Intel Core i7-3930K vs Xeon E5-2630 For Long-duration Financial Calculations

[dima777]

Hello,

I wonder if you can help me to build the perfect PC setup for the following task:

I am going to perform daily financial market analysis on multiple instruments using Excel (large workbooks around 300 and 500 mbs), Matlab and Automation software. The price will be loaded into Excel, the results would be fed into Matlab to create charts (this sequence will be repeated many times). I plan to sell the resultant analysis at my website for a set monthly subscription – so this can be considered a production workstation. The total process would take 5-7 hours daily. I need this process to run as error-free as possible – absolutely predictably on autopilot. So I am not planning to over-clock the CPU.

I am deciding between a workstation built on Intel Core i7-3930K or the one built on single Intel Xeon E5-2630. These CPUs are roughly the same in price (same number of cores, different speed though) with the I7 being much faster one. But I am more concerned with reliability and stability of this setup. Do you think the ECC memory can help eliminate system crashes when the analysis job is running? I need to be able to connect this PC remotely to initiate the analysis jobs as well. I am also thinking of Intel Core i7-3930K which is very fast but not sure about its stability for long-duration number crunching sessions (it can overheat if run at full speed for many hours?).

Please let me know what you think,

Dave

 

[Puppies04]

There are no stability issues with the I7 even running for long periods of time however the ECC memory is going to be the clincher here. Read the first couple of lines in this link...

http://en.wikipedia.org/wiki/ECC_memory

Go for the xeon it has more than enough "balls" to do what you require it to do even with a lower clockspeed compared to the I7 and its ability to be pared with ECC RAM should give you peace of mind that the results it gives you are correct.

 

[dima777]

yes thanks
ECC memory is used in most computers where data corruption cannot be tolerated under any circumstances, such as for scientific or financial computing.
that is exactly what is needed)

 

[crazymonkeyzero]

I have the Xeon E5 2630 in a workstation system with ECC memory (running mathematical computations) and to be honest, it is kind of meh...Don't get me wrong it's not slow by any means, but it really depends on the software you use. In our software, clock speed is very important, so the 4 core xeon E3 1275 v2 at 3.5-3.9ghz is about 25% faster in completing jobs despite have 2 less cores and being MUCH cheaper. I suggest you find out how your software scales before committing. But typically number crunching, such as financial analysis as you're interested in doing, runs better with higher clock speeds (moreso than cores), so 3930k would be my recommendation due to having about a 1ghz clock speed advantage. But the caveat to that is no ecc support. If you have the budget, and really need ECC go for the E5 1660. Also, if you are willing to scale back a bit, the E3 1275 V2 is a great chip, which does support ECC. ivyb architecture does wonders for number crunching, at least based on my experience (also it natively supports ddr3 1600 memory...a big plus)

 

[dima777]

thank you for this fantastically practical anlwer....I will surely look into the CPUS you suggested - I am thinking of adding another xeon donw the path - i was wondering if E3 1275 V2 can be paired with another E3 1275 V2?

as for i7, I am just concerned if i7 will be able to run the continuous number crunching session lasting MONTHS ? I might not power-off the computer for months! you think E3 1275 V2 XEON can handle such workload? and I7?

 

[Markfw]

My 3930k@4.2 runs 100% load for over a year 24/7, so aside from ECC, it can handle the load. Get a good cooling solution however.

 

[tweakboy]

Hello,

I wonder if you can help me to build the perfect PC setup for the following task:

I am going to perform daily financial market analysis on multiple instruments using Excel (large workbooks around 300 and 500 mbs), Matlab and Automation software. The price will be loaded into Excel, the results would be fed into Matlab to create charts (this sequence will be repeated many times). I plan to sell the resultant analysis at my website for a set monthly subscription – so this can be considered a production workstation. The total process would take 5-7 hours daily. I need this process to run as error-free as possible – absolutely predictably on autopilot. So I am not planning to over-clock the CPU.

