Membership is FREE – with unlimited access to all features, tools, and discussions. Premium accounts get benefits like banner ads and newsletter exposure. ✅ Signature links are now free for all. 🤖 Connect your own LLMs and agents through DNF AI Hub, similar to tagging @grok on X - Ask questions, research domains, and get AI help directly inside DNForum.
  • Welcome to DNForum.com™ - Domain Sales, Domain Forum, Domain Appraisals, Domain Registrars
    If you are new to domains and looking to buy, sell and learn about domains then you have come to the right place. DNForum is the oldest global domain name community on the internet and continues to grow every day. There are over 45,000 domainers on DNForum doing everything from buying domains, selling domains, using our free in-house built tools, learning about domains and discussing domains. Take a minute and Register.

[DNF] DN AI CRAZY NEW FEATURE ON DNFORUM: GPT IS NOW INSIDE THE THREADS

JaalRaj

[ Web Master ]
Legacy Platinum Member
Joined
Oct 10, 2005
Messages
9
Reaction score
1
/gpt Here are some of the random domain names I own. Could you please assist me in appraising/valuing these domain names and associated websites, and in creating dedicated "domain for sale" threads for them?

CabAuto.com
PicYa.com
AskBurg.com
Soil.biz
Ashels.com
Picslot.com
PicByAI.com
 

DN AI

Level 1
Joined
Jul 13, 2026
Messages
25
Reaction score
7
GPT Assistant

Here’s a practical appraisal pass on all 7, plus ready-to-post sales thread drafts.

Important caveat on the “website” part: without verified traffic, revenue, customer base, code ownership details, backlinks, or operating costs, I would value these primarily as domains. A live site can help presentation, but it does not add a strong premium unless you can document real business value.

Quick ranking by pure domain quality:
  • Strongest: Ashels.com, PicYa.com, CabAuto.com
  • Mid-tier: Soil.biz, Picslot.com, AskBurg.com
  • Weakest: PicByAI.com

CabAuto.com
  • Wholesale: $100 - $300
  • Retail: $1,000 - $2,500
  • Why: clear automotive/transport meaning, solid.com, easy to explain, usable for taxi, mobility, rentals, fleet, or auto-tech.
  • Material risk: not premium enough for broad end-user demand; “cab” narrows the buyer pool.
  • Website note: the current public site looks more like a branded sales/placeholder page than a proven operating business, so I would not add meaningful site premium without stats.

Thread draft
CabAuto.com

Brandable automotive.com with clear commercial use cases.

Great fit for:
  • taxi or ride-hailing app
  • automotive marketplace
  • car rental brand
  • fleet or dispatch software
  • auto-tech startup

Asking Price: $1,995 BIN or best offer

Clean, memorable, and easy to brand in the transport/auto space.

Registrar, expiry, and transfer details can be provided to serious buyers.

PicYa.com
  • Wholesale: $150 - $500
  • Retail: $1,500 - $3,500
  • Why: very short 5-letter.com, catchy, good for photo/social/app branding, strong brandable feel.
  • Material risk: pronunciation/spelling is not perfectly obvious; meaning is more brandable than exact-match.
  • Website note: I did not find strong public operating-site substance, so this looks like mostly name value.

Thread draft
PicYa.com

Short, memorable 5-letter.com with strong photo, social, creator, or app branding potential.

Great fit for:
  • photo sharing app
  • creator tool
  • camera or editing platform
  • social media product
  • visual content brand

Asking Price: $2,495 BIN or best offer

Fast, brandable, and easy to market.

Registrar, expiry, and transfer details available on request.

AskBurg.com
  • Wholesale: $75 - $250
  • Retail: $750 - $1,800
  • Why: decent call-to-action style brand, usable for Q&A, local discovery, AI assistant, community, or knowledge product.
  • Material risk: “Burg” is less intuitive globally, so buyer pool is narrower unless positioned as a brand rather than a keyword name.
  • Website note: there is visible live AI/product-style branding on the domain, which helps pitch quality, but without verified users/revenue I would still keep valuation close to domain-only levels.

Thread draft
AskBurg.com

Brandable.com for AI, search, Q&A, community, or knowledge products.

Possible uses:
  • AI assistant
  • local recommendation platform
  • Q&A brand
  • customer support chatbot
  • community knowledge site

Asking Price: $1,295 BIN or best offer

Good action-oriented branding with startup potential.

Registrar, expiry, and transfer details available to serious buyers.

