Run Gemma 3 27B on a VPS.
prices as of · Contabo as of · 81 of 524 plans fit, disk included · re-ranked daily
Google's largest open model; chat quality close to much bigger systems, wants 32 GB. On a CPU-only VPS it needs about 22 GB of RAM: 16.6 GB of Q4_K_M weights, 3.9 GB of KV cache for an 8,192-token context and 1.5 GB for the OS and runtime. 81 of the 524 plans in our index fit and include a disk; the cheapest comfortable pick is netcup's VPS 4000 G12 · 12 vCore · 32 GB at $31.43/mo, streaming an estimated 1.4–2.9 tok/s.
The cheapest VPS that runs Gemma 3 27B is netcup VPS 4000 G12 at $31.43/mo, 32 GB of RAM against the 22 GB the model needs, checked 16 Sept 2026.
Best VPS plans for Gemma 3 27B
ranked by estimated tokens/s per dollar, comfortable fits first
- 1runs comfortably~2.2–4.3 tok/s$38.71/moView at netcup ↗
- 2runs comfortably~1.4–2.9 tok/s$28.85/mochecked 10 days agoView at Contabo ↗
- 3runs comfortably~1.4–2.9 tok/s$31.43/moView at netcup ↗
- 4runs comfortably~1.4–2.9 tok/s$34.03/monot available nowView at Hetzner ↗
- 5runs comfortably~1.2–2.3 tok/s$32.31/moView at HOSTKEY ↗
- 6runs comfortably~1.4–2.9 tok/s$42.69/mochecked 10 days agoView at Contabo ↗
Speed = effective memory bandwidth ÷ active weight bytes (4 GB/s per shared vCPU, 6 per dedicated), shown as a band. Real numbers depend on the host CPU generation, AVX-512/AMX support and how noisy the neighbors are, so treat these as order-of-magnitude estimates.
RAM needed · CPU inference
22 GB
- weights · Q4_K_M
- 16.6 GB
- KV cache · 8K context
- 3.9 GB
- OS + runtime headroom
- 1.5 GB
Other quants: Q8_0 weights 28.7 GB (near-lossless, about half of F16); F16 922 MB. Comfortable = 20% headroom over the total.
Model card
- Size
- 27B parameters
- Context
- 128K tokens
- Kind
- chat · vision
- Released
- 2025-03
- License
- Gemma Terms of Use
Facts fetched from Hugging Face on 7 Sept 2026: exact GGUF file sizes, KV geometry from the GGUF header · 374,507 downloads.
Or rent it as an API · 22 providers sell Gemma 3 27B
every price for Gemma 3 27B →Google list price
$0.69/M tokens
Cheapest price we can explain
none below list
Token prices are per million, three parts input to one part output, and exclude tax. A rented box costs the same whether or not you use it and runs whatever else you put on it; an API charges for what you send and nothing when you stop.
Or buy hardware · 8 reference machines fit
all machines →- Mac mini (M6, 32 GB)32 GB · 153 GB/s · ~4.8–7.4 tok/s$1,299= 41 mo of VPS
- Framework Desktop (Ryzen AI Max+ 395, 64 GB)64 GB · 256 GB/s · ~8.1–12 tok/s$1,659*= 53 mo of VPS
- Mac mini (M5 Pro, 64 GB)64 GB · 307 GB/s · ~9.7–15 tok/s$2,299= 73 mo of VPS
- Mac Studio (M5 Max, 36 GB)36 GB · 460 GB/s · ~15–22 tok/s$2,499= 80 mo of VPS
- Framework Desktop (Ryzen AI Max+ 395, 128 GB)128 GB · 256 GB/s · ~8.1–12 tok/s$3,149*= 100 mo of VPS
- GeForce RTX 5090 32 GB (card only)32 GB · 1792 GB/s · ~57–87 tok/s$4,300*= 10+ yrs of VPS
Buy · per month
$37.45
$36.08 hardware + $1.36 power
Rent · per month
$31.43
netcup VPS 4000 G12 · 12 vCore · 32 GB
Break-even
The Mac mini (M6, 32 GB) pays for itself after 43 months of replacing the VPS, and it streams an estimated 4.8–7.4 tok/s against the VPS's CPU-only pace.
* approximate: August 2026 US retail median rather than list price. Local speed = peak bandwidth × 0.7 efficiency (0.35 on CPU-only boards) ÷ active weight bytes; unified-memory machines are assumed to give models 75% of their RAM. GPU cards need a host PC that is not included in the price.
Frequently asked
- How much RAM does Gemma 3 27B need?
- About 22 GB for CPU inference at Q4_K_M: 16.6 GB of weights, 3.9 GB of KV cache at 8,192 tokens of context, and 1.5 GB of headroom. Longer contexts need more KV cache (496 KB per token for this model).
- What is the cheapest VPS that can run Gemma 3 27B?
- netcup VPS 4000 G12 · 12 vCore · 32 GB (32 GB RAM, 12 vCPU) at $31.43/mo excl. VAT runs it comfortably as of 16 Sept 2026. OVHcloud VPS-4 2027 · 8 vCPU / 24 GB · EU at $27.11/mo is a tight fit.
- How fast will Gemma 3 27B run on a VPS without a GPU?
- Roughly 2.2–4.3 tok/s on the top pick. Token generation is limited by memory bandwidth, because each new token reads all of the weights once. More vCPUs help, dedicated ones most, but a GPU or an Apple silicon machine is 10 to 50 times faster.