The Pentium OverDrive is notable for supporting three physically different sockets (that’s not counting the Pentium II OverDrive). The Pentium OverDrive product was sold in variants suitable for Socket 2/3 (486 systems), Socket 4 (early Pentium systems), and Socket 5/7 (second generation Pentium systems).
Pentium OverDrive for 486 Systems
Perhaps the best known member of the OverDrive family is the original Pentium OverDrive, also known as P24T. It could upgrade most newer 486 systems with Socket 2 or Socket 3.
The front of a Pentium OverDrive box.
Two models were available, PODP5V63 and PODP5V83. The nomenclature indicates 5V Pentium OverDrive Processor running at 63 (really 62.5) or 83 (really 83.3) MHz. The processors utilize an unusual (for 486 boards) 2.5× frequency multiplier and are designed for 25 and 33 MHz bus, respectively.
Since the processor core is a 3.3V second generation Pentium (P54C), the OverDrive has a built-in voltage regulator. Because the package dissipates a non-negligible amount of heat, the OverDrive not only has a heatsink like the 486 OverDrives but also comes with a clip-on fan. The advantage of the fan is that it draws power from the socket and does not require any extra wiring. The fan is also easily replaceable in case of mechanical failure.
Pentium OverDrive aka PODP5V83
The installation manual is still available from Intel (note the curious spelling of “processer”).
The socket was a new 238-pin affair. It was essentially an extension of the standard 168-pin PGA socket with an extra row of pins on the outside. The OverDrive socket was very similar in size to Socket 5/7 but with a different pin arrangement.
OverDrive Socket 2
The Pentium OverDrive had fairly predictable problems with compatibility. OEMs were expected to design boards compatible with the Pentium OverDrive without ever having seen one. The upshot was that older boards that would benefit most from an upgrade were the least likely to work, while newer boards worked well with a Pentium OverDrive but also supported other processors with similar performance (DX4, 5×86).
When the Pentium OverDrive was introduced in January 1995, the 63 MHz variant (initially the only one available) was priced at $449 according to contemporary sources. That did not compare very favorably with the $349 100 MHz IntelDX4 OverDrive, which was generally a faster processor.
Internals
The Pentium OverDrive is one of the more technically interesting processors since it bridges two different generations. That presented Intel with a number of challenges.
The voltage and heat dissipation were relatively minor problems that a voltage regulator and a fan easily fixed. A bigger problem was that the Pentium had a 64-bit data bus which no 486 board had. In addition, the L2 caches on 486 boards are much slower than those found on typical Pentium P54C systems. Intel’s solution was to increase the Pentium OverDrive L1 cache size from 8+8K (separate code + data) to 16+16K.
The larger cache equalizes the difference somewhat, although a PODP5V83 still doesn’t perform anywhere near a 90 MHz Pentium.
The cache can operate in write-back (WB) mode, if supported by the chipset/board. Newer 486 boards generally do since many latter-day 486s could run in WB mode as well (while all earlier 486 models were strictly write-through or WT).
To software, the PODP looks like a Pentium although it can be easily distinguished through CPUID, being a model 5 rather than model 4 processor (or having no CPUID like earlier 486s). It also reports itself as an OverDrive processor through the CPUID ‘Type’ field. The Type/Model/Family/Stepping (TMFS) value of an OverDrive with S-Spec SU014 is 1532h.
The PODP supports neither the local APIC nor MCE (Machine Check Exception) features of the P54C because these features are not available in Socket 2/3 systems (not supported by the socket itself and therefore unsupportably by the CPU).
A little-publicized fact is that the PODP is in reality not a superscalar processor because the second pipe (V-pipe) is permanently disabled. This can be seen by attempting to force single-pipe execution (U-pipe only) by setting bit 2 of TR12 (MSR 0Eh). The PODP does not react to the setting, while disabling cache or the BTB through TR12 does have noticeable impact—and likewise, disabling the V-pipe through TR12 on a regular Pentium visibly slows things down.
Performance
An 83 MHz Pentium OverDrive vastly outperforms a 33 MHz 486 and easily bests a 66 MHz DX2. But when comparing with a 100 MHz DX4, the PODP isn’t generally faster… except for floating-point performance.