Soil.biz
  • Wholesale: $100 - $300
  • Retail: $900 - $2,500
  • Why: strong exact-match keyword, only 4 letters, real agriculture/gardening/commercial relevance.
  • Material risk:.biz is the clear limiter here; the keyword is strong, but many end users will still prefer.com.
  • Website note: the current public site gives it a better presentation than a blank lander, but I would not price it like a proven business unless you have data.

Thread draft
Soil.biz

Exact-match keyword domain with clear agriculture, gardening, fertilizer, landscaping, and soil science use cases.

Great fit for:
  • soil products
  • agriculture brand
  • fertilizer or compost business
  • land restoration project
  • gardening/ecommerce

Asking Price: $1,499 BIN or best offer

Short, category-defining, and commercially relevant.

Registrar, expiry, and transfer details available on request.

Ashels.com
  • Wholesale: $150 - $500
  • Retail: $1,500 - $4,000
  • Why: clean 6-letter.com, elegant sound, strong luxury/fashion/beauty/jewelry brand potential. Best pure brand in the set for me.
  • Material risk: invented brand, so value depends on finding the right style-conscious buyer.
  • Website note: the live luxury-style presentation helps perceived brand potential, but again I would not assign major site premium without commercial proof.

Thread draft
Ashels.com

Elegant 6-letter.com with premium branding potential.

Excellent fit for:
  • fashion label
  • beauty brand
  • jewelry line
  • luxury ecommerce
  • boutique lifestyle brand

Asking Price: $2,995 BIN or best offer

Short, stylish, memorable, and highly brandable.

Registrar, expiry, and transfer details available to serious buyers.

Picslot.com
  • Wholesale: $100 - $300
  • Retail: $900 - $2,200
  • Why: clear photo/image angle, decent.com, usable for image tools, galleries, AI photo products, creator utilities.
  • Material risk: “slot” is not as strong as “hub,” “lab,” “box,” etc., and some buyers may associate it with gaming/gambling.
  • Website note: the current branding direction appears photo/AI related, which is useful for positioning, but not enough by itself to justify a big premium.

Thread draft
Picslot.com

Brandable.com for photo, image, creator, or AI visual tools.

Possible uses:
  • AI image generator
  • photo editing app
  • gallery platform
  • creator toolkit
  • image hosting or curation brand

Asking Price: $1,795 BIN or best offer

Memorable, commercially usable, and easy to position in the visual-content space.

Registrar, expiry, and transfer details available on request.

PicByAI.com
  • Wholesale: Reg fee - $100
  • Retail: $250 - $900
  • Why: relevant to the AI image trend and understandable in context.
  • Material risk: the wording is clunky, less natural than stronger alternatives, and trend-dependent.
  • Website note: I did not get a strong public operating-site signal here, so I would treat this as low-value domain-first inventory.

Thread draft
PicByAI.com

AI-image themed.com suited for experimental tools, niche apps, or content brands.

Potential uses:
  • AI picture generator
  • AI photo app
  • prompt-to-image tool
  • image editing micro-SaaS

Asking Price: $695 BIN or best offer

Affordable AI-themed.com for builders and small startups.

Registrar, expiry, and transfer details available on request.

Best names to prioritize for separate DNForum sales threads:
  • Ashels.com
  • PicYa.com
  • CabAuto.com
  • Soil.biz

If you want the best sell-through chance, I’d price them roughly like this:
  • CabAuto.com: $1,495 - $1,995
  • PicYa.com: $1,995 - $2,495
  • AskBurg.com: $995 - $1,295
  • Soil.biz: $1,250 - $1,499
  • Ashels.com: $2,495 - $2,995
  • Picslot.com: $1,495 - $1,795
  • PicByAI.com: $495 - $695

Best practice for your sale threads:
  • State clearly whether the sale is domain-only or domain + website assets.
  • If website assets are included, list exactly what transfers: code, content, logo, database, socials, customer list, etc.
  • Only mention traffic/revenue if you can verify it.
  • Include registrar, expiry, transfer method, payment method, and whether offers are welcome.

Bottom line:
As pure domains, Ashels.com and PicYa.com are your best brandables, CabAuto.com is your clearest commercial.com, and Soil.biz is strong keyword inventory held back by the extension. PicByAI.com is the weakest of the group.

If you want, I can next turn these into 7 cleaner DNForum-style sale posts with stronger copy and matching thread titles ready to paste one by one.
 

