Improved storage benchmark performance with Prime95

[Ao1]

JohnW picked up that a storage benchmark will improve if Prime95 is running in the background on one core. This is repeatable on different 6 and 7 series mobos’ and different SSD’s.

http://forums.anandtech.com/showpost.php?p=34384049&postcount=15

So what gives?

I have all power saving features turned off. I tried running Anvil’s Storage Utility (ASU) with the following settings and it made no difference to the benchmark speed:

· Set Affinity – to a single CPU core
· Set Priority – Realtime

Whilst I ran ASU I monitored the CPU activity without Prime95 running and then with it running. The most noticeable difference is the increase in the maximum frequency when Prime95 is running.

I have no knowledge of how the CPU interacts with I/O performance, but I would really like to understand why this occurs.

Thanks in advance for any ideas.

With Prime95 (Running on one core)


Uploaded with ImageShack.us

Without Prime95


Uploaded with ImageShack.us

 

[groberts101]

I really do wish I was smart enough to understand the mechanisms involved here.. though I can only assume it lies in the ASPM related features of the OS and motherboard itself in how they clock the CPU and interact with the RAM along with the LSPM related features of the PCIe bus.. but I do know that this has been evidenced in HDTune and Windows Experience Index for quite a few years now.

On some systems/config's.. if you run those particular benchmarks with background activity?.. sometimes they will post abnormally high numbers. Seems to be hit or miss and not all systems will react the same way with some showing the expected reduced performance when the system is tasked too hard during those types of tests.

Which kind of makes sense in that we know that not all motherboards/bios/hardware is created equal in mfgring variation.. not to mention that software/drivers/setting variations alone can often have huge impact as well.

I do know for a fact that reducing the number of power saving features and function does reduce that tendency in fluctuation for these types of tests though.

If you have a moment.. could you list the power saving features that you have disabled?(and IMO, John's tests aren't eye opening either unless he's thoroughly dismissed as much power mgmt intervention as possible in those linked tests)

Have you done any ASPM driver related testing?

Any power related reg hacks?(these can be huge for mobile users but of course come at a large cost in power consumption/heat)

 

[ryderOCZ]

I have never done anything to prove it, but it is my belief that the Intel driver somehow uses CPU cache rather than solely using RAM cache.
So, when you disable C1E and in this case get a core to do something else like Prime95, you get better benchmarks. I still feel this is just a synthetic increase, it isn't actually speeding up the drive.

 

[Ao1]

Interesting. Thanks Ryder.