Question about Anandtech SSD reviews: EIST/Turbo On vs. Off

[Jocelyn84]

I just happened to be looking back at some older reviews, and have found that Anand disables EIST/Turbo for them, while Kristian leaves EIST/Turbo enabled. I'm not trying to make a big deal about this, but disabling does boost performance in benchmarking. Anyway, I'm just curious what the reasoning was for this. Thank You

Examples
http://www.anandtech.com/show/6058/corsair-neutron-gtx-ssd-256gb-review/3
http://www.anandtech.com/show/5851/plextor-m3-pro-256gb-review/2

 

[Hellhammer]

AFAIK it affected performance with some older SandForce SSDs but when I started testing SSD and asked Anand for all the settings, he just told me to leave it on since it doesn't matter anymore.

 

[Jocelyn84]

First and foremost, Thank You for the honest reply. Secondly, I don't agree that EIST/Turbo only affected performance with some older SF SSDs. The biggest thing leaving me puzzeled by your reply, is that Anand still continues to disable them for reviews, while you don't. I hope you realize I'm posting this not so much as a complaint, but more so as a request for consistency and for the variables that are not being tested to be controlled. I am not claiming anyone has brand favoritism, but it would be nice to see apples to apples comparisons in Anandtech SSD reviews.

I've done some testing with my Plextor M3 and other drives in the past with EIST/Turbo both on & off and it definitely created large variances when it came to benchmarking. As you can see below, the 4k read/write speeds and access times for writes are very very different and this is directly related to the enabling/disabling of those CPU power saving features.

EIST/Turbo Enabled

EIST/Turbo Disabled

 

[Jocelyn84]

Again, I'm not trying to personally criticize Anandtech reviewers, but the methods used for testing do deserve criticism. Doesn't anyone else care whether or not the tests are consistent with one another?

 

[Elixer]

I would also like to know at what percentage full does the SSD start to have major slowdowns ? Dunno why this isn't tested more...

 

[KingFatty]

I would agree with Jocelyn84. And I think you've seen prior behavior by Anandtech support that position, that you need to be consistent. For example, I seem to recall they decided to revamp testing methodologies in the past, to be consistent, even if it meant tossing out old results? Or maybe that was another site.

Anyway, also consider that with SSDs it's particularly egregious to disregard this as "not really affecting anything" because it appears to have a very tangible effect when you consider this is with SSDs. You'll see on the SSD tests that what separates one drive from another can be much less than the difference shown here, so the importance of being consistent is magnified.

 

[Bill Brasky]

I can't for the life of me understand why why an SSD would be faster with Turbo DISABLED! WTBackwards! Oh I see now. Speedstep was disabled as well.

It does seem that this would be a consistent part of the testbed, though.

 

[Hellhammer]

Sorry for a late reply, I was in Seoul covering the 840 (Pro) release so haven't had time to reply before now.

Let me run some tests tomorrow and I'll get back to you. I'm assuming you're running the drive as a boot drive? That can have an impact on performance and cause more variation, though I'm not saying it's necessarily the cause behind your results.

I agree that consistency is very important, so thanks for bringing this up.

 

[Hellhammer]

I managed to quickly test Turbo/EIST with Plextor M3 Pro (the M3 is my boot drive so I couldn't use it). The drive was secure erased between tests to eliminate possible variables.

Turbo and EIST enabled:

Turbo and EIST disabled:

I'm guessing you're using your drive as the OS drive because it would explain the performance differences (OS adds lots of variables, which is why we don't test drives as OS drives). I would say the differences in my results fall into margin of error.

Also, I asked Anand and he wasn't 100% sure that he has Turbo/EIST disabled. It might be just a typo in the table because Anand most likely used the same table as previously and just switched the CPU and some other components (with our pre-SNB testbed, Turbo and EIST were disabled).

EDIT: Anand just confirmed that he has Turbo and EIST enabled, so it was simply a typo in our tables.

 

[Bill Brasky]

It's very interesting indeed that you found almost no difference between the two settings. What really matters, to me at least, is that the testbed was consistent. Even if you can repeatedly show EIST/ speedstep doesn't make much difference, that's one of those things that can cause internet wars. Thanks again for looking into it.

 

[Jocelyn84]

Actually it was empty and not the OS drive when I ran AS SSD, but I did fail to mention that I also had C states off for the top run and on for the bottom run. Anyway, I wanted to say Thank You for posting your results, but more importantly Thank You for clearing up the typo and proving that you two are controlling the variables that are not being tested. I have a lot more I want to say, but I don't have a ton of time now.

It's very interesting indeed that you found almost no difference between the two settings. What really matters, to me at least, is that the testbed was consistent. Even if you can repeatedly show EIST/ speedstep doesn't make much difference, that's one of those things that can cause internet wars. Thanks again for looking into it.

This was the only thing I cared about. Thank You again Kristian & Anand!

 

[Hellhammer]

I have a lot more I want to say, but I don't have a ton of time now.

Whenever you have time, don't hesitate to post. There is nothing wrong in questioning our tests and you brought up a crucial typo here

The sole reason why I follow these forums is to connect with our readers. Partially to help and clarify things, but also to learn and answer questions about our tests (though I can't reveal everything).

 

[jwilliams4200]

As Jocelyn84 hinted at, C-states are the important setting on several of my motherboard BIOSes. On older ones, there is just one setting, C-states enabled or disabled (SSD benchmarks are faster with C-states disabled). On some newer motherboards like some Z77 ones, there are multiple C-states settings, and to get the highest SSD benchmarks, it is necessary to disable all the C settings (C1, C6, etc.). Or just run one thread of Prime95 or other CPU intensive task.