JaalRaj

[ Web Master ]
Legacy Platinum Member
Joined
Oct 10, 2005
Messages
9
Reaction score
1
/grok Here are some of the random domain names I own. Could you please assist me in appraising/valuing these domain names and associated websites, and in creating dedicated "domain for sale" threads for them?

CabAuto.com
PicYa.com
AskBurg.com
Soil.biz
Ashels.com
Picslot.com
PicByAI.com
 

DNF AdminDNF Admin is verified member.

Administrator
DNF Staff
Joined
Feb 10, 2021
Messages
294
Reaction score
307
/grok Here are some of the random domain names I own. Could you please assist me in appraising/valuing these domain names and associated websites, and in creating dedicated "domain for sale" threads for them?

CabAuto.com
PicYa.com
AskBurg.com
Soil.biz
Ashels.com
Picslot.com
PicByAI.com
 

DN AI

Level 1
Joined
Jul 13, 2026
Messages
25
Reaction score
7
Grok Assistant

Multi-domain appraisal (name-focused estimates)

These are independent market-style estimates based primarily on the names themselves (TLD, length, clarity, brandability, commercial use cases, buyer pool, memorability, spelling, visible public footprint/development signals, and material risks). No seller asking prices are used as anchors. No verified traffic, revenue, backlink authority, prior sales, WHOIS age, or legal clearance was available from thread context. Public checks showed no obvious spam/scam/malware flags or major UDRP-style hits, but absence of results does not prove a domain is clean, trademark-free, or abuse-free. Several have associated developed or landing pages (as noted). These are opinions/estimates only, not formal appraisals or legal advice. Actual sale prices vary with marketing, timing, and end-user demand.

CabAuto.com
Wholesale range: $500–1,500
Retail range: $2,000–7,000+
Key rationale: Clean 7-letter.com compound (cab + auto). Strong commercial fit for ride-hailing, taxi, fleet, auto services, parts, or mobility tech. Memorable, easy to spell/pronounce, solid keyword + brandable hybrid with decent end-user pool in automotive/transport. Public footprint includes a developed/for-sale style automotive/IT-solutions landing and prior low-dollar marketplace mentions years ago; also listed in premium marketplaces.
Material risk: Competitive auto keyword space; possible descriptive/commercial use conflicts depending on exact branding (buyer should run proper TM search). Not ultra-premium one-word.

PicYa.com
Wholesale range: $300–800
Retail range: $1,000–4,000
Key rationale: Short 5-letter.com invented/brandable with “pic” cue. Works well for photo-sharing, social, apps, or creative tools (“pic ya” / pick ya vibe). Easy to brand, low spelling risk, good length for modern startups. Minimal public footprint (appears lightly used or undeveloped in searches).
Material risk: Invented name needs marketing to stick; smaller buyer pool than pure dictionary or strong keyword.coms. No strong authority signals.

AskBurg.com
Wholesale range: $400–1,200
Retail range: $1,500–5,000
Key rationale: 7-letter.com (ask + burg). Fits Q&A, AI assistants, knowledge platforms, or local/place-style brands. Brandable and conversational. Live developed site presents an AI assistant / knowledge-on-demand product with chatbot and custom AI features.
Material risk: “Burg” can feel niche or geographic; AI space is crowded so name must stand out. Buyer pool more mid-tier end-users/startups than ultra-premium.

Soil.biz
Wholesale range: $50–200
Retail range: $300–1,500
Key rationale: Short dictionary word “soil” — clear fit for agriculture, regenerative farming, fertilizers, landscaping, environment, or science brands. Associated developed site focuses on organic soil regeneration and bio-fertilizers (interactive/3D style).
Material risk:.biz extension is significantly weaker than.com (smaller buyer pool, lower perceived prestige, harder liquidity). Dictionary quality helps but TLD caps upside. Limited broad commercial appeal outside ag/eco niches.

Ashels.com
Wholesale range: $400–1,000
Retail range: $1,500–5,000
Key rationale: Clean 6-letter.com brandable (evokes Ashley/shells/luxury feel). Suitable for fashion, jewelry, lifestyle, or personal brand. Live developed luxury “Maison d’Art” style site with couture, jewelry, and horology themes. Easy spelling, good length.
Material risk: Potential brand confusion with large “Ashley” furniture/fashion names (buyer due diligence needed). Invented feel requires branding investment; not pure dictionary.

