Slashdot Mirror
Intel's Pentium Chip Turns 20 Today
girlmad writes "Intel's Pentium processor was launched 20 years ago today, a move that led to the firm becoming the dominant supplier of computer chips across the globe. This article has some original iComp benchmark scores, rating the 66MHz Pentium at a heady 565, compared with 297 for the 66MHz 486DX2, which was the fastest chip available prior to the Pentium launch."
The rest of us made do with 60MHz versions.
It really had to hurt Intel to have to back down on clock speeds for once. They didn't do that again until NetBurst burst.
http://lkml.org/lkml/2005/8/20/95
fdiv bug
The 66MHz original Pentium. What a beast.
It ran on a full TTL +5V. So it sucked down power. Lots of power. I've disassembled first generation Pentium chips, removing the golden cover that protects the die beneath. The die is HUGE! Much bigger than any current production CPU.
In fact, the early models produced so much heat that we boggled at the big fans needed to cool them! It was one of the first Intel x86 chips that REQUIRED a fan for cooling. We used to run our 486DX2/66 and below fanless and they worked great.
All this for only less than twice the performance, at three times the cost.
The vast majority of us skipped the first generation Pentium, instead going for more affordable chips as the i486DX4/100 and the Am5x86/133, which was RIDICULOUSLY popular for several years! In fact, the latter was faster than a Pentium 75MHz for anything that didn't require the FPU. And not much needed the FPU back then.
Then of course we laughed our asses off when the FDIV flaw became known. Clearly the Pentium was the #0.9999999998855 processor on the market!
Ahh, memories.
This article has some original iComp benchmark scores, rating the 66MHz Pentium at a heady 565, compared with 297 for the 66MHz 486DX2, which was the fastest chip available prior to the Pentium launch.
I'm amazed by these scores. I remember having a fairly fast 486 DX4 @ ~100 MHz (probably by Cyrix or AMD perhaps) at the time the Pentiums started to become popular. I got the impression that a Pentium 66 or 75 would actually be a downgrade for me, but maybe that hadn't been the case.
I eventually switched when the Pentium Overdrive came out, so I could keep my 486 mainboard but still have a faster Pentium chip in my machine. That was a pretty sweet deal.
I can't believe this is all 20 years ago, it feels like only yesterday.
Pretty good is actually pretty bad.
Let's not forget the floating point speed improvements that led to Quake running like a beast on Pentiums, but like a dog on 486s.
In all seriousness, the Pentium wasn't the real tipping point imho. Intel didn't really get things perfected until the Pentium 2, if I recall.
From his royal Weirdness...
All About the Pentiums
Mess not in the affairs of dragons, for you are crunchy and good with ketchup.
Not sure I'm interested in such things. My 486SX 25 is just fine for playing Police Quest 2.
Overpowered, you mean. Those old games ran on 8086's...
I had a 486SX/25... overclocked to 33MHz!
I was a total badass. You can feel the badassery radiating from my body! Mwahahaha.
My dad had a 486 SX while I had a mere 386... but it was a DX running at 40 MHz, so it was actually better for playing Doom! Muahahaha.
Pretty good is actually pretty bad.
I had a machine with one of those Pentium-83 Overdrives that plugged into an SX-33's socket. Unfortunately this computer didn't have any secondary SRAM cache so it ran approximately like a DX2-66.
Back in the day, it took about 90 minutes to compile a 2.0.36 kernel, but it ran Duke3D and Descent well (dual boot).
Hail Eris, full of mischief...
E pluribus sanguinem
A 486SX 25? Are you made of money!? My Tandy 1000SX is just fine for playing Police Quest 2!
I remember seeing them but I can't track down the official release.
Well you wanted the EGA Graphics.
If something is so important that you feel the need to post it on the internet... It probably isn't that important.
I had a 486SX/25... overclocked to 33MHz!
I was a total badass. You can feel the badassery radiating from my body! Mwahahaha.
