M17x Stutter

[Lord_Zath]

I'm not sure if I did this right or not. I tried to follow your directions. Attached is a SS of my RMClock graph showing some throttling. The first set was about 5 or so seconds in. The second set was after I stopped Furmark about a minute or two later. Fans went on like mad, but the processor temps only went up to 48C. I do not know what the GPU temps went up to since Furmark showed 0C the entire time.

I ran with the multi gpu option on default settings.

A couple notes about my system:

Win7 64-bit
BIOS A03
My T9600 is overclocked to max in BIOS - 3.5 ghz
Everything's bone stock
LED's and other "power hogs" are all on

Please let me know if I did this right!

I'm wondering if this is happening more to the quad-core users - I'd imagine the quad-core requires more power than the dual-core? Elkay, I did go to the processor monitor tab in your SS and noticed that ODCM was NOT blue like your picture while the test was running...

[Elkay]

Intel states that clock modulation, when used, should be set equally on each core. I sometimes wonder if Dell does this to try and hide what it is doing from monitoring software. If you only sample one core and it is set to 100% then it might miss that clock modulation is still going on. The ThrottleStop log file samples all cores and reports the lowest clock modulation percentage. RealTemp does the same.

I've only seen this on the second and third core. I'm not going to flat out call BS on Dell without absolute proof, but you could be right.

I'm likely going to wait out the end of the week to see if the 5870x2 gets released on the R2, as it runs 10W less TDP total than even the 48070x2 (120W vs 130W, with the 280Ms coming in at a whopping 150W). That could make all the difference, at least under "normal" usage, which to me includes current games like Dragon Age and Crysis. I don't think that's unreasonable to ask of this laptop.

The crazy thing is, Dell officially announced it will *not* be supporting the 5870x2 on the R1. Not sure exactly what to take from that, seeing as the R1 is only 9 months old and uses the exact same MXM 3.0b slots as the R2.

[Elkay]

Thanks for testing, Lord_Zath! Your screenshot isn't pulling up for me, though. Your components (CPU/GPUs) definitely have a bit lower TDP than mine. Would be great if you could put that screenshot up again. Also, are you saying ODCM didn't light at all during your test, or was it only for a moment that you checked? It only lights up when it's currently active.

[trikster]

Thanks for testing, Lord_Zath! Your screenshot isn't pulling up for me, though. Your components (CPU/GPUs) definitely have a bit lower TDP than mine. Would be great if you could put that screenshot up again. Also, are you saying ODCM didn't light at all during your test, or was it only for a moment that you checked? It only lights up when it's currently active.

I checked and his image only loads about a half inch of the image then hangs...

[Lord_Zath]

hmm weird. I'm at work so I'll re-upload it tonight. Must've timed-out during upload.

Elkay, the ODCM didn't light when I was looking at it, which wasn't for long. I'll try doing another test tonight and seeing if it lights up.

[unclewebb]

You can save the image to your desktop when it is trying to load up and then view it that way. There doesn't appear to be any clock modulation going on during the test. The end of the test looks more like SpeedStep kicking in. Not sure.

ThrottleStop can log any incidents of clock modulation so it is easy to go back and look at the log and see exactly what was going on.

http://www.fileden.com/files/2008/3/...rottleStop.zip

Elkay: The most important thing is to keep power consumption under the trigger point for throttling. A Kill-a-Watt meter may not be the most accurate device but it's useful enough and consistent enough to tell you how many watts from the wall triggers clock modulation throttling.

Clock modulation wouldn't be so horrible if it went from 100% to 87.5% but the bios doing 100% to 12.5% in one big step kills performance and turns a great game into a miserable experience when it happens.

Once you know how many watts you are allowed to draw without any throttling then you can try to adjust your performance to keep power consumption under that level. Overclocking used to be the way to maximize performance but because of this, you have to go the other way to under clocking.

I know on the desktop Nvidia GPUs, you can use Nibitor to lower your GPU voltage but flashing the GPU bios is a little risky. Dropping GPU voltage by 0.1 volts can make a significant difference to power consumption. I also wouldn't overclock the GPU while trying to play Crysis.

