← back to Macstudio Specs
public/index.html
434 lines
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Stack — Mac Studio 1 + 2, plus expansion candidates</title>
<style>
:root {
--bg: #0b0d10;
--panel: #14181d;
--panel-2: #1a1f26;
--line: #232a33;
--ink: #e8ecef;
--ink-dim: #9aa4ad;
--accent: #7cd0ff;
--warn: #ffb454;
--good: #74d99f;
--pick: #ffd56a;
}
* { box-sizing: border-box }
html, body { margin:0; padding:0; background:var(--bg); color:var(--ink); font:15px/1.5 -apple-system, BlinkMacSystemFont, "SF Pro Text", system-ui, sans-serif; }
header { padding: 40px 32px 24px; border-bottom: 1px solid var(--line); display:flex; align-items:baseline; justify-content:space-between; flex-wrap:wrap; gap:12px; }
header h1 { margin:0; font-size: 28px; font-weight: 600; letter-spacing:-0.02em; }
header .sub { color: var(--ink-dim); font-size:13px; }
main { max-width: 1200px; margin: 0 auto; padding: 28px 32px 80px; }
section { margin-top: 36px; }
section > h2.section-title { font-size: 14px; text-transform: uppercase; letter-spacing: 0.12em; color: var(--ink-dim); border-bottom: 1px solid var(--line); padding-bottom: 8px; margin-bottom: 16px; }
.grid { display:grid; grid-template-columns: repeat(auto-fit, minmax(280px, 1fr)); gap:16px; margin-top:8px; }
.card { background: var(--panel); border:1px solid var(--line); border-radius:10px; padding:18px 20px; }
.card h2 { margin:0 0 14px; font-size:13px; text-transform:uppercase; letter-spacing:0.08em; color:var(--ink-dim); font-weight:600; }
.card .who { font-size:18px; font-weight:600; margin-bottom: 12px; }
.card .who .who-sub { font-size:12px; color:var(--ink-dim); font-weight:400; display:block; margin-top:2px; }
.row { display:flex; justify-content:space-between; padding:6px 0; border-bottom:1px dashed var(--line); gap:12px; }
.row:last-child { border-bottom:none; }
.row .k { color: var(--ink-dim); flex-shrink:0; }
.row .v { color: var(--ink); font-variant-numeric: tabular-nums; text-align:right; }
.v.mono { font-family: "SF Mono", ui-monospace, Menlo, monospace; font-size:13px; }
.pill { display:inline-block; padding:2px 8px; border-radius:999px; font-size:11px; background:var(--panel-2); border:1px solid var(--line); }
.pill.warn { color: var(--warn); border-color: #4a3618; background:#2a200f; }
.pill.good { color: var(--good); border-color: #1f3a2a; background:#0f2218; }
.pill.pick { color: var(--pick); border-color: #4a3d1a; background:#2a230f; font-weight:600; }
.bar { background:var(--panel-2); border-radius:6px; height:8px; overflow:hidden; margin-top:6px; border:1px solid var(--line); }
.bar > div { height:100%; background: linear-gradient(90deg, #ff7373, #ffb454); }
.bar.ok > div { background: linear-gradient(90deg, #4fa3ff, #74d99f); }
footer { text-align:center; color:var(--ink-dim); font-size:12px; padding:24px; }
.hero { display:flex; gap:24px; align-items:center; }
.hero .badge { font-size:48px; }
table { width:100%; border-collapse: collapse; background: var(--panel); border:1px solid var(--line); border-radius:10px; overflow:hidden; font-size:13px; }
th, td { padding: 10px 12px; text-align:left; border-bottom:1px solid var(--line); }
th { background: var(--panel-2); color:var(--ink-dim); font-size:11px; text-transform:uppercase; letter-spacing:0.08em; font-weight:600; }
td.num { font-variant-numeric:tabular-nums; text-align:right; }
tr.picked { background: rgba(255, 213, 106, 0.06); }
tr.picked td:first-child { border-left: 3px solid var(--pick); }
.legend { display:flex; gap:14px; flex-wrap:wrap; font-size:12px; color:var(--ink-dim); margin-bottom:14px; }
