Which AMD server does the most work for the money?
The EPYC, by a wide margin. 2 × EPYC 7513 costs $2.55 for every thousand PassMark points ($237.59 a month, CPU Mark 93,312). The sixteen-core Opteron costs $9.24 for the same thousand. The Opteron wins on one thing only, the count: it is the least expensive way to have sixteen or thirty-two cores of your own. If the work needs throughput, the EPYC is the cheaper machine per unit of work although its monthly price is higher. The list sorts by this figure.
Is the EPYC 7003 series still a good server processor?
Yes. AMD launched it on March 15, 2021 and has released newer generations since, which we do not build. A 7003 processor has up to 64 cores with two threads each, eight DDR4 memory channels and AVX2, and is on Broadcom’s compatibility guide through ESXi 9. What you give up against AMD’s newest is DDR5 and the highest core counts per socket. What you keep is the price.
AMD EPYC or Intel Xeon for a dedicated server?
By the job. Choose AMD here for the most cores in one machine, for the most cores per dollar, or for a graphics card host that needs many threads. Choose Intel for video encoding on the processor’s own graphics, for four processors on one board, or for the widest choice of models, and for San Francisco and Amsterdam, where we build no AMD line. The estimator tells you each time whether another maker reaches the same cores and memory for less.
How long does an EPYC server take to deliver, and can I get a refund?
It is built after you order. With parts we hold, 4 to 24 hours from the moment the order is approved; a part that has to be bought in takes 5 to 10 working days, and build to order shows which applies. A machine built to order cannot be refunded. An Opteron server that is already racked is yours roughly 30 minutes after we approve the order, and has a three-day refund window from activation: ask in writing and the unused days come back, less 15%. The wording is in the refund policy.
Does the EPYC line take graphics cards?
Yes. The EPYC line is the AMD line that takes graphics cards, with 10 options at checkout: RTX 4090D, RTX 5090, RTX 3080, L4 ADA, L40S, A100 40GB, H100 80GB, AMD Instinct MI210, A100 80GB and RTX PRO 6000 Blackwell 96GB. Each card is priced separately from the server on GPU servers. The Opteron lines take none.
What does Windows Server cost on 64 or 128 cores?
The license is counted per core, so it grows with the processor. Our checkout bills Windows Server Standard at a flat $20.79 a month up to eight cores and per core beyond: $41.44 on sixteen cores, $165.76 on 64 and $331.52 on 128, which is more than the 128-core machine itself. Microsoft’s own floor is eight core licenses for each processor and sixteen for each server. On a per-core license, the eight-core EPYC 72F3 is the cheap one to run, and AMD itself lists that part for databases licensed by the core. How Microsoft counts cores is set out in its Windows Server licensing guide.
Will Proxmox, VMware or a current Linux run on them?
Proxmox VE 9 and current Ubuntu and Debian install on all three lines. Version 10 of Red Hat Enterprise Linux needs a processor at the x86-64-v3 level, which the two Opteron lines do not reach; AlmaLinux publishes a separate x86-64-v2 build of version 10 for such hardware, and version 9 runs as it is. VMware is the one to check before you order: Broadcom’s compatibility guide lists the EPYC 7003 for ESXi 7.0 U1 through 9, and both Opteron families for ESXi 7 only, so for vSphere 8 take EPYC. Windows Server goes up to 2025 on the two rack lines and 2022 on the cartridge.
Does anything count the traffic on an AMD server?
No. The port belongs to your server alone at the access switch, and no meter sits on it: there is no allowance to use up and no overage to pay. The two Opteron lines include 500 Mbps and the EPYC line includes 1 Gbps. The rack lines go up to 20 Gbps; for the cost of each step see unmetered servers, or 10 Gbps servers for that port alone.
Is there DDoS protection, or an uptime guarantee?
No to both, and we would rather say so than leave it blank. We sell no DDoS protection on any server. The service agreement promises no uptime percentage; what it promises is credit, a day back for every hour the network is down once an outage passes fifteen minutes, to a limit of one month. Support is by chat and telephone at any hour on every plan.
Where are the AMD servers located?
New York and Bucharest have all three lines. Miami has the Opteron 6300. San Francisco and Amsterdam have no AMD line; servers there are on the Intel page. You can test a route to any city first with the looking glass.
How do I know which AMD server is racked today?
Not from this page, on purpose: it lists what the checkout sells, which is the same on a full shelf and an empty one. EPYC is never racked ahead of an order, so the question does not apply to it. For the Opteron lines, the units sitting in a rack right now are listed on instant servers; any build not among them is assembled for you.
What is the catch with the $5 server?
There are six, and none is hidden. It is real hardware, one node of a cartridge chassis with its own processor, memory and drive, not a virtual machine; the catches are in what that hardware is. That price is the slowest of three clock settings (1.1 GHz at $5, 1.5 GHz at $10 and 1.8 GHz at $15). The memory is 8 GB and cannot be raised. It has one drive bay and no remote console. The processor is slow. And it carries a one-time $13 setup. In return it is a whole x86 machine of your own with an unmetered 500 Mbps port, root access and the same support as every other server.
Can I add a second processor or more memory later?
Memory, drives and port speed can be raised on the machine you already have, with no migration and no new address. On the two rack lines a second processor can be added too. The cartridge is the exception: its memory and its single drive bay are fixed.
