a/llm_analysis/gpu_configs/v100-sxm-16gb.json b/llm_analysis/gpu_configs/v100-sxm-16gb.json
@@ -5,7 +5,7 @@
     "intra_node_bandwidth_in_GB_per_sec": 150,
     "intra_node_min_message_latency": 8e-06,
     "peak_fp16_TFLOPS": 125,
-    "peak_i8_TFLOPS": 250,
-    "peak_i4_TFLOPS": 500,
+    "peak_i8_TFLOPS": 0,
+    "peak_i4_TFLOPS": 0,
     "inter_node_bandwidth_in_GB_per_sec": 200
-}
+}
diff --git a/llm_analysis/gpu_configs/v100-sxm-32gb.json b/llm_analysis/gpu_configs/v100-sxm-32gb.json
@@ -5,7 +5,7 @@
     "intra_node_bandwidth_in_GB_per_sec": 150,
     "intra_node_min_message_latency": 8e-06,
     "peak_fp16_TFLOPS": 125,
-    "peak_i8_TFLOPS": 250,
-    "peak_i4_TFLOPS": 500,
+    "peak_i8_TFLOPS": 0,
+    "peak_i4_TFLOPS": 0,
     "inter_node_bandwidth_in_GB_per_sec": 200
-}
+}