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

[Arkaign]

If you are truly concerned about your build and want to protect it, build out your power delivery this way :

Wall socket --> Good Surge Protector --> Good Line Conditioning UPS.

The surge protector is first in line, and will be sacrificed immediately should there be a close lightning strike/transformer blowout/etc. Even if it still works, if you see any other electrical damage in your home/office, or see the led on it start flashing/dimming, replace it. You will see stupid instructions saying to plug the UPS directly to the wall, but I have used this exact configuration in many hundreds of locations with absolutely zero problems. It has served well, with many surge protector failures, but no premature UPS failures. After all, a nice $40 surge protector is much easier to replace (or temporarily do without) compared to a $500 mid/high end line conditioning UPS.

As for your build, the mobo you quote is actually a workstation board, though many use workstation boards in custom server builds. What you're doing is classic workstation duty anyway. It will be high quality regardless. I also recommend going raid 10 with 4 1TB 7200RPM Sata Drives, and buy a couple of extra ones that match so that down the line should one or two fail over the course of use, you can go ahead and drop it in with no loss in downtime or going with mismatched drives.

 

[Arkaign]

Of course, if you want something cheaper, you could always do SSD + HDD + External w/Acronis mirroring all data to a mirror.

 

[imported_PowerHouse]

that os very reassuring to hear....I wonder if I have separate parts of the workbook - each of which starts with the column of prices and calculates different parts of the model which are then aggregated by some other processes....does this mean that this all will be calculated by a single thread? Do you need absolutely independed streams of formulas to utilize separate cpus?

I don't know how your workbook is set up, but for the sake of argument let's say you have 16 columns containing prices and you need to sum each column and then average those sums. Simple, I know, but the idea works. Excel can begin the summation of each column independently. So, if your processor can handle 16 threads, then each column can be processed completely in parallel and then at the end, Excel falls back to a single thread to do the average. Now, the next chance it gets it will try to parallelize computation. So just because you have a few places where you can't get parallelized calculations doesn't mean you won't benefit from more cores elsewhere in the workbook.

The worse possible case would be that each subsequent calculation relies on the value of the cell before it and no cell is referenced by more than one cell.

Also, I vote for the Xeon for no other reason than to get ECC memory in this case.

 

[dima777]

While I have no doubt you will find an Excel 2010 multi-threading expert or two here in these forums, I do want to point out that your chances for getting knowledgable answers to your specific Excel 2010 questions will be enhanced if you go to the MS Excel help forum and ask for pointers and feedback in their venue. They will probably be able to help you beyond just the basics, getting into the specifics, in a way that you need for clear-cut answers.

yes I tried to do this before and failed to get any more concrete recommendations that get more ram and a faster cpu...on many excel forums...

 

[dima777]

I don't know how your workbook is set up, but for the sake of argument let's say you have 16 columns containing prices and you need to sum each column and then average those sums. Simple, I know, but the idea works. Excel can begin the summation of each column independently. So, if your processor can handle 16 threads, then each column can be processed completely in parallel and then at the end, Excel falls back to a single thread to do the average. Now, the next chance it gets it will try to parallelize computation. So just because you have a few places where you can't get parallelized calculations doesn't mean you won't benefit from more cores elsewhere in the workbook.

The worse possible case would be that each subsequent calculation relies on the value of the cell before it and no cell is referenced by more than one cell.

Also, I vote for the Xeon for no other reason than to get ECC memory in this case.

thank you very much for this detailed and very practical reply...my workbookms are made from separate subsysems s to speak each working on the same 4 initial price columns...so I guess this can be split into separate threads....I am selecting between the e3 1275v2(4 cores ivy bridge) and i7 3930k (6 cores sandy bridge)....do you think it would be best to go with the maximum number of threads (i7 3930k) to maximize the speed as opposed to having a slightly higher speed per core in e3 1275v2?

also - does excel take a lot of time to pre-calculate the way it will utilize all the cores?

 

[imported_PowerHouse]

thank you very much for this detailed and very practical reply...my workbookms are made from separate subsysems s to speak each working on the same 4 initial price columns...so I guess this can be split into separate threads....I am selecting between the e3 1275v2(4 cores ivy bridge) and i7 3930k (6 cores sandy bridge)....do you think it would be best to go with the maximum number of threads (i7 3930k) to maximize the speed as opposed to having a slightly higher speed per core in e3 1275v2?

also - does excel take a lot of time to pre-calculate the way it will utilize all the cores?

