New on LowEndTalk? Please Register and read our Community Rules.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Comments
The cheapest way to do this is colo.
Grab cheap E5 ES CPUs off ebay, build a dual-socket rig, and colo at a dirt cheap rate.
Will pay itself off in a year or two.
try E3 for good single threaded performance. There are few providers who offer high core box but if you want cheapest, then search for High Core, Low RAM server on google.
I know the cheapest is colo but this is for a client and I don't want to bill to build the server (I'm a software developer not an IT guy). Also he wants scalability so he wants to be able to switch boxes easily, and colo won't allow that.
I'd personally avoid clients that sounds too needy.
You can build a monster rig under a grand this way. If a client wants more than that, he will need to colo his own machine.
I just don't want to build a server, I've never done it before. I don't know how the thing I build will scale horizontally either.
How many cores are you looking for anyway?
We haven't given LET much to chew with. No budget, no location, no-nothing other than you want many cores & high whore (not typo) performance.
16-20 cores or more. no specific budget.
And what ghz (speed) are you looking for per core?
no specific requirement
Not sure about the single core performance, but how about this little IOFlood server?
Dual Intel L5639 CPUs, 72GB ram, 4x 3TB Hitachi HDDs, LSI 9271-4i hardware raid + cachevault BBU
https://manage.ioflood.com/order/?form_id=169
Need it to be dedicated I believe
What?
I am starting to feel for his client.
So what exactly are you building for your client?
Sounds like mining. Hard to do profitably these days. For raw cpu power multiple hetzner auction servers probably beats any E5 box, plus you get tons of disk space you can sell storage plans on.
Just saw cheap and enterprise in one sentence. I need more beer, I’m clearly seeing things.
>
I did some search on colo but I didn't find any of them offers competitive price compare to rented one ! specially if I add extra line speed (above 4Mbit/s ) do you know any recommended colo hosting to start with .
A good Enterprise server will start beyond $500/mo unmanaged.
You are seeing things because it doesn't say "cheap." It says cheapest. As in most affordable.
Something that I need to be able to scale horizontally (hence more cores > clock speed).
No mining at all, just very demanding computational stuff similar to deep learning.
Unless anyone knows of better than this, this is probably what I'll be going with.
This all depend on provider quality which you need,
You can get same specs paying 1/3 of this.
Indeed, also consider if you actually need unmetered :-) That is also a premium in the price calculation.
Technically the OP never said 'cheap'. He said 'cheapest', which is totally reasonable thing to look for among expensive things :-).
What are your other requirements? Like disk and RAM? If you have minimum disk/ram requirements, I can do better in Lithuania.
We have some spare VPS nodes in Dallas if you are interested. Dual E5 2690v2 - 384 gb RAM - 1 gig Unmetered for 700 usd with 4x1 GB SSD or 1200 usd with 4x4 TB ssd
You might want a GPU server then. If you get one, I'll be interested to know what you find. Hetzner EX51-GPU is out of stock and will probably be for a while. I'm using paperspace.com hourly instances which get expensive if you use them nonstop, but I just use a few hours a week so it's not too bad. They are cheaper than OVH public cloud gpu instances and their support is good. No idea at all if that route is of any interest but if it is, I can send you an aff code that gives you a $10 credit.
We're not using a library that can utilize GPU cores unfortunately . I wish though, because it would solve a lot of problems for us. We're using vanilla Python at the moment and I'm not sure what it would require to put our calculations on the GPU as well, other than OpenCL. We have a mathematical algorithm that pounds the crap out of whatever thread it's running on and we have an instance of that algorithm for each concurrent user. Thus 150 concurrent users = 150 threads = 75 cores.
We're in the USA so I'm not sure this would work