Weird SSD behavior, 840 series.

[Chillin1248]

Hey,

A guy over at Guru3D noticed that when his CPU was at 100% load, his 840 SSD would perform much better than when the CPU was at idle. I ran the same test on my 840 Pro SSD and confirmed this observation.

840

840 Pro

Any explanation for this behavior?

 

[wpcoe]

If it's with an Intel Series 5 chipset, it's a well-discussed phenomenon:

Laptops w. Intel Series 5 chipset can not take full advantage of fast SSDs
How To: Improve Low SSD Performance in Intel Series 5 Chipset Environments

 

[Chillin1248]

If it's with an Intel Series 5 chipset, it's a well-discussed phenomenon:

Laptops w. Intel Series 5 chipset can not take full advantage of fast SSDs
How To: Improve Low SSD Performance in Intel Series 5 Chipset Environments

The 840 was tested on an Asus M4A88T-M AMD 880G chipset, the 840 Pro was on an Asrock Z77 Pro4.

 

[Coup27]

Modern Intel CPUs have many power and heat reduction technologies built into them. The two main examples of these are C states and Enhanced Intel Speed Step (EIST). The two technologies combined half the clock of a rated CPU and reduce the vcore when the full CPU power is not required. My Sandy Bridge CPU has a rated frequency of 3.4Ghz with a vcore of ~1.4v will idle at 1.6Ghz with a vcore of ~0.9v due to these technologies.

When you benchmark either the CPU is not being used to it's full or it is fluctuating between idle and full.

If you want to see the maximum possible benchmark scores, disable all C states and EIST in your BIOS. Alternatively you could max your CPU by running a single thread of Prime95 during a benchmark.

There is certainly nothing wrong with your SSD.

 

[bbinnard]

Coup27

Excellent explanation -thanks. The intelligence built in to the latest CPU chips certainly makes the benchmarking process more complex that simply running a piece of software and looking at the results.

It will be interesting to see if the developers of benchmark software start including internal tasks to ensure a CPU is running at full speed during the benchmark data collection process.

 

[Chillin1248]

Modern Intel CPUs have many power and heat reduction technologies built into them. The two main examples of these are C states and Enhanced Intel Speed Step (EIST). The two technologies combined half the clock of a rated CPU and reduce the vcore when the full CPU power is not required. My Sandy Bridge CPU with a rated frequency of 3.4Ghz with a vcore of ~1.4v will idle at 1.6Ghz with a vcore of ~0.9v due to these technologies.

When you benchmark either the CPU is not being used to it's full or it is fluctuating between idle and full.

If you want to see the maximum possible benchmark scores, disable all C states and EIST in your BIOS. Alternatively you could max your CPU by running a single thread of Prime95 during a benchmark.

There is certainly nothing wrong with your SSD.

What I still don't understand is that when my CPU is fully taken up with other stuff (100% load), the SSD performs better than when it access the full resources of the CPU at idle.

Also note that the 840 was tested with an AMD CPU.

 

[tweakboy]

The test are almost identical. Yes I see the difference and I wouldn't care about it. Just use the drive its blazing fast, Wait til you get to SATA 3.0 then you wont worry about this subject again.

I was getting 260mbps at SATA 2.0 my old Q6 rig. Now with SATA 3.0 Im getting 540mbps seq read.

But as far as booting up or launching apps or opening data you will not see a difference between sata 2.0 and 3.0. Only difference you'll see is when you bulk transfer or transfer huge files or load huge maps in a game.

 

[Coup27]

It will be interesting to see if the developers of benchmark software start including internal tasks to ensure a CPU is running at full speed during the benchmark data collection process.

An interesting idea, but if they that it should be optional. This would allow regular users who have EIST and C states enabled to receive a true reflection of their system performance under normal conditions. A benchmark does not always have to be ultimate performance but is often usful to see what performance is available under a real world situation.

What I still don't understand is that when my CPU is fully taken up with other stuff (100% load), the SSD performs better than when it access the full resources of the CPU at idle.

