Registered ECC CL=13 vs Unbuffered NON-ECC CL=9 (PC-14900)

[Kauf771]

Hi
Isn't the CAS Latency in Registered ECC modules too high compared to Unbuffered NON-ECC ones? In other words, is the difference between CL9 and CL13 a noticeable performance hit in real life work time, or is it more of a performance reference for tech analysis, only humanly perceived through memory benchmarks?
Are Unbuffered NON-ECC options enough reliable to be seriously considered when building a workstation for complex 3d modeling and image creation, where rendering times usually are 24h+ and polygon operations require a robust system, even being aware that memory errors will probably abort the rendering process or give you some odd mesh results?

 

[WyomingKnott]

You will see a lot of different opinions on this subject, some of them quite strongly held. My personal opinion, not based in experience, is that ECC should be used for mission-critical or long-running processes. I don't tolerate disks scrambling my data, why should I tolerate it from memory?

Then again, I believe that cosmic rays can flip memory bits. Some people think that this places me in the tinfoil-hat brigade.

 

[Tradesman1]

ECC DRAM has long had a place in high end servers, but even that is slipping by the wayside

 

[Kauf771]

After reading some arguments, I think I will go with ECC. And about some cosmic rays being able to flip some memory bits, well, I think that's odd, but may be possible...
But now I have another doubt.
Does ECC memory modules use their extra chip to correct errors (even in Independent Memory Mode), or they must be set for redundancy (like Mirroring or Sparing) in order to do so?

 

[Tradesman1]

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

 

[Pinhedd]

Hi
Isn't the CAS Latency in Registered ECC modules too high compared to Unbuffered NON-ECC ones? In other words, is the difference between CL9 and CL13 a noticeable performance hit in real life work time, or is it more of a performance reference for tech analysis, only humanly perceived through memory benchmarks?
Are Unbuffered NON-ECC options enough reliable to be seriously considered when building a workstation for complex 3d modeling and image creation, where rendering times usually are 24h+ and polygon operations require a robust system, even being aware that memory errors will probably abort the rendering process or give you some odd mesh results?

CAS latency is not as big of a deal as many people make it out to be. Modern memory controllers (especially Intel's) are good enough to keep the DRAM bus nearly 100% busy when needed and microarchitecture can still keep itself busy while waiting for data from the DRAM to come back. Most servers and high end workstations use memory with looser timings, it's nothing out of the ordinary.

If you are running workflows that take upwards of 24 hours to complete then ECC memory is almost a necessity. A single bit error that crashes or corrupts an application is not a big deal for someone who just plays video games and browses the web but it is a big deal when it interrupts or corrupts a revenue generating process.