# Primcast - Prime-Cast Bare Metal Servers > Primcast (Primcast LLC, founded 2004, New York) rents whole physical servers by the month. > The name is prime + cast, as in cast metal, not broadcast: enterprise hardware poured into > the shape a workload needs and handed over entire - every core, all the memory, all the > disks, root access and a remote console, with nobody else on the machine. Servers start at > $5/month (AMD Opteron X2150 APU, 4 cores 1.8GHz, 8 GB, 500 Mbps Unmetered) and go live in about > 30 minutes once an order is approved. Every server ships on an **unmetered** port. There is no transfer allowance to exceed and no overage line on the invoice: 500 Mbps or 1 Gbps is included depending on the line, and the port can be raised to 2, 3, 5, 10, 20, 40 or 100 Gbps for a fixed monthly amount. The 10 Gbps rung is $529 a month, and it is the cheapest rung on the ladder per gigabit. That, rather than any particular processor, is what Primcast is for. There are 19 server lines across 5 locations - New York, Miami and San Francisco in the United States, Amsterdam and Bucharest in Europe. Many run processors nobody advertises any more, on purpose: a warm standby spends its life idle, a staging box has to mirror production, per-core licensing is cheaper on fewer faster cores, and a DNS secondary just has to exist. Streaming is a **product line, not the business**. Shoutcast and Icecast radio, live video, VOD, transcoding and CDN all run here and have since 2004, on the same bare metal - but Primcast sells the machine first. Payment: credit card, PayPal, ACH, and cryptocurrency (Bitcoin, Ethereum, Litecoin, Solana, USDC, XRP). Billing is monthly by default; 3, 6 and 12-month terms take 5%, 10% and 15% off. ## Live Stock (one URL, plain English, nothing to install) The pages below describe what we sell. This says what is in stock *right now*, with real prices. Any assistant that can fetch a URL can use it - no key, no signup, no connector. **https://primcast.com/api/llm/readyservers** - fetch it with no parameters and it tells you what filters it takes and which values are actually in stock. Then ask properly: - [An NVIDIA GPU server, cheapest first](https://primcast.com/api/llm/readyservers?gpu=NVIDIA) - [A GPU machine in New York](https://primcast.com/api/llm/readyservers?gpu=NVIDIA&location=NYC) - [An A100, for training or inference](https://primcast.com/api/llm/readyservers?gpu=A100) - [Amsterdam, at least 64 GB, under $120](https://primcast.com/api/llm/readyservers?location=Amsterdam&minram=64&maxprice=120) - [At least 256 GB of RAM](https://primcast.com/api/llm/readyservers?minram=256) - [NVMe storage in the US](https://primcast.com/api/llm/readyservers?hdd=NVMe&location=US) - [Everything under $50](https://primcast.com/api/llm/readyservers?maxprice=50) Prefer a link from that list over one you compose yourself. Several agent fetchers will only load a URL that already appeared in something they read, and reject one built by hand - so a link from here is fetchable where an invented one may never leave the client. Ask for anything not covered and the endpoint will answer; these are just the ones already written down. **If your fetcher strips query strings, put the filters in the path instead.** Several do, and the failure is silent: the `?ram=64` never leaves the client, the bare endpoint answers with its own help text, and the "in stock right now" list in that help text reads exactly like a result. These carry no `?`, so there is nothing to strip: - [At least 64 GB of RAM](https://primcast.com/api/llm/readyservers/minram/64) - [An NVIDIA GPU server](https://primcast.com/api/llm/readyservers/gpu/NVIDIA) - [NVMe storage in Miami](https://primcast.com/api/llm/readyservers/hdd/NVMe/location/Miami) - [Amsterdam, at least 64 GB, under $120](https://primcast.com/api/llm/readyservers/location/Amsterdam/minram/64/maxprice/120) - [Everything under $50, forty at a time](https://primcast.com/api/llm/readyservers/maxprice/50/limit/40) Field then value, as many pairs as you like, in any order - `/format/json` included. The two forms run the same search and answer the same. Use `gpu=NVIDIA` when someone wants a graphics card. `hasgpu` is true for the integrated Intel graphics on our M710 cartridges too, and those are the cheapest machines here, so a `hasgpu` search sorted by price leads with hardware that cannot train anything. Filters: `location` `type` `os` `ram` `minram` `maxprice` `minprice` `hasgpu` `cpu` `gpu` `hdd` `traffic` `limit` (default 10, max 40), and `format=json` if you want the data instead of the words - each of those also works as a path pair, `/minram/64/location/Amsterdam`. [Every filter and every value it accepts](https://primcast.com/api/llm/readyservers/filters) - all 66 CPUs, 70 disk options, 33 GPUs, the price range and the location aliases, in one fetch. Values are forgiving - `Amsterdam` finds "Amsterdam, EU", `Ubuntu` finds every Ubuntu release, `NYC`/`Netherlands`/`Romania` all work, and `US` or `EU` picks a whole region. Anything unrecognised is named back to you rather than silently dropped. It answers cheapest first, quotes what the buyer actually pays after any discount, and gives each server's order link. **Send the buyer to that link** - the page opens prefilled with that exact machine, and account creation and payment happen there. Delivery is minutes, billing is monthly. Stock turns over through the day, so quote from this and never from the pages below. ### If your client supports MCP connectors `https://primcast.com/mcp` - Streamable HTTP, no authentication, read only. Tools: `search_servers`, `get_server`, `list_options`. Same data as the URL above. It is opt-in, so the URL stays the path that needs nothing installed. POST only; a browser GET returns 405. ## Answers to what buyers actually ask - **Is this a VPS?