Add AdSense to Jekyll Blog

Add AdSense to Jekyll Blog | Generated by AI

Home 2026.07

Question: How do I add Google AdSense to my Jekyll blog (lzwjava/jekyll-ai-blog) to monetize it?

Answer:

Your site is static Jekyll on GitHub Pages, so AdSense integration is just a script tag + ads.txt. No SDK, no server. Here’s the exact diff based on your repo’s actual structure.

1. Apply for AdSense first — you need approval before ads render. Go to ads.google.com, add lzwjava.github.io as a site, wait for review (can take days-weeks). You’ll get a publisher ID like pub-1234567890123456.

2. Add ads.txt at repo root (required by Google, or you lose revenue to “unauthorized sellers” warnings):

cat > /path/to/repo/ads.txt << 'EOF'
google.com, pub-<PLACEHOLDER_PUB_ID>, DIRECT, f08c47fec0942fa0
EOF

3. Add Auto Ads script to <head> in _layouts/default.html — insert right after your existing GoatCounter script, before </head>:

   <script data-goatcounter="https://lzwjava.goatcounter.com/count"
         async src="//gc.zgo.at/count.js"></script>
+
+  <!-- Google AdSense -->
+  <script async src="https://pagead2.googlesyndication.com/pagead/js/adsbygoogle.js?client=ca-pub-<PLACEHOLDER_PUB_ID>"
+     crossorigin="anonymous"></script>
 </head>

That alone enables Auto Ads (Google decides placement). If you want more control and better revenue, add manual in-article units.

4. Manual in-article ad unit — in _layouts/post.html, drop a unit right before the closing <hr> after `<main id="content" class="main-content post-content" role="main">

85B Model on 4×P40 | Generated by AI

Home 2026.07

Question: Can 4×Tesla P40 (24GB ×4 = 96GB VRAM) directly run an 85B parameter model?

Answer:

Yes, technically you can run an 85B model on 4×P40, but not in FP16. You need quantization, and performance will be limited.

Let’s calculate.

VRAM requirement

An 85B parameter model:

FP16

85B params × 2 bytes
= 170GB

Need:

170GB + KV cache + runtime overhead
≈ 180GB+

❌ 96GB cannot fit.

INT8

85B × 1 byte
≈ 85GB

Add overhead:

85GB + KV cache
≈ 90-100GB

Borderline.

4×P40:

24GB × 4 = 96GB

Maybe possible with aggressive memory management, but tight.

4-bit quantization (realistic)

85B × 0.5 byte
≈ 42.5GB

Add:

+ model metadata
+ KV cache
+ runtime

Usually:

50-60GB

So:

4×P40 96GB

has plenty of room.

Examples:

Llama 3.1 70B Q4
Qwen 72B Q4
DeepSeek 67B Q4

are comfortable.

85B Q4 should also fit.

But speed is the issue

P40 has no Tensor Cores.

Your compute:

1 × P40
≈ 12 TFLOPS FP32

4 × P40
≈ 48 TFLOPS FP32

Compare:

RTX 4090
≈ 83 TFLOPS FP32

But LLM inference is not only FLOPS; memory bandwidth matters.

P40:

346 GB/s × 4
≈ 1.4 TB/s

A 4090:

1 TB/s

So multi-P40 has decent bandwidth.

Multi-GPU inference

You need tensor parallelism:

85B model

        model layers
             |
   ---------------------
   |    |    |    |
 P40  P40  P40  P40

24G  24G  24G  24G

Software:

llama.cpp

Example:

./llama-server \
  -m qwen-85b-q4.gguf \
  -ngl 999 \
  --tensor-split 1,1,1,1

vLLM

CUDA_VISIBLE_DEVICES=0,1,2,3 \
vllm serve model \
--tensor-parallel-size 4

Expected performance

Rough estimate:

Hardware	70B Q4 tok/s
RTX 4090	~20-35 tok/s
2×3090	~30-50 tok/s
4×P40	~5-15 tok/s
CPU only	<5 tok/s

So 4×P40 is a capacity machine, not a speed machine.

For a home AI server:

RTX 4070
    ↓
training / coding / experiments

4×P40 96GB
    ↓
always-on LLM inference server
    ↓
70B-85B quantized models

is actually a reasonable architecture.

The sweet spot is probably 2×P40 or 4×P40 + llama.cpp + GGUF, because the biggest limitation is not VRAM — it is the lack of Tensor Cores and older CUDA architecture.

Back Donate

</main>

` (post-content boundary, decent CTR spot):

   {{ content }}

+  <div class="ad-container" style="margin: 24px 0; text-align:center;">
+    <ins class="adsbygoogle"
+         style="display:block"
+         data-ad-client="ca-pub-<PLACEHOLDER_PUB_ID>"
+         data-ad-slot="<PLACEHOLDER_SLOT_ID>"
+         data-ad-format="fluid"
+         data-ad-layout-key="-gw-3+1f-3d+2z"></ins>
+    <script>(adsbygoogle = window.adsbygoogle || []).push({});</script>
+  </div>
+
   <hr>

Get data-ad-slot from AdSense dashboard → Ads → By ad unit → In-article ad → create one.

5. Deploy — your existing GH Actions workflow (gh-pages.yml) handles the build/deploy, no changes needed there. Just commit and push.

Notes specific to your setup:

google_analytics: "UA-66656236-1" in _config.yml is Universal Analytics — that’s been sunset since 2023 and does nothing now. Worth swapping for GA4 or dropping it since you already have GoatCounter for real analytics.
Since your blog is bilingual (-en/-zh) and AI-translated content, be aware AdSense’s policy on “auto-generated content” — pages flagged as page.generated: true and page.translated: true in your frontmatter could get flagged in a manual review if there’s too much unedited AI output. Not a hard blocker but worth knowing.
At current traffic (2 stars/1 watcher on the repo isn’t a traffic signal, but if your notes pages — 1992 of them — get real search traffic, that long-tail content is actually where AdSense earns most, since it’s indexed niche technical content).

Alternative worth considering given your AI-agent background: skip display ads entirely and monetize via a sponsor/affiliate block (e.g. tools you build — ww, iclaw, zz — or GPU/consulting CTA) since your traffic profile (engineers reading technical notes) converts far better to consulting leads than to $0.50 CPM display ads.

References:

Back Donate