I am deciding between a workstation built on Intel Core i7-3930K or the one built on single Intel Xeon E5-2630. These CPUs are roughly the same in price (same number of cores, different speed though) with the I7 being much faster one. But I am more concerned with reliability and stability of this setup. Do you think the ECC memory can help eliminate system crashes when the analysis job is running? I need to be able to connect this PC remotely to initiate the analysis jobs as well. I am also thinking of Intel Core i7-3930K which is very fast but not sure about its stability for long-duration number crunching sessions (it can overheat if run at full speed for many hours?).

Please let me know what you think,

Dave

That is not fair, Xeon 8 core 16 threads vs Ivy 3960x Extreme 6 core 12 threads OCed to 5Ghz will slap the sh*T out of that Xeon......

 

[Arachnotronic]

thank you for this fantastically practical anlwer....I will surely look into the CPUS you suggested - I am thinking of adding another xeon donw the path - i was wondering if E3 1275 V2 can be paired with another E3 1275 V2?

as for i7, I am just concerned if i7 will be able to run the continuous number crunching session lasting MONTHS ? I might not power-off the computer for months! you think E3 1275 V2 XEON can handle such workload? and I7?

Get the Xeon E5 and pair it with ECC memory. Get a dual socket board so that you can expand to two CPUs later. E3's cannot be run in dual chip configs.

 

[Phynaz]

That is not fair, Xeon 8 core 16 threads vs Ivy 3960x Extreme 6 core 12 threads OCed to 5Ghz will slap the sh*T out of that Xeon......

Would you please get back on your meds? Please?

 

[dima777]

yes..but I am more concerned about stability...I will be doing price analysis in excel using large model files - 200 mbs, 300 mbs and over 500 mbs each -many of these files will need to be processed for each time frame of each currency pair....I need a very fast and very reliable system for this project...I will use the results to help people trade and save time analysing the forex markets....still I am on a budjet for now - around 1500$ - I am thinking of i7 but intrigued by stability and ECC support of the XEON 2600 family......which processors do you think might suit this task the best??? I need to reanalyze longs ot market data daily - to present fresh analysis for the clients each morning...so I need a super reliable system....

 

[Arachnotronic]

If reliability is of the utmost importance, then this is a no brainer - Xeon E5.

 

[crazymonkeyzero]

thank you for this fantastically practical anlwer....I will surely look into the CPUS you suggested - I am thinking of adding another xeon donw the path - i was wondering if E3 1275 V2 can be paired with another E3 1275 V2?

as for i7, I am just concerned if i7 will be able to run the continuous number crunching session lasting MONTHS ? I might not power-off the computer for months! you think E3 1275 V2 XEON can handle such workload? and I7?

Unfortunately, the Xeon E3 can't be paired together to run dual socket, however, I think it is more than sufficient to handle your number crunching needs. IMO, it's more cost effective to build two Xeon E3 systems than one dual socket Xeon E5 system. Dual socket only helps in PARALLEL computations. These involve calculations which a crap ton of variables such as climate modeling and particle analysis such as what physicists use to study the big bang. For pure number crunching, such as financial analysis, you are best off with the fastest single processor system as possible. I think the E3 1275 V2 is perfect for this and should have no problem running months at a time. (My computations are also running months at a time on it with ECC)


The Xeon E3 1275 V2 is essentially the same as the 3770k except that it is locked (no OC) and it supports ECC. In financial analytics benchmarks according to Anand bench, it scores 277 points, while the 3930k (or E5 1660) scores 315 points. This is only about a 10% difference. So I think the E3 is a VERY capable processor.

Take a Look Here at Anandtech Bench Results

http://www.anandtech.com/bench/Product/551?vs=552

 

[Puppies04]

I read this...

That is not fair, Xeon 8 core 16 threads vs Ivy 3960x Extreme 6 core 12 threads OCed to 5Ghz will slap the sh*T out of that Xeon......

Then I scroll back up and read this...

I need this process to run as error-free as possible – absolutely predictably on autopilot. So I am not planning to over-clock the CPU.

Then I realise that it is tweakboy and I move on with my life.

 

[Idontcare]

If reliability is of the utmost importance, then this is a no brainer - Xeon E5.

