What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
For most CPU-only Ollama deployments, start with 16 GB RAM, 4 vCPUs and NVMe storage for an 8B model. A 3B–7B model can run on a 16 GB VPS for personal use, while a 70B model generally needs at least 48 GB of RAM and is usually too slow for interactive CPU-only work. The exact requirement depends on the model tag, quantization, context length and number of simultaneous users.
Need a server to try this on? Antihost offers Linux VPS plans with NVMe storage, full root access and a 14-day free trial.
Model-file size is the lower bound, not the VPS specification. Ollama also needs memory for the operating system, runtime, weights loaded in RAM and the key-value (KV) cache used by the context window. Leave roughly 2–4 GB beyond the model footprint as a practical minimum, and add more for long prompts or concurrent requests.
As an Amazon Associate I earn from qualifying purchases.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →What an Ollama VPS actually needs
Ollama itself uses comparatively little memory. The model you load determines the meaningful requirements. A VPS must have enough RAM to hold the model’s quantized weights plus the operating system, Ollama process, allocator overhead and KV cache. If RAM is exhausted, Linux starts swapping; generation then becomes dramatically slower or the process is killed.
For CPU-only inference, a useful first estimate is about 0.6 GB of memory per billion parameters for a 4-bit model, followed by a 2–4 GB margin for the OS and runtime. This is an estimate from third-party sizing guides, not an official Ollama requirement. Context length, quantization format and implementation details can move the real figure substantially.
#1 Best Overall
- [Ultra-Lightweight Ear Buds Designed for Small Ears] Each earbud weighs only 3.7g and features a compact, ergonomic in-ear design made especially for small ears. Secure, low-profile, and comfortable for workouts, all-day wear or overnight sleeping.The ultra-slim profile sits flush in your ear with no hard edges poking or rubbing, so you can even sleep on your side without discomfort.
- [Immersive Stereo Sound with TOZO OrigX Technology] TOZO OrigX tuning delivers clear vocals, balanced mids, and natural stereo sound for music, podcasts, and videos.
- [Long Battery Life for Daily Use] Get up to 7 hours of playtime on a single charge, with up to 32 hours total using the charging case—ideal for workdays, commuting, and extended listening sessions.
- [Bluetooth 5.3 & Stable Connection] Bluetooth 5.3 provides fast pairing, stable wireless performance, and reduced dropouts as you move around home or office.
- [Deep Bass with Clear Vocals] High-performance drivers produce punchy bass while keeping vocals clean and detailed for everyday listening.
- Personal 3B–7B use: 4 vCPUs and 16 GB RAM is a practical starting point.
- Typical 8B use: 4 vCPUs, 16 GB RAM and NVMe storage is the commonly cited sweet spot.
- Five simultaneous 7B users: a third-party March 2026 estimate calls for 8 vCPUs and 32 GB RAM, at roughly $35–$60 per month.
- 70B use: 48–96 GB RAM is the range cited by third-party guides, with CPU-only speed usually unsuitable for interactive sessions.
There is no single official Ollama hardware-requirements table covering every model and context size. The ranges above should therefore be treated as planning estimates, not guarantees.
Ollama model sizes and the RAM they imply
The following file sizes are the default tags shown in the Ollama library snapshot available on October 14, 2026. The llama3.1 8B and 70B tags are 4-bit builds; the listed 8B tag uses a 128K context. Check the live tag before buying storage because tags and quantizations can change.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →| Model tag or class | Library file size | Typical CPU-only RAM guidance | Practical interpretation |
|---|---|---|---|
| 1–3B, 4-bit | Varies by tag | About 6–8 GB | 8 GB is comfortable for a small single-user model; 16 GB leaves room for larger context and services. |
| llama3.1 8B, default 4-bit tag | 4.9 GB; 128K context | 8–16 GB; 12 GB is often cited for this model, with 16 GB the common VPS target | A 16 GB VPS is the safest general-purpose entry point. |
| 13–14B, 4-bit | Not stated; depends on the tag | 16–24 GB | Choose 24 GB for a comfortable single-user setup and long prompts. |
| 27–34B, 4-bit | Not stated; depends on the tag | 24–32 GB | Expect slower CPU generation and less room for multiple requests. |
| llama3.1 70B, default tag | 43 GB | 48–96 GB (third-party estimates disagree) | Even when it fits, CPU-only throughput is generally too low for a responsive chat service. |
| llama3.1 405B, default tag | 243 GB | Not stated by the cited guides | The file alone exceeds ordinary VPS sizes; plan only with a specialized, very large-memory host. |
A file size is not equivalent to the amount of free RAM you need. For example, a 4.9 GB 8B file still requires memory for the OS and KV cache. Conversely, two tags with the same parameter count can have different footprints because of quantization and context settings.