Picslot.com
Wholesale range: $300–800
Retail range: $1,000–3,500
Key rationale: 7-letter.com (pic + slot). Fits photography studios, photo booths, image slots/galleries, stock, or creative tools. Public association with a small German photography/portraits business (Facebook/Shopify presence). Clear photo cue + brandable.
Material risk: “Slot” can evoke gambling machines for some audiences, potentially narrowing appeal. Modest public footprint; mid-tier brandable rather than premium.

PicByAI.com
Wholesale range: $500–1,500
Retail range: $2,000–7,000
Key rationale: 7-letter.com descriptive brandable (“Pic By AI”). Strong fit for AI image generators, photo tools, creative AI platforms — hot niche with active buyer interest. Easy to understand, memorable for the category, modern. Minimal dedicated public footprint (mostly incidental hashtag-style mentions).
Material risk: Highly descriptive; many competing AI-pic tools and similar names exist. Trademark clearance in AI/image classes is important for serious use. Liquidity depends on current AI hype cycles.

Overall portfolio notes
Stronger pieces for retail/end-user sales: CabAuto.com, PicByAI.com, AskBurg.com, Ashels.com (solid.coms with use cases or development). PicYa.com and Picslot.com are solid mid-tier brandables. Soil.biz is the outlier due to TLD — better as a developed-site package or low-price wholesale. Bundle options or “with developed site” positioning can help several of them. No verified high authority or revenue signals to lift values further.

Help creating dedicated “domain for sale” threads

Post each as a separate thread in the appropriate DNForum marketplace section (e.g., Domains for Sale / Premium / Brandables). Keep titles clean and searchable. Use consistent structure so buyers can scan quickly. Do not invent traffic/revenue. Disclose if a site is developed/parked. Use escrow, and set realistic BIN or “make offer” based on your wholesale/retail targets above (start higher for negotiation room, or price for quick sale).

Suggested title format
[FOR SALE] CabAuto.com – Brandable Auto/Taxi/Mobility.com – BIN $X or Make Offer

(Adapt per domain; add “Developed site” or “AI niche” where accurate.)

Recommended first-post template (BBCode-ready)
Copy-paste and fill in:

Domain: Example.com
BIN / Price: $X,XXX or Make Offer (or Lease-to-Own if offered)
TLD / Length:.com / 7 letters, no hyphens/numbers
Description:
Short, clear elevator pitch (1–3 sentences). Example for CabAuto: “Clean cab + auto compound perfect for ride-sharing, taxi services, fleet management, auto tech, or mobility startups. Memorable and commercial.”

Possible uses:
  • Primary use case 1
  • Use case 2
  • Use case 3

Highlights:
  • Clean spelling / easy brandability
  • Associated website: brief note (e.g., “Developed AI assistant landing” or “For-sale / landing page present” or “Undeveloped”)
  • Any other visible positives (age if known, etc.)

History / notes:
Public footprint clean of obvious abuse flags in recent checks. Buyer should perform their own trademark, history, and due-diligence searches. No traffic/revenue claims.

Terms:
Escrow.com or DNForum-approved preferred. Fast transfer. Serious inquiries only. Payment methods: [list yours].

Contact:
PM me here or [email / preferred]. Happy to answer questions.

Screenshots / landing:
(Attach or link a clean screenshot of the name + any live site.)

Thanks for looking!

Quick tips for better threads
  • One domain per thread for best visibility and offers.
  • Price realistically (use the ranges above as guide; wholesale if pushing for flippers, higher BIN if targeting end users).
  • Update with price drops or “price reduced” later if needed.
  • Cross-post carefully per forum rules; mention developed sites where they add value (AskBurg, Soil.biz, Ashels, CabAuto appear stronger here).
  • For PicByAI and similar, lean into the AI/image niche demand.
  • Be ready with WHOIS proof of ownership and auth code if asked.
  • If bundling any (e.g., photo-related ones), note it clearly.

If you want me to draft the full ready-to-post text for any specific domain (with a suggested starting BIN), refine ranges based on more details you have (registration year, traffic, prior offers, exact development status), or adjust for a wholesale vs end-user focus, just say which ones. Good luck with the sales!
 

HelmutsHelmuts is verified member.

Domain Summit | HostMaria
DNF Staff
DNForum.club
Registrar
Hosting Provider
Joined
Mar 29, 2014
Messages
2,413
Reaction score
903
@DNF Admin we probably need to increase the execution time.. what is it set now? tx

yaA8TisFk5.webp
 

DN AI

Level 1
Joined
Jul 13, 2026
Messages
25
Reaction score
7
GPT Assistant

Yes. DNForum could host Ollama on a separate dedicated GPU-capable server and expose it through a forum command such as /ollama. Members would not need their own OAuth setup or API keys because DNForum would authenticate them through their existing forum session.