Your workbook sounds like it could be well multi-threaded. However, despite the i7-3630K being around 29% faster in pure multi-threaded calculation than the E3-1275 v2, the extra piece of mind from the ECC memory of the Xeon would be worth the lost performance to me. I mean, you are going to be working this processor and its memory sub-system extremely hard for very long periods of time producing data where the correctness is simply paramount. You could also step up to the E3-1280 v2, which is a little faster but costs around the same as the i7. Unless you were going to use its IGP, in which case the 1280 doesn't have it.

No, Excel just starts calculating from what I can tell. I suppose it does it on the the fly.

 

[dima777]

Your workbook sounds like it could be well multi-threaded. However, despite the i7-3630K being around 29% faster in pure multi-threaded calculation than the E3-1275 v2, the extra piece of mind from the ECC memory of the Xeon would be worth the lost performance to me. I mean, you are going to be working this processor and its memory sub-system extremely hard for very long periods of time producing data where the correctness is simply paramount. You could also step up to the E3-1280 v2, which is a little faster but costs around the same as the i7. Unless you were going to use its IGP, in which case the 1280 doesn't have it.

No, Excel just starts calculating from what I can tell. I suppose it does it on the the fly.

yes that makes sense for sure))) thank you))) I wish I could test both cpus on one of my files)) btw - most of my workbooks are over 200 mbs in size - some can approach 500mb - you think both of these cpus will handle the load well????? I am hesitating about the i7 3930k as it has some hidden potential that I might use in my future developments.....and 12 threads.....the 8 threads of e3 sound great but still 4 missing))

 

[imported_PowerHouse]

yes that makes sense for sure))) thank you))) I wish I could test both cpus on one of my files)) btw - most of my workbooks are over 200 mbs in size - some can approach 500mb - you think both of these cpus will handle the load well????? I am hesitating about the i7 3930k as it has some hidden potential that I might use in my future developments.....and 12 threads.....the 8 threads of e3 sound great but still 4 missing))

Your best bet, if you are worried the E3-1275 v2 won't be enough in the future, is to grab a dual-socket LGA-2011 board and and one E5-26xx processor so that you can add another Xeon later. Just don't get the E5-2637 (dual-core); the E5-2603 and E5-2609 are quad-cores but lack Hyper-Threading which can make a difference sometimes. This way you can even go with dual 8-core Xeons later if you find you need more power.

 

[tynopik]

If you are truly concerned about your build and want to protect it, build out your power delivery this way :

Wall socket --> Good Surge Protector --> Good Line Conditioning UPS.

All UPS manufacturers say to plug directly into the wall and you'll void your connected-equipment warranty if you don't.

 

[dima777]

yes thanks...I have been thinking of that...)

I wonder if any one with i7 3930K or XEON E3 127V2 can comment on the multi-threading boost to EXCEL 2010 or EXCEL 2007 performance???? I need this to make the final decision between these two setups...do you think the multi-threaded EXCEL 2010 in the case of i7 3930K or XEON E3 127V2 can be significantly faster than the single threaded EXCEL 2003?
thanks a lot!)

 

[imported_PowerHouse]

yes thanks...I have been thinking of that...)

I wonder if any one with i7 3930K or XEON E3 127V2 can comment on the multi-threading boost to EXCEL 2010 or EXCEL 2007 performance???? I need this to make the final decision between these two setups...do you think the multi-threaded EXCEL 2010 in the case of i7 3930K or XEON E3 127V2 can be significantly faster than the single threaded EXCEL 2003?
thanks a lot!)

I've played around with the different versions and Excel 2007/2010 show almost perfect scaling with threads. Using either of them will be considerably, if not multiple times, faster than using 2003.

 

[dima777]

I've played around with the different versions and Excel 2007/2010 show almost perfect scaling with threads. Using either of them will be considerably, if not multiple times, faster than using 2003.

that is super great to hear....as for the ecc memory - I can theoretically get away with the ecc memory by doing the following steps - I create a single calculation loop and perform it after say each of the 50 loops to see if the results match - to make sure there is no corruption in the files itself......and will always save it clean with no data to made bit by bit comparision with the original file to make sure it was not corrupted while saving.....btw - do you think ECC memory errors are frequent enough to have a hight chance of occurring in a 6-hour calculation session?

bte you played on these versions of excel using 3930 right?

 

[Idontcare]

....as for the ecc memory - I can theoretically get away with the ecc memory by doing the following steps - I create...

Don't re-invent the wheel.

If software-based ECC was a good path to go down then you'd have the option of buying a $50 software app that would take care of all hardware ECC aspects and the entire ECC market itself wouldn't exist today.