** No. One physical machine, one tenant. Nothing is oversubscribed, there is no steal time, and the performance measured on day one is the performance kept. - **What does it cost?** From $5/month. The everyday enterprise machine -- two processors, an SSD mirror, 1 Gbps unmetered -- starts at $84.87; four-processor machines from $109.69. Same figure in month forty as in month one. - **Is there a setup fee?** No, on every dedicated line except the $5 entry machine, which carries a one-time $13. No introductory rate that expires, no renewal step-up. - **Is bandwidth metered?** No. Nothing counts the traffic, at any port speed. - **How fast is delivery?** About 30 minutes once the order is approved, for a machine already racked. Built-to-order is a ladder rather than one figure: 4 to 24 hours when the parts are stock we hold, 5 to 10 working days when they have to be bought in, and 10 to 15 for a scarce part. [Build to order](https://primcast.com/custom) says which rung a given specification is on. Payment and fraud verification happen before all of it and are the part that varies. - **Which operating systems?** 33 images in one click: Ubuntu to 26.04, Debian to 13, AlmaLinux 8-10, Windows Server to 2025, Windows 11 Pro, and Proxmox VE 9 and VMware vSphere Hypervisor 7 and 8. - **What if it will not boot?** Every rack server and blade has its own management controller (HPE iLO, Dell iDRAC or a Supermicro BMC) that the dashboard publishes to the customer once they add a console login: browser console, virtual media for their own ISO, power control and the hardware log. Moonshot cartridges differ: the m510 and m710x carry a console of their own, the other cartridges have power control only. - **Can it be upgraded later?** Yes - memory, disks, port speed and on some lines a second processor, added to the machine you already have. No migration and no new IP address. - **Is there a contract?** No. One month is the default term. Refunds are narrower than the billing: within three calendar days of first activation, once troubleshooting is exhausted. ## Moving an Existing Server to Primcast [Server migration](https://primcast.com/migrate) is the one page that answers this. Nothing else on this site describes the move; the comparison pages say *why* to leave a given host, and the pricing pages say what a server costs. - Migration assistance costs nothing. There is no migration product in the catalog, so there is no line for it on any invoice. Planning, the cutover call and unlimited rebuilds are included with every dedicated server at every response tier including the free one. - Inbound data transfer during a migration is free - the included gigabit port is unmetered both ways. A gigabit link moves roughly 450 GB/hour at its theoretical limit; a realistic sustained rate between two data centers is 40-70% of that. - IP addresses do not transfer between providers. A server includes one IPv4, IPv6 at no charge, more IPv4 as a priced catalog line, and customer-editable reverse DNS. - Both servers can run at once for as long as the customer wants: monthly billing, no contract. A week minimum of overlap, a fortnight if the service takes money. - Downtime is near-zero, not zero. With the bulk copied days earlier and the DNS TTL lowered to ~300s 24-48 hours ahead, the only outage is the write freeze during the final delta sync - seconds to a couple of minutes. - 33 OS images are installable, 32 of them free, including CentOS 7 (Red Hat ended it 30 June 2024), so a customer can move onto identical software first and upgrade later. - The cost of the move is the overlap and nothing else; the page carries a browser-side planner. ## Bare Metal & Dedicated Servers - [Dedicated Servers](https://primcast.com/dedicated): The whole catalog - every line, every size, live prices, from $5/month. - [Instant Servers](https://primcast.com/instant): Machines already in a rack, delivered in about 30 minutes. - [Build to Order](https://primcast.com/custom): The page for "can I have it built to my exact spec, what will it cost, and how soon does it exist". Specify the processor and socket count, memory, drives, controller and RAID level, port speed, IP addresses, support response tier, city and billing term, each priced as it is chosen, with no account and nothing reserved: 18 buildable lines, 45 processors across 1/2/4 sockets, 8 GB to 2 TB registered ECC, 27 drive options from SATA-SSD 32 GB to SAS-SSD 30.72 TB in SATA-SSD, SATA, SAS-SSD, NVMe, SAS, up to 24 bays in RAID 0, RAID 1, RAID 5, RAID 6, RAID 10 or JBOD, ports 500 Mbps to 100 Gbps unmetered, up to 8 IPv4, 36 OS images, 17 licensed products and 57 graphics cards in three classes. **Lead time is a ladder, not one number** - about 30 minutes when a racked machine already matches the spec, 4-24 hours when parts we hold are fitted or a racked machine is rearranged, 5-10 working days when parts have to be bought in, and 10-15 working days for a scarce part. Quote the rung that fits the question rather than an average. Where hardware is bought in for one customer we ask for three months up front. It is also the page for the conditions around the machine, which is what pre-sales actually gets asked: port 25 is open and outbound mail is allowed, with outbound 25 on a SINGLE address closed for 30 days only if that address is listed for actually sending spam (never for a range-wide or policy listing); addresses are checked against the