Detailed How-To – Adrian Allen
This is my first attempt at writing an article for Overclockers.com, though I’ve been a reader since the K6 days (Slot A, we hardly knew ye!)
I have a decent desktop system, consisting of an AthlonXP Barton 2600+ running at 2.4 Ghz with a gig of DDR400, a WD Raptor 36 GB, and my trusty Ti4200 128 MB.
But for the last two years I’ve been spending a ton of time on the road, so in June I got (as an early birthday present) this Hyperdata laptop:
Specs are as follows:
- AMD Athlon64 3000+ DTR @ 1.8 Ghz, 1 MB L2, 800 Mhz HTT
- 512 MB DDR400
- Hitachi Travelstar 7200 RPM 8 MB cache
- ATI M10 (Mobile Radeon 9600) Pro-128 MB
The laptop is basically the Mitac 8355, which I ordered as a barebones from Xoticpc.com and then filled it with my own RAM and HDD.
Performance is pretty decent – UT2K4 runs very nicely at 1024×768 with anisotropic filtering and antialiasing, Doom3 runs at High quality, 1024×768 with no AA or AF at a framerate I can put up with to get the DX9 eye candy like heat blur. I haven’t tried HL2 yet, but since it seems to like ATI hardware, I’m optimistic.
I primarily use this laptop for gaming as I tend to have significant downtime when I’m on the road, whether while actually traveling or between appointments etc., so anything I can do to enhance this performance is beneficial.
I primarily play Desert Combat, and since LCD’s don’t really like running at non-native resolutions, I want as high a resolution as possible. This works out to 1024×768, 16x AF, 2xFSAA for acceptable framerates in online games.
I also run [email protected] (team 32, of course!) pretty much all the time when I’m not gaming, and the laptop is essentially on 24/7 when I’m on the road as I use it for a stereo as well.
Anyway, the upshot is, anything I can do for improved performance, I’ll do. I tried overclocking the graphics card with ATITool, figuring that would be the biggest bottleneck.
My results were AWFUL. The stock speeds on the M10 in this notebook are 350 Mhz core, 190/380DDR RAM. The core was not stable over about 375, the RAM was not stable over even 200! It basically wouldn’t overclock at all. At the best overclocked settings, I got a .7% improvement in 3DMark 2005. Woohoo. Pardon me while I crack open some Dom Perignon.
The CPU was similarly unimpressive – it wouldn’t overclock past about 1850 Mhz, also presumably due to heat issues.
Anyway, I gave up on increasing speed for a bit and looked for other ways to tweak it. I found RMClock (RightMark CPU Clock Utility for AMD64) and found that I could under-volt the CPU very significantly with no performance loss. Eventually I had it running at full speed (1.8 Ghz) at only 1.175 volts (from 1.50), which made it a lot nicer to live with.
The battery life improved to over 4 hours in desktop uses and over 2 hours while gaming, and the system fan was nearly inaudible even while running [email protected] for long periods – it never got hot enough to really spin up, and 100% CPU at 1.8 Ghz was nearly as quiet as idling in Windows at 800 Mhz @0.8v. This was nice, but a few days ago I decided it was time to do something about the poor overclocking, and decided to disembowel my innocent laptop for corrective surgery.
This is essentially a complete teardown of a Mitac 8355, with replacement of the cheesy TIM between CPU and HSF combo, as well as fabricating and adding RAMsinks to the GPU RAM.
This will MOST DEFINITELY void your warranty; don’t even THINK about doing it unless you’re clinically insane and can prove it in court.
If you set your laptop or yourself on fire or cause the end of the world by following the example I’m about to set, neither I nor Overclockers.com will admit we know you.
Here’s the disassembly. I’ve noted as much detail as seems necessary at each point throughout. Note also that I do not have access to any sort of manual or teardown booklet, so I’m basically making this up as I go along – Mitac and Hyperdata have both been equally unhelpful when I tried to find out how to open the thing up. They were SHOCKED that I wanted to install my own RAM rather than pay some schmo at CompUSA $100 to do it, so I was left to my own devices.
Here’s the underside of the laptop:
I found out the first time I took this apart that virtually every screw in it is of a different thread pitch, length, and diameter. I don’t know if this was a deliberate attempt to make it impossible for a home user to service, but the possibility of putting a long screw in a short socket and puncturing the motherboard or something didn’t appeal to me, so I took a piece of paper and marked out where each screw was, then poked a hole in the paper and stuck each screw in its corresponding hole as I removed them to avoid any confusion:
I began by removing the cover of the CPU compartment:
You can see that the heatsink seems that it should actually be pretty good – three heatpipes set in a large copper radiator! I didn’t remove the HSF combo yet. You’ll get a better look at it later.
