Linux

Grab a 4 core Intel i5 with a high clock rate and haswell cores (or wait a bit for prices to settle and go broadwell), a 2x4GB set of RAM modules, any recent nvidia *60 card, a mainboard that can fit those and a good brand PSU (Corsair, Enermax, ..., about 450W maybe - don't try to save money on the PSU). Get a decent ~256GB SSD if you want + a random 1TB hard disk, put that in a pretty case and you are good to go.

Why Intel? - High single core performance makes them king for gaming, no discussion!
Why i5? - Usually, the main difference between an i5 and its i7 counterpart is Hyperthreading, which more often than not doesn't do anything to gaming performance. It's definitely not worth the price difference.
Why nVidia? - It just works better with Linux. The open source AMD drivers with gallium are pretty awesome now, but they aren't quite there yet.
Why *60 - The *60 Segment has the best balance of price and performance for gaming in my opinion. Should be good for at least 2 years.
SSD or not? Totally up to you. Doesn't really affect game performance beyond the loading screen, but in general, having an SSD everything feels so much quicker, I personally wouldn't want to miss it. If you don't get an SSD, you may want to take a better look at the hard disk you buy.

Of course, the whole answer depends on your budget, but this is the baseline I would go from.

The AMD cards usually work well with the open source drivers. The problem is, radeonsi gallium drivers and wine are still somewhat experimental. Even if in theory your AMD card can provide as good performance as an nVidia card, it usually takes a good amount of fiddling around to get there. You will probably want to use bleeding edge mesa and radeonsi builds. This will change, as the radeon drivers are taking huge steps at the moment, but that's what I mean when I say, that they aren't quite there yet.

For a single or MAYBE 2 clients, I could agree, if it weren't for the outrageous price for an Intel chip, aswell as my experiance with stability issues (which may have improved, so we'll set that aside).

But they still run the A64 emulation layer as the Intel architecture simply CAN'T run a true 64-bit process core. (Intel even admited this a while ago)

Because of that, you still run into the emulation bottleneck. In an i5, granted playing eve on 1-2 clients you won't notice it. But if you run 3, 4 or even 5, you will start to notice a performance change between the i5 and it's amd counterpart. I have a FX-4300 clocked at 4.55gHz, my friend has an i5 clocked at 4.50 and running 2 clients he has equal performance to me, running 4 he has about a 4% drop and running 5 he has a 15% drop from my performance level.

Not to mention you can get an 4core FX chip clocked at 4.8-5.2gHz for the same as a 4core i5 clocked at 4.1, lmao.

Because if you're running a 64bit OS, windows or Linux, even running 32bit programs will be faster (esp when running multiple ones) because the main benefit of 64bit vs 32bit is you're able to access twice the amount of RAM at the same time in 64bit. So, let's imagine a VERY VERY simplistic version of the situation.
Your processor handles 32bits of information per milisecond.
A 32-bit program requires the processor to process 32bits of information every 10 miliseconds.
Effectively, you could run 10 of these programs in parallel, all offset by 1-9 miliseconds, and see no change in performance vs having a single running. But you also have to take into account background tasks, the kernel, x-server, WINE itself, the list goes on.
A 64-bit processor can process that 32bits per milisecond TWICE per milisecond, effectively doubling it's speed.
Meaning when you're running multiple instances of a processor-heavy application (like a game, eve) you can effectively handle twice as many as you could with the same processor only running 32-bit architecture, but when running single, or low instance counts of the program, you may not see any performance changes at all.

Also, most 32-bit compatibility libraries for OpenGL (when installed on a 64-bit system) are designed to run multiple libraries in parallel to benefit from the 64-bit architecture, even if the program is only 32bit. I do notice a significant difference in performance when running 5+ clients at once if I run them through a 32bit wine binary and a 32-bit wineprefix, vs a 64bit one, because the 64bit wine binary is designed to call 64bit opengl files, but more importantly because the 64bit wine binary allocates the proper memory and processor times for optimal use of the 64bit architecture, even when the program it's running is 32-bit.

You're also forgetting tho eve IS a 32-bit program, it does have optimizations builtin to the program to detect when it's on a 64-bit OS which re-works several of the games call methods to run in parallel instead of sequence.


Now, having said all that if you're still even reading this by now, even tho intel's IA64 DOES run significantly faster than baseline 32bit processing, it still hits a bottleneck in the emulation layer because all of those 64bit calls, or 32bit calls that were run in parallel, have to still go through a single 32bit channel when leaving the processor.

Basically the way it works, is the IA64 layer receives a 64bit process string, but unlike a true-64 bit processor, it doesn't process that string in a single core, instead it splits the string and processes half of it in 2 separate cores (even in single-core IA64 processors there's 2 cores, yes), then once the process string is done in each core, it's re-united 'on the other side' and sent on it's merrily way. This in itself isn't the issue however, as this kind of string process emulation has been around for years and in itself is quite effective, but at the very end of the process when it is transmitted back out of the cpu, it's split yet again and each half of the 64bit segment is transmitted to the bus line 32bits at a time, where it is then AGAIN reunited and then sent to the destination. This is the problem.

To try and counteract this, most Intel compatible MB's have extra bus lines, to handle getting the split segments out of the cpu and to their destination faster, otherwise they would pile up in the bottleneck. It's a very inefficient way of doing things, but it's the only viable way Intel has to handling 64bit processing as the Intel core's base architecture was never designed to move beyond 32bit processing. Most benchmark tests you see of intel vs amd only take into account the amount of processes per second, the core speed, core sync, etc. They never take into account the extra time it takes for processes to reach their destination from the source once they leave the cpu, and while a few programs at a time isn't generally an issue for Intel, and it is still a solid CPU, when it comes to really CPU heavy multitask processing, nothing can stand up to the speed of a true 64-bit, no matter who makes the chip, and AMD is the only one with a true 64bit architecture.


The OP was asking for recommendations on a PC able to run multiple clients. Most people assume multiple means 2-3 at most, but more often than not people in EVE who have multiple accounts either 2, or 5+. The reason i recommend AMD so adamantly is because wether the OP wants to run 2 clients or 10 clients, the AMD platform would be more equipped to handle it, getting about equal performance on 2 clients, yet significantly better on 10 clients. Plus, cost, you can usually get a high-mid range AMD chip for the same as a low-mid range intel chip. (some exceptions, as always)

There's also a Linux in-game channel you can join to get more immediate advice from people.