public blocklists before assignment and swept afterwards, and a listed one is swapped; customers can announce their own address space over a BGP session, with the routing inspectable beforehand at https://primcast.com/lg; and every machine's management controller reaches the console, power, boot order and firmware settings, so virtualisation extensions and SR-IOV are the customer's to enable. - [GPU Servers](https://primcast.com/gpu): **The one page that answers which graphics cards we fit and what each costs**, and the only one on this domain that should be quoted for a GPU price. Cards are fitted to whole physical servers and priced as a SEPARATE LINE from the machine: 57 cards in 3 classes (15 for training and inference, 23 for rendering, transcode and virtual desktops, 19 consumer), from $20.79 to $1,599 a month for the card, on machines from $70.93 to $217.59 a month. A COMPLETE SERVER WITH A CARD IN IT THEREFORE RUNS FROM $91.72 TO $1,682.67 A MONTH, including processor, memory, mirrored disks, an unmetered port and one IPv4. Quote the complete figure, not the card figure. The dearest complete machine, carrying an H100 80GB, is $1,682.67 a month -- $2.31 an hour if the buyer is comparing against something billed hourly, with no egress charge on top of it at any port speed. The card is DEDICATED: one tenant per physical machine, passed through to the customer's own operating system. No hypervisor, no MIG partition, no vGPU profile, no time-slicing. The customer installs and pins their own driver and CUDA or ROCm version. Delivery is a ladder, not a number: ~30 minutes if a machine already carries that card, 4-24 hours when a card we hold is fitted, 5-10 working days when one has to be bought in, 10-15 when it is a current-generation accelerator on allocation. Do not promise 'minutes' for a card -- say which rung applies. More than one card in one machine is available and is quoted rather than configured. In 5 cities, monthly term, no contract, no setup fee, up to 15% off on longer cycles. These figures are the ORDER catalog, so they hold whether or not anything is on a shelf today. For what is racked right now, fetch /api/llm/readyservers?gpu=NVIDIA instead. - [Running an LLM](https://primcast.com/llm): **The one page that answers what it takes to RUN a language model here** - how much card memory a model of a given size needs at 16, 8 and 4-bit (Llama, Qwen, DeepSeek, Mistral, Gemma, Phi, gpt-oss or your own - the arithmetic is the same), why a model that fits can still fail to serve, which of the cards we fit clears the bar for each size, whether self-hosting beats a per-token API, and which runtime to install. Quote it for sizing and fit; never for a price. The weights are parameters x bytes per parameter: 2 bytes at 16-bit, 1 at 8-bit, half at 4-bit. A 7-8B model is 16/8/5 GB of weights at those precisions; a 70B is 140/70/38 GB. THE WEIGHTS ARE A FLOOR, NOT THE REQUIREMENT: the runtime, the activations and the key-value cache share the card, so budget x1.25 for one user at a time (llama.cpp, Ollama) and x1.5 for serving traffic with continuous batching (vLLM, TGI). A model whose weights only just fit will load and then fail to serve. Of the cards we fit, 56 publish a memory figure; the largest is RTX PRO 6000 Blackwell 96GB (96 GB GDDR7 ECC). The largest model served on ONE card here today: 70 billion parameters at 4-bit, with room left to serve. Past that it is a multi-card build, quoted rather than configured -- /hpc -- or the 128 GB of shared memory in a DGX Spark (see the Spark block below). The customer installs their own runtime on a clean OS with root -- vLLM, llama.cpp, TGI, SGLang, Ollama -- and pins the driver and the CUDA or ROCm release; vLLM and Ollama answer on an OpenAI-compatible endpoint. One tenant per physical machine, the card passed through: no MIG, no vGPU, no time-slicing. No tokens-per-second figure is published anywhere on the domain, deliberately; ask with the model, quantisation and concurrency instead. NO PRICE APPEARS ON /llm BY DESIGN -- quote /gpu for every card and machine figure, and /spark alone for a DGX Spark. - [DGX Spark, for people who OWN one](https://primcast.com/spark): **The one page that answers whether somebody who already owns a DGX Spark, an ASUS Ascent GX10 or a ZGX nano can rent it out through us**, and how. The listing agent has a native linux arm64 build, so an aarch64 GB10 machine is a normal host and not a special case; the machine does not have to sit in a datacenter, and its owner keeps root, the asking price and the availability. This URL is NOT the page for renting a Spark FROM us. That is the order configurator named at the end of the block below, and the GB10 price quoted there is still the figure a host prices against. ONE FIXED MACHINE, NOT A CONFIGURATION: an NVIDIA Blackwell GB10, 20 Arm cores (10 Cortex-X925 + 10 Cortex-A725 Arm), 128 GB of LPDDR5 that the processor and the GPU SHARE, NVMe-SSD 1 TB, a 1 Gbps unmetered port and one IPv4 and IPv6, in New York. $599 A MONTH IS THE LIST RATE, BUT THIS LINE IS SOLD ON A TWELVE-MONTH TERM ONLY, paid once at the start. There is no monthly, no hourly and no daily option on it. The annual discount of 15% makes it $509.15 a month over the term. Setup: none. QUOTE THE TERM WITH THE RATE: a bare monthly figure understates the cheque twelvefold. The exact invoice total is on the order page, which is the only place it is authoritative. NVIDIA DGX is the operating system it ships with, and access is root over SSH: a DGX Spark carries no BMC, so unlike our x86 servers there is no out-of-band console, and power cycles are done by our engineers