The End Is Near
By now we are pretty much at the end of the sweet days of overclocking. The upcoming Intel chips Katmai and the socketed Celeron will not allow any form of overclocking anymore and it would be no surprise if even the well known Pentium II and Celeron chips we know now, will soon also be equipped with Intel’s new overclocking protection. AMD seems to consider the same, so that there will be hardly any new chip available that will allow us to tweak its performance by running it faster than it is supposed to.
In the last three years rumors have been many. You could read many times that Intel would finally put an end to overclocking, but so far nothing much has happened. Intel’s old strategy of locking the multipliers of CPUs has become pretty common these days, but many of us were still well able to run our CPU a lot faster than specified. Now Intel has made reality of what many people foretold and feared already months and years ago. The upcoming CPUs are of course multiplier locked as we already know it, but now there is also a circuit included into the processor that checks the bus speed or FSB clock. If this clock exceeds the specified limit, the CPU simply refuses to run. That will put an end to people running their CPUs for 66 MHz FSB at 75, 83, 95, 100 or even 112 MHz, a 100 MHz FSB CPU will not do 112, 124 or even 133 MHz anymore as well.
A Beetle with a 12 Cylinders Engine
Now why is that so irritating? After all everyone gets the CPU he paid for and if someone wants a faster CPU he will have to pay more. Well, it’s not quite as easy as that.
Imagine you are buying a car. You don’t want to spend that much money, so you go for a small engine. This engine could e.g. a 1.8 liter engine with 4 cylinders and maybe 120 horsepowers. You will have a pretty hard time getting 400 bhp out of this engine unless you get a specialist putting in turbo chargers or compressors, costing you a lot of money.
Your friend spent more money for his car and he got a cool 7.3 liter V12 engine with 580 bhp. This engine is a lot bigger and a lot more expensive to build, so it’s pretty much fair enough that the engine costs a lot more than your 4-cylinders engine. It’s a lot different with Intel CPUs though. If we want to continue with the car example, you would have to imagine that your cheap car would also have the 7.3 liter V12 in it, but the manufacturer was changing the engine so that it produces only 120 bhp, although it could do more. In the past you could open the bonnet and fiddle around with this engine a little, so that your cheap car would finally run as fast as the expensive car of your friend. Why would the car maker do that? Well, for him (Intel) it may be cheaper producing 7.3 liter V12s only, rather than manufacturing a wide spread of different engines. He only includes a few tricks and can adjust this big powerful engine to any kind of performance he likes. He is making good money selling this engine in a cheap car already, but he is making really great money if someone should be so crazy paying double or triple for the same engine with nothing else but a different label on it.
The most recent proof that the above comparison is almost correct, is Intel’s Pentium II 300 processor with the S-spec ‘SL2W8’. Those CPUs carry an almost complete 7.3 liter V12, but are sold as 1.8 liter 4 cylinders engine. In other words, these PII 300 CPUs are almost perfect PII 450 CPUs. Don’t worry that Intel is losing money here, as a matter of fact they just enjoy people paying more than double for the ‘real’ PII 450, because that will increase their quarterly earnings considerably.
This article is meant to advise you on how to avoid paying too much for the so-called ‘real McCoy’. The times of overclocking will be over soon and you may want to grab one of those bargains before it’s too late.
A little history of Intel’s Pentium II
Before we get into the details of the SL2W8, we should first recap the history of the PII 300.
