Why put it in a data centre
Power, heat and noise
A card at full load for a month is the part of mining that domestic wiring, domestic cooling and the people you live with all object to. Here it is a rack, and the monthly figure covers the machine running flat out rather than idling.
Network
Your server sits on a network built for low-latency streaming, and nothing on it is metered: no transfer allowance to exceed, no egress charge, and no overage line on the invoice at any port speed we sell.
The machine is yours
One tenant, root, and the card handed straight through to your own operating system — your driver, your CUDA, your miner. No hypervisor, no shared card, and the server’s own iLO or iDRAC console for when a driver takes the network down.
Somebody in the building
Support answers by telephone and live chat around the clock, and a failed card is replaced under an agreement that puts four hours on it from the moment you report it. The response tiers and the service credits are on the support page.
NVIDIA CMP 170HX
NVIDIA’s CMP line was the only silicon it ever sold that cannot do anything but compute: no display connector on the bracket, no driver support for graphics, no resale into a gaming machine afterwards. The 170HX is the top of that line and it is a cut-down data-centre die rather than a rebadged consumer card, which is why its memory is stacked HBM2e instead of GDDR. Everything that makes it a poor buy for rendering, transcode or a virtual desktop is the reason it is on this page and not on the general catalogue.
Specifications
$189 — the card, per month
$259.93 — the whole server with it fitted, per month
Intel Xeon E5-2600 v1/v2 · Intel Xeon E5-2630L Hex Core 2.00 GHz · 32 GB DDR3 · 2 × SATA 500 GB (RAID 1) · 500 Mbps
Compatible with Linux, Windows, KVM, VMware ESXi and Citrix XenServer.
What we do and do not promise
We do not forecast returns
What a card earns depends on the coin, the algorithm, the difficulty and the exchange rate on the day, and none of those is ours to predict. You will not find a hash rate or a payback period on this page, because any figure we printed would be out of date before you read it.
We do not take a cut
You rent the machine at a monthly figure and keep everything it produces. There is no revenue share, no pool we route you through and no wallet of ours in the middle. If you want the opposite arrangement — hardware you already own, earning without you finding the tenants — that is the marketplace, not this page.
We do not run the miner
The machine arrives with a clean operating system and root. Which miner, which pool and which driver version are yours to choose and yours to keep pinned; nothing upgrades underneath you. Say so on the order if you would rather the GPU driver were installed before handover.
More than one card
Possible, and it is a build we quote rather than a box you tick — what fits depends on the chassis, the physical width of the card and the power budget. Ask, and the reply names the chassis, the count and the lead time.
Why Primcast?
Bare metal since 2004, on a monthly term with no contract and no price step at renewal. Mining has been an ordinary workload here and always has been, which in practice mostly means nobody is going to close the account over what the machine is doing.
Where to go next
This page prices one part. The GPU servers page is the whole catalogue: what each card adds per month, the machine underneath it, and which platform gives it which PCIe generation.
This page shows no stock count on purpose — a card is a part we fit, not a product that comes and goes. Instant servers is the list of machines genuinely standing in a rack this minute.
The marketplace is the opposite trade. You keep the hardware and list it; we verify it, put it in front of the demand and handle the billing, and you set the asking price. It is where a rig that has stopped paying for itself at home can go.
Card, PayPal, ACH and bank wire all work, and so does cryptocurrency. The crypto payments page lists which coins are accepted and what clears fastest on a first order.
