Skip to content

Make the site's benchmark claims match the project's own documentation #651

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
gHashTag merged 5 commits into from
Aug 10, 2026
Merged

Make the site's benchmark claims match the project's own documentation #651

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
855b449
fix(website): 298K is Microsoft's bitnet.cpp, and it is ops/sec, not …
gHashTag Aug 10, 2026
19d7d8b
fix(website): the link preview claimed a chip; the site's own proof p…
gHashTag Aug 10, 2026
295536f
fix(website): make both benchmark tables match the project's own publ…
gHashTag Aug 10, 2026
5b2d60e
docs: commit the 37 documents git could not see
gHashTag Aug 10, 2026
35d3c61
fix(website): apply the site audit's own recommended fixes, four year…
gHashTag Aug 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions apps/website/index.html
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@
<meta property="og:type" content="website" />
<meta property="og:url" content="https://t27.ai/" />
<meta property="og:title" content="TRINITY | Native Ternary Hardware for AI" />
<meta property="og:description" content="100x Faster AI Inference at 99.8% Less Power. First chip with native SU(3) Unitary Core. The Binary Era is Over." />
<meta property="og:description" content="100x Faster AI Inference at 99.8% Less Power. SU(3) unitary core through a SKY130 tape-out. The Binary Era is Over." />
<meta property="og:image" content="https://t27.ai/og-image.png" />
<meta property="og:site_name" content="TRINITY" />
<meta property="og:locale" content="en_US" />
Expand All @@ -32,7 +32,7 @@
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:url" content="https://t27.ai/" />
<meta name="twitter:title" content="TRINITY | Native Ternary Hardware for AI" />
<meta name="twitter:description" content="100x Faster AI Inference at 99.8% Less Power. First chip with native SU(3) Unitary Core." />
<meta name="twitter:description" content="100x Faster AI Inference at 99.8% Less Power. SU(3) unitary core through a SKY130 tape-out." />
<meta name="twitter:image" content="https://t27.ai/og-image.png" />