Now we remove the hard drive. First, remove the plastic cover in the bottom center of the above picture. This only takes one screw, so I figured most people could probably figure out removing it without a separate photo (actually, I lost the picture, but it sounds better my way).
Anyway, once you’ve got the plastic cover off, you’ll need to remove several small screws holding two metal covers in place over the hard drive. First, there is a small retainer bracket held in place with an exceedingly tiny screw (I had to use the tip of a broken X-acto blade to turn it), as well as two larger ones.
Remove the bottom three screws indicated by the arrows in the picture below, then lift out the retainer bracket. The tiny screw is the center one; the two on the sides aren’t hard to deal with. Go ahead and remove the top four at this point as well, as they’ll all need to come out to remove the bracket. Once the screws are out, lift the bracket straight out toward you – it should pull right out with no force needed:
After that, lift off the cover plate to expose the HDD itself:
Now we need to remove the hard drive. Pull it straight back toward you to disengage the IDE interface pins:
Then just lift it straight out.
If you want to do a HDD swap, this is about all you’d need to do (other than removing the CPU cover, obviously). I only removed the HDD to ensure there wasn’t a screw under it that I’d need to access. I also figured that if I fried something, the HDD would probably have a better chance of survival if it were outside the computer and not connected to anything.
Ok, now we flip it over and see about removing everything else. Here’s the top of the laptop:
In order to access any of the screws on the top side, you have to slide the power switch/speaker cover off to the left. Just apply gentle pressure to the LEFT and it will slide off that way about half an inch, and then you can lift it straight up off the laptop:
This will expose the screws to remove the keyboard, as well as the terrible two-watt (no joke) speakers that Mitac stuck in this thing, the worst part of the whole laptop:
The arrows in the above picture indicate the two screws that you need to remove in order to lift off the keyboard. Remove those two, then lift the keyboard up a bit so you can see the little cable that connects it to the mainboard:
To remove the keyboard, you must gently push it in the direction of the arrows above. Apply pressure to both sides at the same time, pushing up toward the back of the laptop and it will pop loose, moving about 3 mm or so. The connector will open up, and then you can slide the cable out – it’s just a bare ribbon cable with no plug on it:
You can also see the square depression which contacts the Radeon 9600 chip to cool it. That sheet of aluminum is all there is to the cooling system for the video chip. No big surprise it doesn’t OC very well, I guess.
Now lift the keyboard out of the unit and set it aside. This is what you’ll see:
There are two foil covers which protect the RAM and a mini-PCI slot from shorting to the keyboard. Fold them back to see what you can see:
You can now see the open mini-PCI slot where I could have added a built-in 802.11x card (external cards get better signal, so I’m using a Belkin 802.11g USB unit with a USB extension cord and a parabolic dish-shaped spaghetti strainer that gets me +20db signal strength…but that’s another article), as well as the antenna for said card, and the single open RAM slot.
If you want to do a RAM upgrade or add a mini-PCI device, all you have to do is pull the keyboard and these slots are accessible. You can also see the black insulator plastic on the inside of the foil sheets which prevents shorts. It seems similar to the anti-static bags a lot of computer components come in.
In order to continue disassembly, we have to remove the LCD, as there are two screws which cannot be accessed without taking it off. Here’s a rear view:
First, remove the two screws I’ve marked in the shot above – these keep the screen from moving and work as setscrews. Then, to remove the LCD, we have to remove those covers over the hinges to access the primary retention screws. The covers just snap in place, but you need something very thin to fit in the seam and pop them loose. Just slide the blade into the seam (CAREFULLY. There are cables in there!) and pry sideways lightly. It will pop right off:
This exposes some of the screws we need to remove:
They’re in the same spot on both sides. On the left side (when facing the rear of the laptop) there is a piece of metallic tape that I assume functions as a ground strap since it’s attached to the chassis of the laptop as well as being screwed down onto the mounting bracket of the LCD. For this reason, I suggest being careful when removing the tape and ensuring that you reattach it in the same fashion during reassembly. Remove the two black screws. You can’t remove the screen yet, though – there are two more screws that come all the way up from the bottom:
Remove those two, and now the LCD will be loose. You still have to remove the two cables that connect it to the body of the laptop, though. Both cables fit into slots in the casing that you’ll have to slide them out of after disconnecting them, or they’ll snag and prevent you from removing the LCD.
