Intel SU2300. Speedstep or not?

Intel says this CPU has Speedstep, but RMClock only shows one multiplier: 6.

Also CPU-Z says it's always running at 1.2Ghz.

I would like to downclock it to get better battery life.

 

your not the only one.

http://forum.notebookreview.com/showthread.php?t=439724

 

Thanks, interesting.

 

Same thing with the su2700 i believe.

 

Yup, same here. In fact, I'm pretty POed that despite what Intel's site claims for the SU2300, it does NOT have SpeedStep. That's one of the main reasons I picked the Aspire with the SU2300: it was dual core, but 10W TDP and WITH SpeedStep. I mean, geez, at least throw in ONE other P-state like 800 MHz at a lower voltage.

Intel's page says it has EIST:
http://ark.intel.com/Product.aspx?id=42779

But the datasheet on the SU2300 basically says it doesn't (page 30, note #10):
http://www.intel.com/design/mobile/datashts/321111.pdf

Right, so that means no SpeedStep! You can't tell me that lowering the voltage by 0.0875V constitutes SpeedStep when in every other Intel processor I've ever seen, SpeedStep means at least 2 different P-states that are not the same clock freq.

Speaking of voltage, I did manage to undervolt my SU2300 to 0.9 using RMClock, which resulted in a 5C temp reduction at full load! SU2300's min voltage is 0.8 according to the Intel datasheet, so it's rock solid and running cooler. Still, it certainly would have been nice to get what I paid for!

Finally, thanks to all of you who have posted such awesome info here on the 1410 and similar laptops. It was like hitting the jackpot!

 

Phil,

What I've noticed is that on my VAIO, speed step isn't used, of course, when running the high performance power profile, but interestingly, using the CrystalCPUID 'real time clock function' (simply press F4 when CrystalCPUID is running), the CPU runs all the way down to 108Mhz via the FSB being dynamically controlled.

Could this possibly be happening also to the CPU you're running?

Download CrystalCPUID for yourself and see:

http://crystalmark.info/software/CrystalCPUID/index-e.html

 

Thanks for the tip Tiller.

In Crystal it seems like it downclocks but when I check with other programs (CPU Z, RMClock, Super PI) it still runs at 1.2GHz.

 

Try pressing F9 on CPU-z. It unlocks the actual clock speed to detect SuperLFM and such

 

Thanks, just tried it. It stays on 6 x 200 = 1.2GHz.

 

I tried CrystalCPUID on my machine. Even though other tools are not reporting a change in the CPU clock speed, my battery life went from 3hr50 on a full charge to 6hr10. The Realtime Clock window does indeed show that the clock is being reduced, and it appears that it IS actually doing something.

When I start doing heavy-duty stuff, the clock jumps up to the full 1.2GHz, and estimated battery life drops. When I'm just sitting there reading a web page, the CPU apparently drops to like 196MHz (???), and estimated battery life jumps back up again.

After a couple of hours of running on battery with CrystalCPUID, I have 4 hrs 20 minutes left at 88% battery capacity (WiFi off, ethernet in use, 40% brightness).

Obviously, due to the severe lack of SpeedStep, the multiplier settings make no difference. Now I just need to figure out if I can force a slower max clock speed when on battery, like 0.8 or 1.0GHz to really maximize battery life without sacrificing performance for things like flash vids.

Thanks for the tip, tilleroftheearth!!

 

I also checked the battery life, mine stayed at 5:20 no matter if I pressed F4 or not.

Maybe I need to give it some more time.

Edit: did more testing: no change in battery life at all. Weird.

 

I have to wait a few minutes for the standard Win7 battery life monitor to update its runtime value. Although when I also run RMClock, its battery life monitor also reports over 6 hours at a full charge.

In my case, the SU2300 doesn't have SuperLFM, unfortunately. It only has a single LFM/HFM value.

After playing around with CrystalCPUID, it seems that I needed to use the following settings:

1. Enable multiplier and set Max, Norm, and Min voltages to 0.9V. For the SU2300, all multipliers will be 6.0, but at least it will enable undervolting.
2. Enable Realtime Clock

If I don't do the F4 thing to load the RTC box (and keep it open), I don't get the extra battery life.

I couldn't get CrystalCPUID to run automagically via Startup folder or Task Scheduler. The RTC function kept stopping and giving me that annoying error message that I have to change the RTC setting to "No Load" in File -> Customize. Which I would do, and then change it back to the second option... Then the RTC feature would work fine.