This.

yes..but I am more concerned about stability...I will be doing price analysis in excel using large model files - 200 mbs, 300 mbs and over 500 mbs each -many of these files will need to be processed for each time frame of each currency pair....I need a very fast and very reliable system for this project...I will use the results to help people trade and save time analysing the forex markets....still I am on a budjet for now - around 1500$ - I am thinking of i7 but intrigued by stability and ECC support of the XEON 2600 family......which processors do you think might suit this task the best??? I need to reanalyze longs ot market data daily - to present fresh analysis for the clients each morning...so I need a super reliable system....

Rare to run into a fellow forex trader/analyst here. I use Excel, Mathematica, and Metatrader 4 for my stuff.

FWIW have you had a chance to run your proposed setup on a system, even as a prototype, to get a feeling for your absolute performance requirements?

I ask because I run timeframe analyses on fairly deep historical files, they are only limited in size by the 32bit aspects of metatrader so we are talking a couple decades worth of M1 data per currency pair, and even then the standard analysis stuff (trend indicators, fib levels, channel, oversold/overbought indicators, etc) only take a matter of minutes.

Now clearly you might be doing some rather deep and complicated timeframe analyses so I can imagine ways you might find yourself truly time-limited to crunch through the statistical analyses of 120 currency pairs within a 24hr looping window, but if you aren't really in that kind of a time-pinch then you could certainly save yourself some money by scaling down your hardware requirements.

For pure number crunching, such as financial analysis, you are best off with the fastest single processor system as possible.

Currency pair analyses can be done in parallel in course grained embarrassingly parallel fashion because you can assign one core (thread) per currency pair and have 100% utilization, and you will typically have a minimum of 16 currency pairs to analyze, and up to around 120-140 currency pairs once you start factoring in the cross currency pairs.

I intentionally went from a 4GHz QX6700 on vaporphase (single-threaded speed demon at the time) to five Q6600 quads running a distributive program because I needed cores more so than pure GHz with currency pair analyses.

Now the OP's specific methods of data analysis might not lend itself to parallelization, so they may still be looking for single-threaded performance. But they should really be looking for ways to run their calcs in parallel, multiple instances of Excel and so forth, to maximize the core utilization IMO.

 

[dima777]

This.



Rare to run into a fellow forex trader/analyst here. I use Excel, Mathematica, and Metatrader 4 for my stuff.

FWIW have you had a chance to run your proposed setup on a system, even as a prototype, to get a feeling for your absolute performance requirements?

I ask because I run timeframe analyses on fairly deep historical files, they are only limited in size by the 32bit aspects of metatrader so we are talking a couple decades worth of M1 data per currency pair, and even then the standard analysis stuff (trend indicators, fib levels, channel, oversold/overbought indicators, etc) only take a matter of minutes.

Now clearly you might be doing some rather deep and complicated timeframe analyses so I can imagine ways you might find yourself truly time-limited to crunch through the statistical analyses of 120 currency pairs within a 24hr looping window, but if you aren't really in that kind of a time-pinch then you could certainly save yourself some money by scaling down your hardware requirements.


Currency pair analyses can be done in parallel in course grained embarrassingly parallel fashion because you can assign one core (thread) per currency pair and have 100% utilization, and you will typically have a minimum of 16 currency pairs to analyze, and up to around 120-140 currency pairs once you start factoring in the cross currency pairs.

I intentionally went from a 4GHz QX6700 on vaporphase (single-threaded speed demon at the time) to five Q6600 quads running a distributive program because I needed cores more so than pure GHz with currency pair analyses.

Now the OP's specific methods of data analysis might not lend itself to parallelization, so they may still be looking for single-threaded performance. But they should really be looking for ways to run their calcs in parallel, multiple instances of Excel and so forth, to maximize the core utilization IMO.

thank you for your interesting reply.....well I have 50 currency pairs to analyse daily - 4 time frames each - each time frame shoudl be analyzed using a few different large excel files - around 300 mbs on average....do you think this process can be speeded up if I break each of the 4 time frames into different virtual machines and place the analysis pipeline into them?.....you think a 4 core E3 1275 V2 might be enough for it???

