updated benchmarks including thunderbolt and configuratuion guides

This commit is contained in:
Donato Capitella
2026-02-25 10:48:42 +00:00
orang tua 6875f62ccf
melakukan b035bcb482
26 mengubah file dengan 820 tambahan dan 83 penghapusan
+21 -3
Melihat File
@@ -473,6 +473,10 @@
style="font-size: 0.9rem; font-weight: 500; display: flex; align-items: center; gap: 4px; cursor: pointer;">
<input type="checkbox" id="toggleTP2Eth"> TP2 (Eth)
</label>
<label
style="font-size: 0.9rem; font-weight: 500; display: flex; align-items: center; gap: 4px; cursor: pointer;">
<input type="checkbox" id="toggleTP2Usb"> TP2 (Thunderbolt)
</label>
</div>
<!-- Attention Group -->
@@ -549,6 +553,7 @@
showTP1: true,
showTP2: true,
showTP2Eth: false,
showTP2Usb: false,
showTriton: true,
showRocm: false
};
@@ -621,6 +626,7 @@
$('toggleTP1').addEventListener('change', e => { state.showTP1 = e.target.checked; render(); });
$('toggleTP2').addEventListener('change', e => { state.showTP2 = e.target.checked; render(); });
$('toggleTP2Eth').addEventListener('change', e => { state.showTP2Eth = e.target.checked; render(); });
$('toggleTP2Usb').addEventListener('change', e => { state.showTP2Usb = e.target.checked; render(); });
$('toggleTriton').addEventListener('change', e => { state.showTriton = e.target.checked; render(); });
$('toggleRocm').addEventListener('change', e => { state.showRocm = e.target.checked; render(); });
}
@@ -643,15 +649,21 @@
results: {
1: { triton: null, rocm: null },
2: { triton: null, rocm: null },
"2_eth": { triton: null, rocm: null }
"2_eth": { triton: null, rocm: null },
"2_usb": { triton: null, rocm: null }
}
};
}
const m = testGroups[testName].models[modelName];
let tp = run.tp || 1;
if (tp === 2 && run.network === "Ethernet") {
tp = "2_eth";
if (tp === 2) {
if (run.network === "Ethernet") {
if (run.tag === "usb") tp = "2_usb";
else tp = "2_eth";
} else if (run.tag === "usb") {
tp = "2_usb";
}
}
if (!m.results[tp]) m.results[tp] = { triton: null, rocm: null };
@@ -766,6 +778,10 @@
if (state.showTriton) cols.push({ id: "tp2_eth_triton", label: "TP2 Eth Triton" });
if (state.showRocm) cols.push({ id: "tp2_eth_rocm", label: "TP2 Eth ROCm" });
}
if (state.showTP2Usb) {
if (state.showTriton) cols.push({ id: "tp2_usb_triton", label: "TP2 TB Triton" });
if (state.showRocm) cols.push({ id: "tp2_usb_rocm", label: "TP2 TB ROCm" });
}
// Thead
let theadHtml = `<thead><tr><th class="col-model">Model</th>`;
@@ -835,6 +851,8 @@
if (colId === "tp2_rocm") return m.results[2]?.rocm;
if (colId === "tp2_eth_triton") return m.results["2_eth"]?.triton;
if (colId === "tp2_eth_rocm") return m.results["2_eth"]?.rocm;
if (colId === "tp2_usb_triton") return m.results["2_usb"]?.triton;
if (colId === "tp2_usb_rocm") return m.results["2_usb"]?.rocm;
return null;
}
+20
Melihat File
@@ -68,8 +68,27 @@ def parse_logs():
# Network
network = "RoCE"
network_prefix = ""
if "_eth" in rest:
network = "Ethernet"
network_prefix = "_eth"
# Tag Extraction
tag = ""
test_type_str = ""
if "throughput" in fname:
test_type_str = "_throughput.json"
elif "latency" in fname:
qps_match = re.search(r"(_qps[\d\.]+)_latency\.json$", rest)
if qps_match:
test_type_str = qps_match.group(0)
else:
test_type_str = "_latency.json"
raw_prefix = f"{tp}{network_prefix}"
if rest.endswith(test_type_str):
tag_part = rest[len(raw_prefix):-len(test_type_str)]
tag = tag_part.lstrip("_")
# Model Name
if "_" in model_part:
@@ -93,6 +112,7 @@ def parse_logs():
"name_params_b": params_b,
"backend": backend_name, # "Triton" or "ROCm"
"network": network,
"tag": tag,
"error": False
}
+358
Melihat File
@@ -10,6 +10,7 @@
"name_params_b": 8.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -25,6 +26,7 @@
"name_params_b": 12.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -40,6 +42,7 @@
"name_params_b": 14.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -55,6 +58,7 @@
"name_params_b": 20.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -70,6 +74,7 @@
"name_params_b": 120.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -85,6 +90,7 @@
"name_params_b": null,
"backend": "Triton",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -100,6 +106,7 @@
"name_params_b": 30.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -115,6 +122,7 @@
"name_params_b": 30.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -130,6 +138,7 @@
"name_params_b": 80.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -145,6 +154,7 @@
"name_params_b": null,
"backend": "Triton",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -160,6 +170,7 @@
"name_params_b": 8.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -175,6 +186,7 @@