.rec { background: var(--panel); border:1px solid var(--line); border-left: 3px solid var(--pick); border-radius:10px; padding:18px 22px; margin-top: 14px; }
.rec h3 { margin:0 0 6px; font-size:16px; }
.rec h3 .num { color: var(--pick); margin-right:8px; font-weight:700; }
.rec p { margin: 0 0 0; color: var(--ink-dim); font-size:13px; line-height:1.6; }
.rec p strong { color: var(--ink); }
.src { font-size:11px; color: var(--ink-dim); }
.src a { color: var(--accent); text-decoration: none; }
.src a:hover { text-decoration: underline; }
</style>
<link rel="icon" type="image/svg+xml" href="/favicon.svg">
</head>
<body>
<header>
<div class="hero">
<div class="badge">🖥️</div>
<div>
<h1>Stack snapshot + expansion picks</h1>
<div class="sub">Mac1 + Mac2 specs · candidate machine for stack addition · May 2026 prices</div>
</div>
</div>
<div class="sub">snapshot 2026-05-12 12:00 PT</div>
</header>
<main>
<section>
<h2 class="section-title">Current stack — Mac Studios 1 + 2</h2>
<div class="grid" style="grid-template-columns: repeat(auto-fit, minmax(320px, 1fr));">
<div class="card">
<div class="who">Mac Studio 1 <span class="pill">mac1</span><div class="who-sub">M1 Max · 2022 · 32 GB · 27d uptime</div></div>
<div class="row"><span class="k">Chip</span><span class="v">Apple M1 Max</span></div>
<div class="row"><span class="k">CPU / GPU cores</span><span class="v">10 (8P+2E) / 24</span></div>
<div class="row"><span class="k">RAM</span><span class="v">32 GB LPDDR5</span></div>
<div class="row"><span class="k">Internal SSD</span><span class="v">512 GB · <span class="pill warn">89% full</span></span></div>
<div class="row"><span class="k">External</span><span class="v">2 TB Henry (USB)</span></div>
<div class="row"><span class="k">macOS</span><span class="v">15.3.2</span></div>
<div class="row"><span class="k">Role</span><span class="v">Ollama models + scrapers</span></div>
<div class="row"><span class="k">Serial</span><span class="v mono">G2H309G5WK</span></div>
</div>
<div class="card">
<div class="who">Mac Studio 2 <span class="pill">mac2 · this box</span><div class="who-sub">M2 Max · 2023 · 32 GB · 3.6d uptime</div></div>
<div class="row"><span class="k">Chip</span><span class="v">Apple M2 Max</span></div>
<div class="row"><span class="k">CPU / GPU cores</span><span class="v">12 (8P+4E) / 30</span></div>
<div class="row"><span class="k">RAM</span><span class="v">32 GB unified</span></div>
<div class="row"><span class="k">Internal SSD</span><span class="v">512 GB · <span class="pill warn">94% full</span></span></div>
<div class="row"><span class="k">Display</span><span class="v">LG UltraWide 6400×2700</span></div>
<div class="row"><span class="k">macOS</span><span class="v">26.3.1 (25D2128)</span></div>
<div class="row"><span class="k">Role</span><span class="v">primary dev + Claude Code</span></div>
<div class="row"><span class="k">Load avg</span><span class="v">5.41 / 6.41 / 6.56</span></div>
<div class="row"><span class="k">Serial</span><span class="v mono">GG2DJP67KR</span></div>
</div>
</div>
<div class="src" style="margin-top:10px">⚠️ Both internal SSDs are critically full (89% / 94%). New box would relieve pressure regardless of which you pick.</div>
</section>
<section>
<h2 class="section-title">$5 K budget — locked top picks</h2>
<div class="rec">
<h3><span class="num">①</span> 2× Techable M1 Ultra 128 GB / 2 TB → $4,998 <span class="pill pick">THE PICK</span></h3>
