Why not buy a Xeon instead of i5 or i7?

[harobikes333]

I don't plan on overclocking at all. I primarily game / browse a billion websites at a time / back up my dvds & CDs.

Would the Intel Xeon E3-1240 V2 be my best bet for around 265~300 range..?

 

[Soulkeeper]

I too was looking at the 1155 xeons
for me the main/only reason for getting one would be ecc support
E3-1245 looks tasty
still cheaper than what I payed for Q9450 when it came out.

It's a shame AMD's options aren't so compelling.
They just can't touch the power usage or performance of the xeons
also things like no pcie 3.0

Also i've heard the xeons can use unbuffered ECC in desktop motherboards if the bios supports it. After googling for awhile it appears many gigabyte and asus boards do.

 

[Jeff7181]

ECC is like driving a car which has an airbag. You won't need it, nor appreciate it, until you have an accident.

Can you drive your car without an air-bag? Yes, millions of people do everyday.

Are you safer with an air-bag installed in your car? No...unless you happen to be in an accident, in that case the answer is yes (except under very rare flukey situations).

Are there even safer ways to travel? Yes. If safety truly is your number one priority then traveling by vehicle with an airbag is still going to be unacceptable.

But if you should happen to find yourself traveling by car and you do happen to find yourself involved in a head-on collision then you are probably going to wish your car has an air-bag.

ECC ram in a consumer box is like that. It is completely unnecessary until it is needed, then you'll probably really wish you had some all along.

Absolutely ridiculous. I dare you to find an instance where ECC saved anyone's life.

*EDIT* In addition, I've been running ESX on the cheapest consumer level junk hardware I could find when I built the box 4 years ago, and haven't had a single memory error. You're fooling yourself if you think you'll benefit from using ECC RAM at home.

 

[aigomorla]

*EDIT* In addition, I've been running ESX on the cheapest consumer level junk hardware I could find when I built the box 4 years ago, and haven't had a single memory error. You're fooling yourself if you think you'll benefit from using ECC RAM at home.

i use ECC ram at home...
Only because my board wont accept anything but ECC.

Now if i was doing massive DB and compiling tho, i dont think i would build a system without ECC.

 

[blastingcap]

i use ECC ram at home...
Only because my board wont accept anything but ECC.

Now if i was doing massive DB and compiling tho, i dont think i would build a system without ECC.

I don't think that Jeff guy knows what silent data corruption means.

 

[Smartazz]

A friend of mine uses the Xeon equivlent of the 2600K. He can't overclock the computer because it's a work PC.

 

[Jeff7181]

I don't think that Jeff guy knows what silent data corruption means.

I don't claim to know all, but I am a storage administrator and I know how to protect data. Data at rest is most prone to corruption... not data in RAM.

I'm not saying don't use ECC, I'm saying (for at least the second time) that seeking out ECC hardware for home use is ridiculous.

If you're concerned enough about data corruption, before buying ECC memory you best get yourself some extra hard drives and run an appropriate RAID level and do regular RAID scrubs, because even CERN's data shows that data on platters is more susceptible to "silent data corruption" than data in RAM: http://www.zdnet.com/blog/storage/data-corruption-is-worse-than-you-know/191

 

[Cerb]

If you're concerned enough about data corruption, before buying ECC memory you best get yourself some extra hard drives and run an appropriate RAID level and do regular RAID scrubs, because even CERN's data shows that data on platters is more susceptible to "silent data corruption" than data in RAM: http://www.zdnet.com/blog/storage/data-corruption-is-worse-than-you-know/191

No doubt. The differences are that RAID and filesystems are cheap and easy to use (for relative values of easy), and that disk corruption has been known as a common occurrence for decades. NTFS checksums most data on the drive, and is usually pretty good about CRC errors, IME (I'm really more disappointed in MS doing half-assed things like ReFS, instead of making real next-gen FS to replace NTFS--it will be more cost on their part, more stress for users, and do less good, over the years). Data I care about also rests on EXT4, which checksums most everything they could without breaking EXT3 (I plan to move to BTRFS once the offline FSCK gets farther along), and static data rests in archives with parity to back them up. All of that cost only the use of my time which was already free to use. The added cost of HDDs should never even be questioned: you either need RAID, or automatic backups to a second HDD (in either case, more HDDs are needed than just enough to fit the data). For most users, that would be a good reason to get a specialized NAS box, and maybe commercial backup software for their desktops (which might do incremental or diff backups with its own CRCs). With CRC checking, backups are generally enough, as long as you can be sufficiently confident of the initial storage effort.

Now, that's mostly taken care of. What's left? ECC RAM is definitely not the first line of defense, but is the last line of defense that hardware functioning as intended lacks, and it has uses in early detection, and diagnosis of, hardware that may be bad or failing. Then, Google's first study came out, and then the follow-up, more recently, showing (a) that hard errors were more common than soft, (b) that error rates correlated positively with utilization (not unexpected, but the old alpha particles hitting resting data does not fit), and (c) that the rate of errors on hardware not otherwise malfunctioning was far greater than previously thought, (d) though many systems never encountered any errors at all.

Add to that, without ECC RAM, you can not state with high confidence that you have not experienced an error, even if the chances are <1/92nd/yr. Since data CRCs make verification checking of resting data fairly easy w/o ECC, I'm of the opinion that ECC on a home server should be done after the desktops, not before, unless said server is also running VMs that do a lot of data manipulation (not the common home server case).