"name_params_b": 12.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -190,6 +202,7 @@
"name_params_b": 14.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -205,6 +218,7 @@
"name_params_b": 20.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -220,6 +234,7 @@
"name_params_b": 120.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -235,6 +250,7 @@
"name_params_b": null,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -250,6 +266,7 @@
"name_params_b": 30.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -265,6 +282,7 @@
"name_params_b": 30.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -280,6 +298,7 @@
"name_params_b": 80.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -295,6 +314,7 @@
"name_params_b": null,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -310,6 +330,7 @@
"name_params_b": 8.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -325,6 +346,7 @@
"name_params_b": 14.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -340,6 +362,7 @@
"name_params_b": 20.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -355,6 +378,7 @@
"name_params_b": 120.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -370,6 +394,7 @@
"name_params_b": null,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -385,6 +410,7 @@
"name_params_b": 30.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -400,6 +426,7 @@
"name_params_b": 30.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -415,6 +442,7 @@
"name_params_b": 80.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -430,11 +458,172 @@
"name_params_b": 12.0,
"backend": "Triton",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
"tps_mean": 159.95620436815713
},
{
"model": "Qwen/Qwen3-14B-AWQ",
"model_clean": "Qwen/Qwen3-14B-AWQ",
"env": "TP2",
"gpu_config": "dual",
"quant": "AWQ",
"params_b": 14.0,
"name_params_b": 14.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 302.43330149293365
},
{
"model": "btbtyler09/Qwen3-Coder-30B-A3B-Instruct-gptq-4bit",
"model_clean": "btbtyler09/Qwen3-Coder-30B-A3B-Instruct-gptq-4bit",
"env": "TP2",
"gpu_config": "dual",
"quant": "GPTQ",
"params_b": 30.0,
"name_params_b": 30.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 291.6914862601304
},
{
"model": "btbtyler09/Qwen3-Coder-30B-A3B-Instruct-gptq-8bit",
"model_clean": "btbtyler09/Qwen3-Coder-30B-A3B-Instruct-gptq-8bit",
"env": "TP2",
"gpu_config": "dual",
"quant": "GPTQ",
"params_b": 30.0,
"name_params_b": 30.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 256.91287782898553
},
{
"model": "dazipe/Qwen3-Next-80B-A3B-Instruct-GPTQ-Int4A16",
"model_clean": "dazipe/Qwen3-Next-80B-A3B-Instruct-GPTQ-Int4A16",
"env": "TP2",
"gpu_config": "dual",
"quant": "GPTQ",
"params_b": 80.0,
"name_params_b": 80.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 178.05541683872298
},
{
"model": "google/gemma-3-12b-it",
"model_clean": "google/gemma-3-12b-it",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": 12.0,
"name_params_b": 12.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 260.10671430704866
},
{
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"model_clean": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": 8.0,
"name_params_b": 8.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 513.235753981134
},
{
"model": "mratsim/MiniMax-M2.5-BF16-INT4-AWQ",
"model_clean": "mratsim/MiniMax-M2.5-BF16-INT4-AWQ",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": null,
"name_params_b": null,
"backend": "Triton",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 99.38780646163637
},
{
"model": "openai/gpt-oss-120b",
"model_clean": "openai/gpt-oss-120b",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": 120.0,
"name_params_b": 120.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 109.82173992256857
},
{
"model": "openai/gpt-oss-20b",
"model_clean": "openai/gpt-oss-20b",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": 20.0,
"name_params_b": 20.0,
"backend": "Triton",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 521.2677328949931
},
{
"model": "zai-org/GLM-4.7-Flash",
"model_clean": "zai-org/GLM-4.7-Flash",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": null,
"name_params_b": null,
"backend": "Triton",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 320.9476287228403
},
{
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"model_clean": "meta-llama/Meta-Llama-3.1-8B-Instruct",
@@ -445,6 +634,7 @@
"name_params_b": 8.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -460,6 +650,7 @@
"name_params_b": 12.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -475,6 +666,7 @@