- Die Size
A few months after the first launch of Intel’s Pentium II as 233 and 266 MHz version in Q2 1997, Intel released the PII 300. At this time the Pentium II was manufactured using the .35-micron technology. The code name of the Pentium II chip from this time was ‘Klamath’. At the beginning of 1998, Intel introduced the 333 MHz PII, followed later by the 350, 400 and 450 MHz version. The 333 MHz PII and up was not using the older .35-micron Klamath core anymore, but the new ‘Deschutes’ core, a chip produced using the .25-micron technology. The ‘Deschutes’ core is not much different to Klamath in terms of its functionality, both chips are Pentium II chips and perform equally. The difference is subtler and is also not simply defined by the core. Klamath has a larger die size, due to the .35-micron technology, which you can see by having a look at the chips. - Klamath
PII 233 – 300
- Deschutes
PII 333 – 400
PII 450
- Core Voltage
The next difference between the chips is the voltage they are running at, the Klamath is using 2.2 V, the Deschutes runs at only 2 V. The die size and the voltage difference are both responsible for the difference in heat production. Klamath becomes a lot hotter than Deschutes, making overclocking of Klamath a lot harder than overclocking Deschutes. - Tag RAM Chip 82459
That is not the only difference between older and new PII CPUs, there is also the L2 cache controller. As you all know, the Pentium II has got a L2 cache within its cartridge, running at half the processor clock. This L2 cache needs a L2 cache controller, because Intel did not include one into the Pentium II core. One of the important things about L2 cache controllers is its cacheable area. The combined L2 cache controller / tag RAM chip used together with Klamath chips on PII 233 – 300 MHz used to be the ‘82459AB’. This chip could only cache up to 512 MB of memory, so that those Pentium II CPUs could not run with more memory than 512 MB. The first Pentium II 333 CPUs came with the ‘82459AC’ L2 cache controller/tag RAM chip, which can run at higher clocks, but it is still only able to cache 512 MB. Later the PII 333 and all faster PII CPUs (350,400,450 MHz) were equipped with the ‘82459AD’, which is not only able to keep up with up to 225 MHz L2 clock speed, but it can finally cache up to 4 GB of memory. Intel’s internal name for the 82459AD is ‘tp6e’. - L2 Cache
The third difference between the PII CPUs is the L2 cache that is used. Klamath (PII 233 – 300) used to come with four L2 cache chips rated at 6.5 or 7 ns, thus good enough for L2 cache clocks of 140 MHz (7 ns) or 150 MHz (6.5 ns). The PII 333 is normally using two 5.5 ns L2 cache chips, rated for up to 180 MHz, the 350 and 400 MHz versions come with two 5 ns chips, good for 200 MHz and finally the PII 450 is equipped with 4.4 ns chips, rated for 225 MHz. This difference if of major importance, since the L2 cache is a sensitive part of the Pentium II. Running it way beyond spec will most likely cause L2 cache errors and thus system crashes, even if the CPU core is doing fine at this speed. - Microcode and Intitialization
Finally the last major difference is the microcode update and the CPU initialization that’s done each time you boot your system. Different chip revisions have different microcode updates (the code that is loaded into a special CPU-area to fix minor CPU bugs) and, more importantly, different chip revisions have different initialization routines at each different CPU clock. When the system is booting, there are several undocumented CPU ‘registers’ read by the BIOS, which then sets the L2 cache timings and other things according to the information read from the CPU. This routine cannot be cheated by anyone except the motherboard maker, because nobody knows what information is read and how to react to it.
So far about the basics.
The Pentium II 300 SL2W8
I already mentioned in the introduction, the ‘SL2W8’ is almost a Pentium II 450. Why do I say that?
Well, first of all, not each SL2W8 is equipped the same. In the past Intel used an S-spec to clearly define a CPU core and its equipment, now it only seems to be a ‘name’ for a family of CPUs, which are not all quite the same. What can be stated about SL2W8 however, is that all CPUs with that S-spec have a ‘Deschutes’ core, thus producing a lot less heat than older PII 300 CPUs and also running at only 2 V. The second fact is that the SL2W8 comes with the ‘tp6e’, the fast 4 GB L2 cache controller / tag RAM.
Those two facts are already quite encouraging for overclockers, because we know that Deschutes is willing to run up to at least 400 MHz in most cases, as well known from the old Celeron chips. You certainly remember, the Celeron without L2 cache was nothing but a pure Deschutes chip, running up to 400 or even 450 MHz when overclocked. The L2 cache controller that’s even used inside the PII 450 is certainly a part that will run up to 450 MHz and having a cacheable area of 4 GB instead of only 512 MB as in the old PII 300 CPUs is certainly a nice thing too.