The big problem with ECC today is that the marginal cost is high, due to Intel's market segmentation, and AMD's lack of competitive CPUs (if BD hadn't been the flop it was, I probably would have upgraded, already). A $400 PC becomes a $500+ PC, and a $1000 PC becomes a $1200+ PC that can't be overclocked much, rather than just needing to spend a little more on the RAM (assuming data and command parity). That makes implementing it a non-trivial value judgement.

 

[know of fence]

Now, that's mostly taken care of. What's left? ECC RAM is definitely not the first line of defense, but is the last line of defense that hardware functioning as intended lacks, and it has uses in early detection, and diagnosis of, hardware that may be bad or failing. Then, Google's first study came out, and then the follow-up, more recently, showing (a) that hard errors were more common than soft, (b) that error rates correlated positively with utilization (not unexpected, but the old alpha particles hitting resting data does not fit), and (c) that the rate of errors on hardware not otherwise malfunctioning was far greater than previously thought, (d) though many systems never encountered any errors at all.

This sums up the results without speculation. But if you were to draw conclusions, would you say that you lean more or less towards ECC ram after seeing the Google study?

Also for those who have experience with this, is there any way to see how many errors were corrected, which would be necessary to use it as early warning system for aging or failing hardware. Google can do it, but can you?

My personal experience with working on a PC with broken RAM (before learning about memtest, lol), confirms that file operations are rather unaffected. Instead there is random system instability, failing installation processes, errors during transcoding, compression. Errors generally occur when a lot of data is written, not copied.

In the end any kind of redundancy comes at a price while horror stories about data loss, can not replace level-headed risk assessment. Something that cannot be expected on a consumer level. My amateurish take on this is that while RAM sizes keep increasing, the overall risk may be declining i.e. the risk that a soft error happens AND hits a critical file which is THEN saved to storage.
Also the size of OSs, apps and their RAM use may be disproportionately increasing due to pretty pictures and video, rather than code critical to system stability.

The reason why memtest86+ runs infinitely, is so that it is possible to make an assessment of just how often errors crop up, once an hour? once a day?. Meanwhile a system with ram that is producing errors every second, is still operational, though not exactly stable.

 

[Cerb]

Also for those who have experience with this, is there any way to see how many errors were corrected, which would be necessary to use it as early warning system for aging or failing hardware. Google can do it, but can you?

Windows and Linux both log such errors. Google might have added some special data collection tools, but they weren't using any special features that you don't have available.

 

[know of fence]

Windows and Linux both log such errors.

I looked a little bit into this, because being able to say, my ECC ram logged X million single-bit errors last year, would make a strong argument for buying ECC ram. The problem with this is that even though there is an error-counter it is reset every so often.

Windows7 actually automatically does what the Google study is suggesting, it's called Predictive Failure Analysis (PFA), it logs errors and once reoccurring ECC errors reach a certain threshold it marks the memory page and disables it upon boot-up.

It is possible to set the time interval as well as threshold or use a driver plug-in to take over this monitoring, but there appears to be no simple way to view these logs.

Explanation
http://msdn.microsoft.com/en-us/library/ff559459.aspx

Why there is no way to look up how and where memory has gone bad.
http://msdn.microsoft.com/en-us/library/ff728892.aspx

 

[Cerb]

I looked a little bit into this, because being able to say, my ECC ram logged X million single-bit errors last year, would make a strong argument for buying ECC ram. The problem with this is that even though there is an error-counter it is reset every so often.

Windows7 actually automatically does what the Google study is suggesting, it's called Predictive Failure Analysis (PFA), it logs errors and once reoccurring ECC errors reach a certain threshold it marks the memory page and disables it upon boot-up.

It is possible to set the time interval as well as threshold or use a driver plug-in to take over this monitoring, but there appears to be no simple way to view these logs.

Explanation
http://msdn.microsoft.com/en-us/library/ff559459.aspx

Why there is no way to look up how and where memory has gone bad.
http://msdn.microsoft.com/en-us/library/ff728892.aspx

The Event Log, found under Administrative Tools, will list any MCA errors that were reported. You get to know that an error occurred, what type, and where.

PFA is actually a pretty advanced feature, attempting to apply historical analysis. Microsoft probably assumes that if you're worried about it to that degree, you're going to have many servers, and will likely be using one or more domain-wide management suites, which conveniently collect and filter various log results, among other duties.

 

[DraGoNsLaYeR786]

I have the option of getting a Xeon E5-2658 from work. Clock speed is low compared to i7 options but it has eight cores. Will be worse than i7 (such as i7 3770) because of the clock speed?

Processor info:
http://ark.intel.com/products/61428/Intel-Xeon-Processor-E5-2658-20M-2_10-GHz-8_0-GTs-Intel-QPI

 

[DPOverLord]

I dont mean to open a new thread but my friend is debating between the i7 3770 or the Xeon 1275 on his AsRock Z75 Pro3.

Are any of the Xeon's able to be overclocked or is the only solution to go for the 3770?

 

[Burpo]

You resurrected a 5 yr old thread! But yes, get the 3770k, no overclocking on the Xeons.

However an E3-1280 v2 is a tick faster than a stock 3770k ,and ~$50 cheaper. No graphics tho.

 

[blackarchon]

...
Are any of the Xeon's able to be overclocked or is the only solution to go for the 3770?

There aren't any overclockable Xeons for sockets 1155, 1150 and 1151. Only some Xeons from the E5-16xx line are overclockable (see also).