Announcement

Collapse
No announcement yet.

The ECC DDR4 RAM Overclocking Potential With AMD Threadripper On Linux

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #11
    Originally posted by ddriver View Post
    But for epyc, with its 8 channel MC, it is also kinda redundant as bandwidth is ample even at lower clocks.
    epyc does have plenty of memory bandwidth, but it can still benefit from the infinity fabric running at higher clock speeds.

    Comment


    • #12
      Originally posted by ddriver View Post
      Overclocking doesn't imply pushing to the limit, merely above the stock clocks. It doesn't need to introduce instability. Crucal has ecc modules rated at 2400 and 2666 that are practically identical, the only thing that differs is the timings, and not surprising in the least, they all cost the same down to the cent. It is just two SKUs of the same hardware product, so overclocking the 2400 one to 2666 is something I doubt will result in the slightest increase in instability.
      You do know of binning, right? Not all silicon is equal. Typically (though probably not always) slower speeds exist because the hardware can't be pushed any harder. When you've got a product focused on stability, the constraints are much tighter.
      I agree going from 2400 to 2666 isn't going to have much (if any) impact on stability, but that's also not much of an overclock either.
      There is quite a margin for ddr4 to safely overclock as many chips are already capable of doing in excess of 4000, and the only reason ecc memory is still stuck at 2666 is jedec never bothered to update their specifications, which they didn't do not because it is unfeasible, but because it is something that they haven't deemed necessary for the time being.
      You say that as though 4000 is a standard for non-ECC, which as far as I'm concerned, it isn't according to JEDEC (2400 was the standard but I think it was increased to 3200). That being said, as far as I'm concerned, JEDEC can't be the reason why ECC is stuck at 2666, since non-ECC doesn't conform to their standards either.
      RAM manufacturers care about profits, which is why they go ahead and release "pre-overclocked" modules. So, the lack of ECC at 3Ghz+ is either because:
      A. There's some industry regulation I'm not aware of that legally prevents them from doing so.
      B. There's no demand (which is BS).
      C. Because they can't guarantee performance/reliability - something ECC users care about since that's the sole purpose of ECC.
      Anyway - I have no doubt ECC is capable of going higher. But can!=should. This is probably why neither AMD nor Intel offer overclocking on their server platforms.
      Power usage is also a concern, especially in high density datacenters, especially when using high capacity fb memory. Enterprise hardware, in general, aims for the optional power\performance ratio, which is also why even low core count server cpus are fairly underclocked, even if they have ample TDP budget to increase per-cpu performance.
      I agree with all of this.

      Originally posted by MaxToTheMax View Post
      I'm not exceeding spec on the actual RAM. If anything, ECC is advantageous for overclocking, because it'll warn you if your OC isn't stable, and may actually correct errors caused by a bad OC until you can ease it back.
      If you are overclocking, by definition, that is exceeding the specs. However, I would agree with your other point.

      Comment


      • #13
        Nice. Like to see people posting about out of the box builds.

        FYI a bottleneck is specifically what it is. It is a physical limitation preventing the flow of something. It really should only apply to the transfer of data in the PC world but people use it to explain everything.

        Comment


        • #14
          Originally posted by schmidtbag View Post
          In general, overclocking ECC is pretty counterproductive. The whole point of having ECC is to improve stability and reliability. When you overclock, you're going beyond the manufacturer's specs, therefore pushing beyond the intended limits of the product. So basically the benefit you're paying extra for is negated.
          Despite all of this though, I think it's great AMD at least gives users an option, and I also think DDR4 ECC is clocked way too low.
          You're right.
          But I wasn't commenting on weather you should overclock your server/workstation or not.
          I was only commenting on the technical aspect of an ECC RAM overclock.

          Comment


          • #15
            Originally posted by schmidtbag View Post
            In general, overclocking ECC is pretty counterproductive. The whole point of having ECC is to improve stability and reliability. When you overclock, you're going beyond the manufacturer's specs, therefore pushing beyond the intended limits of the product. So basically the benefit you're paying extra for is negated.
            Despite all of this though, I think it's great AMD at least gives users an option, and I also think DDR4 ECC is clocked way too low.
            Actually ECC is the most obvious to overclock. It can tell you when it starts to fail, and even correct errors. So it is much better for overclocking than normal memory. And since it is sold underclocked, it makes sense to return it to its original clock.

            Comment


            • #16
              Originally posted by carewolf View Post
              Actually ECC is the most obvious to overclock. It can tell you when it starts to fail, and even correct errors. So it is much better for overclocking than normal memory. And since it is sold underclocked, it makes sense to return it to its original clock.
              I see why you're saying that, and for the most part I agree (though, I wouldn't say it's all that obvious, since you have a very long way to go to squeeze non-ECC performance out of it, so you're going completely blind). In theory, ECC makes overclocking easier and more stable so it's probably easier, but again, I'm focusing on the intent of the product. The blade of a katana very easily cuts vegetables and meat, but that doesn't mean it was meant to be a kitchen knife.