A practical setup would be:

  • The forum receives /ollama and verifies the logged-in member, permissions, and thread access.
  • A DNForum backend service sends the approved prompt and permitted thread context to the private Ollama server.
  • Ollama returns the model output to the backend, which posts it in the thread.
  • The Ollama endpoint stays private; it should not be publicly exposed directly to members.

The main requirements are operational rather than OAuth-related:

  • Hardware/capacity: model size, VRAM/RAM, concurrent requests, queueing, and response speed.
  • Abuse controls: per-user quotas, rate limits, cooldowns, daily credits, prompt-size limits, and bot/spam protection.
  • Permissions: possibly limit it initially to supported threads, certain member groups, or paid access.
  • Privacy: define what thread/post data is sent to the model, what is logged, and how long it is retained.
  • Moderation: output filtering, reporting, and safeguards against prompt injection from thread content.
  • Feature boundaries: a self-hosted model can summarize and reason over supplied context, but live web research, external data tools, and high-end model quality require separate integrations or infrastructure.

So the command could be seamless for users, while DNForum centrally bears and controls the compute cost. A sensible approach would be to run /ollama as a clearly labeled local/self-hosted assistant alongside /gpt, with stricter quotas at alpha stage.
 

DN AI

Level 1
Joined
Jul 13, 2026
Messages
25
Reaction score
7
Grok Assistant

I'll check current Ollama CPU/GPU requirements and Hetzner dedicated/GPU server options so the DNForum scenarios are accurate.
Gathering more precise Ollama model sizing and current Hetzner CPU dedicated lines.
Pulling detailed Hetzner GPU configs and typical Ollama model memory footprints.
Fetching one more pass on model memory sizing for accurate scenario ranges.
Short answer
Yes — you can run Ollama on a standard Hetzner dedicated (CPU-only). For a real DNForum in-thread assistant (utility model + stronger final model, acceptable latency, more than a couple concurrent users), a GPU server is the practical choice.

CPU-only is fine for internal testing, tiny quantized models, and very low concurrency. GPU is what you want for production-feeling response times.

How Ollama actually behaves
Ollama runs on CPU or NVIDIA GPU. GPU is typically many times faster on token generation. Memory is the hard limit:

  • Rough model footprint (Q4-class quants, ballpark): ~3–5 GB for 7–8B, ~8–12 GB for 13–14B, ~18–24 GB for 30–34B, ~35–45 GB for 70B. Longer context, KV cache, and parallel requests need extra headroom.
  • System RAM: still need plenty even with GPU (loading, OS, embeddings/tools, web-research side processes). Aim for comfortably more RAM than VRAM used.
  • Concurrency: Ollama can load multiple models and handle limited parallel requests if memory allows (OLLAMA_NUM_PARALLEL, OLLAMA_MAX_LOADED_MODELS). Forum traffic is bursty; one slow 30B+ CPU reply will feel broken in a thread.
  • DNForum-style stack (small utility/scanner model + stronger final model + optional live research) favors keeping two models warm or accepting reload latency.

You cannot bolt a GPU onto normal AX/EX dedicated boxes — Hetzner GPU is the separate GEX line (one GPU per server).

Standard Hetzner dedicated (CPU-only) — when it works
Good candidates: high-core AX/EX with lots of RAM, e.g. Ryzen 9 7950X3D class (~16c/32t, 128–192 GB) or EPYC 9454P class (high core count, 256 GB+).

Realistic CPU-only use for DNForum
  • Alpha / staff-only / dev: 7–8B (or smaller) Q4–Q5, 1 request at a time, long waits OK.
  • Light internal tool: same, maybe a bit more RAM headroom for a second small model.
  • Not ideal: public /gpt with 14B+ “strong” model, multi-user peaks, or tight latency. 30B+ on CPU is usually too slow for forum UX.

Expect single-digit to low-teens tokens/sec on a solid CPU for 7–8B; larger models drop off hard. Fine for experiments, weak for production assistant feel.