Sooo, I just run it manually with a shortcut with the following options:

/CQ /R /HIDE /RESI

/CQ = enables the Multiplier Management
/R = runs the RTC box
/HIDE = hides main dialog
/RESI = run resident (dunno what this does, but it sounds good)

That was the only way to make it happy with the RTC feature. Your mileage may vary. It also gives that RTC error if I wake from sleep. So, it's kind of annoying that I have to run it manually when I switch to battery power and/or reboot/wake up, but at least it works!

Oh, I also went into File => Customize and unchecked "Top Most" to keep the RTC window from overlapping other open windows.

Hmm... I dunno if this makes a difference, but I have the latest SATA, Chipset, and 4500MHD drivers from Intel installed on Win7 x64. I'm using ethernet for my LAN connection, WiFi off, no bluetooth installed, 40% brightness.

 

I was using Battery Bar for battery life predictions.

Maybe Battery Bar was wrong though, I'm now doing a manual timing.

My SU2300 was already at 0.9V.

Edit: Seems like Battery Bar is wrong. I surfed for 21 minutes on 5% battery capacity. That's more than 400 minutes on a full battery. So seems like it is working.

 

so does it have speedstep? If it doesn't intel has scammed u...

 

No speedstep. Wrong information by Intel.

 

Yes indeed. Not that it will make any difference, but I posted about the SU2300's severe lack of SpeedStep on my blog.

And then I sent some e-mails to some big tech sites. Got a reply from one of them after an hour or two saying thanks and that the info was passed along to their Intel reporter.

I doubt anything will come of it - unless all SU2300 users banded together to demand satisfaction. But it sure beats sitting around and doing nothing about it!

 

agreed... but there's nothing much u can do... they'll just say that the OEM has disabled speedstep....or some excuse...

 

can no one get this to run on start up? If so please provide easy instructions

 

Alright, time to clear this up.

http://processorfinder.intel.com/Details.aspx?sSpec=SLGYW

 

I don't see how that clears things up.

Taken from that website:
Supported Features:
# Enhanced Intel Speedstep® Technology

Reality: No speedstep.

 

It still does voltage scaling, just doesn't scale clock speeds down.

 

Yeah I noticed. But if the clock speed went down too, it would have bigger power savings.

Now I need to use CrystalCPUID, which works too.

But thanks for clearing it up, now I know what Intel means.

 

Yea its a Celeron, Celerons usually never have Speedstep. Surprised they implemented anything at all.

 

As I posted on my blog, that's what I pretty much figured out already - although thanks for the link! I also found another place on Intel's site dated 14 October 2009 that indicates the same thing.

http://www.intel.com/support/processors/mobile/celeron/sb/CS-030964.htm

So, that's nice. At least they aren't being totally dishonest intentionally. I'll amend my blog post.

However, I still maintain that it's rather deceptive to call undervolting "EIST". Sure, they can call it whatever they want and define EIST however they want. But when everyone expects EIST = underclocking and not just undervolting (which normal EIST does as well), then confusion will naturally reign.

Besides, when I undervolted my SU2300 to 0.9V, I saved no measurable amount of power. When I use CrystalCPUID to underclock it, I save TONS of power. So obviously the most important aspect of real EIST is missing from the SU2300's specific "flavor" of EIST. I blame the marketing people.

Having said all that, I'm very happy with my battery life using CrystalCPUID, and I'll by spreading that info around to the "less technically inclined". I know a lot of people who'd love to get more battery life for free, but who would never understand a single word of this forum!!

Anyway, thanks for the info!

 

Hi,
I used the exact same configuration : 0.9V with CrystalCPUID.
But i do not run the RTC option as i'm not sure to understand what it does (i'm just interested in lowering the temperature).

However i got a BSOD after a couple of hours using my laptop.
I would like to set the voltage to say 0.925 to try if it gives a better result. Unfortunetly crystalCPUID seems not to take my configuration into account. When i check with CPU-Z it claims the voltage is 0.900 (same for RMclock )

Is there something wrong, is there something to do with the bios or so ? Have you tried to set it to a higher value ?

Em.

 

I'm not trying to start an argument or anything but based on my knowledge of Speedstep, I'm not entirely sure why you're getting so upset that the SU2300 doesn't have it.
I mean, obviously it's not great if you were lead to believe it did prior to purchasing a laptop that has the SU2300 but it would make no practical difference, you're losing nothing.

Speedstep works by reducing the CPU's multiplier when there's minimal load.
Unless things have changed recently, the minimum multiplier that Intel CPUs can utilise is 6x, so to have Speedstep on a CPU with a maximum multiplier of 6x would do nothing.
So, unless I'm very much mistaken, Speedstep would be pointless on this CPU.

 

Speedstep can lead to greater battery life. Now the SU2300 is stuck at 1.2GHz. If it could down clock to 600 Mhz for example it would consume less power.

