by Sinan Karaca
I purchased the new overclockable Dell XPS M1710 gaming laptop, which ships with a "non-roadmap" part from Intel – the T7600G Core 2 Duo processor. This processor is identical to its T7600 brethren, with the only difference being that it is overclockable (and a $275 price premium, which was $350 at my time of purchase).
Intel probably figured they can squeeze a bit more juice from their Core 2 Duo architecture, so they delivered this part. In Dell’s machine, the overclocking is done through the BIOS, where the following clock speeds are available from a dedicated BIOS screen:
2.33GHz, 2.5GHz, 2.66GHz, 2.83GHz, 3.0GHz, 3.16GHz
Another Dell BIOS setting enables/disables SpeedStep Technology, which is normally used to control processor throttling. Unlike other systems though, this BIOS setting has a diametrically opposite effect on the Dell system: When disabled, it locks your CPU at the 1.0 GHz speed setting – whereas with other laptops, disabling SpeedStep locks the CPU at the maximum clock speed. This fact will become more relevant later on, but is quite surprising in its own right. So the total range of clock speeds supported by the machine comes out to:
1.0GHz, 2.33GHz, 2.5GHz, 2.66GHz, 2.83GHz, 3.0GHz, 3.16GHz
or at least, that’s what it looks like on paper. Hint-hint
The 3.16 GHz speed setting was unstable, even under regular use of the laptop. However, I was pleased to find out that the 3.0 GHz speed setting did appear to be stable, even when the laptop is put under heavy load. While this was initially very encouraging, I ran into some oddities. In particular, the laptop clocked at 3.0 GHz was actually taking 2 hours to complete a compression task which my control system similarly configured but with a regular T7600 processor clocked at 2.33 GHz, took slightly over 1 hour to complete. Repeated testing confirmed this result, so I was forced to undertake a lengthy investigation – which yieldedsome very disturbing results.
I won’t make it a surprise ending for you. Due to a combination of factors, this notebook often ends up performing worse than most of its regularly-clocked counterparts, as we shall soon find out. Some heavy duty tweaking and experimenting is necessary before the overclocking even begins to pay off.
Once the optimal settings are found, Dell’s BIOS still does everything it can to get in your way and reduce whatever speed gains you would otherwise obtain from a regular overclocking process. So let’s see what exactly is going on under the hood here…
The Computer that Lies about its True Speed
I normally use CPU-Z to identify the true clock of a processor. This is a rock-solid product that always reports the accurate GHz/MHz of a processor, and dynamically – it correctly updates the reported speeds as Intel’s SpeedStep technology throttles the core clock up and down. So you can directly see how your computer is performing under various load conditions. Feel free to grab a copy of CPU-Z here, and try out for yourself:
One of the first things that I experienced with this laptop is that the machine actually lies about its own speed. Even when CPU-Z reports 3.0 GHz, the machine would actually be working at a lot less GHz, sometimes as low as 266 MHz! Yes, you read that right – 266 MHz, which I believe was one of the original clock speeds of the Pentium II processor, about a decade ago.
While we’ll soon find out exactly how that disaster happens, the tool that correctly reports the processor speed is called RMClock. This tool seems to be the only one on the entire web which Dell’s XPS BIOS cannot fool.
Leave it to your trusty Russian hackers to figure out the truth As you can see in the screenshot above, the tool captures the reported clock speed, the real clock speed, the temperature, and the voltage of each CPU core, real-time, and graphs it out like Task Manager’s resource utilization charts.
The Computer that Doesn’t Want You Setting its Speed
Another one of my trusty tools which has stood the test of time until Dell’s overclockable XPS is SpeedSwitch XP:
This great tool puts you in charge of your CPU’s speed, instead of having Windows with its cryptic and obscure power management settings run the show. It has 4 modes of operation: Maximum Performance, Maximum Battery, Battery Optimized, and Performance Optimized. With the last two normal throttling does occur, with the first two the processor is locked at the fastest and lowest speeds respectively, so you know what you’re getting. With games and other resource intensive tasks, I’ve always seen performance gains when running under Maximum Performance. And to turn that annoying CPU fan off, or to maximize battery life, nothing beats Maximum Battery.
SpeedSwitch doesn’t use any magic; it just internally creates and manages Windows power profiles, without having you pull your hair out. These profiles produce the desired result, and all you ever do is just choose the performance mode you want in plain English. Unfortunately, with this Dell system, the tool had no effect. In fact, it was so powerless, that even CPU-Z reported idling CPU clocks (1.0 GHz being the lowest for this model) when Maximum Performance was set in the tool (expected GHz being 3.0).