Why the 8B model is the usual starting point
An 8B 4-bit model is small enough to fit on a 16 GB CPU VPS while providing substantially more capability than the smallest 1B–3B models. The 4.9 GB library file leaves several gigabytes for Linux and a moderate context. It is still possible to trigger swapping with very long prompts, multiple loaded models or concurrent requests, so 16 GB should be viewed as a sensible baseline rather than an unlimited configuration.
Rank #2
- 【IP68 Waterproof Multifunctinal Smart Watch】Torjalph fitness activity tracker smart watch is IP68 waterproof technology ,you can wear it when swimming,handwashing,it's rain-proof which means that you perform multiple exercise modes outside without worrying about sweat or rain corroding your fitness smart watch,and this watch equipped with a 230 mAh large capacity battery magnetic charging ,and this sports activity tracker watch can be used 5-7 days If fully charged.
- 【Larger 1.83'' Full Screen Watch & Multi-dial Selection】1.83 inch full HD color touch screen smart watch for men / women provides excellent 240*284 resolution image quality ,This Bluetooth smart watch offers a better touch experience and remarkable immersion and ,more than 200 watch faces for you to choose from App"Da Fit ", you can change the dial style according to your mood to wear keeping the world on your wrist
- 【Accurate Activity Tracker Smart Watch 】sports watch activity tracker uses sweat and dust resistant breathable silicone strap, our Fitness Tracker Smart Watch for Android iOS Phones supports 12 sports modes,walking,running,bicycle,swimming,badminton,basketball,football,rope skipping. This bluetooth calls smart watch for men women is lightweight and stylish, bringing more convenient and relaxing sports experience while also providing more protection.
- 【Your 24 Hour Personal Daily Health Monitor】Fitness watch for women men full touch screen IP67 waterproof bluetooth smartwatch for Android iOS Phones with heart rate blood pressure monitor multifunction Andriod smart watch with steps counter sleep monitor sports fitness watch, monitoring blood oxygen saturation allows you to understand your physical condition in real time to track fitness progress ,to know your health status in time.
- 【More Functions Bluetooth Call Smart Watch】This fitness activity smart watch full screen for iPhone and Android phones includes a bunch of new features including:The bluetooth smart watch can hang up the phone, receive and display WhatsAPP /Text /Facebook /Twitter /Messenger reminder.Support sedentary reminder, alarm clock, stopwatch, timer,weather information,music control, remote camera control compatible with Android 5.0 and above or iOS 9.0 and above smartphones.
Why 70B is a different class of deployment
The default 70B file is 43 GB before runtime overhead. Third-party guides place the complete CPU-only requirement anywhere from 48 GB to 96 GB, largely because they count context and operating-system headroom differently. A 64 GB machine may work for a short single-user context but leave little safety margin; 96 GB is the more conservative target when the workload is important. The estimated generation rate of about 0.8–3 tokens per second makes a 70B CPU VPS impractical for most interactive use.
How quantization and context length change RAM use
Quantization
Quantization stores weights with fewer bits. A 4-bit build is the usual VPS choice because it sharply reduces memory while retaining usable quality. Higher-precision builds require much more RAM: the 8B fp16 tag shown in the Ollama library is about 31 GB, compared with 4.9 GB for the default 4-bit tag. An 8B fp16 model therefore does not belong on a typical 16 GB VPS.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchDo not size a server from the parameter count alone. Before provisioning, identify the exact tag and its quantization. Two tags named for the same model family can have materially different file sizes and memory behavior.
Context length and the KV cache
The context window is the amount of conversation and input the model can retain for a request. As context grows, the KV cache consumes additional memory. The default llama3.1 8B tag is listed with a 128K context, but that does not mean a 16 GB VPS can process a full 128K prompt comfortably. Long-context workloads need extra RAM and can reduce the number of requests that fit concurrently.