The touchiest thing in terms of overclocking a Pentium II is the L2 cache though. Here the SL2W8 varies unfortunately. Some SL2W8 CPUs come with 5.5 ns L2 cache, some with 5 ns, but there is a whole lot that is equipped with 4.4 ns L2 cache modules!!! You can imagine, the SL2W8 chips with 4.4 ns L2 cache are more or less supposed to run at 450 MHz or even more. The ones with only 5 or 5.5 ns are still equipped with a faster cache than what is needed for 300 MHz (6.5 ns), so that you can certainly overclock them, however not all to 450 MHz reliably.
Before you get all worked up about this, please realize that even a SL2W8 with 4.4 ns L2 cache is not a real PII 450. The PII 450 comes with a new packaging of the core chip, called ‘OLGA’ or ‘flip chip’. The PCB of the PII 450 is also different from the PCB of a ‘normal’ Deschutes’ PCB, as found in PII 333 – 400.
Above you can see the PII 400 with its typical core packaging and the PCB with the L2 cache modules left and right of the CPU core. This is what SL2W8 looks like as well. Underneath you can see the ‘real’ PII 450, with the flipchip packaging and both L2 cache modules on the right of the CPU core. The flipchip packaging ensures even better heat conduction than the old packaging above.
Another important difference is the aluminum plate that covers the board on the side you are looking at on the above picture and that’s supposed to conduct the heat between the chips and the heat sink. Intel calls that plate T-plate, and there are three different types available.
That’s what the cartridges look from the outside, looking at the T-plates. The upper T-plate is simply plane and connects only to the CPU core on its other side. This T-plate is used for PII 233 – 333. The T-plate in the middle is for PII 350 and 400. It has two notches, which connect to the L2 cache modules on the other side. The T-plate below belongs to PII 450, touching the two L2 cache modules on the right side of the CPU core. As being an official PII 300, SL2W8 is unfortunately equipped with the T-plate at the very top, thus it’s not cooling the L2 cache modules. This is of course an issue if you want to run it at 450 MHz or more, because your heat sink doesn’t cool the L2 cache, as done in every PII 350 – 450.
Whilst there isn’t a work around for the flip chip issue, you can certainly do something about the T-plate. You have to open up the cartridge though, and then place an aluminum plate of the right thickness between the L2 cache chips and the T-plate, fixing it with some thermal compound. You could also use some coins, but the only ones I found with the right size are coins of 20 Greek Drachmas, 10 French Francs or 50 Hong Kong Dollars and you need more thermal compound to fill the gaps in the coins.
How to Make Sure that you Get a SL2W8 with 4.4 ns L2 Cache?
So far about the theory, but how can you find out if a SL2W8 has 4.4 ns L2 cache? On the quest for finding a definite answer I’ve tried a lot of different things. The result is that there is no reliable way of telling the L2 cache inside the cartridge unless you open it up. If you should already have a SL2W8 and don’t fear opening the cartridge, then you will have to look for either a ‘-44’ (Samsung, Micron) or a ‘-225’ (Toshiba) on the L2 cache chips. ‘-5’ or ‘-200’ means 5 ns L2 cache, ‘-55’ or ‘-180’ stand for 5.5 ns types. Trying to run Andreas Stiller’s ‘ctp2info’, a program that gives you a lot of information about your Pentium II and its L2 cache, doesn’t produce any results that could help you find out the L2 cache type, due to the initializing routines performed by the BIOS. The BIOS can tell a right PII 450 from an overclocked PII 300 SL2W8, even if the L2 cache chips and the L2 cache controller are the same, and it initializes a PII 450 completely different to an overclocked SL2W8. So does a PII 450 give you a L2 cache latency of ‘8’ and a latency mode ’02’, whilst a SL2W8 at 450 MHZ with the very same external chips gives you a latency of ‘3’ and a latency mode of ’00’.
When I was shopping around for SL2W8 I was very lucky. I got 9 CPUs, all of them equipped with 225 MHz (= 4.4 ns) Toshiba L2 cache modules.
They were all produced in Costa Rica, 40th week 1998.
Now there are several people all over the Internet speculating how to make sure that you get a SL2W8 with 4.4 ns L2 cache. Many say that the CPU has to come from Costa Rica, but the week is important too. I will not join this discussion, simply because there is hardly any certainty about it. What I can say however, is that it is extremely likely, that SL2W8 CPUs from Costa Rica, built in week 40, 1998, will have Toshiba 225 MHz = 4.4 ns L2 cache.