I am also thinking - for the price of XEON 2630 -634$ (http://www.newegg.com/Product/Produc...82E16819117270) on a two CPU motherboard? to maximize the number of cores?

 

[dima777]

Unfortunately, the Xeon E3 can't be paired together to run dual socket, however, I think it is more than sufficient to handle your number crunching needs. IMO, it's more cost effective to build two Xeon E3 systems than one dual socket Xeon E5 system. Dual socket only helps in PARALLEL computations. These involve calculations which a crap ton of variables such as climate modeling and particle analysis such as what physicists use to study the big bang. For pure number crunching, such as financial analysis, you are best off with the fastest single processor system as possible. I think the E3 1275 V2 is perfect for this and should have no problem running months at a time. (My computations are also running months at a time on it with ECC)


The Xeon E3 1275 V2 is essentially the same as the 3770k except that it is locked (no OC) and it supports ECC. In financial analytics benchmarks according to Anand bench, it scores 277 points, while the 3930k (or E5 1660) scores 315 points. This is only about a 10% difference. So I think the E3 is a VERY capable processor.

Take a Look Here at Anandtech Bench Results

http://www.anandtech.com/bench/Product/551?vs=552

thank you for your reply....yes it might be reasonable to start with a single Xeon E3 system - do you also think this processor can be used to fire up 4 virtual machines each making the analysis of the subset of the data? with hyperthreading this cpu has 8 cores right? these virtual cpus can be assigned to separate virtual machines?

 

[Idontcare]

thank you for your interesting reply.....well I have 50 currency pairs to analyse daily - 4 time frames each - each time frame shoudl be analyzed using a few different large excel files - around 300 mbs on average....do you think this process can be speeded up if I break each of the 4 time frames into different virtual machines and place the analysis pipeline into them?.....you think a 4 core E3 1275 V2 might be enough for it???

I am also thinking - for the price of XEON 2630 -634$ (http://www.newegg.com/Product/Produc...82E16819117270) on a two CPU motherboard? to maximize the number of cores?

Depending on the type of analysis work you are doing, you could break it down by timeframe and run those in parallel (and serially by currency pair) or you could break it down by currency pair and run them in parallel (and serially by timeframe).

If the nature of the calculations involves analysis of multiple currency pair simultaneously (arbitrage analysis, for example) then you can run those in parallel too, all you need do is duplicate the data files in question to manually avoid locking issues.

For example, if you know your EURUSD1.csv file will be needed for both EURXYZ and XYZUSD currency pair portfolio analyses then you can set it up so the EURUSD1.csv file is physically duplicated (create a new copy) to something like EURUSD1b.csv and have the XYZUSD analysis look for all the *1b.csv file bases.

Then there are no locking concerns.

500MB files really aren't much for a modern system equipped with 16-32GB of ram an a few SSD's setup for raid-0 operation to boost the sequential file read rates.

As I said before, to be sure you could very well be running some sort of deep complex market analysis algorithm that is going to hog the compute power of the computer you end up buying, but unless you are going to high levels of cross-correlation analysis amongst all those pairs I think you are going to find the system you are putting together is going to be more than enough firepower.

And one way you can ensure it will be more than enough firepower is to setup the system to maximize core usage running as much in parallel as possible.

My rule of thumb for approach to a system for this purpose is leave one physical for computations and leave on HT (virtual) core for managing the background computation overhead that the system itself is going to be tasked with.

So if you get 6C/12T cpu only run 6 VMs (or 6 parallel instances of the analysis loop) and leave 6 threads for keeping the background processes running smoothly (and the compute demands from the background processes will scale with your VM/instances loads, so it works out nicely as a scaling rule of thumb).

If you were to go with a bulldozer-based machine I'd tell you to only use one module (2 cores) per VM as well.

I think you are going to have a lot of fun setting this up, I really enjoyed setting up mine. Once you get the hardware in front of you and you start to see ways you can optimize the core loadings and setup the front-end to better parallelize your computations it will be a lot fun for you.

That kind of optimization work is very satisfying and self-rewarding IMO :thumbsup: So don't fret over the hardware all that much because there really is no one perfect answer, whatever you get in the end you will come to realize that you actually make more of a difference in terms of how efficiently and effectively the hardware gets used.