              Comment


              • #17
                Originally posted by schmidtbag View Post
                I see why you're saying that, and for the most part I agree (though, I wouldn't say it's all that obvious, since you have a very long way to go to squeeze non-ECC performance out of it, so you're going completely blind). In theory, ECC makes overclocking easier and more stable so it's probably easier, but again, I'm focusing on the intent of the product. The blade of a katana very easily cuts vegetables and meat, but that doesn't mean it was meant to be a kitchen knife.
                ECC has nothing to do with performance. It stores an additional 8 bits of parity data for data integrity verification purposes. That's quite literally the only difference between an unbuffered ecc memory module that uses the full 72 bits of the interface, and a 64 bit consumer ram module. Not sure exactly how your analogies apply here

                The problem is that ECC ram hasn't really caught up with the changing market. It used to be just this server thing, where top performance wasn't a priority.

                For example, intel has maxed out xeons, which gut the power efficiency to maximize per CPU performance, which would be appealing to workstation users. However, ecc specs and vendors still haven't caught up with that new market segment.

                Whether a memory module is ecc enabled or not doesn't really change the way it operates. Ecc memory is not more delicate. The dram chips aren't any different than those of consumer ram. It just has an additional 12.5% more chips, so as long as the motherboard also provides the full set of traces and the cpu ecc support is not disabled, it just works.

                There is nothing in the nature of ecc ram that dictates it should be any slower than consumer ram. The industry quite literally never bothered to make fast ecc modules.

                The intent of ecc is to detect errors, and if possible - to correct them. If not, to at least inform you, in order to avoid the build up of silent data corruption. That's really all there is to it.

                Comment


                • #18
                  Originally posted by ddriver View Post
                  ECC has nothing to do with performance. It stores an additional 8 bits of parity data for data integrity verification purposes. That's quite literally the only difference between an unbuffered ecc memory module that uses the full 72 bits of the interface, and a 64 bit consumer ram module. Not sure exactly how your analogies apply here
                  Meanwhile, I'm not sure how any of that refutes my point.
                  The problem is that ECC ram hasn't really caught up with the changing market. It used to be just this server thing, where top performance wasn't a priority.
                  And yet, there is a demand for performance ECC. Where there's a known demand that can be tapped into, there's usually a market to follow. So, why isn't there higher-clocked ECC? RAM manufacturers are notorious for gouging people for money. Why wouldn't they make such a product that is guaranteed to sell well at an elevated price?
                  Whether a memory module is ecc enabled or not doesn't really change the way it operates. Ecc memory is not more delicate. The dram chips aren't any different than those of consumer ram. It just has an additional 12.5% more chips, so as long as the motherboard also provides the full set of traces and the cpu ecc support is not disabled, it just works.
                  I'm well aware. I never said ECC was more delicate. My point, which has remained the same since my first post, is that increasing clocks decreases reliability. It doesn't matter if you have ECC or not. Although ECC can help improve that reliability, the purpose of buying it is for the reliability.
                  There is nothing in the nature of ecc ram that dictates it should be any slower than consumer ram. The industry quite literally never bothered to make fast ecc modules.
                  The intent of ecc is to detect errors, and if possible - to correct them. If not, to at least inform you, in order to avoid the build up of silent data corruption. That's really all there is to it.
                  I entirely agree... I never said otherwise. I openly admitted earlier that ECC theoretically has better OC potential.

                  Comment


                  • #19
                    Originally posted by schmidtbag View Post
                    And yet, there is a demand for performance ECC. Where there's a known demand that can be tapped into, there's usually a market to follow. So, why isn't there higher-clocked ECC? RAM manufacturers are notorious for gouging people for money. Why wouldn't they make such a product that is guaranteed to sell well at an elevated price?
                    In the past, Intel kept ECC features out of their highest-clocked processors, so there was no demand for fast ECC. The demand is pretty recent, driven by Ryzen, and still minuscule compared to demand for slow ECC for the datacenter and mediocre non-ECC RAM for OEMs like Lenovo and Dell. It's a pretty small niche and the industry is going to take its sweet time filling it.

                    I'm well aware. I never said ECC was more delicate. My point, which has remained the same since my first post, is that increasing clocks decreases reliability. It doesn't matter if you have ECC or not. Although ECC can help improve that reliability, the purpose of buying it is for the reliability.
                    People can have multiple goals that they try to balance against each other. You don't have to pick just one. What gives ECC its reliability is the error correction, and that doesn't stop working when you overclock it. I'm convinced that my ECC 2933 is going to be more stable than non-ECC 2933 RAM would be. If the stability came from low clocks alone, then Samsung would be charging high-end enterprise prices for their slowest non-ECC DIMMs, and ECC wouldn't exist.
                    Last edited by MaxToTheMax; 24 December 2018, 03:11 PM.

                    Comment


                    • #20
                      schmidtbag - there might also be demand for 5 wheeled cars for all we know, but unless it is strong enough to be economically viable, no manufacturer would rush to meet said demand. Demand for ecc memory outside of the enterprise market is rather weak, and demand for high performance ecc ram is a small niche inside that small niche.

                      Prosumers that care or understand hardware are few and far in between, and most do as they are supposed - overpay for branded systems without giving much thought to the internals.

                      Furthermore, the current market is far, far detached from listening to the demands of consumers. The industry has long been dictating what people should want to buy. Nobody cares about the few people who actually know what they need for themselves.

                      The situation is improving... slowly. 2933 ecc memory is incipient, with a bit of luck, maybe even 3200.

                      Comment

                      Working...
                      X