Overclocking of the SL2W8
Now let’s get down to it. First of all please realize that since mid 1998 all Intel CPUs are multiplier locked to only the one multiplier they are supposed to run at. This means that a 300 MHz PII comes with a x4.5 multiplier, 66 MHz x 4.5 = 300 MHz. Overclocking this CPU means simply increasing the bus speed, or FSB clock. Luckily, the SL2W8, particularly the one with 4.4 ns L2 cache, runs fine at 100 MHz FSB, making it run at 450 MHz instead of 300 MHz. If you want to try exceeding this you can use the harmless 103 MHz setting, which runs in almost as many cases as the 100 MHZ FSB, or you can try 112 MHz, so that the SL2W8 reaches 504 MHz. I strongly advise against any further increase of the bus clock, as e.g. 124 or 133 MHz, simply because hardly any AGP card will let you run any 3D application without crashing. You also require excellent PC 100 SDRAM with a CAS latency of only 2 ns to achieve that.
I have got 9 SL2W8 CPUs, all from Costa Rica, 40th week of 1998, all equipped with 4.4 ns L2 cache, but not all were the same. I can put them in 5 different categories:
Serial Number |
Comments |
08400304-0237, 08400304-240 |
Runs at 504 MHz in almost any BX-board that supports 112 MHz FSB |
08400304-0239 |
Runs at 504 MHz in many BX-boards with 112 MHz FSB-support |
08400304-0238 |
Runs at 504 MHZ in a few BX-boards with 112 MHz FSB-support |
08400304-0369, 08400304-0370, 08400304-0377, 08400304-0378 |
Runs at 450 MHz in any BX-board, runs 504 MHz only when voltage raised to 2.2 V, except BH6 runs all w/o voltage change |
08400304-0376 |
Runs at 450 MHz in most BX-boards, 504 only in BH6 at 2.2 V |
You can see that there are serious differences even though the CPUs are all equipped with the same 4.4 ns L2-cache chips and even though they are have consecutive serial numbers. The 0237 and 0240 are the best chips, the 0376 is the worst of them all.
The result is pretty encouraging, at 450 MHz almost each CPU ran in each board, only 504 MHz are somewhat difficult to reach. The only board that ran all of the CPUs at 504 MHz was Abit’s BH6. The reason for that is pretty simple, it lets you raise the core voltage to more than the 2 V default core voltage. Abit’s BX6 wasn’t as successful as its younger brother, the 0376 CPU failed even at 2.3 V core voltage, which looks as if the BX6 isn’t quite as stable as the BH6. However, the BH6 does still not run with the Samsung 6ns 256 MB registered DIMMs.
The Core Voltage Issue – Try It, Don’t Fry It!
In overclocker circles it’s the only real thing, Abit’s BH6. The reason for this is certainly its stability, but the main success of the BH6 is based on its ‘SoftMenu II’, giving you the chance of changing the core voltage of your CPU.
Now how is this possible and why don’t other board members do it? Since the days of the Pentium Pro, Intel thought that it would make sense if the motherboard could adjust the core voltage for a CPU automatically. After starting the system, the motherboard checks a few pins of the CPU and detects the voltage, so that nobody would have to adjust it with jumpers, dip switches or in the BIOS setup menu. This is a very convenient as well as safe solution, you don’t take the risk of frying your CPU for any reason. Intel does not want this to change, so most board vendors comply and stick to the specifications. It’s rumored that Abit had to make a ‘trade’ with Intel, they promised not to make any Super 7 boards so that Intel still supplies them with chipsets although they ship SoftMenu II.