<!-- Canonical URL -->
Expand Down
74 changes: 36 additions & 38 deletions apps/website/messages/de.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,7 +111,7 @@
"proof": "Gewichte {-1, 0, +1} benötigen nur 1,58 Bits vs. 32 Bits für float32",
"result": "20× Speicherkompression, multiplikationsfreie Inferenz",
"source": "BitNet b1.58 (arXiv:2402.17764, Microsoft Research)",
"advantage": "KI-Inferenz ohne GPU, bis zu 3000× Energieeffizienz auf FPGA",
"advantage": "KI-Inferenz ohne GPU, bis zu Bis zu 3000× Energieeffizienz auf FPGA (projiziert, FPGA)",
"verified": {
"metric": "15,7× Kompression",
"value": "98KB → 6KB mit 43% Sparsity-Bonus",
Expand Down Expand Up @@ -304,7 +304,7 @@
},
{
"title": "Exponentieller ROI",
"desc": "578x Energieeffizienz-Sprung ist der einzige Weg zum Mining-Überleben nach 2026."
"desc": "Bis zu 578x Energieeffizienz-Sprung ist der einzige Weg zum Mining-Überleben nach 2026 (projiziert, FPGA)."
}
]
},
Expand Down Expand Up @@ -346,7 +346,8 @@
"label": "Singularität",
"color": "green"
}
]
],
"projectedNote": "* Projiziert auf kundenspezifischer FPGA-Hardware. Auf CPU verifiziert: 20× Speicherkompression."
},
"whynow": {
"title": "Warum <span class=\"grad\">Singularität</span> Jetzt?",
Expand Down Expand Up @@ -431,7 +432,7 @@
"color": "green"
},
{
"value": "578x",
"value": "578x *",
"label": "Energie-Sprung",
"color": "blue"
},
Expand All @@ -455,7 +456,8 @@
"label": "Entropie",
"color": "green"
}
]
],
"projectedNote": "* Projiziert auf kundenspezifischer FPGA-Hardware. Auf CPU verifiziert: 20× Speicherkompression."
},
"publications": {
"badge": "WISSENSCHAFTLICHE PUBLIKATIONEN",
Expand All @@ -473,8 +475,8 @@
"note": "Echte RunPod-Benchmarks - Pod IDs: bhhnrrdothqm54 (RTX 3090), ydk41ymp8uoeyp (A100). Keine Simulation.",
"gpus": [
{
"name": "RTX 3090",
"tokens": "298K",
"name": "RTX 4090 / L40S",
"tokens": "141K-608K",
"noise": "70,2%",
"power": "348W",
"cost": "$0,10",
Expand All @@ -492,9 +494,9 @@
},
"metrics": [
{
"value": "298K",
"label": "Tokens/s (RTX 3090)",
"desc": "Auf RunPod GPU verifiziert",
"value": "141K-608K",
"label": "Ops/sek (RTX 4090 / L40S)",
"desc": "Roher Kernel-Benchmark, RunPod",
"theorem": "GPU"
},
{
Expand All @@ -510,7 +512,7 @@
"theorem": "1"
},
{
"value": "3000×",
"value": "3000× *",
"label": "Energieeffizienz",
"desc": "Projektion auf Custom-FPGA",
"theorem": "2"
Expand Down Expand Up @@ -592,7 +594,8 @@
]
},
"verifyLink": "Benchmarks auf GitHub Verifizieren ↗",
"url": "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/gHashTag/trinity/blob/main/deploy/trinity-nexus/docs/research/BITNET_MATHEMATICAL_PROOF.md"
"url": "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/gHashTag/trinity/blob/main/deploy/trinity-nexus/docs/research/BITNET_MATHEMATICAL_PROOF.md",
"projectedNote": "* Projiziert auf kundenspezifischer FPGA-Hardware. Auf CPU verifiziert: 20× Speicherkompression."
},
"mining": {
"title": "Mining <span class=\"grad\">Industrie</span> 2026",
Expand Down Expand Up @@ -1144,44 +1147,37 @@
"rows": [
{
"company": "TRINITY",
"arch": "RTX 3090 (verifiziert)",
"val": "298K",
"ternary": true,
"open": true
},
{
"company": "TRINITY",
"arch": "A100 80GB (verifiziert)",
"val": "274K",
"arch": "CPU (RunPod)",
"val": "35-52",
"ternary": true,
"open": true
},
{
"company": "Groq",
"arch": "LPU (binär)",
"val": "276-840",
"arch": "Llama-70B, LPU cloud",
"val": "227-276",
"ternary": false,
"open": false
},
{
"company": "NVIDIA",
"arch": "H100 TensorRT",
"val": "1000-2000",
"company": "GPT-4o-mini",
"arch": "Cloud API",
"val": "~100",
"ternary": false,
"open": false
},
{
"company": "Lokale GPU",
"arch": "RTX 4090 (binär)",
"val": "100-300",
"company": "Claude Opus",
"arch": "Cloud API",
"val": "~80",
"ternary": false,
"open": false
},
{
"company": "OpenAI",
"arch": "Cloud API",
"val": "50-100",
"ternary": false,
"company": "B200 BitNet I2_S",
"arch": "B200 GPU",
"val": "52 (CPU)",
"ternary": true,
"open": false
}
],
Expand All @@ -1202,7 +1198,8 @@
"power": "293W",
"cost": "$0.20"
}
}
},
"unitNote": "End-to-end tokens/sec, from this project's own benchmark table. Cloud providers are faster; Trinity's advantage is cost — $0.01-0.35 against API pricing — and running without a GPU at all. The earlier version of this table put Trinity's raw kernel ops/sec beside everyone else's tokens/sec."
},
"hlsCompetition": {
"title": "Universeller <span class=\"grad\">FPGA-Compiler</span>",
Expand Down Expand Up @@ -1559,7 +1556,7 @@
"whyNow": [
"Ternäre KI ist bewiesen und nativ realisiert",
"Wir sind die einzigen mit nativem Hardware-Stack",
"578x Vorteil — kompetitives Endspiel",
"Bis zu 578x Vorteil — kompetitives Endspiel (projiziert, FPGA)",
"Funktionierendes Ökosystem von 20.645 Spezifikationen",
"Klarer Weg zur KI-Marktdominanz"
],
Expand Down Expand Up @@ -1603,7 +1600,7 @@
"label": "Gründeranteil"
},
{
"value": "578×",
"value": "578× *",
"label": "Energieeffizienz"
},
{
Expand All @@ -1622,7 +1619,7 @@
"why": [
"Absolute Triade erreicht",
"Binäre Ära offiziell beendet",
"578x Energievorteil",
"Bis zu 578x Energievorteil (projiziert, FPGA)",
"20.645 native TRI-Dateien",
"99% Gründeranteil = Alignment"
],
Expand Down Expand Up @@ -1718,7 +1715,8 @@
"email": "admin@t27.ai",
"website": "https://trinity-site-one.vercel.app",
"github": "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/gHashTag/trinity"
}
},
"projectedNote": "* Projiziert auf kundenspezifischer FPGA-Hardware. Auf CPU verifiziert: 20× Speicherkompression."
},
"bitnetProof": {
"badge": "MATHEMATISCHE BEWEISE",
Expand Down
Loading
Loading