It is thus unsurprising that an AMD 5×86 (133 MHz) or a 100 MHz Cyrix 5×86 beats a Pentium OverDrive in integer performance. However, the PODP is king when it comes to floating-point performance, much like RapidCAD was in a 386 socket.
In general the 83 MHz Pentium OverDrive performs a lot like a 60 or 66 MHz first generation Pentium, but much depends on the chipset and board used.
The relatively poor performance is presumably caused by inadequate memory bandwidth coupled with the loss of the second execution pipe.
Pentium OverDrive for 5V Pentium Systems
Not well known, the PODP5V (or Pentium OverDrive processor 5V) was used to upgrade first-generation 60 and 66 MHz Pentium systems. These OverDrives were somewhat rare, which is not surprising since the original 5V Pentium systems were never very widespread due to their high price and reliability problems.
Pentium OverDrive family (PODP5V133 bottom right)
The PODP5V is essentially a clock doubler. It upgrades a 66 MHz Pentium to a 133 MHz Pentium. As such, it brings a sizable performance kick.
The 5V OverDrive uses the second generation Pentium core (P54C), just like a regular 133 MHz Pentium. It comes with a now-obligatory voltage regulator since the newer core runs at 3.3V.
A notable but unsurprising difference between the 133 MHz PODP5V and a standard P54C 133 MHz Pentium is that the former lacks the built-in APIC, or more likely has it permanently disabled. This is a consequence of the fact that the original 60/66 MHz Pentium processors had no integrated APIC and Socket 4 cannot support it.
Given that it’s nearly identical, a 133 MHz PODP5V performs more or less exactly like 133 MHz P54C. However, a Socket 4 board is almost certain to have an older and slower chipset and cache, therefore overall system performance cannot be expected to match a 133 MHz P54C system.
Pentium OverDrive for 3.3V Pentium Systems
The PODP3V is the consequence of a design oversight. Early P54C systems ran with 50/60/66 MHz bus and supported only 1.5× and 2× multipliers. The 100 MHz Pentium CPU was normally run with 66 MHz bus and 1.5× multiplier but could also be used with 50 MHz bus and 2× multiplier. This design limited the internal processor speeds to 133 MHz (2× 66 MHz).
Pentium OverDrive voltage regulators (left, right), and filtering (center)
That’s where the PODP3V comes in. This is an upgrade processor with a fixed 2.5× multiplier for Socket 5 systems. It does not have a voltage regulator because it doesn’t need one, but it does add voltage filtering. It was available at speeds up to 166 MHz.
The PODP3V was not nearly as popular as earlier 486 OverDrives. One apparent reason is that while 486s OverDrives doubled (and later tripled) the clock speed, the PODP3V couldn’t do that (typically upgraded from 100 to 166 MHz or from 90 to 150 MHz).
It is worth noting that at least the 166 MHz PODP3V166 model (S-Spec SU125) is not distinguishable from a regular 166 MHz Pentium by software. Unlike most other OverDrives, this processor does not set the OverDrive type in CPUID.
The 3V Pentium OverDrive processors have a built-in local APIC (just like regular P54C CPUs) and are capable of dual-processor operation.
Pentium OverDrive MMX
The final member of the Socket 5 (and Socket 7) OverDrive family is the Pentium OverDrive MMX or PODPMT. It was available in several variants with a fixed 2.5× or 3× multiplier. The fastest model runs with a 66 MHz bus and 3× multiplier to achieve 200 MHz internal clock speed.
PODP models with fans removed
Just like a regular Pentium MMX (P55C) processor, the OverDrive MMX chip has larger 32 KB (16 + 16) L1 cache (compared to 16 KB total in earlier models). And just like regular Pentium MMX, the PODPMT has a built-in APIC and can be used in dual-processor systems.
Unlike the PODP3V, the PODPMT does report itself as an OverDrive part and is recognized as such by BIOS. Model PODP66X200 (S-Spec SL2RM) reports TFMS value of 1543h, indicating an OverDrive CPU family 5, model 4, stepping 3.