No doubt you can convince yourself you've found a way to negate the need for ECC, but do you really think you've outsmarted the rest of the planet and discovered a software-based trick that no one else thought to commercialize?

If your financial calculations are worth money to your customers such that the risk of inaccuracy in the results compels you to replicate ECC via software cleverness then I put forward to you that your financial calculations are worth enough money that you ought to prioritize budgeting the hardware as necessary to purchase hardware-based ECC.

Remember your cleverness in this equation is supposed to be in the unique market analysis, not in your ability to build 100% uptime computers on the cheap.

If your cleverness really lies in being able to build 100% uptime computers on the cheap then your skills are wasted on financial analyses, the computer IT world alone could make you a very wealthy person.

Focus on your expertise, and rely on the expertise of others to build you reliable tools for your trade. I might be able to grow the best wheat on the planet but I'm not about to sidetrack my career to become a wheat farmer just so my eggs and toast in the morning is the best money can buy I'll leave it to the farmers to make the eggs and bread I buy in the supermarket.

 

[dima777]

Don't re-invent the wheel.

If software-based ECC was a good path to go down then you'd have the option of buying a $50 software app that would take care of all hardware ECC aspects and the entire ECC market itself wouldn't exist today.

No doubt you can convince yourself you've found a way to negate the need for ECC, but do you really think you've outsmarted the rest of the planet and discovered a software-based trick that no one else thought to commercialize?

If your financial calculations are worth money to your customers such that the risk of inaccuracy in the results compels you to replicate ECC via software cleverness then I put forward to you that your financial calculations are worth enough money that you ought to prioritize budgeting the hardware as necessary to purchase hardware-based ECC.

Remember your cleverness in this equation is supposed to be in the unique market analysis, not in your ability to build 100% uptime computers on the cheap.

If your cleverness really lies in being able to build 100% uptime computers on the cheap then your skills are wasted on financial analyses, the computer IT world alone could make you a very wealthy person.

Focus on your expertise, and rely on the expertise of others to build you reliable tools for your trade. I might be able to grow the best wheat on the planet but I'm not about to sidetrack my career to become a wheat farmer just so my eggs and toast in the morning is the best money can buy I'll leave it to the farmers to make the eggs and bread I buy in the supermarket.

thank you very much for this direct and valuable reply.....he thing is I woudl be more than happy to go the ecc route if I had a local reseller in the vicinity of the place I am staying in - in case of a memory failure or a motherboard - I would need to replace these with an order that would take from 2 to 4 days to arrive....I might stack double the repository with these parts but that would be quite expensive...therefore I am on the crossroads.....do not you think that checking the file to see if it has the same number of bytes as the original version can be a measure to protect against the file corruption at saving time?

 

[Idontcare]

thank you very much for this direct and valuable reply.....he thing is I woudl be more than happy to go the ecc route if I had a local reseller in the vicinity of the place I am staying in - in case of a memory failure or a motherboard - I would need to replace these with an order that would take from 2 to 4 days to arrive....I might stack double the repository with these parts but that would be quite expensive...therefore I am on the crossroads.....do not you think that checking the file to see if it has the same number of bytes as the original version can be a measure to protect against the file corruption at saving time?

That is a measure of error-detection, not error-correction. And you need your results to be correct, not just known to be wrong.

There are studies out there that show that the vast majority of errors corrected by ECC methods are systematic errors (meaning errors produced by failing hardware that hasn't quite died, or never quite fully functioned) versus the truly random errors caused by background radiation events.

In those cases, even with your bit-wise checkpoint approach you are just going to find out you have some walking wounded hardware that needs replacing because the errors aren't going to go away. At least with ECC you'll have error correction so you can rely on the results, and if you track the reporting of the errors then you'll know if you need to replace something in a timely manner rather than waiting until it truly dies on you.

 

[imported_PowerHouse]

that is super great to hear....as for the ecc memory - I can theoretically get away with the ecc memory by doing the following steps - I create a single calculation loop and perform it after say each of the 50 loops to see if the results match - to make sure there is no corruption in the files itself......and will always save it clean with no data to made bit by bit comparision with the original file to make sure it was not corrupted while saving.....btw - do you think ECC memory errors are frequent enough to have a hight chance of occurring in a 6-hour calculation session?

bte you played on these versions of excel using 3930 right?

Like the guys before me have said, ECC memory is worth it in your situation.

I was using an i7-980X EE which is a 6-core Gulftown vs. a 6-core SB-E for the i7-3930. Both address 12 threads via Hyper-Threading, so the architectural differences do not matter.