Rank #3
- [Low Noise High Fidelity] New bee guitar cord with 22AWG oxygen-free-copper center conductors provides enhanced signal transfer, the PVC insulation material and the high-density stranded copper double shield provide noise rejection for quiet operation
- [Ideal for Choice] New bee 1/4 straight and right angle metal connectors, ensures a secure connection in any instrument, pedal, or amp jack. L shape great fro on stage guitar stands
- [Wide Compatible] Suitable for mixer, power amplifier, microphone, effector, decoder, equalizer, electoronic organ, electric guitar, electric bass and other electronic equipments. Widely used in the installation of smart home, karaoke speaker, professional stage audio, etc
- [Easy Disassemble] Making incidental repairs quick and easy! Very nice, flexible cable
- [Dont Hesite to Order] 24 hours customer services and professional technology team are standing by
If a deployment is close to its memory limit, cap the context to the smallest value that meets the application requirement, avoid unnecessarily large retrieved documents and unload models that are not in use. Treat context-related memory as workload-dependent; no fixed multiplier applies to every model.
CPU, tokens per second and user concurrency
CPU capacity controls how quickly tokens are generated once the model is in memory. A third-party 2026 comparison estimates roughly 8–20 tokens per second for 7B models on CPU-only VPSs. The same table estimates about 0.8–3 tokens per second for 70B. Actual speed varies with processor generation, memory bandwidth, thread settings, quantization, prompt length and whether other users are active.
| Workload | Suggested starting VPS | Expected behavior |
|---|---|---|
| One person, 3B–7B model | 4 vCPUs, 16 GB RAM, NVMe | Good entry point; enough headroom for the OS and moderate context. |
| One person, 13–14B model | 4–8 vCPUs, 24 GB RAM | Fits more comfortably, but generation is slower than with 7B–8B models. |
| One person, 27–34B model | 8 vCPUs, 32 GB RAM | Memory is workable; CPU latency may be high for chat. |
| Five concurrent users on 7B | 8 vCPUs, 32 GB RAM | Third-party March 2026 estimate; budget roughly $35–$60 per month. |
| 70B CPU-only | 48–96 GB RAM, as workload allows | Very slow token generation; generally unsuitable for interactive service. |
Concurrency is not simply “one model multiplied by the number of users.” Ollama can share loaded weights, but each active request needs its own prompt and KV-cache memory. More users therefore increase both RAM pressure and CPU contention. If requests queue, adding vCPUs may help throughput; if the host swaps, add RAM first.
Storage, disk performance and network considerations
Provision disk space for the model file, the operating system, logs, updates and any additional tags you intend to keep. A 70B deployment needs more than the 43 GB file because deleting old layers, downloading updates and retaining a fallback tag all require free space. The 405B tag’s 243 GB file exceeds the storage allocation of many general-purpose VPS plans.
Rank #4
- Digital Family Calendar: BSIMB digital calendar wall touch screen puts an end to your chaotic life. Designed for busy families, the digital calendar keeps you organised and in step with your family. Assign different colours and tasks to each family member to make family scheduling easier and more intuitive.
- Easy Setup and Auto-Sync: Plug in, connect to WLAN and auto-syn your electric calendar in minutes. Digital wall calendar compatible with Google calendar, iCloud calendar, Outlook calendar, Cozi calendar and Yahoo calendar. Add events directly on the device or through the ecalendar app. Electronic calendar stays current whether you're at work or traveling—no manual updates needed.
- All-In-One Home Assistant: Smart calendar builds habits and encourage independence with personalised chores. Display daily meal plan to reduce the daily ‘What's for dinner?’ stress. To-do lists keep track of everything you want. Turning chores into valuable achievements with stars and rewards, making life easier and more enjoyable.
- Manage Anytime Anywhere: BSIMB electronic calendar for family stay connected with the powerful ecalendar app. Access and manage your digital wall calendar anytime, anywhere with free features like stars & rewards, photo screensaver, chore charts, meals plans, to-do lists etc. Electronic planner paid features like Magic Import tailor advanced tools for users with special needs.
- Stay in Touch with Loved Ones: Share photos or videos from your phone to electronic wall calendar with app, instantly update your life moment, and keep in touch with your family, especially those who are far away; you can also surprise your smart calendar digital wall by sending him a wish card; it's also great for keeping tech-illiterate elders in touch with the youngsters.
