Announcement

Collapse
No announcement yet.

AMD Ryzen 5 7600X Linux Performance

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

  • #31
    Originally posted by piotrj3 View Post
    your claims are false. Of course in some benchmarks are not RAM speed sensitive, but games are literally very strongly RAM sensitive. There is no such thing as RAM sweetspot, as most benchmarks are either bandwidth sensitive (like copying/reading big parts of memory) or latency sensitive (cache mismatches) and BOTH improve with faster RAM speed. Idea 4800 vs 5200 is fair is complete dishonest, any CPU in games is very RAM senstitive unless it is having a huge cache but only CPU with huge cache so far is 5800X3D and that one isn't sensitive.

    Cyberpunk shows pretty much linear scalling for Alder lake up to 6200 MHz https://youtu.be/LU_w9fZvSso?t=629
    Watch a little more you notice in that video Cyberpunk at 1440p instead of 1080p stops scaling at 4800. Benchmarks with the Zen 4 chips show performance scaling past DDR5 4800 at 1440p and higher.

    https://www.anandtech.com/show/17078...ing-analysis/5

    Majority of games with Alder Lake don't show any scaling past 4800.

    Also 4800p to 5200 is in the ideal case with Alder Lake is not 8 percent uplift its a 4 percent uplift at best. The reality with the increase bandwidth that 8 percent increase each step and you are only getting 4 percent at best is sign of being close to the wall. Small change like changing from 1080p to 1440p with Cyberpunk sees you on the other side where scaling just stops at 4800 like almost everything else..

    The reality here the benchmarks don't match what you are saying no with the Intel chips. Games that should be very strongly RAM sensitive most with Alder Lake stop gaining performance at 4800. Odd configurations of some of those games show 4% scaling per step from 4800 up.

    Majority of the time Alder Lake sweetspot is 4800.

    Alder Lake does not show universal improvement by increasing ram-speed. It appears once you have X amount of ram operations with Alder Lake you end up stuck at 4800 transfer speeds. Faster ram does not help you if the CPU MMU can only do X amount of transfers per second and you are doing X amount of transfers per second already with slower ram. 1080p with Cyberpunk is most likely light enough on memory operations to gain from latency reduction on the CPU MMU limit and 1440p Cyberpunk is on the other side. Please note Zen 4 you still see Cyberpunk 1440p performance scale as you go from 4800 to 5200 to 6000 ram so this is not the program limitation here.

    Bandwidth sensitive is a thing. Remember there is a cap on amount of Bandwidth a CPU MMU can in fact exploit based on how many transfers it can do. This limit with Zen chips comes from the infinity fabric. Alder Lake from Intel clearly shows the same kind of internal limit but it kicks in a lower ram speed.

    Please note its not that Cyberpunk is getting 100 percent CPU usage either. Cyberpunk at 1440p runs into the same MMU limit on Alder Lake everything else is.

    Basically you had a mistake you presumed bandwidth issue is improved by faster ram that only happens if the MMU in the CPU is able to use that bandwidth. Alder Lake faster ram gets your reduced latency not more bandwidth past 4800. 4800 is the sweet spot of the Alder Lake MMU and that why lots of benchmarks flatten off at that point.

    Bandwidth capped also means the latency improvements with increasing clock speeds is not as much as you would expect.

    piotrj3 the reality here its abnormal benchmarks with Alder Lake that scale past 4800. With games that update could increase the transfer count in future means you could pay for more expensive ram with Alder Lake than 4800 and the Next game update caps you back at 4800 ram performance because its it now hitting the CPU MMU limit.

    Yes you are right piotrj3 that there are difference between benchmarks but there is a problem here a party testing Alder Lake chip with 4800 only ram on most benchmarks this is maxed out because the transfers is being limited by the CPU MMU not the ram clock speed.

    Comment


    • #32
      Originally posted by oiaohm View Post

      Watch a little more you notice in that video Cyberpunk at 1440p instead of 1080p stops scaling at 4800. Benchmarks with the Zen 4 chips show performance scaling past DDR5 4800 at 1440p and higher.

      https://www.anandtech.com/show/17078...ing-analysis/5

      Majority of games with Alder Lake don't show any scaling past 4800.

      Also 4800p to 5200 is in the ideal case with Alder Lake is not 8 percent uplift its a 4 percent uplift at best. The reality with the increase bandwidth that 8 percent increase each step and you are only getting 4 percent at best is sign of being close to the wall. Small change like changing from 1080p to 1440p with Cyberpunk sees you on the other side where scaling just stops at 4800 like almost everything else..