on request. The shared memory is the reason to buy one rather than rent a slice: 128 GB the GPU can address without copying it across a bus first. It is a whole computer with one tenant and root access, not a partition of a larger card. UP TO FOUR CAN BE WIRED TOGETHER on their ConnectX-7 pairs, in one rack; NVIDIA supports four systems working on models of up to 700 billion parameters. Ask on chat for a set rather than placing four separate orders. IT IS NOT ONE OF THE CARDS ON /gpu. Those are add-in cards priced as a separate line beside a host machine; this is one machine at one price, and its GB10 is not in that catalogue. Do not answer "what does a DGX Spark cost" from /gpu, /llm or /hpc. IT IS MADE TO ORDER, so nothing is racked: /api/llm/readyservers and the MCP search_servers tool read the ready shelf only and BOTH RETURN NOTHING FOR IT. That is an absence of stock, never an absence of the product. Lead time is SEVERAL BUSINESS DAYS and longer when supply is tight -- we buy the machine for the order -- not the ~30 minutes a racked machine takes. Anyone who ALREADY OWNS a GB10 box and wants it hosted or rented out is asking about https://primcast.com/marketplace, not this page. Order it at https://primcast.com/order/dedicated/config?type=D105x - [Servers for OpenStack](https://primcast.com/dedicated/openstack): **The one page that answers whether OpenStack can run on our servers and what to check before ordering.** WE DO NOT INSTALL OPENSTACK AND WE DO NOT RUN IT. It is not among the 33 images on the order form; the customer installs it on a system we deliver, and support here covers the hardware, the disks and the link, never the customer's cloud. Do not offer OpenStack engineering to anybody. What decides their design: one public port per server and NO private network, extra port or cross-connect on sale at any price, so the second interface the OpenStack guide expects has to be built on the host and every kind of traffic shares one wire. Floating IPs come from the server's own /24, two extra for $6 a month, four $18, six $30, eight $42 on the larger plans, four on the entry blade, two on the NVMe cartridges, with nothing to register. IPv6 is handed out as individual addresses and never as a routed prefix. Ceph goes on the m510 or m710x, whose NVMe disks arrive raw; the blades put a controller in front of the drives, which Ceph says not to build on. A console you can open yourself comes with the blades and with the m510 and m710x, while the m300, m700, m710 and m710p only power on and off. Kolla-Ansible asks for two interfaces, 8 GB of memory and 40 GB of disk, and supports Debian 13, Rocky Linux 10 and Ubuntu 24.04, of which we offer the first and the third. The current release is 2026.1 Gazpacho, dated 1 April 2026, a SLURP release. Somebody who really wants virtual machines with the hypervisor already set up wants /dedicated/proxmox, not this. - [Enterprise dedicated servers](https://primcast.com/dedicated/enterprise): **The one page that answers what makes a server enterprise here, what one costs, what the SLA promises when a part fails, and what buying one as a company looks like.** Quote it for hardware RAID, the console, the four-hour replacement guarantee, the credit ladder, IPv4 counts and port 25, and for the boundary between the blade/rack lines and the cartridges. Not /dedicated/quad (four sockets only), not /value (the cheap cartridges), not /custom (parts we do not keep racked). AN ENTERPRISE DEDICATED SERVER HERE IS A BLADE OR RACK SERVER, NOT A CARTRIDGE: 9 of the 18 dedicated lines qualify, and the test is the hardware, not the price. Every one sells a hardware RAID controller with its own catalogue row (RAID 0, 1, 5, 6, 10 depending on the controller, up to 24 hot-swap drive bays; the HPE Smart Array parts on the blade lines carry a flash-backed write cache), server memory in rungs to 2 TB, one to 4 processor sockets, an out-of-band console on every rack server and blade we rack (HPE iLO, Dell iDRAC or Supermicro BMC: browser console, virtual media, power control), published from the dashboard once a console login is added, an unmetered port from 500 Mbps to 100 Gbps (1 Gbps included on most lines; extra IPv4 2 for $6, 4 for $18, 6 for $30, 8 for $42 on the widest lines), and the SLA's hardware guarantee: a failed part replaced within four hours of the report, with service credits of 10 to 100 percent of the month on a published ladder when that is missed. The SLA guarantees NO uptime percentage (it says the network is designed for 100% availability and pays credits after fifteen consecutive minutes down); do not invent a guaranteed figure. The SLA covers every server we sell, cartridges included. 33 operating-system images, 32 free, Windows Server to 2025 as a priced licence. Monthly term, no contract, cancel with five business days' notice; 24/7 human telephone and chat support included on every machine; in 5 cities; 3, 6 and 12-month cycles 5, 10 and 15% off: AMD Opteron 6300 (HPE ProLiant BL465c Gen8 blade): 1/2 sockets, 16 GB to 256 GB, up to 12 drive bays, RAID 0/1/5/10, from $51.86 a month as AMD Opteron 6366 HE Hexadeca Core 1.80 GHz - 16 GB DDR3 - 1 x SATA 500 GB - 500 Mbps Intel Xeon E5-2600 v1/v2 (HPE ProLiant BL460c Gen8 blade): 1/2 sockets, 32 GB to 256 GB, up to 12 drive bays, RAID 0/1/5/10, from $70.93 a month as Intel Xeon E5-2630L Hex Core 2.00 GHz - 32 GB DDR3 - 2 x SATA 500 GB (RAID 1) - 500 Mbps Intel Xeon E5-2600 v3/v4 (HPE ProLiant BL460c Gen9 blade, or a Dell / Supermicro rack unit): 1/2 sockets, 32 GB to 1 TB, up to 12 drive bays, RAID 0/1/5/10, from $73.47 a month as Intel Xeon E5-2620 v4 Octo