<p><strong>The right answer for scale-out architecture.</strong> 256 GB aggregate RAM, 1,600 GB/s aggregate memory bandwidth, true parallelism across 2 new nodes. Combined with Mac1 + Mac2 that's <strong>320 GB across 4 nodes</strong>. Each new node hosts its own Ollama instance — no time-slicing, failure isolation, drop-in to the existing tailnet. 2-year reseller warranty on each unit.</p>
</div>
<div class="rec" style="border-left-color: var(--accent)">
<h3 style="color: var(--accent)"><span class="num">②</span> 1× M1 Ultra 128 GB at Techable → $2,499 <span class="pill">MINIMUM VIABLE</span></h3>
<p><strong>Half the spend, pocket $2,500.</strong> 128 GB on a single new node — same per-node spec as Pick ①. See how it slots in for a month, buy the second unit later if it earns its keep. Lowest-risk path into the upgrade.</p>
</div>
<div class="rec" style="border-left-color: var(--good)">
<h3 style="color: var(--good)"><span class="num">③</span> 1× M2 Ultra 192 GB (Swappa, used) → ~$4,939 <span class="pill">SINGLE-BOX</span></h3>
<p><strong>If you'd rather have one bigger box than two matched ones.</strong> 192 GB unified is the largest single-machine pool at this budget. M2 Ultra is one generation newer than M1 Ultra. Loses to Pick ① on aggregate bandwidth, true parallelism, and failure isolation — but wins if you specifically want a single 192 GB model to fit on one node.</p>
</div>
<div class="legend" style="margin-top: 18px">
<span><strong>Skip-list at this budget:</strong></span>
<span>· DGX Spark (lower per-stream BW than Mac, CUDA software bet only)</span>
<span>· RTX 5090 single-card build (only 32 GB VRAM)</span>
<span>· M4 Max refurb 128 GB at $3,139 (M1 Ultra 128 GB is $640 cheaper for same RAM)</span>
</div>
</section>
<section>
<h2 class="section-title">LA-local marketplace — outbound search links</h2>
<div class="legend">
<span><strong>Inspect-in-person tier.</strong> No shipping risk, no scams, see the box before paying.</span>
</div>
<div class="grid" style="grid-template-columns: repeat(auto-fit, minmax(300px, 1fr));">
<div class="card" style="border-left: 3px solid var(--pick);">
<h2>Craigslist LA — known finds</h2>
<div class="row"><span class="k">M1 Ultra 128 GB / 8 TB <span class="pill pick">deal</span></span><span class="v">$2,800 · Culver City</span></div>
<div class="row"><span class="k">M1 Ultra 64 GB / 2 TB</span><span class="v">$1,999 · LA</span></div>
<div class="row"><span class="k">M1 Max + monitors bundle</span><span class="v">$2,000 · West LA</span></div>
<div class="row"><span class="k">M4 Max 64 GB / 1 TB</span><span class="v">$2,650 · WeHo</span></div>
<div class="row" style="border-bottom:none; padding-top:10px;">
<a href="https://losangeles.craigslist.org/search/sya?query=mac+studio+ultra" style="color:var(--accent); font-size:13px;">→ Search Craigslist LA: mac studio ultra</a>
</div>
</div>
<div class="card" style="border-left: 3px solid var(--accent);">
<h2>Facebook Marketplace LA</h2>
<div class="row"><span class="k">Requires Facebook login</span><span class="v">listings ≠ scrapeable</span></div>
<div class="row" style="border-bottom:none;">
<span style="font-size:13px; color:var(--ink-dim); line-height:1.7;">
<a href="https://www.facebook.com/marketplace/la/search?query=mac%20studio%20128gb" style="color:var(--accent); display:block;">→ Search: mac studio 128gb</a>
<a href="https://www.facebook.com/marketplace/la/search?query=mac%20studio%20m1%20ultra" style="color:var(--accent); display:block;">→ Search: mac studio m1 ultra</a>
<a href="https://www.facebook.com/marketplace/la/search?query=mac%20studio%20m2%20ultra" style="color:var(--accent); display:block;">→ Search: mac studio m2 ultra</a>
