local-frontier / app.html
osolmaz's picture
osolmaz HF Staff
Fix MoE expert traffic, surface search limits, retire dennard naming (#2)
5ed97d7
Raw
History Blame Contribute Delete
10.3 kB
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<title>Local Frontier</title>
<meta
name="description"
content="Interactive local AI hardware and model-bound database."
/>
<link rel="stylesheet" href="/assets/app.css" />
</head>
<body data-page="index">
<div class="site-shell">
<header class="masthead">
<nav class="top-nav" aria-label="Primary">
<a class="wordmark" href="/">Local Frontier</a>
<div class="nav-links">
<a href="#frontier">Charts</a>
<a href="#database">Database</a>
<a href="/accelerators">Accelerators</a>
<a href="/systems">Systems</a>
<a href="/models">Models</a>
<a href="/bounds">Bounds</a>
<a href="#method">Method</a>
</div>
</nav>
<div class="masthead-grid">
<div>
<h1>Local Hardware and Model Bounds Database</h1>
</div>
<div class="lede">
<p>
This is the interactive version of the local hardware writeup. It
tracks Apple, NVIDIA, and AMD systems from 2020-2026 by
complete-machine price band, memory capacity, bandwidth, and
complete-system USD per GB.
</p>
<p>
The local copy adds model metadata and theoretical memory-side
decode bounds: single-session, KV-aware aggregate, and loose
memory-power ceilings.
</p>
</div>
</div>
<div
class="stat-strip"
id="statStrip"
aria-label="Dataset summary"
></div>
</header>
<main>
<section
class="frontier-panel"
id="frontier"
aria-labelledby="frontier-title"
>
<div class="panel-heading">
<div>
<h2 id="frontier-title">Chart Explorer</h2>
</div>
<p>
Switch metric and price band. Hover for source rows; click a point
to open its item page when the point comes from a specific device.
</p>
</div>
<div class="control-grid">
<label
>Metric
<select id="metricSelect">
<option value="annual_memory">
Annual memory by accelerator vendor
</option>
<option value="annual_price_per_gb">
Annual complete-system USD/GB
</option>
<option value="annual_bandwidth">
Annual memory bandwidth
</option>
<option value="best_memory">Best-so-far memory</option>
<option value="best_price_per_gb">
Best-so-far complete-system USD/GB
</option>
<option value="best_bandwidth">
Best-so-far memory bandwidth
</option>
<option value="projection">
Consumer memory projection to 2030
</option>
</select>
</label>
<label
>Price band
<select id="bandSelect"></select>
</label>
<div
class="producer-toggle"
id="producerToggle"
aria-label="Accelerator vendor toggles"
></div>
</div>
<div id="frontierChart" class="chart-host" aria-live="polite"></div>
</section>
<section
class="section-grid intro-grid"
aria-labelledby="what-this-tracks"
>
<div>
<h2 id="what-this-tracks">What This Tracks</h2>
</div>
<div class="prose">
<p>
The data tracks accelerator-accessible memory in one physical
machine, grouped by complete-machine price band. It covers Apple,
NVIDIA, and AMD from 2020 through 2026.
</p>
<p>
The core question is practical: what can someone buy or configure
locally, and how do affordability, capacity, bandwidth, and vendor
tradeoffs move over time?
</p>
</div>
</section>
<section class="symbol-band" aria-labelledby="symbols">
<div>
<h2 id="symbols">How To Read The Symbols</h2>
</div>
<dl class="symbol-list">
<div>
<dt>U</dt>
<dd>Unified or coherent memory available to the accelerator.</dd>
</div>
<div>
<dt>V</dt>
<dd>VRAM on one discrete GPU.</dd>
</div>
<div>
<dt>&Sigma;</dt>
<dd>
Aggregate installed VRAM across multiple GPUs in one chassis.
</dd>
</div>
<div>
<dt>*</dt>
<dd>
Price band inferred from a documented component bill of
materials.
</dd>
</div>
<div>
<dt>&asymp;</dt>
<dd>
Estimated, reconstructed, or bill-of-materials inferred price.
</dd>
</div>
</dl>
<p class="note">
Aggregate VRAM is useful, but it is not the same thing as one
unified memory pool. Treat aggregate points as chassis capacity, not
a guarantee that one process can use it as one contiguous model
memory space.
</p>
</section>
<section class="section-grid" aria-labelledby="main-read">
<div>
<h2 id="main-read">Main Read</h2>
</div>
<div class="prose multi-column">
<p>
Under $3,000, AMD changes the shape of the local market in 2025
and 2026 with 128GB unified memory through Framework Desktop-class
hardware. Apple is strong on usable unified memory, but the
current sub-$3,000 Apple ceiling in this dataset is 64GB. NVIDIA
stays at 24GB in this band because the data tracks complete
machines, not used cards or component-only builds.
</p>
<p>
In the $3,000-$5,000 band, both AMD and NVIDIA reach 128GB unified
or coherent memory by 2025-2026. This is the band that matters
most for consumer and prosumer local model work because it is
expensive but still within a serious personal hardware budget.
</p>
<p>
In the $5,000-$10,000 band, Apple has the most dramatic historical
point: the 512GB M3 Ultra Mac Studio launch configuration in 2025.
The best-so-far memory charts keep that 512GB point through 2026
because the historical record should not go down just because the
current Apple configuration changed.
</p>
<p>
Above $75,000, NVIDIA dominates the memory and bandwidth records.
DGX Station-class systems are local in the physical sense, but not
local in the affordability sense.
</p>
</div>
</section>
<section
class="database-panel"
id="database"
aria-labelledby="database-title"
>
<div class="panel-heading">
<div>
<h2 id="database-title">Local Hardware Database</h2>
</div>
<p>
A TechPowerUp-style catalog for complete local systems and
accelerator configurations, including Apple unified-memory
machines.
</p>
</div>
<div id="hardwareReactRoot"></div>
<div class="table-controls hardware-legacy">
<label
>Search
<input
id="deviceSearch"
type="search"
placeholder="Mac Studio, DGX, W7900, 128GB"
/>
</label>
<label
>Accelerator vendor
<select id="producerFilter">
<option value="all">All accelerator vendors</option>
</select>
</label>
<label
>Price band
<select id="tableBandFilter">
<option value="all">All bands</option>
</select>
</label>
</div>
<div id="deviceTable" class="table-wrap hardware-legacy"></div>
</section>
<section
class="section-grid"
id="method"
aria-labelledby="method-title"
>
<div>
<h2 id="method-title">Method</h2>
</div>
<div class="prose">
<p>
The editable source of truth is <code>data/items/*.json</code>.
The generator expands each item price history into CSVs, derives
annual and best-so-far records, and renders the Space from those
generated artifacts.
</p>
<p>
The complete-system price-per-memory metric is total configured
machine price divided by accelerator-accessible memory. The
extracted component memory-price file is separate and should not
be confused with complete-machine affordability.
</p>
<p>
The projection view fits simple least-squares linear trends to
2020-2026 best-so-far memory series. It is a scenario view, not
verified future product data.
</p>
</div>
</section>
</main>
</div>
<script src="/assets/local-frontier-data.js"></script>
<script src="/assets/local-frontier-model-data.js"></script>
<script src="/assets/local-frontier-profile-data.js"></script>
<script src="/assets/bounds-engine.js"></script>
<script src="/assets/app.js"></script>
<script type="module" src="/src/main.tsx"></script>
</body>
</html>