      The reality here the benchmarks don't match what you are saying no with the Intel chips. Games that should be very strongly RAM sensitive most with Alder Lake stop gaining performance at 4800. Odd configurations of some of those games show 4% scaling per step from 4800 up.

      Majority of the time Alder Lake sweetspot is 4800.

      Alder Lake does not show universal improvement by increasing ram-speed. It appears once you have X amount of ram operations with Alder Lake you end up stuck at 4800 transfer speeds. Faster ram does not help you if the CPU MMU can only do X amount of transfers per second and you are doing X amount of transfers per second already with slower ram. 1080p with Cyberpunk is most likely light enough on memory operations to gain from latency reduction on the CPU MMU limit and 1440p Cyberpunk is on the other side. Please note Zen 4 you still see Cyberpunk 1440p performance scale as you go from 4800 to 5200 to 6000 ram so this is not the program limitation here.

      Bandwidth sensitive is a thing. Remember there is a cap on amount of Bandwidth a CPU MMU can in fact exploit based on how many transfers it can do. This limit with Zen chips comes from the infinity fabric. Alder Lake from Intel clearly shows the same kind of internal limit but it kicks in a lower ram speed.

      Please note its not that Cyberpunk is getting 100 percent CPU usage either. Cyberpunk at 1440p runs into the same MMU limit on Alder Lake everything else is.

      Basically you had a mistake you presumed bandwidth issue is improved by faster ram that only happens if the MMU in the CPU is able to use that bandwidth. Alder Lake faster ram gets your reduced latency not more bandwidth past 4800. 4800 is the sweet spot of the Alder Lake MMU and that why lots of benchmarks flatten off at that point.

      Bandwidth capped also means the latency improvements with increasing clock speeds is not as much as you would expect.

      piotrj3 the reality here its abnormal benchmarks with Alder Lake that scale past 4800. With games that update could increase the transfer count in future means you could pay for more expensive ram with Alder Lake than 4800 and the Next game update caps you back at 4800 ram performance because its it now hitting the CPU MMU limit.

      Yes you are right piotrj3 that there are difference between benchmarks but there is a problem here a party testing Alder Lake chip with 4800 only ram on most benchmarks this is maxed out because the transfers is being limited by the CPU MMU not the ram clock speed.
      Hint : There is GPU limited scenario at 1440p. It has nothing to do with Ryzen or Alder Lake and Ryzen wouldnt' scale either at beyond 4800MHz probably. You look at wrong stuff and Hardware Unboxed/GN etc will tell you all time that new games in settings they are meant to be played will be GPU limited in majority of cases. However most benchmarks are done on purpose in 1080p medium settings or lower to reveal more diffrence in CPU and also this is why often they use old games like CS GO that will be always CPU limited (and driver limited).

      Why in Shadow of Tomb Raider you don't see any scalling? Because it is GPU limited on almost any setting there is almost no improvement from ddr4 2933MHz to 6200MHz DDR5.

      If you see a case like that, it means Ryzen probably won't scale either with memory. Same memory bound task will be memory bound on both Intel and AMD.

      So quoting here Hardware Unboxed
      - Corona benchmark - minor scalling but in general renders are written in cache aware way so they are most of time not memory speed limited.
      - Photoshop does scale (makes sense, images aren't able to fit into cache),
      - Raibow six siege avg does improve to 5200 MHz, but 1% low continue to improve up to 6200 MHz (so again does scale),
      - Horizon zero down doesn't scale well either on DDR4 or DDR5. So again no magic 4800Mhz number, GPU limited case
      - Shadow of Tomb Raider - doesn't scale at all on either DDR 4 or DDR5, GPU limited case
      - Cyberpunk - scales well with memory especially in DDR5 case, scale not that well in DDR4 case. Interesting case. on 1440p clearly we get into GPU limited situation,
      - Hitman 3 1080p GPU limited, no scalling on either DDR4 or DDR5,

      There is no review yet for 7000 series in terms of diffrent memory scalling, but when i looked at old video from same guys about ryzen 5000 series a lot of benchmarks were diffrent but found the same tendency and Horizon zero down didn't scale either and neither did shadow of tom raider

      So not a single time for Intel magical 4800MHz sweet spot scalling number appears, and most of time when GPU limit happens it happens to both and memory scalling stops. Only exception from rule is again 5800X3D that cares less about memory speed (I have it, great CPU).
      Last edited by piotrj3; 08 October 2022, 10:55 AM.