 

Speedstep just varies the multiplier though, it doesn't touch the FSB.

Your CPU speed is made up of your FSB x a multiplier.
My desktop's Q6600 for example; it's FSB is 266MHz, it has a max multiplier of 9 and, like all other Intel CPUs, a minimum multiplier of 6. So the stock speed is 2.4GHz (9x266MHz)

All that happens when Speedstep is active is that the multiplier gets dropped from the max ( or a user-specified value that is less than the CPU's upper limit) to 6.
So, when Speedstep kicks in on the Q6600, it drops from 2.4GHz to 1.6GHz (6x266MHz) but the FSB doesn't change.

On the SU2300, it would drop from 1.2GHz (6x200MHz) to 1.2GHz (still 6x200MHz) - since the FSB isn't touched by Speedstep, you would see no drop in CPU speed and, consequently, no increase in battery life.

 

Thank you for the lesson but I know how clock speed is made up.

By the fact that Intel advertises Speedstep on the SU2300 I assumed it can lower the clock speed.

It can't. False advertising in my opinion. Feel free to disagree.

I never said it did.

 

Well, I wasn't trying to be condescending but after pointing out that Speedstep cannot drop the CPU's multiplier below 6x, you made your comment about dropping the clock speed from 1.2GHz to 600MHz, something which could only happen if the FSB was dropped, which is not how Speedstep works.

As has been mentioned previously in this post, it looks as though any mention of Speedstep with regards to the SU2300 was a genuine mistake.
All I'm saying is that even if it did have Speedstep, it wouldn't actually make a blind bit of difference.

 

Speedstep to me means the multiplier can be lowered. I think we all agree on that.

So if it would have Speedstep, that would mean the multiplier could be lowered. If the multiplier could be lowered, that would mean power consumption would be lower.

I hadn't heard of a hard limit of 6 for the lowest multiplier and I don't see why Intel could not implement a lower multiplier.

As pointed out by Scottie and Inteluser earlier in the thread the mention of Speedstep does not seem to be a mistake.

 

Right, yea. Speedstep does have x6 multiplier minimum on all mobile Intel CPUs. I don't know what's up with that, but that's how it is.

Anyways, the SU2300 doesn't support Super LFM which would allow the FSB to go to half frequency and result in 600MHz.

 

It's just the way it's always been.
There's no technical reason to it from what I can work out, it's just Intel being awkward.

 

read thread, very informative, thanks to all contributions!

so... consensus is manual undervolt (rmclock, etc.) on su2300 has no significant change for battery life

 

I guess it depends. Doesn't seem to have any effect on mine. Pressing F4 in Crystal CPUID does seem to have effect.

 

Hmmm... that stinks... wonder if it's even worthwhile. Scottie seems to have favorable results, but then ~6 hours w/o WiFi is probably close to what I'm getting now, though I haven't done a real world test... prolly because I'm never around an ethernet jack that long. Heck though, Win monitor could be reporting wrong, but it somtimes says 6 hrs and some odd minutes and that's with WiFi on. Maybe I should check Batterybar again, it always reported lower life with my previous 1410

 

So, maybe Speedstep is not all it's cracked up to be?

This is an very interesting article that states Speedstep doesn't affect power consumption as much as we may think. It may even be detrimental to getting maximum performance too - especially for AMD platforms.

See:
http://it.anandtech.com/IT/showdoc.aspx?i=3722

Cheers!

 

I don't suppose you'd care to be a little more specific as to where it states that Speedstep may not be all it's cracked up to be?

Also, the article does seem to be written with regards to servers, not laptops or desktops and some of the initial points on the first page don't really translate to consumer usage/needs.

Lastly, perhaps a touch pedantic but technically there's no way Speedstep can be detrimental to AMD platforms as it's an Intel-specific technology.
True, AMD do have their own equivalent technology but it's different to Speedstep ( no 6x lower limit on the CPU multiplier from what I can work out) and the differences in the way it's implemented could well be why it has a bigger negative effect on AMD's platforms.

 

Step666,

Yes, you're right in regards to Speedstep on AMD platforms. Sorry I wasn't more clear: I meant PowerNow! and the like for AMD systems.

This is biased towards servers, but Speedstep and PowerNow! are not server only technologies.

Basically (without skimming the whole 12 pages again), it states that clock gating and power gating is more practical/effective than Speedstep and PowerNow! and that the Intel's use of more aggressive C6 states and faster waking/up of sleeping CPU cores than AMD's platform offers gives it a performance and power advantage when using Balanced and Performance power options on the different platforms.