<a href="https://www.facebook.com/marketplace/la/search?query=mac%20studio%20ultra" style="color:var(--accent); display:block;">→ Search: mac studio ultra (broadest)</a>
</span>
</div>
</div>
<div class="card">
<h2>Other LA-local</h2>
<div class="row"><span class="k">MacEnthusiasts (Pico Blvd)</span><span class="v">M1 Ultra · "call"</span></div>
<div class="row"><span class="k">OfferUp LA</span><span class="v">login required</span></div>
<div class="row" style="border-bottom:none; padding-top:10px;">
<a href="https://shop.macenthusiasts.com" style="color:var(--accent); font-size:13px; display:block;">→ MacEnthusiasts · (310) 287-2777</a>
<a href="https://offerup.com/explore/sck/ca/los_angeles/macbooks" style="color:var(--accent); font-size:13px; display:block;">→ OfferUp LA macbooks</a>
</div>
</div>
</div>
<div class="rec" style="margin-top: 18px;">
<h3><span class="num">⭐</span> The LA-local standout — Craigslist Culver City M1 Ultra 128 GB / 8 TB / $2,800</h3>
<p><strong>$301 more than the Techable refurb buys 4× the storage (8 TB vs 2 TB) and zero shipping risk.</strong> No reseller warranty — but you boot it, run a memory test, and check the serial against <a href="https://checkcoverage.apple.com" style="color:var(--accent)">checkcoverage.apple.com</a> on the spot before paying. The 8 TB SSD alone is a ~$400 value if you'd otherwise need external storage. Hard rule for any private-seller pickup — confirm the seller signs out of iCloud in front of you (avoid Activation Lock).</p>
</div>
<div class="src" style="margin-top:14px; line-height:1.7;">
<strong>Inspect-before-paying checklist:</strong>
boot → About This Mac (verify chip + RAM + storage) →
Terminal: <code>memory_pressure</code> (confirm no swap) →
serial in <a href="https://checkcoverage.apple.com" style="color:var(--accent)">checkcoverage.apple.com</a> →
System Settings → Apple Account → Sign Out (seller does this) →
erase + reinstall macOS (you do this) → pay.
</div>
</section>
<section>
<h2 class="section-title">$2 K tier — sub-$2,500 finds (May 2026, verified)</h2>
<div class="legend">
<span><span class="pill pick">PICK</span> standout at this budget</span>
<span>resellers checked: Swappa, Techable, OWC, iPowerResale, Apple Refurb, Amazon</span>
</div>
<table>
<thead>
<tr>
<th>Listing</th>
<th>Chip</th>
<th class="num">RAM</th>
<th>Storage</th>
<th class="num">Price</th>
<th class="num">$/GB</th>
<th>Source</th>
</tr>
</thead>
<tbody>
<tr class="picked">
<td><strong>Refurb Mac Studio (2022)</strong> <span class="pill pick">PICK</span></td>
<td>M1 Ultra · 20-core CPU · 48-core GPU</td>
<td class="num">128 GB</td>
<td>2 TB SSD</td>
<td class="num">$2,499</td>
<td class="num">$19.5</td>
<td><a href="https://techable.com/collections/refurbished-mac-studio" style="color:var(--accent)">Techable</a> · 2yr warranty</td>
</tr>
<tr>
<td>Refurb Mac Studio (2022)</td>
<td>M1 Max · 10-core CPU · 32-core GPU</td>
<td class="num">64 GB</td>
<td>2 TB SSD</td>
<td class="num">$1,799–1,999</td>
<td class="num">$28–31</td>
<td>Techable · "Excellent" · 2yr warranty</td>
</tr>
<tr>
<td>Refurb Mac Studio (2023)</td>
<td>M2 Max · 12-core CPU · 30-core GPU</td>
<td class="num">64 GB</td>
<td>1 TB SSD</td>
<td class="num">$2,099</td>
<td class="num">$33</td>
<td>Techable · "Excellent" · 2yr warranty</td>
</tr>
<tr>
<td>Used Mac Studio (2023) · mint</td>
<td>M2 Max</td>
<td class="num">64 GB</td>
<td>4 TB SSD</td>
<td class="num">$2,457</td>
<td class="num">$38</td>
<td><a href="https://swappa.com/listing/view/LAEZ07894" style="color:var(--accent)">Swappa LAEZ07894</a> · Latham NY</td>