I am also thinking - for the price of XEON 2630 -634$ (http://www.newegg.com/Product/Produc...82E16819117270) on a two CPU motherboard? to maximize the number of cores?

That is a very reasonable plan and approach.

Now for me, since I had control over the local network, I actually avoided the 2S tax (mobo is more, cpus are more, etc) and I built two separate boxes for get my dual-cpu system on the cheap and then I just managed farming the computations out on a per box basis. In the end I went with five quad-core boxes and I spent about $600/box...the problem with more boxes being the added expense of software licensing though.

So you have lots of options, and I'm serious when I say I don't think you can make a bad decision here. Whatever you decided to go with, in the end it will be you that makes all the difference and I'm convinced you know what you are doing so I don't see a downside to the hardware options you are entertaining.

The 2S mobo w/1CPU for now option is a winner in my mind. Make sure you get LOTS of ram and SSDs for some really fast file IO.

Get it and see how well your computations can be broken up into separate parallel analyses and run with the ball once you are at that stage.

I'm a bit jelly, sounds like you are about to embark on something that is going to be super fun and super cool at a personal level for you, I wish you all the best in your escapades

 

[dima777]

Depending on the type of analysis work you are doing, you could break it down by timeframe and run those in parallel (and serially by currency pair) or you could break it down by currency pair and run them in parallel (and serially by timeframe).

If the nature of the calculations involves analysis of multiple currency pair simultaneously (arbitrage analysis, for example) then you can run those in parallel too, all you need do is duplicate the data files in question to manually avoid locking issues.

For example, if you know your EURUSD1.csv file will be needed for both EURXYZ and XYZUSD currency pair portfolio analyses then you can set it up so the EURUSD1.csv file is physically duplicated (create a new copy) to something like EURUSD1b.csv and have the XYZUSD analysis look for all the *1b.csv file bases.

Then there are no locking concerns.

500MB files really aren't much for a modern system equipped with 16-32GB of ram an a few SSD's setup for raid-0 operation to boost the sequential file read rates.

As I said before, to be sure you could very well be running some sort of deep complex market analysis algorithm that is going to hog the compute power of the computer you end up buying, but unless you are going to high levels of cross-correlation analysis amongst all those pairs I think you are going to find the system you are putting together is going to be more than enough firepower.

And one way you can ensure it will be more than enough firepower is to setup the system to maximize core usage running as much in parallel as possible.

My rule of thumb for approach to a system for this purpose is leave one physical for computations and leave on HT (virtual) core for managing the background computation overhead that the system itself is going to be tasked with.

So if you get 6C/12T cpu only run 6 VMs (or 6 parallel instances of the analysis loop) and leave 6 threads for keeping the background processes running smoothly (and the compute demands from the background processes will scale with your VM/instances loads, so it works out nicely as a scaling rule of thumb).

If you were to go with a bulldozer-based machine I'd tell you to only use one module (2 cores) per VM as well.

I think you are going to have a lot of fun setting this up, I really enjoyed setting up mine. Once you get the hardware in front of you and you start to see ways you can optimize the core loadings and setup the front-end to better parallelize your computations it will be a lot fun for you.

That kind of optimization work is very satisfying and self-rewarding IMO :thumbsup: So don't fret over the hardware all that much because there really is no one perfect answer, whatever you get in the end you will come to realize that you actually make more of a difference in terms of how efficiently and effectively the hardware gets used.


That is a very reasonable plan and approach.

Now for me, since I had control over the local network, I actually avoided the 2S tax (mobo is more, cpus are more, etc) and I built two separate boxes for get my dual-cpu system on the cheap and then I just managed farming the computations out on a per box basis. In the end I went with five quad-core boxes and I spent about $600/box...the problem with more boxes being the added expense of software licensing though.

So you have lots of options, and I'm serious when I say I don't think you can make a bad decision here. Whatever you decided to go with, in the end it will be you that makes all the difference and I'm convinced you know what you are doing so I don't see a downside to the hardware options you are entertaining.