The Socket 7 processors saw the voltage games (later repeated for Socket 370) where the socket remained mechanically the same but processor voltage kept decreasing. Pentium MMX required split voltage, but a given Socket 7 board might or might not be able to supply it. The PODPMT adds the required voltage regulation and as a side effect can also be used in Socket 5 systems, since the onboard voltage regulator takes care of the biggest difference between Socket 5 and Socket 7.
There are clear parallels between PODPMT and the IntelDX4 OverDrive. In both cases the processor core is more or less completely unmodified newer model, but the processor package includes a built-in voltage regulator and heatsink (and in the Pentium OverDrive case, also a fan).
Summary
The Pentium OverDrive product line enjoyed moderate success, especially early on. But it ran into what would soon kill off the OverDrive entirely: Too little, too late, too expensive. Part of the problem was that the x86 platform started evolving faster and a mainstream socket design wouldn’t last 5+ years anymore. A 4-5 year old PC couldn’t be effectively upgraded with an OverDrive CPU, and while a 2-3 year system could, the cost/benefit ratio wasn’t attractive.
OverDrives pin side
The original Pentium OverDrive was an excellent upgrade for a 25 or 33 MHz 486, but not for a 100 MHz 486 (let alone the 133 MHz AMD parts). The PODP5V133 was a solid upgrade (doubling the clock speed) but Socket 4 systems were never widespread. The PODPMT would have been again somewhat interesting for older Pentium systems, but couldn’t keep up with the Pentium II processors available at more or less the same time. For those reasons, the Pentium OverDrive processors were never as widespread as the 486 OverDrive CPUs.
But there was still one last OverDrive processor to come…
Nice reading, so this series will continue with Intel’s Pentium II Overdrive for Pentium Pro systems?
Yes, that’s the plan 🙂 I just need to dig up a Socket 8 board and run a few tests on the PII OverDrive before I can call the final part in the series done.
PODP3V Overdrives do not have voltage regulator.
All socket 5 Pentiums have the same voltage up to Pentium 200.
You need VRM only when you want to use Pentium MMX in socket 5.
Afaik the only reason of existing non-MMX Socket 5 Overdrives is multiplier setting.
Early Pentiums (75-100) have only 1.5x and motherboards of that era
might not able to set correct one via CPU pins BF0,BF1.
They are hardvired on Socket 5 Overdrives and do not depend on motherboard.
Funny trivia from “AMD-K5 Processor Technical Reference Manual”:
“On a locked and aligned CMPXCHG8B operation, the AMD-K5
processor performs two 32-bit locked reads followed by two 32-
bit locked writes, all with SCYC asserted. The Pentium processor
combines these 32-bit reads and writes into one 64-bit read
and one 64-bit write for the quadword-aligned case. “
Which also reminds me that the Enhanced Am486 and thus AMD 5×86 don’t have VME or PSE.
I think you’re right about that. Socket 7 specified higher wattage, 3.3V CLK and PICCLK pins, dual voltage for MMX CPUs, required additional space for cooling, and added a key pin. The thing about the BF0/BF1 pins is that original Socket 5 processors only had a single BF pin for 1.5 or 2.0 multiplier. And for the initial batch of P54C processors, the 2.0 multiplier was only used for the 100 MHz part running with 50 MHz bus. Old Socket 5 boards definitely did not have more than one jumper selecting between 1.5/2.0 multiplier. I’ll update the article accordingly.
This might explain why my PODP3V does not even report itself as an OverDrive… it’s not different enough.
They had to be lucrative enough at the time for Intel to bother at all. Similar to how Nvidia started a trend of >$1K graphic cards, you wont sell a lot of those, but apparently you will sell enough.
It’s well known that the margin on Intel Extremely Expensive CPUs and Xeons is insane. And obviously that’s where Intel makes its money. So as long as such parts do not require too much engineering, they’re terrific… for the vendor at least.
I think the OverDrives were expensive enough that Intel made money on them, but they did need extra engineering. The biggest problem was no doubt the fact that the OverDrive targeted customers who did not want to shell out a lot of bucks for a top-notch system.