</tr>
<tr>
<td>Used Mac Studio (2022)</td>
<td>M1 Max</td>
<td class="num">64 GB</td>
<td>2 TB SSD</td>
<td class="num">$2,472</td>
<td class="num">$39</td>
<td><a href="https://swappa.com/listing/view/LAET47467" style="color:var(--accent)">Swappa LAET47467</a></td>
</tr>
<tr>
<td>Mac mini (new)</td>
<td>M4 Pro · 14-core CPU · 20-core GPU</td>
<td class="num">64 GB</td>
<td>1 TB SSD</td>
<td class="num">$2,199</td>
<td class="num">$34</td>
<td>Apple direct · ⚠️ "currently unavailable" May 2026</td>
</tr>
<tr>
<td>Single RTX 3090 build</td>
<td>RTX 3090 · 24 GB GDDR6X</td>
<td class="num">24 GB VRAM</td>
<td>n/a</td>
<td class="num">~$2,000</td>
<td class="num">$83</td>
<td>eBay used GPU + host build</td>
</tr>
</tbody>
</table>
<div class="rec" style="margin-top: 18px;">
<h3><span class="num">⭐</span> The $2,499 winner — Techable M1 Ultra 128 GB / 2 TB</h3>
<p><strong>Same 128 GB unified as the M4 Max refurb at $3,139 — but $640 cheaper.</strong> You give up some per-token speed (M1 Ultra ~8–11 tok/s vs M4 Max ~14–18 tok/s on 70B Q4), but for your workload — many small/medium models loaded concurrently for adversarial debate panels — the RAM matters far more than peak single-stream speed. 2-year reseller warranty. 20-core CPU + 48-core GPU. <strong>This is the box.</strong></p>
</div>
<div class="src" style="margin-top:10px">
Note: Apple's own refurb store currently has <em>no</em> Mac Studios listed; Apple is funneling them through authorized resellers (Techable, OWC, iPowerResale). Verified by direct fetch May 12 2026.
</div>
</section>
<section>
<h2 class="section-title">$3 K+ tier — for reference</h2>
<div class="legend">
<span><span class="pill pick">PICK</span> top recommendation</span>
<span><span class="pill good">value</span> best $/GB</span>
<span><span class="pill warn">caveat</span> needs care</span>
</div>
<table>
<thead>
<tr>
<th>Candidate</th>
<th>Memory</th>
<th class="num">Price (USD)</th>
<th class="num">$/GB</th>
<th>70B Q4 tok/s</th>
<th>Where</th>
<th>Fit notes</th>
</tr>
</thead>
<tbody>
<tr class="picked">
<td><strong>Mac Studio M4 Max</strong> <span class="pill pick">PICK</span></td>
<td class="num">128 GB unified</td>
<td class="num">$3,139</td>
<td class="num">$24.5</td>
<td>~14–18</td>
<td>Apple refurb</td>
<td>Drops into Ollama instantly · 1yr warranty · 4× capacity of Mac1/Mac2</td>
</tr>
<tr>
<td>NVIDIA DGX Spark</td>
<td class="num">128 GB unified</td>
<td class="num">$3,999–4,699</td>
<td class="num">$31</td>
<td>~? (new arch)</td>
<td>NVIDIA Marketplace / Micro Center</td>
<td>CUDA stack (vLLM, ExLlamaV2) · ARM Linux quirks · brand new</td>
</tr>
<tr>
<td>Asus Ascent GX10 (DGX Spark variant)</td>
<td class="num">128 GB unified, 1 TB SSD</td>
<td class="num">$2,999</td>
<td class="num">$23</td>
<td>~? (same as Spark)</td>
<td>Asus / Micro Center</td>
<td>Cheapest path to Blackwell + 128 GB unified · same Spark silicon</td>
</tr>
<tr>
<td>Mac Studio M2 Ultra</td>
<td class="num">192 GB unified</td>
<td class="num">$4,939–5,293</td>
<td class="num">$27</td>
<td>~10–14</td>
<td>Swappa (used)</td>
<td>Biggest unified pool · runs 70B at FP16 · older chip than M4 Max</td>
</tr>
<tr>
<td>RTX 5090 (build cost)</td>
<td class="num">32 GB VRAM</td>
<td class="num">~$3,658 GPU + ~$1,200 host = ~$4,858</td>
<td class="num">$152</td>
<td>~25–35</td>
<td>eBay / scalper market</td>
<td><span class="pill warn">caveat</span> low VRAM for parallel debate · still hard to find at MSRP</td>
</tr>
<tr>
<td>RTX 4090 used (build cost)</td>
<td class="num">24 GB VRAM</td>