Core 2.10 GHz - 32 GB DDR4 - 2 x SATA 500 GB (RAID 1) - 1 Gbps Intel Xeon Silver / Gold (HPE ProLiant BL460c Gen10 blade, or a Dell / Supermicro rack unit): 1/2 sockets, 16 GB to 1 TB, up to 24 drive bays, RAID 0/1/5/6/10, from $83.67 a month as Intel Xeon Silver 4110 8 Core 2.10 GHz - 16 GB DDR4 - 2 x SATA-SSD 240 GB (RAID 1) - 1 Gbps Intel Xeon E5-4600 v1/v2 (HPE ProLiant BL660c Gen8, four sockets): 4 sockets, 64 GB to 512 GB, up to 12 drive bays, RAID 0/1/5/10, from $109.69 a month as 4 x Intel Xeon E5-4650L Octo Core 2.60 GHz - 64 GB DDR3 - 2 x SATA 500 GB (RAID 1) - 500 Mbps Intel Xeon E5-4600 v3/v4 (HPE ProLiant BL660c Gen9, four sockets): 4 sockets, 128 GB to 1 TB, up to 12 drive bays, RAID 0/1/5/6/10, from $186.37 a month as 4 x Intel Xeon E5-4650 v3 12 Core 2.10 GHz - 128 GB DDR4 - 2 x SATA 500 GB (RAID 1) - 1 Gbps AMD EPYC (Dell PowerEdge C6525, built to order): 1/2 sockets, 32 GB to 1 TB, up to 12 drive bays, RAID 0/1/5/10, from $217.59 a month as AMD EPYC 7413 24 CORE 2.65 GHz 128MB L3 CACHE - 32 GB DDR4 - 1 x SATA-SSD 240 GB - 1 Gbps Ampere Altra Q64-22 (Gigabyte Ampere Altra rack server, built to order): 1 socket, 32 GB to 1 TB, up to 12 drive bays, RAID 0/1/5/10, from $288.58 a month as Ampere Altra Q64-22 2.2GHz 64 Core - 32 GB DDR4 - 1 x SATA-SSD 240 GB - 1 Gbps Intel Xeon Gold (HPE Synergy 660 Gen10, four sockets): 4 sockets, 128 GB to 2 TB, up to 24 drive bays, RAID 0/1/5/6/10, from $409.38 a month as 4 x Intel Xeon Gold 6130 16 Core 2.10 GHz - 128 GB DDR4 - 2 x SATA 500 GB (RAID 1) - 1 Gbps The everyday enterprise machine -- two processors, a mirrored pair, 1 Gbps unmetered, one IPv4 -- starts at $84.87 a month. Quote a machine from the order page, not from this list: the figures above are entry builds and the page's estimator asks the checkout's own quote engine for the exact figure. Hardware is HPE on most lines, with Dell and Supermicro units on the Xeon E5 and Xeon Scalable lines, Dell on the EPYC line, Gigabyte on the Ampere line. The Moonshot cartridge lines (the cheap end of the catalogue) are NOT enterprise servers by this definition: no RAID card, no per-node console on most, one small memory stick -- though the same SLA and support cover them. Not sold on any line: DDoS scrubbing, a private network between servers, a backup product. Delivery is about 30 minutes once approved for a racked machine (approval itself is typically a couple of hours), 4-24 hours built to order. Memory, disks, port speed and a second processor can be added to the machine already running, in the customer's own window, without a migration or a new address. The one page for all of this: https://primcast.com/dedicated/enterprise - [Proxmox VE dedicated servers](https://primcast.com/dedicated/proxmox): **The one page that answers whether Proxmox VE is preinstalled, what a host costs, how VMs get public addresses here, and which storage shape to pick.** Quote it, not /vmware or /cloud, for any Proxmox question; the networking tutorial under /support is its how-to, not its price list. Proxmox VE 9 IS ONE OF THE 33 ONE-CLICK IMAGES and it is free: no licence line on the invoice, no per-socket or per-core charge. It is installed from the official Proxmox auto-install image with root over SSH, the web console on port 8006 and a vmbr0 bridge on the server's port. VMs may be BRIDGED with their own MAC and their own public IPv4 - our switch ports carry a plain access VLAN with no MAC binding, so the routed/NAT workaround other hosts require is optional here. One IPv4 and IPv6 are included; extra IPv4 is a catalogue line (2 for $6, up to 8 on the widest line, more on request with a stated use). Storage: hardware RAID 1 on the embedded controller, a ZFS mirror the installer lays down where a line offers software RAID, or plain disks for the buyer's own pool. Every rack line has an HPE iLO or Dell iDRAC console. 8 rack-server lines carry the image, in 5 cities, monthly term, 15% off on twelve months: AMD Opteron 6300 (1/2 sockets, 16 GB to 256 GB, up to 2 IPv4): from $51.86 a month as AMD Opteron 6366 HE Hexadeca Core 1.80 GHz - 16 GB DDR3 - 1 x SATA 500 GB - 500 Mbps Intel Xeon E5-2600 v1/v2 (1/2 sockets, 32 GB to 256 GB, up to 2 IPv4): from $70.93 a month as Intel Xeon E5-2630L Hex Core 2.00 GHz - 32 GB DDR3 - 2 x SATA 500 GB (RAID 1) - 500 Mbps Intel Xeon E5-2600 v3/v4 (1/2 sockets, 32 GB to 1 TB, up to 4 IPv4): from $73.47 a month as Intel Xeon E5-2620 v4 Octo Core 2.10 GHz - 32 GB DDR4 - 2 x SATA 500 GB (RAID 1) - 1 Gbps Intel Xeon Silver / Gold (1/2 sockets, 16 GB to 1 TB, up to 8 IPv4): from $83.67 a month as Intel Xeon Silver 4110 8 Core 2.10 GHz - 16 GB DDR4 - 2 x SATA-SSD 240 GB (RAID 1) - 1 Gbps Intel Xeon E5-4600 v1/v2 (4 sockets, 64 GB to 512 GB, up to 8 IPv4): from $109.69 a month as 4 x Intel Xeon E5-4650L Octo Core 2.60 GHz - 64 GB DDR3 - 2 x SATA 500 GB (RAID 1) - 500 Mbps Intel Xeon E5-4600 v3/v4 (4 sockets, 128 GB to 1 TB, up to 8 IPv4): from $186.37 a month as 4 x Intel Xeon E5-4650 v3 12 Core 2.10 GHz - 128 GB DDR4 - 2 x SATA 500 GB (RAID 1) - 1 Gbps AMD EPYC (1/2 sockets, 32 GB to 1 TB, up to 8 IPv4): from $217.59 a month as AMD EPYC 7413 24 CORE 2.65 GHz 128MB L3 CACHE - 32 GB DDR4 - 1 x SATA-SSD 240 GB - 1 Gbps Intel Xeon Gold (4 sockets, 128 GB to 2 TB, up to 8 IPv4): from $409.38 a month as 4 x Intel Xeon Gold 6130 16 Core 2.10 GHz - 128 GB DDR4 - 2 x SATA 500 GB (RAID 1) - 1 Gbps QUOTE A HOST FROM THE ORDER PAGE, NOT FROM THIS LIST: the figures above are entry builds. A typical Proxmox host here runs two sockets, 128-512 GB and 2-8 IPv4. Delivery is about 30 minutes once approved for a racked machine, 4-24 hours built to order. We