Also note that the 840 was tested with an AMD CPU.

How do you know that your benchmark is accessing 100% of the CPUs power? Unless you are running something in the background like Prime95 I doubt your benchmark is taxing the CPU enough.

The use of an AMD CPU makes no difference. C states are universal across both Intel and AMD platforms and AMD's equivilent to EIST is "Cool 'n' Quiet".

 

[tweakboy]

What up Coup27 long time. Great input ; Well if you have dual monitor just launch up speedfan or taskman on that monitor and play game on the other.

 

[Chillin1248]

An interesting idea, but if they that it should be optional. This would allow regular users who have EIST and C states enabled to receive a true reflection of their system performance under normal conditions. A benchmark does not always have to be ultimate performance but is often usful to see what performance is available under a real world situation.


How do you know that your benchmark is accessing 100% of the CPUs power? Unless you are running something in the background like Prime95 I doubt your benchmark is taxing the CPU enough.

The use of an AMD CPU makes no difference. C states are universal across both Intel and AMD platforms and AMD's equivilent to EIST is "Cool 'n' Quiet".

I had Intel Burn in Test running in the background.

 

[It's Not Lupus]

Some recommend running benchmarks while running one thread of prime 95.

 

[tweakboy]

Speaking of which some peep should try AIDA64 stability test. Its a reputable company and top notch software. ty

 

[rgallant]

because of this I have EIST set so my 3770k idles at 3.5 then to 4.6 when a heavy load kick's in , like moving my mouse.
-what would be nice if I could slow the down clocking down , have it stay at 4.6 for a few minutes then down clock and not every micro second.

 

[groberts101]

because of this I have EIST set so my 3770k idles at 3.5 then to 4.6 when a heavy load kick's in , like moving my mouse.
-what would be nice if I could slow the down clocking down , have it stay at 4.6 for a few minutes then down clock and not every micro second.

although I haven't tested that specifically( I run my systems heavily OC'd with max power at all times).. I believe that can be set in the advanced power options under "processor power states" which allows "min & max" settings to be adjusted. IIRC, and you use the high-performance power plan it will set both minimum and maximum at 100% for less fluctuation.

There is also a reg hack to do the same thing as well.. but it requires some testing/resetting to get it where you want it.

 

[Emulex]

I set all of my servers to MAX POWER and COOLING (no o/s control) and found massive increases in IOPS with ssd (raid/fastpath) for the 30watts of extra wasted power.

Default esxi setting is if cpu is < 60% throttle the piss out of it. (Advanced settings, even in esxi 5.1). Setting this value to 1 or 0 helped a lot but honestly disabling the o/s power management was the best.

What would work I suppose is a design where it spikes the CPU instantly to turbo when i/o starts happening, then settles a little, then perhaps after a minute or two it clocks back down. This is pretty much how smartphones achieve maximum performance and power savings.

Wake -> Balls out max turbo no power savings instantly for a few seconds then
run -> max realistic without overheating
start to sleep -> wait a minute before going back to powersave mode

 

[wpcoe]

From the long threads I read on notebookreview.com it was said something as "minor" as playing an audio file while benchmarking would trigger the higher CPU power state that would spike the benchmark results. I thought I was going overboard by playing a video file, but sounds like people run Intel BurnTest?

Again from the notebookreview.com thread: by doing only one or two Registry hacks affecting Windows power management, it would cause the CPU to run at a higher state, but the problem (especially since it is a notebook forum) was higher heat/faster battery drain from disabling power management features. One very patient and enterprising fellow explained the various settings that could be activated in Windows Power Settings (by default, they are hidden, but can be unhidden by Registry entries) and how they could provide a user-variable range of power savings while still increasing SSD benchmark results, allowing users to set a balance between SSD improvement vs power savings.

The bottom line for me was that the Windows power settings were okay at their default settings. I did see an increase in the SSD benchmarks by tweaking, but really, nothing that would justify using more power.