<td class="num">~$2,375 GPU + ~$1,200 host = ~$3,575</td>
<td class="num">$149</td>
<td>~20–30</td>
<td>eBay used</td>
<td>Same trap — only 24 GB; doesn't fit 70B Q4 cleanly</td>
</tr>
<tr>
<td>Dual RTX 3090 build (NVLink)</td>
<td class="num">48 GB VRAM pooled</td>
<td class="num">~$1,800 GPUs + ~$1,200 host = ~$3,000</td>
<td class="num">$63</td>
<td>~18–28</td>
<td>eBay used + custom build</td>
<td><span class="pill good">value</span> Best $/GB CUDA · <span class="pill warn">caveat</span> 700 W heat · water cooling · noise</td>
</tr>
<tr>
<td>Single RTX 3090 build</td>
<td class="num">24 GB VRAM</td>
<td class="num">~$900 GPU + ~$1,100 host = ~$2,000</td>
<td class="num">$83</td>
<td>~15–22</td>
<td>eBay used</td>
<td>Cheapest CUDA entry — but doesn't fit 70B Q4</td>
</tr>
</tbody>
</table>
</section>
<section>
<h2 class="section-title">Recommendation</h2>
<div class="rec">
<h3><span class="num">①</span> Refurbished Mac Studio M4 Max, 128 GB — $3,139</h3>
<p><strong>Best fit, best $/GB unified, lowest friction.</strong> Drops straight into the existing Ollama stack — same OS, same tooling, same fleet conventions. 128 GB unified is <strong>4× per-machine capacity</strong> of your current M1/M2 Max boxes, which means qwen3:72b, deepseek-r1:70b, and Kimi-K2-Q4 fit in RAM instead of swapping. Frees Mac1 to specialize in smaller fast models. 1-year Apple warranty.</p>
</div>
<div class="rec" style="border-left-color: var(--accent)">
<h3 style="color: var(--accent)"><span class="num">②</span> Asus Ascent GX10 (DGX Spark variant) — $2,999</h3>
<p><strong>Strategic CUDA bet.</strong> Same Blackwell + 128 GB unified silicon as DGX Spark at $1 K less. Unlocks the CUDA-only side of the stack (vLLM, ExLlamaV2, exact tensor parallelism, NVIDIA NIM containers, TensorRT-LLM). Diversifies you off pure Apple Silicon for the debate panels. <strong>Caveat:</strong> ARM Linux ecosystem is still settling; expect some Ollama-on-DGX-Spark teething pain.</p>
</div>
<div class="rec" style="border-left-color: var(--good)">
<h3 style="color: var(--good)"><span class="num">③</span> Dual-3090 NVLink build — ~$3,000 total</h3>
<p><strong>If you want maximum CUDA VRAM for the dollar.</strong> 48 GB pooled, runs Llama-3.3-70B Q4_K_M with full context. Requires water cooling, ~1000 W PSU, custom case. Best raw CUDA value but highest operational tax — heat, noise, build time. Worth it only if you specifically want NVLink experiments.</p>
</div>
<div class="src" style="margin-top:18px; line-height:1.8;">
Sources (May 2026 verified):
<a href="https://www.apple.com/shop/refurbished/mac/mac-studio">Apple Refurb Mac Studio</a> ·
<a href="https://swappa.com/listings/mac-studio-2023">Swappa M2 Ultra listings</a> ·
<a href="https://www.nvidia.com/en-us/products/workstations/dgx-spark/">NVIDIA DGX Spark</a> ·
<a href="https://www.microcenter.com/product/699008/nvidia-dgx-spark">Micro Center DGX Spark</a> ·
<a href="https://bestvaluegpu.com/history/new-and-used-rtx-5090-price-history-and-specs/">RTX 5090 price tracker</a> ·
<a href="https://bestvaluegpu.com/history/new-and-used-rtx-4090-price-history-and-specs/">RTX 4090 price tracker</a> ·
<a href="https://bestvaluegpu.com/history/new-and-used-rtx-3090-price-history-and-specs/">RTX 3090 price tracker</a> ·
<a href="https://www.xda-developers.com/used-rtx-3090-still-best-for-local-ai-in-value/">XDA — 3090 still best for local AI</a>
</div>
</section>
</main>
<footer>Live at :9779 · pm2 process <code>macstudio-specs</code> · regenerate with <code>pm2 restart macstudio-specs</code></footer>
</body>
</html>