Skip to main content

Adata Teams Up with Gigabyte & MSI for 64GB DDR5-8400 Memory Modules

Adata
(Image credit: Adata)

Adata has teamed up with Gigabyte and MSI to ensure that its upcoming DDR5 memory modules will be able to hit an 8400 MT/s data transfer rate on Intel's next-gen Alder Lake platform. 

The DDR5 memory promises a rather massive performance increase compared to DDR4 due to a tangible increase in data transfer rates as well as efficiency improvements. When the DDR5 specification was published in mid-2020, it was expected that the first desktop memory modules based on the new standard would hit DDR4-4800 or DDR5-5200 data rates, whereas eventually the standard will reach up to DDR5-8400. As it turns out, even the first-generation DDR5 platforms will be able to hit an 8400 MT/s data transfer rate.  

Adata on Wednesday said that it was testing its 64GB DDR5-8400 memory modules on the 'latest Intel platforms' for consumers, which probably means Intel's upcoming Alder Lake processors with up to eight high-performance Golden Cove cores and eight energy-efficient Grace Mont cores. The motherboards that Adata uses right now are from Gigabyte and MSI. 

DDR5 memory supports several technologies to enable high data rates, long-term I/O scalability, and real-world efficiency, including on-die single error correction (SEC) ECC, DFE (decision feedback equalizer) to eliminate reflective noise at high frequencies, improved training modes, on-die termination, and two independent 32/40-bit I/O channels (non-ECC/ECC) per module. Micron once estimated that DDR5 memory would be 28% ~ 36% more efficient than DDR4 at the same I/O speeds. 

Intel's 12th Generation Core 'Alder Lake' processors based on the next-gen Golden Cove microarchitecture promise further instructions per cycle (IPC) performance gains compared to Willow Cove and Cypress Cove microarchitectures. Using DDR5-8400 modules with such CPUs will further increase their real-world performance thanks to a higher peak bandwidth and improved efficiency, giving them a competitive edge over DDR4-based platforms. 

Hitting DDR5-8400 using the first-generation DDR5 platform and memory modules is an impressive achievement. Given all the advantages that DDR5 already has on an architectural level, it remains to be seen how significantly real-world data rates can be improved with the evolution of memory controllers, platform designs, and memory module designs.  

  • tennis2
    Will AMD or Intel handle the DDR5 transition the best?
    Reply
  • hannibal
    Intel market share is still 80% vs 20% of AMD, so memory makers will tune their new memory first to Intel platforms. But amd 20% is still big enough that memory makers are also interested in getting to AMD rigs. But the pure mass of Intel makes it so that they will get the first dip, even AMD is leading in the technology wise at this moment.
    The transition is easy to both. They both need new platform for ddr5, so no difference in there. in AMD platform the infinity cache speed ratio vs memory speed still remains to be important. So the best memory speed is one that is near 1800 Mhz to 2000 Mhz internally. Because ddr4 is douple rate the "good" amd memory for current platform is 3600 to 4000 (douple the io speed speed)