NVMe storage is preferred over slower network-backed disks because model downloads, startup and swapping (which should be avoided) are storage-sensitive. Disk speed does not replace RAM: a VPS that constantly swaps may remain unusably slow even with fast NVMe.
Network bandwidth primarily affects download time and API latency; it does not increase tokens per second once the model is running. For a private single-user endpoint, ordinary VPS networking is usually adequate. Public APIs also need authentication, TLS termination and rate limits, which consume some CPU and memory beyond the model itself.
A repeatable VPS-sizing procedure
- Choose the exact model tag. Record parameter count, quantization, file size and advertised context. Do not size from a family name alone.
- Set a RAM floor. Start with the model file’s size, then add at least 2–4 GB for Linux, Ollama and runtime overhead. Increase the margin for long context or more than one active request.
- Account for context. If the application sends long documents or maintains large conversations, select more RAM or cap the context window.
- Estimate concurrency. Count simultaneous generations, not registered accounts. Each active request adds KV-cache and CPU demand.
- Select vCPUs for latency. Use about 4 vCPUs for light 3B–7B personal use; 8 vCPUs is a reasonable starting point for five concurrent 7B users.
- Reserve storage. Fit the model file plus the OS, logs, updates and any second tag. Keep free space available so updates do not fail.
- Measure after deployment. Watch RAM usage, swap activity, load average, response queueing and tokens per second under the real prompt sizes. Resize based on the bottleneck you observe.
Common failure symptoms and fixes
The model will not load or the process is killed
This usually indicates insufficient RAM. Stop other models, reduce context, select a smaller quantized tag or move to the next memory tier. Adding vCPUs alone will not solve an out-of-memory failure.
Generation starts quickly but becomes extremely slow
Check whether the VPS is swapping. If swap usage rises, add RAM or reduce the model/context footprint. If RAM is available but CPU utilization is saturated, move to a faster or larger-vCPU plan, accepting that CPU-only 70B inference may still remain too slow.
Requests queue when several users connect
Concurrent prompts can exhaust CPU and KV-cache memory even when the model itself fits. Reduce simultaneous generation limits, shorten context, add RAM and vCPUs, or deploy a smaller model for routine requests.
Storage fills during an update
Remove unused tags and logs only after confirming that no service depends on them, then expand the disk. Keep enough free capacity for a complete new layer set; a model update may temporarily require more space than the final installed file.
Free tools Windows power users keep installed
One-click scans. No signup required.
When a GPU VPS is justified
This guide focuses on CPU-only sizing because that is the common VPS question. A GPU changes the bottleneck: compatible model layers can be placed in GPU memory, improving throughput and reducing CPU pressure, but the host still needs system RAM for the OS, runtime and any layers that remain on the CPU. GPU pricing, available VRAM and Ollama’s support for the specific device must be evaluated separately. Do not assume that a GPU VPS eliminates the need to size system RAM.
How much should you budget?
For a five-user 7B service, one third-party guide priced an 8-vCPU, 32 GB VPS at approximately $35–$60 per month in March 2026. That is a market estimate for that period, not a universal price and not a quote for every region or provider. Smaller 16 GB instances may cost less, while high-memory 70B machines can cost substantially more. Compare the advertised RAM, sustained CPU performance, NVMe capacity and bandwidth rather than choosing solely by monthly price.
Limits of the published estimates
The 1–3B, 7–8B, 13–14B, 27–34B and 70B ranges come from third-party guides, several of which are published by hosting companies. Their counting methods differ: some include KV-cache headroom and the operating system, while others describe only the model process. The 70B range is consequently broad at 48–96 GB.
The Ollama library figures are useful for identifying the default tags, but the captured page may be about a year old relative to a later deployment. Recheck the live tag’s file size and quantization before committing disk or RAM. Treat every number as an estimate tied to the stated model, date and workload, not as an official Ollama guarantee.
Recommended Free Tools
Bottom line
For an ordinary Ollama VPS, choose 16 GB RAM, 4 vCPUs and NVMe storage for a 7B–8B 4-bit model. Move to 24 GB for 13–14B, 32 GB for 27–34B, and at least 48 GB—often closer to 96 GB—for 70B, while expecting very low CPU-only throughput. Verify the exact model tag, quantization and context length, then leave memory headroom for the OS, KV cache and every concurrent request.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