So again, I had to scout around and find another tool which could force the right speed setting. Notebook Hardware Control to the rescue:
Just like RMClock, this great tool is able to bypass whatever custom protection Dell’s BIOS has, and imposes one of the 4 performance modes mentioned earlier forcibly. As an added benefit, it also adjusts the core CPU voltage…which became a vital need down the road. So, kudos to the makers of Notebook Hardware Control!
First Accurate Results, Recorded without Voltage Regulation
Armed with the right tools for the task, let’s take a look at some results. All screenshots were captured live on the Dell machine. The program used to stress the machine is 7-Zip, which is unique in that it can perform dual core compression (or n-core compression, for that matter). If you’re still using RAR, or – gasp – ZIP, 7-Zip provides better compression with faster speeds thanks to its dual core features. For our purposes however, we don’t really care about the compression so much, just the fact that it will fully utilize both processor cores. I used the CompreXX shell around 7-Zip which makes the tool a little easier to configure:
As you can see in the following screenshots, without voltage regulation, the following happens:
Rising temperatures and drops in clock speed directly correlated
Peak temperature appears to be 93.2 Celsius before BIOS-forced throttling begins
The lowest speed I was able to observe was 266 MHz – yes, .2 GHz!
The average clock speed during full-power compression comes out to slightly less than 1.0 GHz!
- The CPU seems to have a peak temperature tolerance of about 93(C).
- This temperature is achieved within the first two minutes of full system load (for both cores).
- As soon as this temperature is reached, the BIOS forces the CPU to throttle down.
- No tool that I have been able to find can prevent this forced throttling.
- Only the RMClock tool is able to even detect that the throttling is occurring at all.
- While the throttling occurs, the voltage fed to the CPU remains the same.
- Thus temperatures stay more or less constant.
- CPU speeds fall dramatically and linearly over time, as low as 266 GHz.
- As the CPU cools off, Dell’s BIOS slowly increases the core clock speed again.
- As the core clock goes up, so does the heat. Another cool-off cycle begins, and the same mechanism repeats thereon after.
- The CPU returns to its maximum speed only when the processor starts idling again, which is quite useless since when idle, the CPU isn’t even being used!
This explains why the laptop performed twice as bad as its lower clocked, on-the-roadmap 2.33 GHz counterpart! But what it doesn’t explain is:
- How is Dell achieving this magical under performance? My guess is its some kind of BIOS magic – the BIOS not only lies about the true system speed, gets in your way of setting its performance to a desired level; it also actively and un-overridably underclocks the machine.
- Is Dell not aware that at best, this is a bug? I can understand that whatever mechanism they have put in there is to protect the machine from overheating. However, the mechanism is flawed, because even at 266 GHz clock speeds, it is still being fed the full voltage of the 3.0 GHz clock speed. Needless to say, this doesn’t help much with the cooling.
- Has Dell never tested this system in the real world? At worst, this can be thought of as false advertising, because while the BIOS and a plethora of other Windows tools report a 3.0 GHz clock, the system is running at extremely low speeds, resulting in extremely undesirable performance.
In the last analysis, especially with an overclockable system, it’s extremely disappointing to see that Dell is "doing what’s best" for you, deciding on your behalf. While they obviously do this to reduce the number of fried computers and support calls, their implementation is buggy. This results in an almost unusable system for all practical purposes.
It is also important to mention that with most synthetic benchmarks, and especially those that exercise only one processor core, the heating problem isn’t as bad, and the underclocking isn’t as bad. This might be why Dell’s own engineers failed to discover this major issue with their system – yet, this excuse would be worse than the disease, if indeed the cause
Methodology for a Cure
Since the issue seems to be heating, and how Dell is forcibly lowering the CPU clock to countermand that, I figured the only chances of getting more juice out of this computer is lowering the CPU voltage. That way it might be possible to delay heating and actually receive some kind of performance gain.
Initial tests with lowered voltage
System idling before beginning work
After barely two minutes of dual core compression, system takes to underclocking
Extended compression consistently lowers processor speed, even beyond its rated clock speed of 2.33 GHz, despite voltage control
Occasionally, the CPU does return to faster clock speeds – even touching its peak of 3.0 GHz – but only briefly
The system returns to the true speed it always reports, 3.0 GHz, only after the CPU has been idled – not much use at that point!