"name_params_b": 14.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -490,6 +682,7 @@
"name_params_b": 20.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -505,6 +698,7 @@
"name_params_b": 120.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -520,6 +714,7 @@
"name_params_b": 30.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -535,6 +730,7 @@
"name_params_b": 30.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -550,6 +746,7 @@
"name_params_b": 80.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -565,6 +762,7 @@
"name_params_b": null,
"backend": "ROCm",
"network": "Ethernet",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -580,6 +778,7 @@
"name_params_b": 8.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -595,6 +794,7 @@
"name_params_b": 12.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -610,6 +810,7 @@
"name_params_b": 14.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -625,6 +826,7 @@
"name_params_b": 20.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -640,6 +842,7 @@
"name_params_b": 120.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -655,6 +858,7 @@
"name_params_b": 30.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -670,6 +874,7 @@
"name_params_b": 30.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -685,6 +890,7 @@
"name_params_b": 80.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 2,
@@ -700,6 +906,7 @@
"name_params_b": 8.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -715,6 +922,7 @@
"name_params_b": 14.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -730,6 +938,7 @@
"name_params_b": 20.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -745,6 +954,7 @@
"name_params_b": 120.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -760,6 +970,7 @@
"name_params_b": 30.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -775,6 +986,7 @@
"name_params_b": 30.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -790,6 +1002,7 @@
"name_params_b": 80.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
@@ -805,10 +1018,155 @@
"name_params_b": 12.0,
"backend": "ROCm",
"network": "RoCE",
"tag": "",
"error": false,
"test": "Throughput",
"tp": 1,
"tps_mean": 267.99881204205957
},
{
"model": "Qwen/Qwen3-14B-AWQ",
"model_clean": "Qwen/Qwen3-14B-AWQ",
"env": "TP2",
"gpu_config": "dual",
"quant": "AWQ",
"params_b": 14.0,
"name_params_b": 14.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 311.59947769256274
},
{
"model": "btbtyler09/Qwen3-Coder-30B-A3B-Instruct-gptq-4bit",
"model_clean": "btbtyler09/Qwen3-Coder-30B-A3B-Instruct-gptq-4bit",
"env": "TP2",
"gpu_config": "dual",
"quant": "GPTQ",
"params_b": 30.0,
"name_params_b": 30.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 299.24102990342374
},
{
"model": "btbtyler09/Qwen3-Coder-30B-A3B-Instruct-gptq-8bit",
"model_clean": "btbtyler09/Qwen3-Coder-30B-A3B-Instruct-gptq-8bit",
"env": "TP2",
"gpu_config": "dual",
"quant": "GPTQ",
"params_b": 30.0,
"name_params_b": 30.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 267.6893183038276
},
{
"model": "dazipe/Qwen3-Next-80B-A3B-Instruct-GPTQ-Int4A16",
"model_clean": "dazipe/Qwen3-Next-80B-A3B-Instruct-GPTQ-Int4A16",
"env": "TP2",
"gpu_config": "dual",
"quant": "GPTQ",
"params_b": 80.0,
"name_params_b": 80.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 174.78952518165474
},
{
"model": "google/gemma-3-12b-it",
"model_clean": "google/gemma-3-12b-it",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": 12.0,
"name_params_b": 12.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 342.7783704164132
},
{
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"model_clean": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": 8.0,
"name_params_b": 8.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 550.8856190566602
},
{
"model": "mratsim/MiniMax-M2.5-BF16-INT4-AWQ",
"model_clean": "mratsim/MiniMax-M2.5-BF16-INT4-AWQ",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": null,
"name_params_b": null,
"backend": "ROCm",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 98.10548978313048
},
{
"model": "openai/gpt-oss-120b",
"model_clean": "openai/gpt-oss-120b",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": 120.0,
"name_params_b": 120.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 111.32822174858649
},
{
"model": "openai/gpt-oss-20b",
"model_clean": "openai/gpt-oss-20b",
"env": "TP2",
"gpu_config": "dual",
"quant": "BF16",
"params_b": 20.0,
"name_params_b": 20.0,
"backend": "ROCm",
"network": "Ethernet",
"tag": "usb",
"error": false,
"test": "Throughput",
"tp": 2,
"tps_mean": 519.7896706376232
}
]
}