E5-2690 vs 3960X

[KoSoVaR]

First thing I need to mention, price is not an issue here. I've been searching for information regarding head to head tests of the E5-2690 vs overclocked SB-E systems but have not had much luck.

I'm aware that you can not overclock the 2690s. What I'm mainly interested in is getting the fastest, most stable CPU. Something that will go head to head in a benchmark test and beat the 2690s.

In a mission critical environment when speed is everything, but stability is essential, how would you feel about using non ECC memory and a highly overclocked machine to get a huge gain over the competition? Ups, downs?

If anyone can direct me to some good information it would be much appreciated.

 

[InvalidError]

In a truly "mission-critical" environment, substituting desktop parts for server parts is simply not an option because the liabilities of any failure or failure to prove that every reasonable step to prevent or mitigate such failures have been taken usually far outweigh any possible savings.

Imagine a bank or stock exchange accidentally losing track of a few billion dollars due to a single-bit error on an overclocked transaction server running non-ECC RAM...

 

[orca12345]

Gaming based- i7 3960x; acctually at this point if you are gona buy that why not just pay the same price or about $25 more for the 3970x.

Server-Xeons are better, BUT in a game environment they are inferior.


Honestly, xeons will always win a benchmark test. If you want a benchmark computer get the xeons. Its stupid however to build a computer around benchmark results. You build it based on what you are going to do with it.

 

[KoSoVaR]

Gaming based- i7 3960x; acctually at this point if you are gona buy that why not just pay the same price or about $25 more for the 3970x.

Server-Xeons are better, BUT in a game environment they are inferior.


Honestly, xeons will always win a benchmark test. If you want a benchmark computer get the xeons. Its stupid however to build a computer around benchmark results. You build it based on what you are going to do with it.

Let's say it's folding 100% of the time. Constantly crunching numbers.

As for InvalidError, I am completely aware of the consequences of 1 bit flipping and the havoc it could cause. I believe in this case I am trying to figure out exactly how much more productive a desktop part can be versus the Xeon E5-2690 in a setting where we're talking raw performance. No what-ifs.

 

[InvalidError]

where we're talking raw performance. No what-ifs.

In your OP, you described your application as "mission-critical" and in truly mission-critical applications, working out the "what-ifs" takes precedence over raw performance.

If you can afford the risks of "highly" overclocking (which "highly" eats into the parts' safety margins) and using non-ECC then your application must not be mission-critical at all since there seem to be little to no consequences aside from lost time (mostly yours or whoever uses the workstation) should there be a crash or data corruption.

 

[KoSoVaR]

In your OP, you described your application as "mission-critical" and in truly mission-critical applications, working out the "what-ifs" takes precedence over raw performance.

If you can afford the risks of "highly" overclocking (which "highly" eats into the parts' safety margins) and using non-ECC then your application must not be mission-critical at all since there seem to be little to no consequences aside from lost time (mostly yours or whoever uses the workstation) should there be a crash or data corruption.

You are correct and I apologize about the information I provided being ambiguous. I use high end CPUs for mission critical work - and yes, I would never want to jeopardize the integrity of the results that my CPUs are providing. However, I would like to build a few systems that are highly overclocked, but only knowing that they can truly offer a large increase in, for example, "crunching numbers".

 

[mcarey2249]

Have you looked into MPICH2 applications (I believe its message passing interface). I use this program to cluster together computers to perform radiation transport simulations. So instead of spending all that money on one computer, you could split it between a few and have overall much more power. Currently, I have 32 cpus joined together... I would recommend looking into this option.