 

[imported_PowerHouse]

I just threw together a sample workbook that contains millions of cells; apart from some static cells that the computing cells reference, they are completely independent of each other meaning it should multi-thread perfectly. I tested it using varying numbers of threads and recorded the time it took to complete the calculation and this is what I got using Excel 2007:

3.333GHz Core i7-980X EE 6-Core / 12-Thread
* 2-Threads - 91.54 s
* 4-Threads - 46.50 s
* 6-Threads - 31.69 s
* 8-Threads - 24.24 s
* 10-Threads - 19.85 s
* 12-Threads - 16.89 s

Remember that you need to divide the number of threads by 2 to get the number of cores being utilized because of Hyper-Threading technology.

So, that works out to 542% increase in performance when using all 6-cores. As you add more threads, the linearity decreases probably due to Amdahl's Law and Turbo-Boost skewing things. Just for reference, on my dual Xeon box the increase from 2-threads to 32-threads showed a 1284% increase. As you can see, the performance gains can be immense if the workbook calculations can be threaded well.

 

[dima777]

I just threw together a sample workbook that contains millions of cells; apart from some static cells that the computing cells reference, they are completely independent of each other meaning it should multi-thread perfectly. I tested it using varying numbers of threads and recorded the time it took to complete the calculation and this is what I got using Excel 2007:

3.333GHz Core i7-980X EE 6-Core / 12-Thread
* 2-Threads - 91.54 s
* 4-Threads - 46.50 s
* 6-Threads - 31.69 s
* 8-Threads - 24.24 s
* 10-Threads - 19.85 s
* 12-Threads - 16.89 s

Remember that you need to divide the number of threads by 2 to get the number of cores being utilized because of Hyper-Threading technology.

So, that works out to 542% increase in performance when using all 6-cores. As you add more threads, the linearity decreases probably due to Amdahl's Law and Turbo-Boost skewing things. Just for reference, on my dual Xeon box the increase from 2-threads to 32-threads showed a 1284% increase. As you can see, the performance gains can be immense if the workbook calculations can be threaded well.

thank you for this amazing test you did for me))) very much appreciated!))) yes the speed does seem to rise proportionately to the number of cores used in the calculation....I wonder what XEONS make up your dual Xeon setup???

btw I have done some investigation and found this twin-brother of i7 3930k - INTEL Xeon E5-1650 (similar price as well)….it runs at the same frequency and support all kinds of features as well as the ECC memory…I wonder if you might comment on how this CPU works in practice? I have seen comments that this CPU is out of stock everywhere…..do you think it can be hard to find??? Thanks!)

 

[dima777]

That is a measure of error-detection, not error-correction. And you need your results to be correct, not just known to be wrong.

There are studies out there that show that the vast majority of errors corrected by ECC methods are systematic errors (meaning errors produced by failing hardware that hasn't quite died, or never quite fully functioned) versus the truly random errors caused by background radiation events.

In those cases, even with your bit-wise checkpoint approach you are just going to find out you have some walking wounded hardware that needs replacing because the errors aren't going to go away. At least with ECC you'll have error correction so you can rely on the results, and if you track the reporting of the errors then you'll know if you need to replace something in a timely manner rather than waiting until it truly dies on you.

interesting to learn that ECC memory helps to correct hardware related problems.....that must be very helpful for the stable operation of pc....

 

[imported_PowerHouse]

thank you for this amazing test you did for me))) very much appreciated!))) yes the speed does seem to rise proportionately to the number of cores used in the calculation....I wonder what XEONS make up your dual Xeon setup???

btw I have done some investigation and found this twin-brother of i7 3930k - INTEL Xeon E5-1650 (similar price as well)….it runs at the same frequency and support all kinds of features as well as the ECC memory…I wonder if you might comment on how this CPU works in practice? I have seen comments that this CPU is out of stock everywhere…..do you think it can be hard to find??? Thanks!)

No problem. I am running dual 8-core 2.0GHz Xeon E5-2650's. Just built it less than a week ago!

Well, the i7-3930K and that Xeon should perform exactly the same. However, it does seem to be hard to find. Personally, I would much rather have it than the E3-1275 v2. It might even be around the same speed even in single-threaded apps due to the larger cache.

 

[dima777]

thanks)...those xeons are pretty expensive....))) for now I could go the
E5-1650 route but of course it will be interesting to find a dealer here) I wonder what kind of work you submerge your dual xeon beast into?)

 

[crazymonkeyzero]

No problem. I am running dual 8-core 2.0GHz Xeon E5-2650's. Just built it less than a week ago!