The maximum tolerated temperatures before BIOS-forced underclocking occurs is about 82 degrees Celsius – about 10 degrees lower than when native CPU voltages are used
Lowering voltage has the undesired side effect of making the system unstable. If you go too low, you will start seeing odd crashes with programs that are currently running. Fortunately, the Notebook Hardware Tool can be used to bump up voltage dynamically as this happens, so at least you don’t need to reboot your system every time that happens. Even then though, the process is time consuming and error prone.
Another issue that affects voltage is the number of USB devices that you have connected to the system. In particular, devices which do not have their own external power source seem to cause some issues. They dramatically reduce the tolerance of low voltages, making your system unstable a lot sooner. Unplugging these devices seems to let your system go down to lower voltages safely.
A final issue that I ran into was that my voltage which had worked well initially "degraded" over time. The initial voltage I settled at, after substantial trial and error, was 1.3125V. This yielded, in my real world usage, only a 4 minute time savings compared to the 2.33 GHz control machine. However, after a couple weeks I found out that even this voltage was causing rapid overheating. The new "stable" low voltage seems to be 1.2750V. It’s interesting to note that this voltage was previously unstable, but now it appears to be stable, and any higher voltage still causes dramatic heating issues. Again, the real world benefit is only about 4 minutes, the same that the previous "stable" voltage had yielded. I don’t know if this last issue means the CPU is fried, or something else. I find that unlikely, as I run the processor at the lowest speed setting most of the time (forced with Notebook Hardware Control). So it might just be another mystery.
Best results I was able to obtain – an average clock of 2.4 GHz with a very low voltage of 1.275V
Low voltage set through Notebook Hardware Control
A momentary lapse of BIOS interference
Throttling resumes, inevitably
Best-case long term powergraph; sustained speed is barely 2.4 GHz
As seen in the screenshots, even with optimal voltage settings, the system averages a core speed of 2.4 GHz at best. It keeps fluctuating between 3.0 GHz and 2.0 GHz in an attempt to "cool" its heat. Our reduced voltage certainly helps with this, but the net result is just about 60 MHz of real overclocking.
The system simply performs as a 2.4 GHz machine overall – and that is, with multiple utilities, tweaking, and crashes in between on the road to getting there. I’m going to have to call this a disappointment.
I tried two other types of experiments. The first was to attempt to get 3.16 GHz to run stably through reduced voltage. This never worked. The second was to try out the lower speeds, starting with 2.83, and scaling slowly down, to find a sweet-spot where the BIOS might not interfere, and where one might see more than a 4 minute performance improvement on a task that takes 60 minutes on a 2.33 GHz machine. Unfortunately, even at the lower speed settings, the BIOS continued stepping in to "cool" the machine.
It seems inevitable that, when you are utilizing both CPU cores, heating will occur. Heating is a natural process that cannot be prevented with any processor, after all. The moment that happens, Dell’s BIOS steps in and under-clocks the system, without reducing its voltage. The slower your original clock, the worse your overall performance. So 3.0 GHz appears to be the sweet-spot after all.
Closing Remarks (to Dell)
It is my hope that someone from Dell would read this article, and publish a BIOS update, or something similar, to solve the problem. I get the feeling they had the best intentions in trying to prevent people from frying their systems, but their implementation is so poor, that without special tools, the CPU throttles all the way down to 266 GHz (hey, that might be an interesting battery life technology there). And even with lots of patience and tweaking, the CPU barely manages 2.4 GHz on average. This is not the CPU’s fault – it seems like one solid piece of hardware – but the BIOS.
Most programs fooled by Dell’s BIOS into reporting incorrect clock speeds
If the Dell engineers manage to at least feed lowered voltages to the CPU as they forcibly underclock it, I suspect it will be a lot quicker for the CPU to cool down, and therefore a quicker return to faster speeds. At a minimum this obvious bug in their implementation must be fixed. Above that, I would like to see them stop lying about the real clock speed of the processor, and to add in an option that disables the BIOS throttling mechanism altogether, leaving the machine to Intel SpeedStep’s native devices, which seem to work just fine on every other system. I’d rather suffer system crashes in exchange for predictability in clock speeds – compared to the BIOS lowering my system clock all the way down to 266 MHz in exchange for “crash-free reliability”!
If you have one of these machines, please install the tools, and report your own findings. Hopefully with this article you can squeeze every bit of performance from your XPS M1710 machine – despite Dell’s best efforts to not let you do so!