What is fun is that I wonder why Conroe was not designed to be compatible with older LGA775 mobos.
I think once Intel cornered the chipset market, there was no upside (for them) in letting customers keep using existing boards. I don’t know how much of a concern that was in the Pentium OverDrive times when Intel sold a lot of chipsets but there were numerous alternatives as well.
Seeing this reminded me of something I saw years ago…
… and yep, it’s still there at http://www.redhill.net.au/c/c-6.html – something of an “evil twin” (way more snark, way less technical info) of this article.
I prefer this far deeper and more intelligent post, though. Thanks, Michal. 🙂
Indeed, I think a price drop in Core 2 Duo processors back in 2007 came with the requirement to move to a new 1333Mhz FSB and thus new chipsets (you could still buy the 1066Mhz FSB processors but only at the old prices).
But back in 2006, while the new features of the 965 chipset was useful for example for 64-bit users or IGP users, clearly not everyone needed them.
Designing a chip to work with an older socket and chipset is hard. It wasn’t just Intel failing with that; AM2 motherboards ability to be upgraded to use Phenom was rather hit or miss. Best not to bother and change chipsets and sockets every few CPU releases to save on customer service expenses.
The boxed retail price for an Overdrive tended to match the street price for the slightly faster tray box processor: i.e. the 83MHz Overdrive was introduced at $299 while the budget retailers offered the Pentium 90MHz at $299. The fast 486s were better values but yields were lacking for the 133 MHz variation. In 1995, one could buy a cheap 486-100 or an expensive super fast Pentium but between the two was either overpriced slow Pentiums or getting on a waiting list for AMD 133MHz. Not a good time.
Except in this case they didn’t change the socket or even the chipset, the only thing they changed was the VRM.
Yep, it’s hard, because it requires predicting the future. Tough business to be in 🙂 It worked well for 486s (and PODP3V I assume) when only the internal clock multiplier changed, but with the retrofits it was a lot more difficult. Some of the blame surely was on the board and BIOS vendors too, but again designing for hardware you’ve never seen is difficult.
As for speeds, it’s tricky comparing the Socket 2/3 OverDrives with their Socket 5 brethren. The disabled V-pipe made a difference, and the 486 boards tended to have much slower L2 cache and main memory. Then again the PODP5V had double the L1 cache. The 83 MHz OverDrive performs more like a 66 or at best 75 MHz Pentium in my experience, not really close to 90 MHz.
I remember the pricing of that era very well. A 90 MHz Pentium was great, but very expensive. A 75 MHz Pentium was barely better than a fast 486 (if at all) but cost a lot more. Consequently many customers went for the almost-as-good but much cheaper option.
Slower but cheaper: Overdrives had the retail markup baked in and tray CPUs don’t.
One thing I never saw anyone check was what voltages the Overdrive ran at internally. Running at a lower clock is a good way to palm off some bum silicon but it looks easy to also have the Overdrive operate at a slightly higher voltage and thereby scavenge some marginal chips.
AMD even labeled the 5×86 as “P75”.
Don’t forget the cache interposer and the associated class action lawsuit, BTW.
Atleast on some early Pentium 3,3V motherboards you could simply solder in a wire to ground the pin that selects the multiplier that’s not available on a jumper. Remeber trying that to run a P166 on a HP Vectra XU 5/90 dual cpu Pentium socket 5 computer (officially upgradeable to 133MHz IIRC).
P.S. the HP Vectra XU 5/90 is a strange beast. I haven’t investigated why but I’ve never got Linux to boot on any such machine. Probably some BIOS issue.
So basically a homemade 3.3V Pentium OverDrive 🙂
One benchmark back then was “can it play 128k MP3 without hiccups”… the OverDrive 83 MHz and the P66 did it (as long as nothing else was running, not even a screen saver), but the DX4-100 had a hard time (“please don’t move the mouse!”). Otherwise they felt very similar indeed.
As for the longevity of sockets: Slot 1 came close to Socket 2/3. I bought my Asus P2B-B mainboard in 1999 (with a PII-350 for the 100MHz bus clock). 3 years later, it got a Celeron (Coppermine) 1 GHz upgrade for tripled performance.