Well, there is SOMETHING radiating from your body but it sure isn't badassery. :P
No tributes should be paid to this abomination of micro architecture. The x86 ISA should have been shot dead and discarded of 20 years ago already. Porting code is more and more turning into a matter of just selecting a different architecture before you build. Will we ever move on? It's time to let go of your 30 year old corpses of x86 object code.
http://www.youtube.com/watch?feature=player_detailpage&v=jwg4mbGL4JE#t=36s
Ahh, the AMD 386DX 40, friend of mine back then introduced me to DOOM running properly on one of those, which he had at the time. His lawyer father presented him with a brand new Intel 486DX33 and he told me I could have his AMD 386DX 40 with a then mind blowing 16MB of RAM to take home with me and keep so he would have room to set up the new one. Had just finished re-installing DOS, Windows, DOOM etc and grabbed a backpack when he called saying he wanted his AMD386DX40 back. He brought it back to me after later getting 486 DX2 66 with VLB video. My son loved it when I handed it over to him.
That's way too fast for Police Quest 2. Hell, that's probably too fast for Wing Commander. But not fast enough for DOOM, at least for someone who is used to smooth frame rates on modern FPS. I find DOOM barely playable on my 486dx2/66 with 256k cache. Gets around 25fps.
Give me Classic Slashdot or give me death!
If you performed a calculation that took a week to complete on a modern Core i7 2600k, you'd still be waiting for your Pentium 1 to finish the same calculation even with a 20 year head start!
Source
I feel so old now :(
The post fails to mention the real reason that this line was significant: The Pentium and its successors were the first computer chips designed by the marketing department rather than the engineering department!
I put 8MB of 30-pin SIMMS into a (then old) 386-DX40 with no math coprocessor and got it to launch Quake under Linux.
It was unplayable, but didn't crash, so I launched a 2nd instance.
The OS churned away at about 30 seconds per frame - still no crashing and easy task switching to other shells.
Linux is solid as a rock.
There's a nice New Yorker podcast from a couple of years ago that discusses what went into picking the name: http://www.newyorker.com/online/2011/10/03/111003on_audio_colapinto . It was done by Lexicon Branding, who actually write code to break up words into phonems and then remix those sounds into new words. The program spits out lists of candidates that are then vetted by the linguists at Lexicon. I found it a really interesting discussion.
The PowerPC eats Pentiums for lunch.
Circumcision is child abuse.
If only he had realised he could've overclocked his 486 DX33 to 40Mhz ;)
i didn't realize that the Pentium processor is this old. I had a Pentium 4 that ran about 2.6 Ghz or something before I bought a new computer. The Pentium 4 chip ran hotter than my dual core processor too. The P4 ran at 140 degrees F. Yes, the CPU and case fan were running. lol Yeah, 140 degrees was hotter than my old IDE drive running at 110 degrees.
Oh yeah, the P4 had hyper threading and MMX. Fun stuff. Motherboard had an AGP x4 video card, no PCI express. I don't think it even had USB 2.0. interesting how different older computers are compared to the new dual and quad core and eight core computer systems.
never heard of a 486 until now. thanks for posting.
Back in the day, most engineers wish IBM had gone with motorollas 68000 rather than the address hobbled 8086 series. Oh how we hated paging 640k hell.
46137
Was the calculation done on a Pentium?
(Sorry, had to make the obvious joke.)
If 66 MHz was state of the art in 1993 and Moore's Law predicts a doubling of density (effectively clock rate) every 18 months, then 20 years/1.5 years = 13.3. So clock rates should be 2^13.3 times faster today, about 10,000x, or 66 MHz x 10,000 = 6.6 GHz, which is actually twice as fast as today's max of 3.2 GHz.
Given that 3.2 GHz Pentiums arrived on the market about 10 years ago (2002) but haven't moved since, it seems Moore's Law was clearly pessimistic for the Pentium's first 10 years and wildly optimistic for the last 10.
Of course the introduction of multicores Pentiums was a hack borne of desperation. The amount of speedup per app from a 2013 quad core Sandy Bridge @ 2.5 GHz over a 2002 P6 @ 3.2 GHz is... Ugh.