The one on the left side when facing the front of the laptop is for the three LED’s in the base of the LCD’s frame. Just pull straight up with gentle pressure and it should come out easily. Next you’ll need to remove the data cable, which is on the right side, and held down with the same section of tape which I think doubles as a conductive ground strap. Peel the tape back carefully, then use a flathead screwdriver to gently work the plug out of the socket. Bring it up a few mm at a time, first one side, then the other, until it’s loose:
Now that the connectors are unplugged, pull the cables gently out of their recesses in the body of the computer. You’ll also need to peel out the 802.11x antenna that goes up into the LCD frame in order to remove the screen. Once you’ve got the cables loose, just pull the LCD straight up and it will slide out.
Now you can see the two screws that are the reason for all this trouble, as marked above. If you can figure out a way to remove these without removing the LCD, be my guest – I couldn’t. Anyway, remove these two screws, then flip over the laptop and remove all the screws you can find. The only one you don’t need to remove is a small one that’s inside the battery compartment.
Other than that, as I recall, they all have to come out. You MUST, MUST ensure that you keep track of which screw goes where or you will not be able to get the machine back together properly. At the same time you remove the screws holding the bottom of the case on, you will need to remove the four spring-loaded retention screws that secure the HSF combo, as these go through the motherboard and attach to the top half of the case. Make sure you unplug the fan first:
With the screws removed, the HSF combo will lift right out:
Take a look at that crummy foil TIM joint! When I was peeling it off, it seemed to consist of a layer of thin foil with a thick waxy thermal compound under it. No thermal compound actually contacted the CPU.
Anyway, now that you’ve got all the screws out and you’ve removed the HSF, pop the two halves of the case apart. You’ll have to be careful doing this as the front audio plugs and volume control wheel, as well as the card reader slot, protrude into the upper half of the front of the case and you’ll have to apply a bit of pressure to them to let the top half come off without snagging on them and possibly damaging them or even breaking them off the board entirely. I’m pretty brave, but I’m not keen on the idea of resoldering components to my laptop board. You should now be presented with this view:
This is probably the most delicate part of disassembly. The connector for the touchpad is very fragile. You must be GENTLE when loosening it. I actually cracked the connector, but fortunately it was still in good enough shape that it retains the touchpad cable.
I was trying to open it in the same manner as the keyboard cable, but this applied too much stress. What you want to do is push it lightly toward the rear of the laptop while, at the same time, pulling up a bit and trying to rotate the far edge of the connector toward you. Below is a shot of the connector AFTER removal of the cable, for clarity:
Once you’ve loosed the connector, the cable will slide out in the same way the keyboard cable did. Now you can completely separate the halves of the case – you will see this:
You can see that there is no cooling at all on the graphics RAM. The Radeon chip is cooled by a rather THICK thermal pad that connects it with the piece of aluminum sheet metal pictured earlier. There is a small amount of airflow across this section of the board due to a small fan near the back of the case which pulls air across both halves of the motherboard and the voltage regulators, so it’s not entirely passively cooled, but it might as well be.
This essentially completes the disassembly. I did not unscrew the motherboard from the case itself. I only realized AFTER reassembling everything that since there were only three RAM chips visible, there were quite likely more on the other side of the motherboard for the Radeon. I suppose I’ll get to those later. :/
Now, to the modding!
First, let’s get that terrible TIM off of the HSF combo. I peeled off the foil component with my fingernail, revealing the waxy thermal compound beneath it:
After this, I used acetone-free nail polish remover and a cotton ball to remove the remainder of the stuff from the heatsink. The CPU itself was clean, since it had never touched anything but the foil, but I cleaned it with some alcohol just to be safe.
I considered lapping the copper slug that contacts the CPU core, but it would have been very difficult, since the cast aluminum standoffs you can see surrounding the slug are all higher than it is and would have interfered with lapping to a considerable degree. I decided to leave it alone for now and perhaps come back to it later if needed.
Now to the Radeon and its memory.
I didn’t want to remove the thick TIM joint that was attached to the aluminum sheet, as it was so thick that I was concerned that if I removed it, there would be no contact between the GPU and the aluminum. For lack of anything better to do, and with no replacement thermal pad, I put some Arctic Silver on the surface of the GPU to maintain decent contact between the GPU and the TIM compound, hopefully making up for any degradation in the contact due to removing the pad and replacing it (thermal pads are generally only supposed to be used once).
I toyed with the idea of melting some aluminum and pouring it into the depression in the aluminum sheet where it contacts the GPU to at least give it a bit more mass to distribute the heat through, but didn’t want to possibly deform the contact surface. Another project for another day. I decided the best I could do at the moment was to add some heatsinks to the graphics memory. I found an old low-profile heatsink from a Stone Age computer (P-90, if I recall correctly) and measured out three heatsinks for the RAM.