What is also interesting is that using the Performance option on an Intel platform (a modern one) does not negate any power saving effects - that's how effective power gating and clock gating is.

I can confirm that running my VAIO (definitely no server) on the Performance option does not use appreciably more battery life than when it is on Balanced. This is why I think this link was applicable to this thread.

Cheers!

 

Well, as I said, it was a bit pedantic the whole 'Speedstep being for Intel only' point but if it is that much worse for AMD platforms, it suggests that their implementation is in some way flawed.


Obviously, there will be some points in the article that are as applicable to personal computers ( be they laptops or desktops) as Speedstep, PowerNow! and the like are technologies found on both consumer- and enterprise-class platforms.

But, given the tone of the start of the article and the points it makes regarding the whole performance/watt issue and why that doesn't apply to servers, that marks a reasonably clear difference between the requirements of servers and those of consumer systems.
So, it does make me wonder whether this supposed loss of performance is really that much of an issue for consumers, for whom it isn't as necessarily as imperative to squeeze the best performance out of their PCs.

I'm not saying that it's definitely not a matter that people should be concerned about, I just think it would be useful to read a similar article written from the viewpoint of an individual's needs, not the requirements of a server.

 

Well, 15 to 25% less powerful is too much of a drop in performance for anyone.

Otherwise, we should just get the lower end/cheaper CPU and be done with it. (And forget about better chips/platforms).

You really should read the article, it is an eye opener for how Intel and AMD implement these technologies and how useful they are in real world use.

Cheers!

 

I honestly don't believe that there's any way that Speedstep can cause a 15-25% decrease in performance.
I use it on my Q6600 to keep temperatures and noise down and I haven't seen any noticeable performance decrease - certainly nothing close to 15%, let alone 25%.

I fully intend to read the article at some point, it's just a little late to be getting quite so technical.

 

Not Speedstep, AMD's PowerNow! is causing the decrease in performance.

Cheers!

 

Ah sorry, my mistake.

 

Don't take a server Windows review and read it same as home PC review. They are not one and the same.

On server platforms, it looks like Balanced mode disables Turbo Boost. On the home versions of Windows 7, Turbo Boost works just as well and there is no performance loss.

 

neither here nor there but in the spirit of energy savings, bought a 320GB Hitachi HDD with the uber low W requirements (1/2 of industry standard) and i can get windows power management to actually say 7 hrs, 6 min now on full charge... pretty cool though don't know about real world results. I haven't done the undervolt listed here as results seem mixed and overall, not worthwhile.

 

Hi all,

Have found this thread interesting and slightly disconcerting as I wish I had read it before buying a UL20A with a SU2300 - but here we are! I upgraded from an Atom and wanted the power, but was really afraid of the battery life drop.

So, while i'm seeing most of the results as reported over the past pages in this thread with CPUID-Z and Crystal's CPUID, the latter in F4 (realtime) mode DOES seem to dynamically clock with the load I put the CPU under.

For example, I'm just typing in the forum now and it shows 'Q 299.Mhz 49.94x6.00'

First, what does the Q stand for? This is not documented in the readme.

Second, the speed, dropping as low as 25 (150Mhz) and of course as high as 206 (1236Mhz). I can force it up by watching you tube in other browser tab. It seems to hover as low as possible and speeds up in relation to what i am doing. Now the interesting thing is that there is no relation that i can see for the base speed (if this is fsb). THe multiple does stay fixed at 6, but the other number seems really random: 52.79 or 41.69 or 36.61 or 43.14 or 40.41, etc.

Now i'm assuming (hoping in fact) that this is the processor dynamically undervolting. The thing that makes me think so is that i also have BattStat running and see the wattage rise totally in relation to the increase in MHz. Currently draining at about 8-9W and under load up to 16W. So on a full charge, running full tilt, it predicts something like 4hours, but sitting idle, this quickly increases to 7-8hours. (5600ma battery). Have been playing around with the computer most of the day and seem to have 6-7 on its first full charge which i'm generally satisfied by (whew!).

Now in Win7 power settings, you can set the minimum and maximum power states as a percentage value. I've been fiddling with this and it doesn't seem to change anything, so i wonder if this only works with a multiplier.

So those are my observations (and hopes) on this processor. As nobody else in the thread mentioned this seemingly dynamic clocking, am I alone in observing this?

EDIT: discovered that the max was 206 because the BIOS has a default 3% overclock in the BIOS! (gotta love asus) I set this to 0% and now both CPU progs report 200x6=1200.

EDIT2: so i've now seen the base speed reading go as low as 14.97 (89.82). am now almost certain of the correlation between this number/cpu load/and wattage use