Has the speedup of the Pentium really gone negative? Has multicore-ing it caused the demerits of Amdahl's Law to finally overtake the merits of Moore's Law?
You've got it Amdahl's vs Moore and it has always been there. Density is roughly doubling as per Moore. Amdahl pretty much means we have a limited opportunity to find more work to parallelize so we need to find more different things to do. You won't be able to parallelize the heck out of everything if for no other reason than some logic is inherently serial and waiting on slower I/O. So They split off to multicore but still doesn't create more parallel work. At some point silly things like predictive branching or computing both sides of a branch ahead of time and then keeping the winner only gets you so far especially since in the first case you can be wrong sometimes, and the second you are guaranteed that you won't need one side of the calculation (so you are creating extra work just to be sure you have the result when needed and to avoid idle hardware).
I had one of those P24T Pentium overdrives. I got it at Best Buy. It had been returned a few times and every time someone returned it they cut the price more.I ran it at 100mhz. It did perform a lot better than a DX2-66 on Quake which was the big thing back then. For the most part it was like a P75 or better.
I had tried it in two different montherboards and in both cases it made the floppy fail to work, but otherwise it rocked. VLBUS all the way :)
Intel should have never made an 83mhz. Thing ran fine at 100mhz :)
If 66 MHz was state of the art in 1993 and Moore's Law predicts a doubling of density (effectively clock rate) every 18 months...
The problem with what you are saying is that doubling of density != doubling clock rate! nothing could be further from the truth. Moores law only pertains to the number of transistors that can be squeezed into a given space and does not at all speak to clock frequency. An i7 + gpu is over a one billion transistors vs about 3 million for a pentium 1.
I had a 486DX33. I was unintentionally badass for quite awhile because I had the AT-bus multiplier set wrong on my motherboard for a long, long time. I don't remember the exact figure, but I was running the clock on the IO channel about 33% faster. I didn't notice it until I eventually plugged in a card that wouldn't run faster than the design and figured out why.
the 486 dx4/120. I loved that processor.
Some people are only alive because it's against the law for me to hunt them down and kill them.
If AMD hadn't suffered management-inspired design failures or low-power hadn't displaced performance as a priority then Intel would be at 6.6GHz today. The Ivy Bridge can easily be run at far higher speeds, infact Intel used a duff heat disappation system in the chip to stop extreme overclocks. Haswell is just around the corner too.
In fact, clock rate and density have grown inseparable. Compare these plots of Clock speed (p. 61) with Moore's Law (p.67):
The two curves are doppelgangers.
Clock rates aren't a measurement of density or processing ability. Moore's 'law' is that the number of transistors that can be (and are being) stuffed onto chips doubles approximately every 18 months.
Persistent Volume manager for Kubernetes - https://github.com/dwimsey/openshift-pvmanager
66Mhz x 10,000. = 660,000Mhz = 660Ghz
Not that Moores law predicted clock speeds.
Yikes. Time for me to repeat 6th grade math.
As to whether Moore's Law fails to predict clock rate, I'm not so sure. In the first decade of the Pentium, ML accurately predicted clock rate within 2X. (Starting at 66 MHz, rising 2x each 1.5 years, that's 10/1.5 = 6.66, or 2^6.66, or 100x, or 66 MHz x 100x = 6.6 GHz, which is only 2x larger than life. That's pretty accurate, IMHO.
But Moore's Law fails to predict clock rates of the Pentium only during the past decade, where at 660 GHz, its estimate is 200x too large. Clearly the difference between those two decades reveals a sea change in the life of the Pentium (and CMOS in general).
Moore's law specifies transistor density, and clock speed is just a side effect. Also, a Core i7 probably outperforms whatever a Pentium 6.6gHz would be.