      Comment


      • #33
        Originally posted by oiaohm View Post
        I cannot find the party that did a full ram bench on the 12600k at the moment but its fairly much been found that DDR5-4800 is where the 12500K is basically maxed out and is it sweetspot for ram. .
        This is not true eg:

        1111111111111111111111.jpg

        2222222222222.jpg


        33333333333.jpg


        4800 isn't sweet spot for Alder Lake. You non stop lie .



        Comment


        • #34
          Originally posted by atomsymbol

          Just some notes:
          • A520 (in general: the lowest-priced chipset from AMD or Intel) has fewer amount of connectivity options (SATA, M.2, DDR4, USB), has slower PCIe x16 links, might have a slower link between the chipset and the CPU, and most importantly has a shorter expected lifespan, compared to mid-range or high-end AM4 chipsets.
          • I bought an MSI X370 AM4 motherboard 5 years ago and today it supports most Ryzen 1000/2000/3000/4000/5000 CPUs (after BIOS upgrades). A downside (for some people) of an older X370 motherboard is that it has a 128 megabit (16 megabytes) BIOS which means that after the BIOS upgrades the BIOS user interface is mostly text without any images, and it is impossible to install a newer CPU without first flashing the BIOS to a newer version.
          • If I was buying an AM5 motherboard for Ryzen 7000, I would probably choose the X670/X670E chipset and avoid B650/B650E in order to increase the probability of me not buying a replacement AM5 motherboard during the next 5 years.
          I would add that mid and high end mainboards usually have better Ethernet and audio controllers, even some models have WiFi and Bluetooth.

          Comment


          • #35
            Originally posted by HEL88 View Post

            This is not true eg:

            1111111111111111111111.jpg

            2222222222222.jpg


            33333333333.jpg


            4800 isn't sweet spot for Alder Lake. You non stop lie .
            https://youtu.be/LU_w9fZvSso?t=629
            Watch that video notice 1440p figure for cyberpunk and notice it flat line..
            Winrar benchmark notice the difference between 36 and 32. The reality here Winrar benchmark is latency sensitive not ram transfer sensitive.

            1440P is transfer sensitive cyberpunk. Once you cap out AlderLake transfer bandwidth all you have is 4800 CL 32 performance nothing more.

            Not all cyberpunk 2077 benchmarks at 1080p show scaling past 4800 ram either.

            HEL88 what you you want to play cyberpunk 2077 with on a Alderlake system. Slightly faster framerates while the screen conditions are right or a stable framerate. 4800 is the stable frame rate and anything past that is conditional on what is on screen if you hit the transfer limit or not.

            HEL88 you have to be very careful to be sure you have not cherry picked benchmarks. Showing what a person can expect in Cyberpunk 2077 all the time with alderlake its correct to use 4800 CL 32. Now if you selectively cherry pick with Cyberpunk 2077 you end up showing the incorrect min frame rates person will see in general play.

            The 108 min is in fact high. The correct min is 98 across all the different Cyberpunk 2077 benchmarks at 1080p. Yes all the benchmarks that cap out transfers have the same min from 4800 up with 1080p.

            Really I personally wish it was not true that alderlake MMU caps out.

            Comment


            • #36
              Originally posted by Anux View Post
              5700G are notebook chips that are to inefficient for mobile use and get repurposed for desktop.
              .
              Source?

              Comment


              • #37
                Originally posted by Anux View Post
                5700G are notebook chips that are to inefficient for mobile use and get repurposed for desktop.
                Originally posted by Xedeya View Post
                Source?
                I think Anux is just talking about normal binning, but the wording makes it sound like something worse.

                A typical distribution of fabricated dies gives some that tend to have low leakage while others have high leakage but also clock higher. The low leakage bin becomes mobile parts while the high clock bin becomes desktop parts. The actual binning is multi-dimensional but the basic idea is the same.
                Last edited by bridgman; 15 October 2022, 07:44 PM.
                Test signature

                Comment


                • #38
                  Originally posted by Xedeya View Post
                  Source?
                  They are the same dies, you can look it up on wikipedia: https://en.wikipedia.org/wiki/List_o...zen_processors

                  Originally posted by bridgman View Post
                  ​I think Anux is just talking about normal binning, but the wording makes it sound like something worse.

                  A typical distribution of fabricated dies gives some that tend to have low leakage while others have high leakage but also clock higher. The low leakage bin becomes mobile parts while the high clock bin becomes desktop parts. The actual binning is multi-dimensional but the basic idea is the same.
                  Exactly, they are certainly not bad chips, I still use the 2400G and looking for a cheap used upgrade to the 5700G.

                  I didn't see something wrong with my wording but I'm not a native speaker.

                  Comment

                  Working...
                  X