TR Forums

I built my system with gaming in mind and my current i7-3570k is great for that. But I have since gotten into distributed computing in a big way and would like to upgrade to a processor with more cores. Either via more cores or via hyper threading. After taking a look around I have come to the conclusion anything with more physical cores is going to cost too much and probably require a new motherboard as well.

Since I am using a Asus P8Z77-V LK LGA 1155 motherboard the logical step would be to get a i7-3770k to take the place of my i5-3570k. But, I was talking to a friend who wondered if a Xeon wouldn't work. Looking around a Xeon E3-1275 is in the same price range as the 3770k and it's supposed to work on a 1155 motherboard.

I have no experience with Xeon's. Would the Xeon work with my motherboard, I have my doubts, and windows 8? Is there even any advantage to the Xeon over the 3770k?

Hyperthreading is a rather dubious feature that brings little to the table if you plan on running at 100% load, you're unlikely to see much of an improvement beyond the slightly faster clock speed. It's one thing to spend a little extra to get it when upgrading everything, and quite another to buy another $300 chip when you've already got a perfectly good i5.

I'm not an expert on Xeons but they go into workstations as well as servers (my work PC has one), and whether it's supported is likely based on the specific motherboard.

The main thing that Xeon gets you compared to the 3770K is support for ECC RAM, which is kind of pointless since your motherboard only supports non-ECC. Performance will probably be slightly *worse* since the maximum core and RAM clocks on the Xeon are lower (with the rest of the specs otherwise being quite similar). Also, the Xeon is not on Asus' list of officially supported CPUs for that motherboard (though a few other Xeons are); so compatibility is a bit of a crapshoot as well.

In short, don't bother with the Xeon.

Edit: What NovusBogus said too. You're not increasing the number of real cores by going from the 3570 to the 3770; you're just increasing the number of threads. Depending on the workload, this could result in anything from a moderate improvement, to no improvement at all. Unless you can find specific benchmarks indicating that the DC projects you're running benefit significantly from extra threads (as opposed to extra cores), it's probably not worth it.

With my i5 I can only run four tasks at a time. With the i7 in my laptop I can run eight at one time. Granted the laptop is slower because of its mobile i7 chip running at 2.2GHz. With a i7-3770k I can double the work load and keep the speed of my i5. At least that's my theory and it seems to hold true when you compare the work I get done with my i5 compared to others running a i7.

Hyperthreading is worth something like 20-30% in multithreaded workloads that can take full advantage of it. Don't let task manager fool you into thinking that the additional logical cores = 'doing 4 more things".

If you want more threads and are willing to spring for a 3770K or 4770K, then go for it. If your app can run off more threads and if those threads play well with Intel's HyperThreading, then you're good. Remember, HT works by allocating unused execution resources to a second thread, meaning, it takes advantage of underutilized resources. This means that if all the threads are practically using only a few very specific set of execution resources within a core (say, they all use only ALU1 and FPU1 all the time), the first thread would leave little for the second thread, which means HT won't help you a lot.

Of course there's the option of springing for a Xeon, and while many desktop boards MAY work with Xeons, it would be best to consult your motherboard's manual or web site for more information. These server-class CPUs support ECC and Registered DIMMs too, which your desktop-class motherboard may not support. These LGA115x Xeons work best with server-class boards specifically tailored to run these server-class CPUs, of course, but server-class boards can cost quite a bit too.

Then of course there's the desktop-class LGA2011 chips which can give you a maximum of 6 strong cores. i7-3960X, i7-4960X... all those pricey chips. These would obviously be the best chips to get without going Xeon, but you probably know that they're very expensive.

Lastly, there's the AMD option. If your workload is primarily integer-focused, then according to AIDA64's CPU Hash test, even a relatively cheap FX-8350 can edge out even the i7-3960X which costs 5x as much. Unlike Intel's HyperThreading, AMD gives each thread dedicated execution resources, albeit some resources such as the fetch, decode, and branch prediction circuitry are shared, as well as the L2 cache within a module. The link above, however, gives you an idea how an app that's able to use as many cores available and purely stresses integer execution can wring better performance out of the AMD option compared to an Intel option with HT.

MadManOriginal wrote:

Hyperthreading is worth something like 20-30% in multithreaded workloads that can take full advantage of it. Don't let task manager fool you into thinking that the additional logical cores = 'doing 4 more things".

Khali , what MadMan said. Hyperthreading doesn't equal more cores, it only better utilizes the cores the chip has by keeping more parts within each core working at the same time. Not all software can exploit hyperthreading either, such as if the underlying software is well-optimized to fully utilize the cores then HT isn't going to help at all. Conversely if the software only does one specific type of calculation, then it will be bottlenecked by how many function units the cores have to process that type of calculation and so HT won't help with that either. Ya should always check the specific workloads you plan to run first to make sure they will benefit sufficiently to warrant any upgrade.

For Folding@Home it used to be around 20% higher PPD when comparing a quadcore with HT to one without, but that was only for a perfect 100% load distribution. With partial loads (say 6 threads instead of 8 ) the thread-thrashing used to eat into most of the performance gain. Not checked recently though, given GPU folding now makes CPU folding performance look like small potatoes.