I too had a 486 SX/25mhz CPU as my 1st PC I ever bought (after I traded in a comic book collection I had built since 1972) in 1991. However - I didn't consider it 'suffering' though - I considered it LEARNING to deal with PC's @ the hardware level in fact, & enough so I was employed doing it in a year or two later, simply because of what I ended up doing in upgrading my own rig, + learning doing it:
---
1.) It had only 64kb of cache on the motherboard that I upgraded to the 256kb limit possible on that motherboard, & to a FASTER type too (from 15ns to 10ns iirc).
(Which was doable IF you could stand the price of the "Tag SRAMs" that is (& IF you didn't mind pulling up those chips + replacing them)).
2.) I ended up upgrading that system to a Dx/4 133mhz CPU too, but to do so, I also had to change the tiny 'clock' oscillator also (25mhz -> 33mhz) doing some soldering for that to work out right.
3.) The system also originally only had 4mb 30-pin "fastpage" RAM on it - & that ended up being boosted to 32mb of the same type, just more of it (faster internal clockings on the newer RAM too).
4.) That rig also ended up with a TekRam DC-600 (iirc) 16mb Caching ISA RAID 1-0 controller too (on dual Western Digital 212-424mb IDE disks @ 5,400rpm iirc) where it had none before that, just a "Super I/O" card that controlled floppies, CD-Rom, & IDE disks. I used the DC-600T Vesa Local Bus model (VLB) on rigs later that featured that faster bus circuit too (good stuff - caching controllers help speed up the slowest things - hard disks, & on both reads/writes!).
5.) Combined in the end with a Diamond Stealth 1mb "VRAM" 24 "Windows Accelerator" videocard (coming up from a 256k stock SVGA videocard).
---
* "Old Chevy's NEVER die - they just get FASTER..."
(For its day, it was a FAST system @ the end of all that - probably fast as its type, ISA bus, could be in fact...)
Those upgrades ended up making it VERY "competitive" for things vs. Pentium 60 class newer rigs (with VLB/Vesa Local Bus too on them) vs. that old ISA system (packed with tons of caching RAM types on it + upped from stock clocks on CPU & better video by far)...
APK
P.S.=> I even patterned my current home system after that one's design (& I have ever since in fact on every machine SINCE that original one which I still have no less), by using:
---
A.) Fast CPU's (MultiCore Intel Core I7 920)
B.) Fast video (NVidia GeForce 470 GTX)
C.) Fast disks (Dual WD 10,000rpm "Velociraptors" with 16mb cache on them)
D.) The disks are combined with a Promise Ex-8350 PCI-Express 128mb ECC Ram Caching RAID controller
E.) A Gigabyte IRAM PCI Express "True SSD" based on DDR2 RAM for storing:
Pagefile.sys
%temp% ops
%comspec%
Logging @ BOTH the OS + application level
Webbrowser caching & browsers on it
(For fastest possible seek/access/load-store + to offload the slower mechanical disks of those tasks & more...)
This latter parts in SSD application ends up offloading the MAIN "C" drive used to be done on separate HardDisks other than C, now I do that on SSD's since they're quicker thus, 'speeding up' the slower HDD's by them NOT having to do those tasks (reducing fragmentation on them too, bonus, which slows them even moreso)!
---
I kept that type of design thru ALL of my future systems, where applicable that is!
(E.G.-> For years, I did without SOME parts: Such as the case of when I couldn't FIND a caching disk controller during the PCI 2.2 bus era, & for a LONG time in fact, + I haven't always used SSD's of non-FLASH design either - 1st was in 2002 iirc with a CENATEK RocketDrive PCI unit with PC-133 SDRAM - before that, I used software ramdisks to compensate that much @ least...).
I design s
I ember I had got a 386sx33 in 1993 and used that for all the programming until 2004 when I upgraded to k6 amd
Barely playable on your 486? You probably had an ISA video card! VESA baby!
Nope, I'm using a VLB Diamond Stealth 64 DRAM T. I'm just spoiled by getting 60FPS on my modern machine.
Give me Classic Slashdot or give me death!
That was I think due to ISA bus. Had to buy a PCI card as I noticed Doom was being held down by my Trident. I bought a Diamond Stealth64 for ~200$.