The 2S mobo w/1CPU for now option is a winner in my mind. Make sure you get LOTS of ram and SSDs for some really fast file IO.

Get it and see how well your computations can be broken up into separate parallel analyses and run with the ball once you are at that stage.

I'm a bit jelly, sounds like you are about to embark on something that is going to be super fun and super cool at a personal level for you, I wish you all the best in your escapades

thanks a lot for your supportive and such detailed reply)))) yes this is a very large project and I wish to have as much power under my fingertips when it comes to getting it to the production level - firing up all the cores to get it it done as fast as possibel daily..........I like the freedom of EXCEL which does not constrain you in the way you can analyze the prices...I will surely let you know how it goes.....I also thought - if I use the Intel Xeon E3-1275 V2 3.5GHz (3.9GHz Turbo) - it is rougthly close to i7 3930 but has only 4 cores but still supports teh ECC memory....do you think it I set up 4 VMs each running a single 3.5GHz core - each running excel Matlab and automate by Network Automation - and leave the rest of the virtual cores to handle the load on the system - that can pull off?) thanks again for your supportive words))

PS: I do plan to calculate cross-correlation and correlation-dependent analysis on all the pairs on top of the traditional techs)

 

[Idontcare]

I also thought - if I use the Intel Xeon E3-1275 V2 3.5GHz (3.9GHz Turbo) - it is rougthly close to i7 3930 but has only 4 cores but still supports teh ECC memory....do you think it I set up 4 VMs each running a single 3.5GHz core - each running excel Matlab and automate by Network Automation - and leave the rest of the virtual cores to handle the load on the system - that can pull off?) thanks again for your supportive words))

I do think the E3-1275 could pull it off with four cores. The extra Ghz will not be wasted with your calculations whereas if you went with the hex-core the extra cores might be wasted unless you find a way to put them to work. So going with fewer but faster cores is a good way to get more performance upfront with less being dependent on you to make it all work.

PS: I do plan to calculate cross-correlation and correlation-dependent analysis on all the pairs on top of the traditional techs)

You will definitely find yourself chewing through the cpu cycles then.

One other thing that I meant to mention earlier as just an anecdotal comment is that I noticed you are keen to maximize system stability and reliability (uptime), something you will have to put some thought into managing are the windows auto-updates because some of them invariably cause your system to reboot and it doesn't care if the computer is busy doing stuff when it initiates the reboot.

I found out the hard way :\ So now all that windows is allowed to do is download the updates but not install them. Then I just make it a point to remote desktop into the systems periodically (every other week, or so) and manually initiate the windows update process complete with reboot when I know the system is safely at idle (usually on a weekend).

 

[dima777]

Unfortunately, the Xeon E3 can't be paired together to run dual socket, however, I think it is more than sufficient to handle your number crunching needs. IMO, it's more cost effective to build two Xeon E3 systems than one dual socket Xeon E5 system. Dual socket only helps in PARALLEL computations. These involve calculations which a crap ton of variables such as climate modeling and particle analysis such as what physicists use to study the big bang. For pure number crunching, such as financial analysis, you are best off with the fastest single processor system as possible. I think the E3 1275 V2 is perfect for this and should have no problem running months at a time. (My computations are also running months at a time on it with ECC)


The Xeon E3 1275 V2 is essentially the same as the 3770k except that it is locked (no OC) and it supports ECC. In financial analytics benchmarks according to Anand bench, it scores 277 points, while the 3930k (or E5 1660) scores 315 points. This is only about a 10% difference. So I think the E3 is a VERY capable processor.

Take a Look Here at Anandtech Bench Results

http://www.anandtech.com/bench/Product/551?vs=552

I was wondering if you can recommend a proper setup (motherboard+ram+disk) for the Xeon E3-1275 V2 CPU that you mentioned? I am truely gravitating toward it!
thanks a lot!)
D

 

[dima777]

I do think the E3-1275 could pull it off with four cores. The extra Ghz will not be wasted with your calculations whereas if you went with the hex-core the extra cores might be wasted unless you find a way to put them to work. So going with fewer but faster cores is a good way to get more performance upfront with less being dependent on you to make it all work.



