Strange SSD behavior!

[bathrobehero]

Hi, I'm just going to start with a picture:

http://screenshotuploader.com/i/01/6uu0lcv1h.png

Explanation:
The benchmarks on the left side are made while the CPU was not stressed, as in idle environment, but here's come the fun part because the benchmarks on the right side are made while the CPU was kept at 100% - by Everest's stress test (CPU only) but it's not exclusive to Everest, any CPU hogging application seem to make the SSD faster.


This happens both on my new Samsung 840 (non-Pro) and on my older SiliconPower V20 40GB SSD's.
Also, it's the same with both the amd_sata (1.2.1.331) and the Standard AHCI drivers.
I have an Athlon II X2 3.4 @ 4.0 CPU and a fresh Win7 x64 HP and the minimum processor state is set to 100% in the power options so it's not even about it being downclocked to around 800mhz when being idle.
My MB (M4A88T-M) only supports Sata2.


So why is my SSD is faster in every test I tried when my CPU is stressed to 100%?

 

[Peter1986C]

I doubt whether you notice the difference in real life. Do not care about those numbers in that synthetic benchmark please.

And I assume you know the reason why pro testers always do multiple runs of a test and use the average results?

 

[bathrobehero]

Yes, I know, but I need to figure this one out because it's just keep bugging me.
And I've seen a bunch of benchmarks about the very same SSD's I have and people are getting the numbers in these synthetic tests as I am when my CPU is at 100%.

I thought of everything, but couldn't figure out this yet.

 

[bathrobehero]

Ok, I found the problem and the solution.

So apparently the CPU has more idle modes than I thought. In other words, the CPU can be idle even if it's not downclocking itself. Which is great, but some chipsets - like mine (AMD 880G/SB710) - is not "waking up" the CPU in case of heavy I/O operations.

So there's a registry key with which we can disable the cpu ever being idle by chaning the Attributes key from 1 to 0 under:
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\PowerSettings\54533251-82be-4824-96c1-47b60b740d00\5d76a2ca-e8c0-402f-a133-2158492d58ad]

which unlocks an option in the Power Options which looks like this:

So with this option I have awesome speeds - for sata2 that is - but this option has some disadvantages as well. One of which is higher CPU temperatures, more power consumption because it appears disabling this idle state is acting like the CPU is at 100% but it is not, as you can see marked on the picture with red (cpu usage).
So if you have a dumb chipset and a CPU which gets hot under heavy load than it's not for you.
My CPU's idle temperature is around 30°C with stock cooler at 10% (ten) fan speed @4.0 (from 3.4) but with this "tweak" it's around 38°C.


Source: http://www.storagereview.com/how_to...rmance_in_intel_series_5_chipset_environments

 

[drdeathx]

your sequential read/write speeds are very low BTW/ Post your system specs

 

[Athlon2K15]

He said in the Op he is on Gen 2 SATA. So his numbers are normal

 

[drdeathx]

He said in the Op he is on Gen 2 SATA. So his numbers are normal

ahhh thanks

 

[Athlon2K15]

It even worse being AMD based

 

[bathrobehero]

It even worse being AMD based

Why is that?

 

[Aquinus]

The only reason this is happening because I/O doesn't create enough CPU usage to warent coming out of an idle state to do the I/O. Lowering the minimum CPU usage threshold might solve this without disabling idle states.

 

[bathrobehero]

The only reason this is happening because I/O doesn't create enough CPU usage to warent coming out of an idle state to do the I/O. Lowering the minimum CPU usage threshold might solve this without disabling idle states.

That's what I thought except it's not about the CPU clock itself, meaning that setting the minimum processor state to 100% (and checking it with CPU-Z that it is 4.0 ghz at all times, http://valid.canardpc.com/2667419) is not helping, so if I have to guess its about a feature other than CPU clock/FSB/Bus Speed/Multiplier which is getting cut back due to being idle.
I don't know which feature it is, but I mean it has to be something like that.

Also, notice that the main difference in the benchmarks is about 4k R/W which has the highest IOPS action, meaning that something has to manage those high IOPS actions which would be our missing element (CPU/FPU/cache/whatever) which was not required for HDD's due to low IOPS so it's understandable why chipsets are not "waking up" this missing component for heavy I/O actions - even if they should for SSD's.

 

[Aquinus]

That's what I thought except it's not about the CPU clock itself, meaning that setting the minimum processor state to 100% (and checking it with CPU-Z that it is 4.0 ghz at all times, http://valid.canardpc.com/2667419) is not helping, so if I have to guess its about a feature other than CPU clock/FSB/Bus Speed/Multiplier which is getting cut back due to being idle.
I don't know which feature it is, but I mean it has to be something like that.

Also, notice that the main difference in the benchmarks is about 4k R/W which has the highest IOPS action, meaning that something has to manage those high IOPS actions which would be our missing element (CPU/FPU/cache/whatever) which was not required for HDD's due to low IOPS so it's understandable why chipsets are not "waking up" this missing component for heavy I/O actions - even if they should for SSD's.

It's because CPU frequency scaling depends on load not latency. IOPS is low because there is no real load on the CPU to get it to clock up.

 

[Peter1986C]

I actually doubt it is a real issue, TBH.

 

[Aquinus]

I actually doubt it is a real issue, TBH.

This. Should have said that before. I wouldn't worry about it.

 

[bathrobehero]

I would call it an issue, but wouldn't call it serious.

Even though the performance difference is debatable, something is not working as intended.

 

[Peter1986C]

I think that when loading heavy applications, you should be getting max i/o. Light apps might be another story but than you would not notice it anyway.