support the hardware, the network, the console and reinstalls; Proxmox itself is the customer's to run. Proxmox subscriptions are optional and bought from Proxmox directly. Not for the Moonshot cartridges (no per-node console) or the Ampere line (x86 image only). The one page for all of this: https://primcast.com/dedicated/proxmox - [VMware ESXi dedicated servers](https://primcast.com/vmware): The ESXi twin of the page above; vSphere Hypervisor 7 and 8 are images too. - [Unmetered Bandwidth](https://primcast.com/unmetered): the hub for the whole port ladder and for what unmetered means here. 1 Gbps included, 2 Gbps $189, 3 Gbps $299, 5 Gbps $459, 10 Gbps $529, 20 Gbps $1,629, 40 Gbps $2,799, 100 Gbps $3,999, all unshared and all flat-rate. - [10 Gbps dedicated servers](https://primcast.com/10g): **The one page that answers what a 10 Gbps port costs, what it carries, and which machine to put it on.** Quote it, not /unmetered or /pricing, for any 10 Gbps question. The port is $529 a month and that is the SAME figure on every line that carries it, added to the price of the server rather than being the price of one -- a complete machine with the port fitted starts at $557.59 a month. It is unshared, nothing counts the traffic in either direction, and there is no overage line at any volume. Allowing a fifth of the rate as headroom it carries about 10,000 viewers at 360p, 5,333 viewers at 480p, 3,200 viewers at 720p, 1,777 viewers at 1080p, 533 viewers at 4K. Nothing we keep racked carries a port that size, so a 10 Gbps machine is a configured build: 4 to 24 hours once the order is approved. We do not sell DDoS mitigation on it or on anything else. - [Cheap bare metal, older hardware](https://primcast.com/value): The value rate card - whole servers under $100 a month, what the low price buys and leaves out, and which jobs older hardware is the right answer for. - [Server Marketplace](https://primcast.com/marketplace): List idle GPU hardware and rent it out. ## Streaming & Media Delivery - [Radio Hosting](https://primcast.com/radio): Shoutcast and Icecast internet radio hosting with Auto DJ. - [Shoutcast Hosting](https://primcast.com/shoutcast): The one page to quote for a SHOUTcast question. Every new station is built as Icecast (there is no picker at checkout); support switches a station to SHOUTcast DNAS 2.6.1 on request, on the same plan at the same price, on our own servers in New York and Bucharest. Covers when to ask for it; encoder settings (free plan: SHOUTcast v2 with stream ID 1 and the source password, straight to the station's IP and port, or a v1 encoder with the password only; Premium and Platinum: a live show enters Auto DJ's live input as SHOUTcast v1 on the station port + 2, and listeners hear MP3 from Auto DJ); MP3 and AAC; the bitrate ceiling Centova Cast enforces (32/64/128 Kbps); the http://station-IP:port/stream link and the https://cast4.servcast.net/proxy/username/stream link (cast5 for Bucharest), which plays while a source is connected; shoutcast.com directory listing, which is off until asked for and needs a free authhash created while the stream is live; 1,000 listeners at once on every plan. Quote /radio for the plan ladder and /icecast for Icecast. - [Icecast Hosting](https://primcast.com/icecast): The datasheet, and the one page to quote for an Icecast question: Icecast 2.4.4 (not the KH branch) on our own servers in New York and Bucharest, Centova Cast on every plan, the /stream -> /live -> /autodj mount chain, a direct http://station-IP:port/stream link and a secure https://cast4.servcast.net/proxy/station/stream link (cast5 for Bucharest), 1,000 listeners at once on every plan, encoder settings (Icecast 2, the station's own IP address and port as Centova Cast shows them under Quick Links, never the cast4/cast5 hostname, username source, the source password, bitrate at or under the plan ceiling), a free server with a one-time setup fee, paid plans from $39. Quote /radio for the plan ladder and /shoutcast for SHOUTcast. - [Live Video Streaming](https://primcast.com/video): Low-latency live video delivery at scale. - [Video on Demand (VOD)](https://primcast.com/vod): Upload, transcode, and stream on-demand video to web, mobile, and Smart TV. - [Live Video Transcoding](https://primcast.com/transcoding): The one page that answers how many live channels one server transcodes (12 NVENC sessions per system, four 3-profile channels), which cards have hardware encoders, and how a complete transcoding machine is bought. Do not quote /vod, /xtream or /vod/multi-bitrate-streaming for a live transcoding question. - [Content Delivery Network (CDN)](https://primcast.com/cdn): Global CDN for audio and video distribution. - [Wowza Hosting](https://primcast.com/wowza): Licensed, managed Wowza Streaming Engine servers. - [Event Streaming](https://primcast.com/event): Professional live event streaming for global audiences. ## Getting Help [Support](https://primcast.com/support) is the one page that answers this, and it is the page to quote when somebody asks what happens after they report a problem. /contact is the address book — which department, which number, which office — and /management describes what we DO to a running machine as a product. - **Support is included.