Memory Bandwidth

AIDA64

This program has many tools for determining memory bandwidth as well as various latency values.

Memory Read

Intel Core i9-10900K Aida64 Memory Read

Here we can see solid performance out of all the test systems. The memory controller on the 3900X is very competitive here, turning in a slightly higher score than our 10900K. However, the difference is fairly small. We do see a marked improvement in the 10900K compared to the outgoing 9900K.

Memory Write

Intel Core i9-10900K Aida64 Memory Write

The AMD and Intel mainstream CPU’s trade blows here. The Intel 10900K comes out with a slight lead. Again, Intel has definitely improved the memory controller of the 10th generation CPUs.

Memory Copy

Intel Core i9-10900K Aida64 Memory Copy

Here, the AMD Ryzen 9 3900X is considerably faster than the Intel Core i9-10900K. Still, Intel does improve on the previous iteration of its Skylake architecture by a good margin. One thing to point out here is that while Intel may win and lose in these tests, it’s CPU’s do not have the same Infinity Fabric clock divider, which limits the Ryzen 3000 series after about DDR4 3800MHz. Overclocked to DDR4 3866MHz, Intel can easily win these memory tests.

Sandra Memory Bandwidth

This latest iteration of Sandra shows us aggregate memory performance, rather than splitting up the values as we did with AIDA64.

Intel Core i9-10900K SiSoftware Sandra 2020 Memory Performance

This test shows us raw memory bandwidth for each test configuration. You can see the strength of Intel’s memory controller as it produces the highest results overall.

Join the Conversation

8 Comments

  1. So the memory on the Zen system was 3200 or 3600? I know the kit was 3600 but I am just double checking.
  2. It was set to DDR4 3200MHz speeds which is our testing standard for everything unless otherwise noted. If you look at the specification table, I list the part number for the RAM and then the speed used. That’s how I do it for all of these.
  3. It was set to DDR4 3200MHz speeds which is our testing standard for everything unless otherwise noted. If you look at the specification table, I list the part number for the RAM and then the speed used. That’s how I do it for all of these.

    Oof missed that. Was the platform unable to hit 3600?

  4. Oof missed that. Was the platform unable to hit 3600?

    Yes, it can easily hit DDR4 3600MHz speeds and more. I’ve addressed the Ryzen 3000 / X570 memory speeds in previous CPU and motherboard review articles. Given the time allotted for getting the 10900K review done by the embargo date, I was not able to retest the 3900X and 9900K under overclocked conditions. Even if I had, memory overclocking is handled separately as we try to keep that variable out of the benchmarks unless that’s what we are testing.

  5. Good review and well written. Nothing stood out as a glaring inconsistancy.

    It will be interesting to see what happens to code that has been heavily optimized for a 10+?? year old instruction set actually has to run on something new.

    This is what AMD is doing and I think that is a large reason so many of the normal work and Gaming examples were performing better on Intel. (Other than raw execution speed)

    I might be way off base in thinking that coders are using older optimizations that simply don’t exist on the newer AMD silicone.

  6. Intel has always pushed software companies to optimize for Intel silicon going back at least as long as I’ve worked with computer hardware. There are all kinds of SDK’s and programs for doing that. Intel even mentions this in the product brief we got. What little there was of it anyway. But this is one reason why I think that Intel achieves so much despite the lack of cores and threads compared to AMD. Sure, clock speed and cache are part of that too, but I think that optimization for Intel silicon comes into play in cases where we know something is multi-threaded, but Intel still manages to pull a big win vs. AMD.

    It’s worth noting that Ghost Recon Breakpoint was optimized for AMD silicon and it shows. The results between the 9900K and the 3900X are quite similar. The only reason why the Core i9 10900K beats either of them comes down to clock speed and additional cache. That and the extra threads don’t really matter. If I recall correctly, Ghost Recon Breakpoint only sees 12t or at least, that’s all it shows in the in-game performance metrics. Something like that.

  7. I find the 400 fps difference in Doom quite huge for the little difference between the CPU’s but I guess the average tells another story and the min’s are even stranger.

    Any chance of a quick retest when the new doom patch hits next week orso to see if that did anything?

  8. I find the 400 fps difference in Doom quite huge for the little difference between the CPU’s but I guess the average tells another story and the min’s are even stranger.

    Any chance of a quick retest when the new doom patch hits next week orso to see if that did anything?

    Yes. I’d have looked more into the anomalous performance if I had the time. That said, its easily something I could have done differently. Those are Frameview captures of manual run throughs. I could have done something with the camera, or did something slightly different that caused that in some of the runs. If you run into a wall and stare at it in most games your FPS shoots up, or if you explode an enemy at point blank, it can drop substantially. That’s why I prefer canned benchmarks for these types of things, but not every game that people are interested in has built in tools for that.

Leave a comment