Although this is a low-profile unit, the fins were still too high. I eyeballed the space available and used my Dremel tool to cut about half the height off of the fins, leaving perhaps 3 mm of height and about 1.5 mm of thickness. The baseplate is perhaps 2 mm thick. Not much, but certainly better than nothing. Here is the result after cutting the section and using a wire brush to remove the black anodized finish:
This is obviously before cutting into thirds. After sectioning it into thirds, I used a grinding wheel and the wire brush to deburr all the edges and remove any bits of loose metal that might have fallen off inside the case. Then I used 150 grit aluminum oxide sandpaper to smooth the bottom of my new RAMsinks. I worked my way down to 220 grit and then used some rubbing compound (1500 grit) to take out the last bits of ridges.
When looking at the right-hand RAM chip, I saw that a piece of plastic film was covering part of the chip and would interfere with the fit of the RAMsink:
I used my pen knife to carefully saw around the edges of the RAM, cutting through the plastic film and then grabbed it with some needle nose pliers and removed it:
After this I used thermal adhesive to attach the newly minted RAMsinks:
They all have no less than 0.5 mm clearance to any electronic component around them, and are solidly attached.
Ok, now to reassembly.
No surprises, it’s basically just doing everything we just did all over again in reverse order. Don’t forget to put Arctic Silver on your CPU HSF combo. The only thing that’s kind of a pain is reinserting the cable for the touchpad, since it tries to pull itself out of the socket as you’re putting it in. You’ll have to have the laptop on your…well…lap. Then use one hand to hold the cable flat and slide it into the connector while you hold the connector open (GENTLY!) with a fingernail. Once the cable bottoms out in the slot, just press the connector down and it will snap into place (or possibly in half, you know, whatever).
Once I had it back together, I immediately fired up ATITool to see if I’d made any headway.
The short answer is, no, not at all.
I saw absolutely no improvement in the graphics card overclock whatsoever. The RAM may actually have gotten a touch worse, but it was so bad before that it’s really hard to tell. In any case, it didn’t improve, so I assume one of the following:
- My RAM is already running at its maximum rated speed, and simply has no headroom
- There are three more RAM chips on the other side of the board that also need cooling
- There is insufficient airflow to make the RAMsinks effective
- God hates me
So, one of those is probably it. I’ll tear the thing down again when I start having performance issues and actually NEED the extra speed and see if there are, in fact, other chips in there on the other side of the board to cool. If so, I’ll post an update.
Now, “What about the CPU?”, you ask? Well, THAT is a whole ‘nother story.
Using ClockGen (a wonderful utility for tweaking your laptop which I thoroughly recommend), I’ve been working my way up slowly and carefully the whole time I’ve been writing this, and SO FAR I’m up to 2.0 Ghz @ 222 Mhz (888QDR) HTT bus with RAM @ 182 (364DDR)! This is probably the biggest improvement I’ve ever gotten by simply tweaking an existing cooling system.
The voltage is at 1.325 not because I found that I NEEDED it, but I figured 1.175 was just asking for trouble. When I hit a ceiling (I haven’t yet, I’m taking it slow) I’ll try increasing the voltage and see if that improves things. Once I find the absolute max for this setup, I’ll find out what the lowest voltage it’ll work at is, and probably leave it that way.
Even right now, the fan isn’t even running at full speed – I can barely hear it. The air coming out of the exhaust is cooler than when running at stock speed at 1.50 volts, so I’m pretty darned happy, even if I didn’t accomplish my original goal.
For stability testing while overclocking the CPU, I used the [email protected] textmode client running 100% CPU utilization in the background. The CPU itself remains stable up to something over 2.0 Ghz – I haven’t continued to push it past this point, because once I hit 2.0 Ghz, I wanted to see how the 3D performance was affected, and I found that although Windows was perfectly stable at this speed (even still at only 1.325v), the AGP subsystem is unhappy at 222 (888) HTT bus speeds.
I don’t believe ClockGen is correctly reporting the PCI and AGP speeds – as far as I know the K8T800 chipset doesn’t have an AGP/PCI lock, so at 222HTT bus, the AGP subsystem is running at 74 Mhz, which is obviously well out of spec. I believe the M10 chip is AGP8X (anyone?), and generally the higher the AGP mode you’re using, the less tolerant it is of overclocking.
Eventually, with some tinkering, I found that 1910 Mhz was the highest speed that the system was completely stable at in all operating modes. I then ran some before/after benchmarks to see what I gained. I ran 3DMark 2001SE, 3DMark 2003, 3DMark 2005, and PCMark 2004. I ran PCMark at both 1.9 Ghz and 2.0 Ghz, since it ran fine at 2 Ghz, presumably because it doesn’t make much use of the AGP subsystem.