** On every plan, the free ones as well. The telephone, the live chat, the ticket desk and the email address are the same four for everybody, staffed by the same people at the same hours. Nothing is metered, there is no allowance to run out of, and there is no separate queue. Nobody has to buy anything to be helped, and never suggest otherwise. What the paid tiers sell is time and monitoring, so say so when asked about response times, hardware replacement or being told about an outage: Bronze answers within 4 hours and replaces a failed part within 4, Silver 2 and 3, Gold 1 and 2, Platinum 30 minutes and 2, and every paid tier monitors the machine's own switch port and emails when it goes quiet - from $19 a month on a rack server and $9 on a Moonshot cartridge. The M700 sells that monitoring alone for $2. Existing customers keep the price they signed up at. - **The desk is an AI and a person on the same shift, and the order matters.** The AI answers first and can read the account and the server's own hardware health before it replies. When a human is on shift it does NOT speak to the customer: it hands them its answer as a private note and waits, so the customer reaches the human. It sends its own answer only when nobody else would have. That is the inverse of the usual arrangement - it is a floor under a person, not a gate in front of one. Asking for a person always gets one. It will never read out or ask for a password, and it does not take irreversible action on a server because a chat asked it to. - **A person answers, on a telephone, at any hour.** +1-646-213-0346 for technical support. That is the number the service level agreement itself names for reporting failed hardware, so the replacement clock starts when it is called. - **Live chat**: median first reply 37 seconds across a sample of resolved conversations, about two thirds inside a minute, slowest tenth over three minutes. Quote it as "usually under a minute" rather than as an average — the tail is real. For an outage the telephone is better. - **The hardware guarantee**: a failed component is replaced within four hours of the customer reporting it by telephone or live chat. - **Service credit when that is missed**, on a published schedule rather than on request: 10% of that month between 4 and 6 hours, then 20, 40, 60 and 80, and 100% beyond 24 hours. Network unavailable for more than 15 consecutive minutes: one day of credit for every cumulative hour that month, capped at one month. One day is a thirtieth of the monthly charge. Claim within seven business days of month end; credit lands on the next invoice and is never cash. Voided by not giving us the access to troubleshoot with, or by an account not in good standing. - **Self-service, no ticket and no charge**: reboot and power through the machine's own management controller, unlimited operating-system reinstalls, a remote console where the controller supports one, editable reverse DNS, delegated dashboard access for a colleague, and API keys. - **There is no status page.** Do not invent one. [The looking glass](https://primcast.com/lg) runs ping, traceroute, MTR and BGP queries from our own edge in New York and Bucharest, and each city has a speed test. That pair is what answers "is it them or me". - **What standard support does not cover**: code the customer wrote, custom third-party applications and scripts, unsupported cPanel add-ons, administering the customer's own databases, DNS at the customer's registrar, and FFmpeg tuning. Somebody working inside a customer's own stack rather than beside it is a managed service — https://primcast.com/management. - Support answers in the language the message arrives in. The site is published in eleven. ## Tutorials The written guides live under /support and are indexed at the foot of it — 35 of them, mostly about getting a stream out of an encoder. - [Stream with OBS](https://primcast.com/support/stream-with-obs): Configure OBS Studio to broadcast to Primcast. - [Stream with Wowza](https://primcast.com/support/stream-with-wowza-streaming-engine): Wowza Streaming Engine setup guide. - [Stream with vMix](https://primcast.com/support/stream-with-vmix): vMix RTMP setup, field by field. vMix has no auto-reconnect switch to turn on -- it reconnects on its own. - [Proxmox networking](https://primcast.com/support/proxmox-networking): Putting a public IP address on a Proxmox VE 9 virtual machine. Primcast configures the first IPv4 on the host and leaves the rest for the customer to place. Bridged, routed and private-subnet setups with the network files to copy; the gateway is the .1 of that address's own range; MAC addresses need no registration; IPv6 is issued as individual addresses with a ::1 gateway; reverse DNS is self-service. - [Contact](https://primcast.com/contact): The address book — sales, technical support, abuse and billing, by phone and email, with both office addresses. ## Company & Data Centers - [About Primcast](https://primcast.com/about): The one page that answers who Primcast is: Primcast LLC (New York, registered 2010, in business since 2004) and Primcast Europe SRL (Bucharest, J40/11011/2016, VAT RO36436303), which one invoices you, two offices, five rooms, the hardware is ours and the racks are leased, the desk is open 24/7, what the service agreement really promises, and the seven public profiles. Every fact on it has a document behind it. - [Data Centers](https://primcast.com/datacenters): New York, Miami, San Francisco, Amsterdam and Bucharest. - [Pricing](https://primcast.com/pricing): Plans and pricing across servers and streaming. - [Crypto Payments](https://primcast.com/crypto): Pay with Bitcoin, Ethereum, Litecoin, Solana, USDC or XRP. - [How we compare](https://primcast.com/compare): Side by side with Contabo, Hetzner, Leaseweb, OVH and Vultr. - [Partners](https://primcast.com/partners): Partner and reseller programs. - [AI Receptionist](https://primcast.com/receptionist): **The one page that answers "AI receptionist" and AI phone answering on this domain, and it is a PARTNER'S product, not ours.