Quad processor cores vs. 4 processors.
 

Hi!

I've heard from a few of my friends that an Intel Xeon Quad Core is actually slower than 4 individual processors. Is this true? I was considering buying a Xeon Quad core server, or six inexpensive ones. I'm not sure if these inexpensive computers would be as efficient, but they would be more affordable. The cheap servers would be VIA C7's 1.0GHz mini-ITx servers. The Intel server is roughly $1500 and is 2.67GHz, I could buy six mini-ITx servers at that price. This isn't gonna be a Enterprise setup, the servers are going to be hosting random things at my house, (I've been messing around with Home Automation).

That is a very complicated question with no simple answers.

But I think that one does have a simple answer.

I can't be certain, because I didn't find any really solid performance data on the VIA C7 1.0GHz. But I read a few pages with reasonable and consistent data and the comparison seems to be so lopsided that all possible misunderstandings and mis measurements combined shouldn't be enough to reverse the conclusion:

Those six pitiful CPUs don't add up to a decent fraction of the performance of one 2.67GHz quad core Intel.

For most home uses, I think one dual core 3.2 GHz AMD system will cost less and get more work done than a quad core 2.67Ghz Intel. But that one is certainly not a simple question. It really depends on the type of work you're trying to get done.

Having completely separate processors is indeed more efficient than having multiple cores on one die, but there is more to it than that.

The C7 processor is not built for power, but for energy efficiency. Performance can no longer be measured in clock speed, you need to take other aspects (such as architecture, cache size, bus speed, etc) into account.

There is also inherent inefficiency involved in interconnecting separate computers, and greatly increased complication in terms of load balancing and control.

So the real world performance of a quad core Xeon versus 4 VIA C7s in a cluster would be a complete joke. Even with six, I would be surprised if they managed to muster half the computational power of the Xeon.

That said, it is entirely possible (even likely) that the Xeon is well overpowered for your needs. In that case, we would be talking about inefficiency of another sort: monetary.

I'll bring up just two of the sub issues that make that a very complicated question:

A) Memory access: In many cases, memory access is the dominant part of total system performance. It is really a four way comparison:
1) Single package with uniform shared memory.
2) Multiple package with uniform shared memory.
3) Multiple package with non uniform shared memory.
4) Multiple package with non shared memory.
As you move from (2) to (4) (keeping many other factors constant) you increase the total memory performance, but because of issues of cooperation and load balancing, you decrease the fraction of total memory performance that is realized as effective memory performance. Does the effective memory performance go up or down? That depends on lots of other things we don't know in this abstract a question. Choice (1) is probably (but there are always other factors) a tiny bit better than choice (2) both ways (raw performance and ability to cooperate), but (1) is hard to compare to (3) or (4). The later question of the OP was obviously choice (4) in memory access.

B) L2 cache performance: In many cases L2 cache performance is the dominant factor in total system performance.
What exactly are you comparing? A typical fairly high end Intel Quad core CPU has two 3GB L2 caches. I think each cache is used by two of the four cores. Would you compare that to four 1.5GB caches? Two cores sharing a 3GB cache will typically perform a little better than two cores each having their own 1.5GB cache. But that really depends on what tasks you're running. They might perform a lot better or a lot worse or anywhere in between.
Anyway, where would you find a CPU with a 1.5GB L2 cache? So what exactly are you comparing when you compare a quad core cpu to four individual cpu's?

So did you intentionally ignore the rest of the sentence/post, or did you just start typing your response when you read that far?

Sorry if it seemed that way. I thought your "more to it" applied to just the "energy efficiency" and "separate computers" issues that you followed with and thus implied a simple answer if those two issues didn't apply.

Even with comparable processor design (eliminate that energy efficiency issue) and across the whole spectrum from uniform shared memory through separate computers, I think the quad core vs. multiple CPUs question is too complicated to say things like "completely separate processors is indeed more efficient". "more to it" doesn't say "the preceding statement is meaningless without overwhelmingly more context and information". Without that, I disagree with it.

