Are there any ECC laptops?

[wild_bill]

I have a requirement for a laptop which can utilize ECC RAM, does anyone know of any current laptop or past model from the last 3 years than can use ECC? Thanks! this really means a lot to me if you can help me find some candidates!

(PS - Please, do me a favor and save the "you don't really need ECC" comments, I know what I need, just having trouble locating a laptop that can do it! )

[RTAdams89]

Originally, I thought some of the ToughBooks had ECC memory as an option, but after checking, I can't find any evidence of this. As far as I know, there are no laptops that support ECC memory. I did find some suppliers of ECC SODIMMs, which at first I thought might indicate the existence of an ECC supporting laptop, but all indications are they were designed for embedded systems and Telecom equipment. Good luck on your search, and please let us know if you find anything.

[rkawakami]

Assuming that you already have some ECC SODIMMs, I'd contact the manufacturer and ask them which laptops their product(s) were designed for. For example, http://www.virtium.com/ makes DDR2 SO-CDIMMs so I would contact their sales/marketing/app support departments. I would expect that any laptop which accepts ECC modules is going to cost $$$$$$$$$$$ .

Are you planning on operating a laptop atop a mountain and want protection against random alpha particle hits on the memory?

[hkazemi]

I have looked for laptops with ECC several times. I think it would be great if Lenovo decided to add ECC support to their W series laptops...they do market them as mobile workstations afterall.

ECC memory is only part of the equation, though, as you also need chipset support, and probably BIOS support too. At one point there was an initiative to create software-based ECC. Search for 'SoftECC' or see this article:
http://pdos.csail.mit.edu/papers/softec ... n-meng.pdf

I've only found one laptop that supported ECC. It was the Sun Ultra 3 Mobile Workstation which came with Solaris 10:

From http://www.sun.com/desktop/workstation/ ... atures.xml

Processor choices:
1.2 GHz UltraSPARC IIIi processor
Up to 2 GB DDR RAM, error correction SDRAM
64-bit SPARC Solaris OS applications in a laptop
Sun quality laptop screens for better visibility

UltraSPARC Performance
Power to spare
You can specify your choice of a 550 MHz or 650 MHz UltraSPARC IIi processor or a 1.2 GHz UltraSPARC IIIi processor. UltraSPARC IIi processors deliver impressive power savings and throughput with on-chip Level 2 cache as well as 64-bit wide memory interfaces and internal busses. UltraSPARC IIIi processors also conserve power but add the processing muscle to run web and application servers and CAD/CAM applications. Either way, you'll have complete compatibility with more than 12,000 Solaris applications developed for the SPARC family.

Memory Choices
Some customers want a low-cost notebook and others need the equivalent of a mobile server for the road. Now you have the choice. By specifying the 1.2 GHz UltraSPARC IIIi processor and the full 2 GB DDR RAM (error correction SDRAM), you make your large applications fly at full speed.

From http://www.sun.com/desktop/workstation/ultra3/specs.xml

550 MHz or 650 MHz UltraSPARC IIi processor or 1.2 or 1.28 GHz UltraSPARC IIIi processor
Up to 2 GB DDR RAM, error correction SDRAM
80 GB IDE internal disk storage

[sktn77a]

PS - Please, do me a favor and save the "you don't really need ECC" comments, I know what I need, just having trouble locating a laptop that can do it!

Tell us...... we're all dying to know!

[wild_bill]

Tell us...... we're all dying to know!

I need a laptop that will be able to correct the greater than 1 per week errors that I (and all of you) get from cosmic rays and other sources - the accuracy of my data is often critical, especially when permanently written to disk or SSD or USB flash drive.

you might want to read this for a refresher

[perbilse]

Hi,

I came across this thread while researching essentially the same question, and thought I'd post here what I found, in the hope some people may find it useful.

There are in fact several manufacturers of ECC SO-DIMMs. Such DIMMs are correctly referred to as SO-CDIMM, and your favourite search engine should provide plenty of links if you look for SO-CDIMM. On the downside, these items are intended and priced for an uncompromising professional market, and cost $200-$300 per G, except that I just came across an ebay listing that (as of this date) offers 1G modules at $50; look for SG572288FG8CWDGME2 (I have no connection with the seller). I have ordered two modules in the hope they'll be what I want.

An entirely different issue is motherboard support, but I have seen it claimed that all AMD Opterons and derivatives do support ECC even if disabled in the BIOS, given that their MMU is on the CPU chip. Hence, on Linux ECC can be forcibly enabled, though mileage may apparently vary.

I'll report back later.

Best,

-- Per

[Pete B]

I worked in CPU design back in the 1980s first for a custom super computer for the government, and later for a small company doing a mini computer sized clone of the IBM 370 architecture. My main task as an EE was to design the complex pipeline control logic for the CPU but I also helped with the DRAM soft failure rate analysis and I find many issues with that paper. DRAM design has changed to protect the capacitors from Alpha particle radiation and this is partly addressed in a follow up to that original analysis:
http://lambda-diode.com/opinion/ecc-memory-2

The author there talks about the radiation coming from externals sources but I remember at the time that the manager of our group found some claim that it came from impurities in the packaging materials. That sounded far fetched at the time, but remember the capacitors holding the charge in these DRAMs are beyond microscopic so it does not take much to disrupt the charge.

I did read this part of the OP:
"(PS - Please, do me a favor and save the "you don't really need ECC" comments, I know what I need, just having trouble locating a laptop that can do it! )"
but at least read part 2 of the paper, if you are going to reference part 1.