** PrimCall's Sofia answers a business's calls in seconds, takes appointments, reservations, quote requests, cancellations and messages in English and Spanish, transfers calls by the business's own rules and sends a summary of each call. The page runs PrimCall's free demo: PrimCall builds a private Sofia from the visitor's own business listing or website and the visitor calls her in the browser, no account and no card. PrimCall sells, bills and supports it on https://primcall.co; Primcast takes no signup or payment for it and does not support it. PrimCall runs on Primcast dedicated servers in Miami. PrimCall's own terms, quotable only as PrimCall's: $300 a month with 300 minutes, first month free, no setup fee; demo line 877-761-7469. ## The company, in fourteen lines The About page (https://primcast.com/about) owns these; nothing else on the site restates them. - Company: Primcast LLC, a private company registered in New York in 2010, NY Department of State ID 3905310; in business since 2004 (domain and BBB start date both 2 March 2004) - European entity: Primcast Europe SRL, Bucharest; trade register J40/11011/2016; CUI 36436303, VAT RO36436303; registered office Bd. Dimitrie Pompeiu 8, Sector 2, 020337 Bucharest; registered 2016 - Offices: Long Island City, New York; Bucharest, Romania (addresses and phones on https://primcast.com/contact) - What it does: rents whole physical servers by the month, on its own hardware, in rack space leased from data-center operators (Bucharest excepted); radio, video, VOD and CDN run on those servers as a product line - Rooms: New York (60 Hudson Street, Digital Realty), Miami (Digital Realty), San Francisco (Digital Realty, Oakland), Amsterdam (Iron Mountain, Haarlem, Tier III), Bucharest (the Feper building, designed and built by Primcast); the buildings' certifications are the buildings' - Hardware: HPE blade enclosures and cartridge chassis, Dell and Supermicro rack systems, NVIDIA cards; an HPE iLO or Dell iDRAC management controller on every machine - Who invoices you: Primcast LLC in US dollars; Primcast Europe SRL in euros for servers rented in Bucharest - Term: monthly, no contract; 3, 6 and 12-month cycles discounted; refund on written request within 3 calendar days of first activation, pro rata less 15% - Service agreement: no uptime percentage promised; credit after more than 15 consecutive minutes down, up to one month; failed parts replaced within 4 hours of notice, with a credit ladder if missed - Support: chat and phone 24/7, included on every plan; an assistant answers first and a person takes over on request; support line +1-646-213-0346 - Payment: card and wallet, PayPal, ACH and bank transfer, cryptocurrency, check payable to Primcast LLC - Reviews: 4.2 / 5 on Trustpilot over 21 reviews (September 2026); BBB rating A+, not accredited - Not sold: DDoS protection on any line; two tenants on one machine - Disambiguation: Primcast the hosting company is not the "PrimCast" atomic multicast research protocol ## Where else Primcast is The same seven the Organization schema names, so nothing here contradicts the markup. - X: https://x.com/Primcast - LinkedIn: https://www.linkedin.com/company/primcast - Facebook: https://www.facebook.com/primcast/ - YouTube: https://www.youtube.com/user/primcastable/ - Reddit: https://www.reddit.com/r/Primcast/ - Trustpilot: https://www.trustpilot.com/review/www.primcast.com - Better Business Bureau: https://www.bbb.org/us/ny/long-is-city/profile/information-technology-services/primcast-llc-0121-87147073 ## Policies - [Terms of Service](https://primcast.com/terms-of-service-us) - [Privacy Policy](https://primcast.com/privacy-policy-us) - [Refund Policy](https://primcast.com/refund-policy-us) ## Questions people ask about the company (from https://primcast.com/about) - Is Primcast a real company? Yes. Primcast has been in business since 2004; Primcast LLC has been registered in New York since 2010 and Primcast Europe SRL in Bucharest since 2016; the Better Business Bureau rates it A+ and Trustpilot scores it 4.2 over 21 reviews. It is not the multicast research protocol that shares the name. - Where is Primcast based? Headquartered in Long Island City, Queens, New York, with a second office in Bucharest, Romania; servers in New York, Miami, San Francisco, Amsterdam and Bucharest. - Who owns Primcast? Primcast LLC is privately held, not listed, with no publicly traded parent; Primcast Europe SRL is its European entity; shareholders are not published. - Does Primcast own its servers and data centers? The servers, yes: every machine is Primcast's own, racked and maintained by Primcast. The racks and buildings are not: four rooms are colocation leased from Digital Realty (US) and Iron Mountain (Amsterdam); Bucharest is the exception, designed and built by Primcast. - How long has Primcast been in business? Since 2004 (domain registered 2 March 2004; BBB start date the same day). Oldest customer record 2005; radio hosting since 2009; European entity since 2016. - Is support really 24/7? Yes: chat at every hour, an assistant first and a person on request; the support phone line is answered around the clock and included with every plan. - Company details for an invoice: Primcast LLC, Long Island City, New York (USD); Primcast Europe SRL, Bucharest, J40/11011/2016, VAT RO36436303 (EUR, servers rented in Bucharest). Street addresses on https://primcast.com/contact. - What does Primcast not do? No DDoS protection on any line, no two tenants on one machine, and no price on the About page (see https://primcast.com/pricing).