Jump to content

Memory Write Speed Much Lower Than Read and Copy


Recommended Posts

I have noticed that my memory write speed is much lower than my read and copy.  At first, I thought perhaps my memory was defective.  I was using the following kit:

https://www.gskill.com/en/product/f4-3600c16q-32gtzr

When overclocked to 3800MHz and at CL16 17-16-36 1T and tighter secondaries than XMP, I'm getting the following benchmark results (see attached image).  I tried one of my older Corsair 3200MHz kits and that kit was also hitting 76-78K MB/s write with the same secondaries and teriaries, but at 3200MHz.  It's as if there is some odd threshold on my memory write speed.  Everything else scales up correctly with an overclock, but my write speeds eventually get stuck in the same place.  I'll note that much weaker CPUs on this platform with slower memory are able to break 90K, with similar overclocks to my 3800 profile breaking 100K without issue.  

Aside from trying different kits, I've also used different BIOS and I performed a clean install of Windows.  The issue persists.  Any ideas as to what could be causing this poor memory write performance?  

 

postsecpatch.png

Link to comment
Share on other sites

We're not sure why that happens.  Our best guess is that there must be a bottleneck about memory write bandwidth in the IMC (Integrated Memory Controller) of Skylake-X processors.  And that bottleneck may work differently in 12+ core versions (HCC) than 10-core and lower SKUs of Skylake-X (LCC).

Link to comment
Share on other sites

14 hours ago, Fiery said:

We're not sure why that happens.  Our best guess is that there must be a bottleneck about memory write bandwidth in the IMC (Integrated Memory Controller) of Skylake-X processors.  And that bottleneck may work differently in 12+ core versions (HCC) than 10-core and lower SKUs of Skylake-X (LCC).

The 7980XE does not have this issue either.  Various 7980XE owners I've come across are hitting over 100 GB/s in memory write with 4000MHz and decent timings on their memory.  I haven't seen any results for the 7960X or 7940X.  So perhaps it is a worse IMC on the cut-down HCC parts, whereas the cut-down LCC parts do not have this issue.  It may also explain why Intel never sent any 7920Xs out for review...

Link to comment
Share on other sites

  • 3 years later...
On 1/10/2021 at 8:58 PM, Roman Tereshin said:

What's more interesting is how copy speed could possibly be higher than the write speed. Makes me doubt Aida's results entirely.

It could happen when the read and write channels are independent which could mean that their bandwidth could be aggregated in a copy benchmark -- but that advantage cannot be exploited in a read or write benchmark.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...