    NamesMemory clockI/O bus clockTransfer rateTheoretical bandwidthDDR-200, PC-1600100 MHz100 MHz200 MT/s1.6 GB/sDDR-400, PC-3200200 MHz200 MHz400 MT/s3.2 GB/sDDR2-800, PC2-6400200 MHz400 MHz800 MT/s6.4 GB/sDDR3-1600, PC3-12800200 MHz800 MHz1600 MT/s12.8 GB/sDDR4-2400, PC4-19200300 MHz1200 MHz2400 MT/s19.2 GB/sDDR4-3200, PC4-25600400 MHz1600 MHz3200 MT/s25.6 GB/sDDR5-4800, PC5-38400300 MHz2400 MHz4800 MT/s38.4 GB/sDDR5-6400, PC5-51200400 MHz3200 MHz6400 MT/s51.2 GB/s
    So good ddr5 to amd are about ddr5 7200 to ddr5 8000 range. That means also that first ddr5 4800 are no good for amd zen prosessors! You need to have higher end variants. At least ddr 6400 before it start to get even with ddr4 platform that can run infinity cache at higher speed.
    Reply
  • salgado18
    Sounds very good on paper, but it's been a long while since CPUs were bottlenecked by RAM. The real world benefit of going from DDR4-2666 to DDR4-4000 (a 50% increase) is around 10% in the best cases. I like fast RAM? Definitely, but I hope it doesn't lead to unnecessary cost increases, since the benefit is hardly there.
    Reply
  • gg83
    salgado18 said:
    Sounds very good on paper, but it's been a long while since CPUs were bottlenecked by RAM. The real world benefit of going from DDR4-2666 to DDR4-4000 (a 50% increase) is around 10% in the best cases. I like fast RAM? Definitely, but I hope it doesn't lead to unnecessary cost increases, since the benefit is hardly there.
    Thats exactly what I was just thinking. Nobody recommends more than 16gb at normal speeds for the average user. Will software have to change in order to utilize the speed and bandwidth?
    Reply
  • tennis2
    salgado18 said:
    The real world benefit of going from DDR4-2666 to DDR4-4000 (a 50% increase) is around 10% in the best cases.
    People get all giddy about MHz and MT/s, but forget to take CAS latency into the equation. Sure, it still doesn't relate 1:1 with real-world performance, but:
    DDR4-2400, CAS 16 = 13.33ns latency
    DDR4-3600, CAS 18 = 10ns latencyAside from the 50% transfer rate increase, the actual latency only improves by 33%. In the real world, yes, the actual performance difference is far below 10%.
    Getting past my level of knowledge, but as I understand, you're not throwing a bit of data into system RAM and then retrieving it immediately. That stuff should be caught in the CPU cache(s). This is further exacerbated by multi-threaded workloads where waiting on one specific RAM call doesn't stall the whole pipeline.
    Reply
  • dan1991Ro
    Let me ask you guys a few questions.
    1.If the density is increased,does that mean you could get a 32 GB set for cheaper than right now?Does it mean we will have more dram at a lower price or not?
    2.Will a single 16 gb ddr5 stick be equal to 2x8 ddr4 ram?(they say its supposed to be split in half ?,2 channels per stick,i dont really understand thatm sorry if my question is bad)
    3.Will increased dram density or whatever they call it mean that phone ram will increase because they can cram more of it in,less space concerns?And 6-8-12 Gb ram for phones will become the standard?
    Reply
  • Jesse_20
    dan1991Ro said:
    Let me ask you guys a few questions.
    1.If the density is increased,does that mean you could get a 32 GB set for cheaper than right now?Does it mean we will have more dram at a lower price or not?
    2.Will a single 16 gb ddr5 stick be equal to 2x8 ddr4 ram?(they say its supposed to be split in half ?,2 channels per stick,i dont really understand thatm sorry if my question is bad)
    3.Will increased dram density or whatever they call it mean that phone ram will increase because they can cram more of it in,less space concerns?And 6-8-12 Gb ram for phones will become the standard?
    Density does not really impact price/GB. It's simply how much you can fit on a single chip.
    Depends on what you are measuring. 16GB of DDR4 will hold the same amount of data that 16GB of DDR5 will hold. In that respect, they are equal. As for speed, DDR5 is going to be much faster transferring that data in and out of the memory chip. In this respect, DDR5 is better.DDR5 will likely cost more than DR4, so in this respect, DDR4 will be better. Ultimately, your system will be designed to use one or the other, but not both. Older systems will stay with DDR4 (or DDR3/2 if old enough), or use DDR5 for the brand new chipsets and processors coming out next year.
    3) The memory talked about here is for desktop and laptop computers, and does not correlate to phone memory at all. They are entirely different beasts. Phone memory is more like a usb flash drive/SSD/NVME/HDD than a DDR4 (or 5) stick of Ram in that it "Stores" information even if the device is turned off, whereas DDR4 and DDR5 loses all data in them if the power is cut. You will see larger capacity storage in phones eventually, but that cannot be derived from this article.
    Reply