|3 posts found|
Freeloading mooches are the scourge of the gaming community.
OP 4/12/12 2:47:01 AM#1
Why are computers with Xeon processors so much more expensive? Are they that much faster
4/12/12 6:42:54 AM#2
Originally posted by cybertrucker
They are server processord with a more limited supply and demand than standard CPUs used in desktops. They also aren't as good for games as other processors.
4/12/12 1:16:05 PM#3
Xeon is Intel's server brand. Some of them are the same processors that desktops use, but merely with some server features (e.g., ECC memory) enabled that desktops disable. For example, toward the low end, this:
is basically the same as this:
Moving up the chain some, this:
is basically the same as this:
That last one is the extreme high end in a desktop. But it's nowhere near the extreme high end in a server. For starters, it only has six cores. Few desktops have any real use for six cores, let alone more than that.
But some servers run programs that will scale to as many cores as you've got. So instead of just one processor, you can get two. The top end of Intel's recent Romley platform will let you get this:
Or perhaps rather, get two of them in a single server. That's eight cores as 20 MB of L3 cache. But for many purposes, you can't just have two processors sitting there in two separate boxes next to each other. They have to be able to communicate in real time, and quickly. As in, tens of nanoseconds. That takes special stuff in the chipsets and motherboards.
But that's not Intel's high end, either. They're soon going to launch something similar that lets you get four processors in a server with the Sandy Bridge architecture. They've already done that with older architectures. Right now, Intel's top end is Westmere-EX, which will let you get eight processors in a single server, each with 10 cores and 30 MB of L3 cache. That's huge, huge processor dies, plus eight of them, plus a ton of special sauce on the motherboard and in chipsets to make it all work. And yeah, that gets expensive.
For what it's worth, AMD's top of the line is roughly this:
Though there are other bins for higher and lower TDP. Take an FX-8150, put two of the dies in a single package, and you've got a 16-core processor. In desktops, the extra HyperTransport links to let the two dies communicate with each other are disabled. And then you can put four of the 16-core processors in a single server for 64 cores in a server.
So why is that so much cheaper than Intel's top of the line? Because it's not nearly as good. If you read reviews on the FX-8150, it was quite a fiasco. Put eight of the dies in a server and clock them lower and it doesn't magically become a great architecture. It's still a fiasco. That will get you a lot of performance in things that scale well to many cores, but will also burn a lot more power than something that gets you the same performance from Intel.
For servers, the heat problem often isn't so much one of getting it out of the case, as it is one of getting it out of the building. A single 4P server that puts out 400 W isn't so hard to cool. Put a thousand of them in a great big room and now you're looking at 400 kW, which is a lot harder to cool. That's why some data centers spend more to cool the room than they do to run the hardware. And why some try to locate far north so that they can get a lot of cooling by blowing air through the room from outside.
For some enterprise purposes, the price tag isn't such a big deal, but performance and reliability are really, really important. Imagine what would happen if some MMORPG lost its database, for example. All your progress, all your account info, gone. Lots of people would quit over it. How much do you think that would cost them? While the answer depends on the game, it's a whole lot more than a $10,000 server.