But I think on the specific comparison (six VIA C7 1.0GHz), we gave basically the same answer. Maybe that is all the OP wanted. I didn't mean to imply any disagreement on that.

Thank you for your responses. MSG3FX was right, the VIA C7 even at 2.0 GHz, is a complete joke (I didn't actually look at the extended specs, they have L2 caches of 128 KiB :eek: terrible! ). I needed the resources and was just looking for a cheaper alternative to a full-blown Xeon server. I now know how badass the Xeon is.
I have been looking at the OpenMicroServers. They weren't that bad, but I considered the mini-ITx systems because I could get one of those for 1/4 of the price.

For memory-intensive multi-threaded programs, the Quad Core will be more efficient due to the shared L2 cache (both cores have equal and fast access to the cache).

Are you sure you really need that much computing power, though? I run a web, FTP, SSH, SAMBA, printing (CUPS), IMAP, SMTP server on my P3-600mhz, and the CPU usage is rarely above 10%. Any of the cheapest Celerons out there will outperform it by orders of magnitude.

Random! If you can find a use for all that computer power for home automation, please tell me what you are automating. Or maybe you'd rather not tell us, if it really turns out that your home is the Pentagon.

The comment by cyberfishee really seems to most appropriate here; you should be able to perform any reasonable home automation task on a celeron (or a Via or an AMD...) unless you do something very, very inefficient with the coding. You might decide to take, say, a Pentium 2160 or a low end Athlon X2 as the purchase price is not much higher. But remember that there is also the power consumption to think about and high (computing) power solutions like the Xeon will be unnecessarily hot and noisy.

Oh, and the C7 isn't VIA's latest design. The new nano is considerably more efficient on a 'work done per clock' basis and would be competitive would something like an atom (both m/fact will argue that theirs is best in some way or another, of course...).

True, the Xeon would be overkill for my purposes, but it just so happens that I do have a P3 running at 564 MHz, and having to compile all the code on it is slow, and is a real pain in the ass to burn that much time (glibc is compiling now, and it is taking forever). The P3 is also a terrible choice for a media server that serves multiple frontends and occasionally does some transcoding. I wanted something that no matter what I add to it, it will keep on going with no noticeable performance decrease.
The kind of home automation I'm doing is controlling various features around my home with voice commands. So far, I've only managed to get lights to turn on and off, I plan to move on to more complex configurations soon. But the P3 I have now has been riced to the max, and I still can't get it to serve fast PHP code on my blog, or do anything else very fast. I'll try a VIA C7, or a Nano. The C7's are faster, but the Nano's have larger L2 caches and are about 200MHz slower, this difference is probably unnoticeable.

An expensive overkill at that :). Of course the development machine needs to be fast. I run a Core 2 Duo @ 3.36ghz as my development machine, where I compile all my code. I only move the final binary to the P3 to be executed. All the development and testing are done on the Core 2 Duo. As for real-time media transcoding/streaming, a Pentium Dual-Core (low end version of Core 2) at 2ghz will sure suffice (at ~1/10 the price of a Xeon). If you really need the power and are into overclocking (I am myself, my Core 2 Duo runs at 1.86ghz stock), just get a ~$120 Core 2 Duo E7200 and overclock it to 4ghz. It will do anything you throw at it at blazing speeds.

I am running my blog on my server, too, a WordPress blog. It does take some time to load, even when I am on the LAN. However, I just did a little experiment, and it revealed that the CPU is not the bottleneck (top reporting 40% CPU usage max, and only for an instant), meaning it will perform nearly the same even if the CPU is 10 times more powerful.

Thats pretty clever, moving the code over. I probably wasn't going to invest in a Xeon anyway, but I considered it. Thanks for all of your responses.

Instead of either of those get 10 alix boards (about $130 each)(500mhz Geode for the cpu). They make great little boxes for doing all kinds of things and they use very little power. I have one at home that I use as my router (has a minipci slot if you want it to do wifi stuff) and its running an asterisk voip server for my phones, a squid server for caching http, vpn software. Its definitely not something your going to do video encoding on but for running home servers or X10 control its fantastic and not power hungry.