7 Replies Latest reply on Apr 23, 2013 3:43 AM by yurtesen

    Does HD 7970 have ECC RAM?

    RezaRob

      I find contradictory information about how the 7970 and W9000 differ.  Many people say they are the same thing, but the official specs page indicates ECC RAM only for W9000.  Which cards have the ECC RAM support?

       

      Thanks.

        • Re: Does HD 7970 have ECC RAM?
          himanshu.gautam

          AFAIK, the information is correct. I can also see ECC support for W9000 card, but not for AMD 7970 card on official special pages of these cards.

          • Re: Does HD 7970 have ECC RAM?
            iqbal51

            I think you choose the rite thing. You can use 75 series cards for ECC RAM.

              • Re: Does HD 7970 have ECC RAM?
                yurtesen

                AFAIK AMD cards support EDC which checks data validity transferred between memory and chip. But W9000 additionally ECC if you want to detect if data was changed within the memory chip after storage. Do you really need ECC?

                1 of 1 people found this helpful
                  • Re: Does HD 7970 have ECC RAM?
                    RezaRob

                    I clicked "Helpful Answer" because you mentioned "data validity transferred between memory and chip" and I read about that somewhere else last night.  I'm starting to feel like this might be the case with the gaming cards(7970 etc.)

                     

                    Nevertheless, what I was hoping for was definitive documentation, and not what people "feel" might be true.

                     

                    Do I need ECC?  This is how (VERY) unreliable RAM is(according to Wikipedia/Google/UToronto):

                    "A very large-scale study based on Google's very large number

                    of servers was presented at the SIGMETRICS/Performance’09

                    conference.[4] The actual error rate found was several orders of

                    magnitude higher than previous small-scale or laboratory studies, with

                    25,000 to 70,000 errors per billion device hours per megabit (about

                    2.5–7 × 10−11 error/bit·h)(i.e. about 5 single bit errors in 8

                    Gigabytes of RAM per hour using the top-end error rate), and more than

                    8% of DIMM memory modules affected by errors per year."

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

                    ---------------------------

                    However, my brother says he's not convinced.  He tells me to run memtest overnight, and  _prove_  to him that it really is that bad.  (I think according to the Google paper, only  _some_  modules are affected by higher error rates.)

                     

                    PS: 

                    ECC price shouldn't be so high.  

                    Even "real people" run physics simulations these days.

                     

                    Reza.

                      • Re: Does HD 7970 have ECC RAM?
                        yurtesen

                        Reza, you can run MemTestCL on your GPU. However I have to warn that the 1.0 version had a bug which caused false failure reportings on GCN architecture. you may want to build it from latest sources (there was a thread here about the bug in memtestcl also):

                        https://github.com/ihaque/memtestCL

                         

                        Obviously it is difficult to put every tiny feature listed, but here are some mentions of EDC in the following documents:

                        http://www.amd.com/us/Documents/GPGPU-Embedded.pdf

                        http://developer.amd.com/wordpress/media/2012/10/Heterogeneous_Computing_OpenCL_and_the_ATI_Radeon_HD_5870_Architecture_201003.pdf

                        http://developer.amd.com/wordpress/media/2012/10/GPU%20Computing%20-%20Past%20Present%20and%20Future%20with%20ATI%20Stream%20Technology.pdf

                        Maybe it is not mentioned because EDC is one of the default features in AMD GPUs.

                         

                        I wouldnt trust Google's study, not because I am paranoid They run barebone servers with ultra cheap hardware, probably without good shielding and close to each other. Your card will be inside a well shielded PC case at the very least. That said, I ran memtestcl on some radeon cards for over 24 hours and got 0 errors (with the bugfixed memtestcl program)

                         

                        To be honest. 5 errors per 8GB ram per hour is quite high in any scale. I can today run normal memtest86 on my desktops (16/32GB RAM) and get no errors at all for 24+h run. In my experience errors only occured when the RAM was defective. (perhaps I would see some errors if I let the machine running memtest for several years)

                         

                        I do program and run physics simulations and not having ECC is not a big problem for small simulations. If your simulation will take 1+month to run, perhaps then it may make more sense to run it with a device capable of ECC. It would be well worth the investment in that case. But especially if you are simply developing a program then you can start with a 7970 and you shouldnt have any problems.

                         

                        I agree with your brother but the pricing for cards with ECC is high, because they are not sold/produced in so large quantities, require additional design changes, often have more memory than cards targeted for gaming etc.

                         

                        Thanks,

                        Evren

                        1 of 1 people found this helpful