You will definitely find yourself chewing through the cpu cycles then.

One other thing that I meant to mention earlier as just an anecdotal comment is that I noticed you are keen to maximize system stability and reliability (uptime), something you will have to put some thought into managing are the windows auto-updates because some of them invariably cause your system to reboot and it doesn't care if the computer is busy doing stuff when it initiates the reboot.

I found out the hard way :\ So now all that windows is allowed to do is download the updates but not install them. Then I just make it a point to remote desktop into the systems periodically (every other week, or so) and manually initiate the windows update process complete with reboot when I know the system is safely at idle (usually on a weekend).

thank you for your reply)))) yes I guess this processor shoudl work charmingly for this task.....coupled with the ecc memory.....I am now thinking what motherboard might be the best for Xeon E3-1275 V2....
thank you for your suggestion about the automated windows updates)) will take that into account)

 

[Markfw]

You know the 3930k is now $499 at newegg ?

 

[crazymonkeyzero]

thank you for your reply....yes it might be reasonable to start with a single Xeon E3 system - do you also think this processor can be used to fire up 4 virtual machines each making the analysis of the subset of the data? with hyperthreading this cpu has 8 cores right? these virtual cpus can be assigned to separate virtual machines?

The E3 1275 V2 (it has 4 cores, 8 threads due to HT) will be more than able to handle 4 VMs comfortably given you have enough system memory. I suggest 16gb DDR3 1600 ECC memory. Typical rule of thumb is allocate 2gb of ram per vm, so you'd need 8gb just for 4 VMs, and the rest to run your system.

(32gb bit overkill, but something you may want to consider for adding many more vms in the future).

If you want data integrity, ECC is a must. So the 3930k is kind of out of the picture.
The E5 1660 or the E3 1275 V2 are your best bet. I highly advise you NOT to go with the E5 2630~clock speeds are too slow, and you may end up regretting it like I did. Also keep in mind 2 socket systems do not scale exactly by doubling performance. Best case scanarios typically see about a 75% increase, which can lead you to question if you want to spend so much if you could have easily bough two separate single socket systems cheaper which will double productivity.

@Markfw900

Although that is a great deal, OP seems to indicate need for ECC memory, so only Xeons are feasible. Or I'd be all for it as well, especially at that price.

 

[crazymonkeyzero]

I was wondering if you can recommend a proper setup (motherboard+ram+disk) for the Xeon E3-1275 V2 CPU that you mentioned? I am truely gravitating toward it!
thanks a lot!)
D

Here's a set up very similar to mine:

I use the same core parts (cpu ram and mobo), so they should be 100% compatible.

CPU: Xeon E3 1275 V2
http://www.newegg.com/Product/Produc...82E16819117282

MOBO ASUS P8C WS

(don't listen to troll reviews,it's a great board, that has a uefi bios and many premium features
you wont find on boards by supermicro or tyan but make sure to only use UNBUFFERED ecc memory)
http://www.newegg.com/Product/Produc...82E16813131849

RAM 16GB DDR3 1600 ECC Unbuff. (4x4gb) you can go for 32gb as well, but 16 is definitely more than enough for 4vms
http://www.newegg.com/Product/Produc...82E16820239226

Case : (great silent case ) Fractal Design Define R4
http://www.newegg.com/Product/Produc...INE+R4&x=0&y=0

Power Supply: Seasonic G Series 550w modular (on sale for BF $55 AR till tomorrow)
http://www.newegg.com/Product/Produc...C%20G%20SERIES

Hard Drive 2TB Samsung
http://www.newegg.com/Product/Produc...82E16822152245

SSD : samsung 830 : 128gb (May or may not need this? Will definitely make system snappier over all)
http://www.newegg.com/Product/Produc...82E16820147163

Monitor: BENQ 21.5 Inch 1080p (on sale for $99 till tomorrow)

http://www.newegg.com/Product/Produc...82E16824014298


I opted not to select a graphics card, because Integrated graphics on ivy are good enough for me. But this should be evaluated on a case by case basis.

 

[dima777]

You know the 3930k is now $499 at newegg ?

yes thanks)