First, I ran 3DMark 2001SE in a few different modes. I ran it with everything at stock speeds to begin with, 1.8 Ghz CPU, 350 GPU, 189 graphics RAM. This produced a score of 9673. Then I ran it with just the CPU overclocked and got 9734. Since I had increased the speed of the CPU by about 6% and got only a (slightly less than) 1% improvement, it shows that the CPU is definitely not the bottleneck in this system. That is, for every 1% increase in CPU speed, I can only expect a 0.2% increase in 3D performance.
I then returned the CPU to stock speed and set the graphics card to 360/193 (maximum stable settings). This got 9901. I then ran it with both the CPU and graphics card overclocked to their maximum stable settings (1910 Mhz CPU, 360 Mhz GPU, 193 Mhz graphics RAM) and got 9961, which is consistent – the CPU contributed about the same gain percentage wise. You can see that the performance scales perfectly linearly with increase in the graphics subsystem – that is, increase GPU/GRAM speed by 1% and get about a 1% increase in performance. The CPU still only gains about .2% per 1% increase in speed.
In 3DMark 2003, the scores were 2614 “stock” and 2700 with both CPU and AGP system overclocked to the same speeds as above. This shows that the performance increase of about 3% seen in 3DMark 2001 holds true here as well.
In 3DMark 2005, the scores were 1082 “stock” and 1119 at max stable OC – so basically the same results. 3% OC gives 3% increase in performance, or thereabouts.
In PCMark 2004 the initial score was 3285. The 1.9 Ghz score was 3425, which is a bit less than expected (should’ve been about 3432). The 2.0 Ghz score was 3601, where I predicted 3580, so the scaling seems to improve a touch in this arena. The results were repeatable within a few points when I ran it a couple more times to ensure this wasn’t a fluke.
For some reason I didn’t think to take screenshots of the scores, but it would’ve just taken up a lot more space and been pretty boring anyway.
Here’s a table of the scores:
3425 @ 1.9 Ghz 3601 @ 2.0 Ghz
3384 @ 1.9 Ghz 3482 @ 2.0 Ghz
I ran 3DMark 2001SE four times with different settings to establish whether there was a “synergy” effect, where the overclocked CPU + overclocked GPU gained more than they would separately. Since this turned out not to be the case, I only ran the other two 3DMark programs and PCMark in two modes – “stock” speeds and with both the CPU and GPU overclocked.
To see how I established this, look at the row of 3DMark 2001SE scores. Overclocking only the CPU resulted in a gain of 61 3DMarks. Overclocking only the GPU resulted in a gain of 228 3DMarks. Given this, you’d expect that the total gain with BOTH overclocked should be 9962 (ie 9673 + 61 + 228 = 9962). The result I got was 9961, which demonstrated that there was no “cooperative” effect.
All the remaining benchmark scores were within a fraction of a percent of where I predicted they would be, given the expected results from the runs of 3DMark 2001, with the exception of the PCMark scores, which scaled better than expected due to increased HTT bus bandwidth and higher RAM speed which do not affect the 3D scores as much, since the CPU is waiting for the GPU a lot more than the GPU is waiting for the CPU.
It should be noted that I’ve been able to reach much higher speeds with the GPU and graphics RAM, but they’re not stable by any stretch. I’ve had the graphics core up to about 410 and 225 DDR on the RAM, but after about two or three minutes of Desert Combat, it freezes. I could probably get through a benchmark at the max speeds if I let everything cool off for a while first, but I don’t consider those “real” scores as the system won’t work at that speed for extended periods. If it’s not stable FOREVER, I don’t think of it as a “real” overclock.
A good overclock should be exactly like a stock system in terms of stability, only faster. So far, I can’t exceed 360 core / 193 mem and fit that criterion. For now, the system is staying at a very mild overclock of 1853 Mhz, 205.9 HTT bus/168DDR RAM. There appears to be a slight inaccuracy in the divisor for RAM speed, and at the stock CPU speed, RAM was only running at 162 Mhz, when it should have been at 166. So, I’m leaving it bumped up a touch.
I’ll continue tinkering with it, and when I eventually figure out a way to keep it cool enough to work at higher speeds I’ll send in an updated article.
Until then, I hope you enjoyed reading this as much as I enjoyed writing it. Happy overclocking!
Credit to Dylan Bennett for fixing my atrocious Photoshop work, and to Joe Wain for help with proofreading.