Those 440BX boards supported 233 to 1100 MHz (but the 1,1 GHz CPUs were rare and expensive). Don’t think that the designers had anticipated that…
re Slot 1: My impression is that the voltage regulator on the motherboard sometimes didn’t like the slotket adapters for using newer processors, resulting in sadness when trying to upgrade (I.E. an attempt at running a faster processor might even kill the motherboard).
That was a thing, yes… I was lucky to have bought this particular Asus board, and they still supported it years later with expanded compatibility lists. So I ordered the Asus S370 adapter board – and then I also ordered a new BIOS chip (after bricking the original one; Flash Update wasn’t necessary stable back then).
Funny: As Intel also produced Slot 1 versions of the 100 MHz Pentium III Coppermines, they surely _could_ have repackaged them and sold as PIII OverDrive back then… but they were fully committed to the P4 at that time; and those GHz Slot 1 chips always were expensive (I remember seeing those only in servers), so they would have sold even less than the Pentium OverDrives. Lesson learned 🙂
The PIII-S CPUs were very nice, up to 1.4 GHz. Back in the day Intel sold them for “internet servers” because P4-based 1U servers were just not gonna happen, especially not dual socket ones (which many of the PIII-based ones were).
Right. It’s sad that consumer mainboard support for the Tualatins was limited at best (everybody wanted to sell you a P4 room heater or an Athlon back then).
I have some of them (mostly in Slot 1 format, which just makes it more complicated – my S1 boards don’t even feature 133 MHz), pulled from old servers, but can’t do much aside from putting them on a shelf. Surely won’t start to collect servers 🙂
Going off on a tangent, I wonder why there never were half width rack servers? Compare with pro audio equipment, studio/musician stuff, where half width is a common thing and afaik you use brackets to join two devices to fit one 19″ slot. That would likely had solved the P4 server thing, I.E. 2U but half width, and thus two servers would take the same space as full width 1U servers.
> One benchmark back then was “can it play 128k MP3 without hiccups”… the OverDrive 83 MHz and the P66 did it (as long as nothing else was running, not even a screen saver), but the DX4-100 had a hard time (“please don’t move the mouse!”).
Was windows scheduler really that bad?
I remember that 100dx4 was the first cpu to play 128 stereo cbr mp3’s without hiccups, and you could even do light work, like reading text or editing small files. No X11 or compiling, of course.
This was on freebsd 2.2.7-8, with mpg123 player.
@Vlad:
Re windows scheduler: The classic ridiculous thing was how screen savers especially in Windows 3.x but IIRC maybe also in Win 9x got “hiccups” 1-2 times per second…
Windows 3.x did not even have what one would call a scheduler. Completely cooperative multitasking. So yes, it really was that bad.
Well, TIL. I didn’t know that fraunhofer released player for win3.x, after all, the mp3 format became popular in 96 – 97.
It took a while for Windows 95 to take over the consumer market.
As for Windows 3.x, I would think that a MP3 player might had ran kernel code to be able to run more seamless than otherwise, perhaps?
> It took a while for Windows 95 to take over the consumer market.
Well, it’s actually reasonable to have player for win3x. The first mp3 encoder was available in 1993, the first player was released in sep’95, the format started to become popular in the middle of 1996 when pirates ripped audio CDs in it, so it was not an instant success, it just looked like one.
> As for Windows 3.x, I would think that a MP3 player might had ran kernel code to be able to run more seamless than otherwise, perhaps?
No, winplay3 was used (pretty slow) l3audio.dll: http://old-dos.ru/dl.php?id=14700
Hmm, I’m quite sure I used an early WinAmp (2.2?) or sMP3P on Windows 95 on those machines. Both have a timestamp of 1999 on my server, so that fits (my collection of “old junk” really grew in this timeframe) – I think I wanted a (background or DJ) music server to replace the FM radio or CD player in our local youth club.
Otherwise there was DosAmp 0.8, but that lacked the comfort. As for 3.x… no, that was out of fashion.