You should definitely be undervolting your CPU and for gaming, I wouldn't be overclocking it either. There's no point in overclocking 20% if this helps to send you over the edge and you end up losing 90% of your CPU's performance.

On a 45nm Dual Core I only save about 6 watts at the wall by undervolting while running two threads of Prime95 Small FFTs. On a Quad that should be closer to a 12 watt savings which might help out and maybe a couple of more if you get your CPU back to stock clocks.

If you prefer RM Clock, that's fine. With a simple registry adjustment it can be setup and used with CPUs that have half multipliers. You don't have to worry about losing that. Here's a P95 run with a 10.5 multiplier while using RM Clock.



RMC 45nm Half Multiplier Fix
The "Undervolting" Guide

Hope you don't mind Lord Zath, I uploaded your pic to imageshack.

[Elkay]

Look at how rock solid his throttle is in the top graph. I've never seen mine do a straight line. It's always had at least teeny tiny dips and fluctuation, as I mentioned previously. Very interesting. Really makes me think the power supply is being pushed beyond its limits in my setup. I should also mention that I'm running RAID 0 (stock from Dell), so I have 2 hard drives also sucking juice. Googling seems to show an average draw of 10W per drive, though I can't vouch for accuracy on that.

All of the last tests (TS logs and RMC screenshots with all of the huge, long modulation dips) with Crysis and Dragon Age were 100% *stock* clocks. There was no CPU or GPU overclocking or overvolting. That's how the machine runs coming from the factory.

Dell provides no voltage control for the CPU. That's been one of the sour points about the BIOS, and the BIOS is encrypted so nobody has been able to mod it. I'm no newbie at flashing BIOSes, actually last night I tried bringing both GPUs down from 1V to .85V since .85V was already defined, but once it got to Windows it started flickering and barely showing the screen so I just went back to 1V for today at work. Tonight I can experiment some more, .90V being another pre-defined voltage that I can just try, and from there on I have access to the voltage tables anyway and can do them in .01V increments. I've already built a .95V BIOS, actually.

Regardless, why did I pay all of this money for an unlocked, overclockable CPU straight from Dell if they're forcing me to underclock and undervolt it just to play games stable?

[Mickbt26]

I have the same specs as you Elkay and mine stutters a lot aswell whilst gaming when throttlestop is not running.
It seems to me as soon as the gpu fans come on full speeds it will start throttling.
I want to know if I am risking breaking the psu or motherboard by using throttlestop over the long term.

[unclewebb]

If the GPU voltage is already down to 1.0 volt, I think you will lose GPU stability if you try to go any lower.

Dell provides no voltage control for the CPU.

A couple of quick mouse clicks in ThrottleStop can drop your CPU core voltage and save you some watts. For 2300 MHz, I need somewhere around 1.00 to 1.05 volts to maintain stability. You might be able to run your Quad reliably at default clocks at 1.05 volts or very close to it.

I agree with your last comment. Dell should not be selling products if they can't perform at the level that their website promises and leads you to believe. No one would pay extra money for SLI and an expensive QX CPU option if they knew that it was going to kill gaming performance. Dell.com doesn't mention this problem at all and the sales reps won't admit to this either.

Mickbt26: Invest in a Kill-a-Watt meter. I have no idea where Dell has set the bar here. They tend to use clock modulation and limit power output from their adapters to far less than the rated DC output specification. Any adapter that passes certification is supposed to be able to run at 100% of the rated DC output for extended periods of time without damage or failure.

I don't have access to any Dell testing so I can't guarantee you that ThrottleStop is 100% safe to use. I have no idea what under designed component or components Dell is trying to protect with this throttling scheme.

[Elkay]

Good to know I'm also not the only one seeing this still. As far as breaking something, I think it's been made obvious through my testing that the system will shut down before it gets to a threshold that would break something, as long as you're not doing it constantly. The CPU isn't even getting to 100 degrees, and it's not set to shut down until 125 degrees and would itself throttle long before that. As far as the PSU goes, that is sort of up in the air. We're definitely drawing more than it's speced out to (some say they see it get to 295W), so I can't speak for the longetivity of it. It definitely gets almost scalding hot. I have a Kill A Watt meter arriving tomorrow so I can get a more accurate case of real-world current draw.