Let me offer this quote for people who might not want to bother reading part 2:
"The most interesting thing that we learn from that paper is that cosmic ray-induced errors are negligible. Indeed, if we are to believe them, two-thirds of their machines ( 67.8% ) detected absolutely no memory errors in two and a half years, in spite of having gigabytes of memory. The exact number of machines is undisclosed, but is believed to lie between 10 000 and 500 000 ; the average amount of RAM per machine is also undisclosed, as is the average RAM utilization."

The last quote is not the entire story and therefore I hope people read the entire paper. This was also quite interesting:
"The difference is that a large fraction of their DIMMs experience no errors at all for years. This seems to rule out cosmic rays as a cause for most of the errors. Indeed, the coefficient of variation accross machines of the error rates is huge, meaning that there is a strong machine effect. Maybe some of their machines are contaminated by radioactive materials, subject to abnormally strong electromagnetic fields, or simply substandard."

It is possible that some DRAM manufactures have found ways to protect against alpha particle radiation (what they are referring to as cosmic rays) or that the problem is not even real and that the soft error rate is due to noise related flaws in the electrical design. I'd sure like to know who manufactured the RAM and the systems that showed "no errors at all for years" - they are obviously doing something right. It is also possible that they are doing something very wrong, lol, perhaps the RAM error detection interrupt is erroneously and permanently disabled - even when set to enabled.

The analysis in part 1 of the paper is IMO flawed in many, many ways.

[rkawakami]

Back in the day (circa 1987) when I was an SRAM and DRAM test engineer, I had to write a test program to monitor soft bit errors that resulted when a radioactive source was placed in close proximity to the DRAM. Indeed, the molding compounds used to encapsulate the chip (typically plastic, but ceramic materials are also used) were the main source of alpha particles. Naturally occurring radioactive elements (I forget which) end up in the package material and were bombarding the IC. In order to reduce the chance that a random strike would disrupt the memory cell contents, a polyimide coating was applied to the top surface of the wafer, prior to packaging. I would assume that this is still being done with today's DRAM processes, in addition to circuit designs which might minimize the chance of any data disturbance due to cosmic rays.

Speaking of random memory failures, one of the main problems with current DRAM cell design is the inherent reduction in data retention time caused both by utilizing smaller capacitors in the memory cell and lower operating voltages. This data retention time is generally known as the "refresh" interval in DRAMs. Back in the mid-80's, wafer process technology was at the 1 micrometer (um), or 1000 nanometer (nm), minimum feature size level and typically were operated at 5V. Today, we're around 30nm and 1.5V, or a 35 times reduction in size and less than 1/3 of the voltage. Being able to pattern smaller features on the wafer is what has enabled memory density to vastly increase over these years. However as a result, the potential electrical charge stored in the memory cell's capacitor has decreased. This charge is what allows the memory to store a 0 or a 1. By nature, a capacitor is able to keep this electrical charge only a finite amount of time. Think of it this way: a capacitor is simply a very poor battery. All batteries will eventually lose their charge over a period of time; most take years. The capacitor in a DRAM memory cell will self-discharge in a matter of seconds, if not sooner, unless you periodically "re-charge" them. Again, back in the mid-80's I was testing DRAM (64Kb) which had more than 4 seconds of retention time. Several years later after a couple of process shrinks had occurred and memory density increased, retention time was down around a second for 1Mb devices. Following an industry-wide memory standard, DRAMs are specified to have a refresh interval of 64 milliseconds (ms). This means that data is guaranteed to remain intact because after every 64ms, all of the capacitors are recharged (refreshed) by the memory controller in the computer. Memory manufacturers usually test their products to guarantee this specification by testing them using a longer refresh interval, say 128ms, higher temperatures (typically refresh gets worse the hotter the chip gets), voltages outside the "normal" operating range or any combination of these things.

As you suggest, some of these system errors could also be due to voltage fluctuations (noise) or they could be bits that have a marginal refresh. It's very hard to determine which it is unless you have a logic analyzer/oscilloscope hooked up to the system and monitor every signal and voltage pin.

This white paper touches on some of the causes for these randomly occurring errors: http://www.smartm.com/files/salesLitera ... er_sbe.pdf

[Pete B]

Perhaps we should break this off into a new thread but I do think that it is an interesting discussion.

I believe that it was the deep trench DRAM cell architecture, or perhaps a similar precursor to it that helped protect the cell simply as a result of the physical configuration. Google shows up many hits on deep trench DRAM as there are constant improvements to this basic configuration. There were probably many varied advantages for this configuration.

Interesting that you confirmed the claim for impurities in the packaging materials. It is good to know what is actually going on in order to avoid invalid assumptions.

[perbilse]

Hi,

I can now report that unfortunately SO-CDIMMs are an entirely different form factor. I'm a bit puzzled that none of the sources I originally found mentioned this, but there you have it.

http://www.flickr.com/photos/13067182@N07/8815306799

SO-CDIMM on top, regular SO-DIMM on bottom.

If anybody ever finds a laptop that supports SO-CDIMMs I'd love to hear about.

Best,

-- Per

[perbilse]

Correction to myself! The photo shows in fact a Mini-DIMM on top; it had been labelled as a 200-pin SO-CDIMM, but is a 244-pin Mini-DIMM.

So there's still hope, if I find SO-CDIMMs at a reasonable price I'll update here.

Cheers!

-- Per