Well, the i7-3930K and that Xeon should perform exactly the same. However, it does seem to be hard to find. Personally, I would much rather have it than the E3-1275 v2. It might even be around the same speed even in single-threaded apps due to the larger cache.

I actually now have both chips running at my work (as of last week) and in my tests, the i7 3930k is 25-29% faster in multitreaded tasks than the E3, and the Xeon E3 1275 V2 is about 5-10% faster in single threaded tasks (Which is expected due to ivy bridge architectural advantages over sandy)

@Dima777

I think this thread has kind of been going around in circles since last week. So I'm going to cut to the chase. After reading everything, you really have two effective choices in terms of cpu, assuming you won't even consider 8 cores due to cost.

1. Hex Core Xeon E5 1660

or

2. Quad Core Xeon E3 1275 V2

Here is my conclusion

You need ECC memory, since the data you work with is valuable.
Thus do not consider anything except xeons. (3930k is out)

In my opinion, if you have the budget, go with option 1. You will have an overall faster system with the hex core. And multi threaded works will be finished with ease. Most software is multithreaded these days, and this will be moreso in the future. Furthermore, you can upgrade to dual socket in the future if you decide to.

If you have limited budget go with option 2, the quad core E3 is the better route. Save some dough, and in the future, you can build another similar PC which together will probably be faster than any dual socket can offer. This chip is very capable.

I don't really think there is anything more to consider.... Most of applications you use are multithreaded and even if you have that one single threaded app, that will perform 5% slower on the hexcore than the quad, I don't think it's worth fretting over~I willing to bet either of these chips are probably much faster than what you have now.

 

[dima777]

I actually now have both chips running at my work (as of last week) and in my tests, the i7 3930k is 25-29% faster in multitreaded tasks than the E3, and the Xeon E3 1275 V2 is about 5-10% faster in single threaded tasks (Which is expected due to ivy bridge architectural advantages over sandy)

@Dima777

I think this thread has kind of been going around in circles since last week. So I'm going to cut to the chase. After reading everything, you really have two effective choices in terms of cpu, assuming you won't even consider 8 cores due to cost.

1. Hex Core Xeon E5 1660

or

2. Quad Core Xeon E3 1275 V2

Here is my conclusion

You need ECC memory, since the data you work with is valuable.
Thus do not consider anything except xeons. (3930k is out)

In my opinion, if you have the budget, go with option 1. You will have an overall faster system with the hex core. And multi threaded works will be finished with ease. Most software is multithreaded these days, and this will be moreso in the future. Furthermore, you can upgrade to dual socket in the future if you decide to.

If you have limited budget go with option 2, the quad core E3 is the better route. Save some dough, and in the future, you can build another similar PC which together will probably be faster than any dual socket can offer. This chip is very capable.

I don't really think there is anything more to consider.... Most of applications you use are multithreaded and even if you have that one single threaded app, that will perform 5% slower on the hexcore than the quad, I don't think it's worth fretting over~I willing to bet either of these chips are probably much faster than what you have now.

thank you very much for clearing this out for me...

I think I have finally nailed a very attractive offer that might well suit my needs perfectly....well nearly perfectly....
I have found a workstation by HP which features the much coveted Intel Xeon E5-1650 processor!!!!!!! and its price is within the constraints

that I have -

HP Workstation Z420 - Xeon E5-1650 3.2 GHz

take a look here - http://www.amazon.com/HP-Workstation...d_sim_sbs_pc_6

I am slightly weary of building my own custom build so this one looks very attractive....it features the e5-1650 as well as the 8 gbs of ram

and is all factory packed and of course ecc and everythign super fail proof I guess...please let me know what you think!!!!!)




I have a couple of more questions...do you think it might be possibel to add more 8GBs of Ram and a video card without voiding the warranty?
thanks!!
feel so close to the final step!)

 

[imported_PowerHouse]

thanks)...those xeons are pretty expensive....))) for now I could go the
E5-1650 route but of course it will be interesting to find a dealer here) I wonder what kind of work you submerge your dual xeon beast into?)

Yeah, and what's crazy is they are the cheapest 8-core models! I just built it for fun as an earlier Christmas and graduation present.


That HP looks nice. I would almost certainly say you could upgrade the RAM and add in a GPU without warranty trouble.

 

[dima777]

Yeah, and what's crazy is they are the cheapest 8-core models! I just built it for fun as an earlier Christmas and graduation present.


That HP looks nice. I would almost certainly say you could upgrade the RAM and add in a GPU without warranty trouble.

yes....thanks))) I wonder if there are some happy owners of hp z420 here who can comment on their workstation??