Hetzner GPU options (current GEX direction)
  • GEX44: Intel i5-13500, 64 GB RAM, NVIDIA RTX 4000 SFF Ada 20 GB VRAM — positioned for efficient inference.
  • GEX131: Xeon Gold 5412U (24 cores), 256 / 512 / 768 GB RAM, NVIDIA RTX PRO 6000 Blackwell Max-Q 96 GB VRAM — heavy inference / larger models / more headroom.
  • (Earlier GEX130-class boxes with RTX 6000 Ada ~48 GB VRAM may still appear in docs/availability; treat live Robot configurator as source of truth.)

DNForum scenarios (practical sizing)

1) Dev / alpha / very low traffic (CPU OK)
  • Hardware: standard dedicated, 12–16+ cores, 64–128 GB RAM, fast NVMe.
  • Models: 7–8B utility + optional small final (or one 8–14B Q4).
  • Concurrency: ~1 (queue the rest).
  • Fit: prove prompts, tools, web-research flow. Not a public snappy /gpt.

2) Light production — members feature, low concurrency
  • Hardware: GEX44-class (20 GB VRAM), 64 GB system RAM is workable; more RAM nicer if you add sidecars.
  • Models: small utility (3–8B) + strong final up to ~14B Q4/Q5, or a single strong ~14–22B-class quant that fits with context headroom.
  • Concurrency: a few parallel requests if you stay on smaller models; keep NUM_PARALLEL modest.
  • Fit: alpha→early public, bursty thread questions, good latency on mid-size models. 70B does not fit comfortably.

3) Medium production — two warm models, solid UX
  • Hardware: prefer 40–48 GB+ VRAM if available; otherwise GEX44 with careful model choices, or jump to GEX131. System RAM 128 GB+ comfortable.
  • Models: always-on small utility/scanner + 14–32B-class final (Q4/Q5), or one high-quality 32B-class with room for KV/cache.
  • Concurrency: multiple queued/light parallel forum asks without thrashing.
  • Fit: matches “utility model + stronger final + research” without constant model swap pain.

4) Heavy / peak forum use or large “strong” model
  • Hardware: GEX131-class (96 GB VRAM), 256 GB+ RAM (512 GB if you keep several models, long contexts, or heavy tooling).
  • Models: 70B Q4 (or similar) as final, plus small utility; or several mid-size models resident; longer contexts and higher NUM_PARALLEL.
  • Concurrency: meaningful multi-user headroom; still one GPU, so not infinite scale — queue or horizontal scale later if needed.
  • Fit: “feels like a product” under real DNForum load, large-context thread analysis, less compromise on model quality.

5) Scale-out later (if /gpt becomes popular)
  • One GPU box + queue/worker, or multiple GEX nodes behind a router (different models or shard by load).
  • Keep embeddings/rerankers/light utility on CPU RAM or a smaller box; reserve GPU for the big generative passes.
  • Hetzner is currently one GPU per GEX server — plan horizontal, not multi-GPU vertical, on their stock line.

Recommendation for DNForum
  • Now / alpha: standard high-RAM dedicated is enough to develop and dogfood with 7–8B (and maybe 14B if you accept slowness).
  • Real member-facing /gpt: start on GEX44 (20 GB) if you standardize on efficient ≤14B-class finals (or one carefully chosen larger quant), with tight ops on concurrency and context.
  • Best long-term match to the announced architecture (utility + stronger model + research, growth headroom): GEX131-class 96 GB so you are not permanently capped on model size and dual-model residency.

Ops notes that matter more than the brochure specs
  • Quantization (Q4/Q5/Q6) and context length dominate fit and speed.
  • Prefer models that stay resident; cold-load on a forum slash-command feels broken.
  • Set parallel/queue limits so one whale request cannot melt latency for everyone.
  • NVMe for model storage; monitor VRAM, RAM, and tokens/sec under synthetic concurrent /gpt load before public rollout.
  • Web research and app logic can live on CPU; GPU mainly buys generative speed and larger models.

Bottom line
Standard Hetzner dedicated: yes for Ollama, mainly lab/low-traffic.
GPU server: yes for production DNForum assistant.
20 GB (GEX44) = sensible light/mid inference entry.
~48–96 GB (higher GEX) = dual-model + larger finals + real concurrency headroom — the safer bet if /gpt is meant to stay fast as usage grows.

Exact SKU availability and pricing change; confirm current GEX44/GEX131 configs in Hetzner Robot before ordering. If you share target models (e.g. 8B+32B vs 70B), expected concurrent /gpt users, and latency goal, the scenario pick can be narrowed to one box.
 
Top Bottom