It’s also well known for a long while though, that raising the core voltage of a CPU can improve its stability at clock speeds that exceed its specification. This trick has often been used with Pentium and later Pentium MMX CPUs, and you can still be successful with K6 or K6-2 CPUs. The only problem is that once you exceed the specified range of a CPU, you can just as well damage it. Raising the voltage does also not necessarily have to lead to success. Due to the higher voltage, a CPU gets hotter, because the power dissipation increases just the same. Thus it can reverse the beneficial effect of the higher voltage and your CPU does still not run at the higher speed. When raising the voltage you need to cool your CPU as good as in any way possible. This cooling issue is misunderstood by most of the overclockers out there. They buy fans and heat sinks of crazy dimensions, but finally they close their computer case and nothing has improved. It does not help placing huge heat sinks and fans on the CPU unless the heat can get out of the case. You will get a much better effect when leaving your heat sink as it is, but placing a large (and hopefully quiet) fan the way that it blows over your CPU and then OUTSIDE THE CASE. This cools a million times better than having a monster cooler on the CPU in a closed case.
The Deschutes core is supposed to run at 2 V, so this is what SL2W8 gets by default. Intel’s specifications give you a range of +/- 0.1 V. Thus 2.2 V are already way over the limit. Please consider that 2.2 V may sound little, but compared to 2 V it’s an increase of 10%. The Pentium usually ran at 3.3-3.5 V and I guess you wouldn’t have felt that great, running the Pentium at 3.6-3.8 V. So be careful when raising the voltage to 2.2 V or even higher.
Changing the Core Voltage on Any Slot1-Motherboard
Now what do all the people do who don’t have, want, like the Abit BH6? Well, if you want my honest opinion, I’d recommend them to run their CPU at 450 MHz and be happy. If this should not comply to you, there’s still no reason to trash your current board and go buy the BH6. You can change the voltage to at least 2.2 V with a little trick. The voltage detection of the motherboard works via 5 CPU pins, which are either connected to Vss or to nothing. Now covering a CPU pin with tape, as discussed in my B21 article, simulates a ‘not connected’ or ‘open’. Unfortunately you cannot simulate a ‘connected to Vss’ or ‘closed’ condition, which impairs you a little bit and makes the solution in the BH6 still more flexible.
First the theory. The 5 pins used for the voltage detection are pins VID0 – VID4, ‘VID’ standing for ‘Voltage Identification’. The following table shows where to find the pins on the Slot1-connector:
Pin Description |
Pin # |
VID0 |
B120 |
VID1 |
A120 |
VID2 |
A119 |
VID3 |
B119 |
VID4 |
A121 |
The next table shows you how the pins have to be set up for voltages between 2 and 2.3 V.
VID4 |
VID3 |
VID2 |
VID1 |
VID0 |
Core Voltage |
connected |
connected |
connected |
connected |
not connected |
2.00 V |
connected |
connected |
connected |
connected |
connected |
2.05 V |
not connected |
not connected |
not connected |
not connected |
not connected |
No CPU plugged in, BIOS error message |
not connected |
not connected |
not connected |
not connected |
connected |
2.10 V |
not connected |
not connected |
not connected |
connected |
not connected |
2.20 V |
not connected |
not connected |
not connected |
connected |
connected |
2.30 V |
As you can see, the default 2 V setting that is supplied by the CPU has already got VID0 ‘not connected’, which cannot be changed by taping the pin. This is why you cannot reach 2.1 V or 2.3 V, both need VID0 ‘connected’. However, you can achieve 2.2 V, by covering the pins for VID2 – VID4 = A119, B119, A121.
I do not recommend increasing the voltage over 2.2 V, the next possible step for a non-Abit-board-owner would be 2.4 V, which is definitely way over the limit. Please also keep in mind that this operation here is certainly only for people who know what they are doing. I do not take any responsibility for damaged CPUs or motherboards. As a matter of fact, sticking to the default voltage and running a PII 300 at 450 MHz is already cool enough.
Motherboards for Overclockers
So far I’ve tested 20 boards, but I’m still testing. I have not tested with increased voltage on any motherboard except the Abit ones. I found out that there are two ways of making you system fail unless it’s absolutely stable. I have got a recording of me playing my favorite DM64-level (sewer64) on my favorite server (which is down now, bummer!) for 24 minutes, actually recorded by my clan brother [F3]Eike (thanks buddy!). Running this demo for the whole length is excellent to cause crashes, it heats up the CPU as well as the 3D card by a serious amount. The additional test I ran was Winstone99. It loves crashing in the Microsoft Office part when the system isn’t stable. Both tests were ran twice, over at least 2h.
Here is what I got:
Motherboard |
Number of SL2W8 CPUs running at 450 MHz |
Number of SL2W8 CPUs running at 504 MHz |
Comments |
Abit BH6 |
all |
all |
excellent stability, at 504 MHz the voltage had to be raised to 2.2 V only for one CPU (0376), memory problem with 256 MB DIMM still not sorted out with latest BIOS, but LG memory works now |
Abit BX6 |
all |
8 |
Board would not work at 504 MHz with CPU 0376, even when voltage was increased to 2.3 V. Four CPUs needed 2.2 V to run at 504 MHz. |
Aopen AX6B |
8 |
none |
|
Aopen AX6B Plus |
8 |
none |
|
Asus P2B |
all |
2 |
|
Asus P2B-DS |
all |
3 |
|
Asus P2B-LS |
all |
4 |
|
DFI P2XBL |
all |
2 |
|
DFI P2XBL/S |
all |
none |
|
ECS P6BX-A+ |
all |
4 |
B21 procedure required, memory problems sorted out with new BIOS |
EpoX EP-61 BXA-M |
all |
2 |
|
FIC VB-601 |
all |
4 |
|
Gigabyte GA-6BXC |
all |
4 |
|
QDI Brilliant I |
all |
4 |
B21 procedure required |
Soltek SL-67B |
all |
3 |
B21 procedure required |
Soyo SY-6BA |
all |
none |
|
Soyo SY-6BA+ |
all |
2 |
Soyo Combo Setup, even latest BIOS (Oct 26, 1998) still has memory problem, only half the memory modules get recognized |
Soyo SY-6BE |
all |
3 |
B21 procedure required |
Superpower SP-P2BXA |
all |
none |
B21 procedure required |
Tekram P6B40-4AX |
all |
2 |
B21 procedure required |
As you can see, the Abit BH6 is by far the most stable board, and I can understand that everyone is recommending it for overclockers and the use with SL2W8. I only wish it wouldn’t have any other problems. Abit’s BX6 cannot quite compete to BH6, although it comes with the same ability to change the CPU core voltage in the BIOS. The other great overclocker boards are Asus P2B-LS, Elitegroup P6BX-A+, FIC VB-601, Gigabyte GA-6BXC and QDI Brilliant I. Those boards ran 4 of 9 SL2W8 CPUs at 504 MHz without any core voltage change. I will add more boards as I go along, there are still about 20 to come.
Are there any Other S-specs Good for Overclocking?
Yes there are. All of those CPUs have definitely got a Deschutes core and the ‘tp6e’ L2 cache controller / tag RAM chip. I cannot say which L2 cache chips they have, but it’s most likely less than the old 7 ns.
S-spec |
Comments |
SL2W8 |
PII 300, multiplier locked to x4.5, OEM CPU |
SL2YK |
PII 300, multiplier locked to x4.5, Intel boxed CPU |
SL2W7 |
PII 266, multiplier locked to x4, OEM CPU |
SL33D |
PII 266, multiplier locked to x4, Intel boxed CPU |
SL2VY |
PII 266, multiplier locked to x4, Intel boxed CPU (??) |
Some people also recommend PII 333 for overclocking. I would advise against getting a PII 333 for that purpose. It comes with a x5 multiplier and would have to run at 500 MHz with a 100 MHz FSB. This is a very touchy speed, so that you are taking a very big risk of being unsuccessful.
Tom’s 2 cents
A long time ago I was one of the people who started the overclocking scene on the Internet. At this time it was something special and people stayed within reason.
Today overclocking seems to become a topic like what you can listen to in male changing rooms. Who’s got the longest? Well, it’s, ‘who overclocked highest?’.
This may be a crazy sport for some, other people are scared off however. I would like to send the message to the more cautious ones of you, who know what they are doing.
The SL2W8 is a great bargain. You get a PII 450 for the price of a PII 300. This may be the last chance that this is possible. Katmai will close the door for overclockers, the overclocking scene will mostly be dead by Fall 1999. Grab one of those bargains while you can and stay within reason.