Intel's abandoned Pentium 5 project...bought on eBay! (with info from Intel engineer)
Vložit
- čas přidán 3. 06. 2024
- Patreon: / fullybuffered
I shed light on Intel's cancelled project which aimed to take Netburst to even more extreme heights - called Tejas and Jayhawk. After obtaining engineering samples years ago, I have been searching for more information about this project. Just recently I managed to get in touch with a former Intel engineer who worked on the projects and who was willing to disclose information never published before!
My Twitter: FullyBuffered
0:00 - Intro
0:45 - Frequency was king
1:27 - Enter Netburst
1:42 - Pipelining even deeper
2:57 - Issues with deep pipelines
3:31 - Enter Tejas and Jayhawk
4:58 - Jayhawk chips surface 14 years later
5:43 - Contact with Intel engineer
6:08 - Q1: Recalling Tejas and his involvement
7:45 - Q2: What was it like within Intel during that time?
8:49 - Q3: Are the samples real?
9:31 - Q4: How far did development get?
10:00 - Testing the Jayhawk samples
11:09 - Q5: Would it have been the Pentium 5?
11:37 - Q6: How did the people react who worked on it?
12:01 - Q7: Were any changes made afterwards?
12:54 - Q8: Was anything reusable?
13:34 - Q9: Where did the names come from?
14:05 - Conclusion - Věda a technologie
I think you should look for bios modders to help you mod your motherboard's bios to get them running. They probably need a microcode to be running (now finding the right microcode might be tricky, maybe it can be found on the intel leak that happened few years ago ? Or maybe it can be reconstructed? I'm not sure what's supposed to be in those so maybe it's not that simple)
from what i can gather, QBGC was a thermal sample, which means the chips just flat out will not work, sadly
While not impossible, I would say it unfortunately very unlikely that the microcode ever made it outside of Intel labs..
i offer the option that all test chips of a certain generation should share the same id codes. I have no evidence for this other than ease of testing. If you had to put each id for each cpu it might pose some problems. If all the test chips are the same id , then that id is input once and covers all Qualifying samples and one code for engineering samples. all speculation.
Perhaps there are some "removed from intel bins" motherboards, where that bios is still attached. Just find it! 🤣
his cpus are not real dude
The greatest CPU that Crysis never had.
Crysis is easy to run. My system crushes the game. Never understood the meme.
@@helium5912 the 2007 original build relied heavily on single thread calculations.
@@helium5912 that's because you're doing it 15 years late -_-
@@helium5912 do it on a 2007 system and you'll understand.
@@helium5912 Zoom Zoom
Absolutely fascinating. I worked in IT during the turn of the millennium and I was a hardcore P3 fanatic - when my friends dumped all their money into the P4 I doubled-down and bought a dual-P3 system. It ended up being my daily driver well into the XP era.
Thanks for putting this together!
Thanks you! I'd love a dual P3 system - that would have been quite a system back in the day.
That was actually a really intelligent double down if you ask me.
Did you ever test SMP Quake III Arena?
@@nexxusty asking the questions that matter!
@@nexxusty
Mickey Mouse bro. It was designed to exploit the original ultra weak logical cores.
@@eclipsegst9419 I ALWAYS wanted to test Q3A with Dual CPU's of that era.
I stupidly threw out a Dell Dual Tualatin board that used RDRAM in 2017. I still feel like an idiot for doing that.
I remember Tejas and its demise, but never thought there are actual chips in the wild. Honestly, if they dont work, I would risk a de-lidding to see if its actual silicon, a die shot would be awesome. Maybe someone like FritzchensFritz or der8auer could help? In any case an amazing video, thank you for your in-depth research!
Second this, but for delidding please contact der8auer as he can delid it non-destructively and has access to industrial level electronic microscopes to take die shots.
Thanks for the recommendations! I am looking at the options for delidding and photography
Yeah +1 for contact de8auer, he loves this sort of stuff and has the expertise, connections, and equipment to potentially make them work.
@@FullyBuffered Maybe you can get in touch with Roman Hartung aka der8auer from Germany. I'm sure he'll be interested in this kind of stuff and he has lots of friends and resources for such projects.
@@DJKr15py LOL, just wrote the same thing without even reading your comment. It must be true then. 😂
Just need an Intel 771/775 "Lab" board which are available with the right microcode/BIOS, or a custom BIOS with the right microcode on one of ASUS 775 boards. Just need to shave the substrate so the board's socket doesn't need to be cut. People in Russia have custom written BIOS for 775 boards that made 771 XEON chips work, so i'm sure it is possible to get these chips running.
Thanks for the comment! I would love to get hold of an ES 771 board for this, but those are possibly even harder to find than these chips.
There *HAS* to be someone with connections at intel to help you get that cpu up & running :/ fingers crossed!
They could get those chips working because they had access to microcode from other BIOS's. It's a very different story when you have no reference. Especially when the chips might just be so different that they don't work with the northbridge.
Yeah but the 775 and 771 are essentially the same CPU internally, the 775 has just SSE4.1 fused away.. The difference between the chips is a 90° rotation of the socket and the exchange of two pins.
The 771 xeons will even boot on the 775 mainboards without a bios mod, but then SSE4.1 is not availiable.
771 chips work out of the box on a 775 board (given it has an suitable chipset - p45 is best and the socket mod has been done already). you only lose a couple of instructions if stock bios is used
Intel in 2004: "we just cant ship processors at 150 watts."
Intel in 2022: " 12900KS is hot, but its damn fast, you can figure out how to power and cool it."
EDIT: omg thanks for so many likes.
Nvidia: hold my beer
I always think about the Tejas TDP when I read about the i9. It's a habit I can't shake
Actually, Intel can handle it all now
coolers evolved a lot throughout those years though
@@warmike excuse me? Coolers evolved because they needed to, it's not like that were a breakthrough in cooling technology
I still remember writing an article immediately following Prescott release calling it Preshot and predicting this would be the last of its kind. I was actually called out by Intel as I was according to them spreading nonsense 🙂
That's amazing, thanks for sharing. What publication did you write for?
I remember buying prescott and hating it because it ran hot
I remember buying an Athlon XP because their IPC was better and the P4's where just too hot to handle. Luckely Intel got wise and continued with developing the P3 architecture.
ah yes presshot
it even fit the name being named after a place in AZ lmao.
@@FullyBuffered source " trust me bro"
I remember those days, that's the 1st time AMD pulled ahead. Intel was making higher frequency processors because people mistakenly thought more hertz meant more speed. But the lower clock speed AMD were faster at that time.
The 1st time AMD pulled ahead was when AMD 1st hit the scene making Intel CPUs better than Intel.
@@bogartwilley Yeah the Athlon line in '99. AMD has been making CPUs for awhile.
@@Ittiz 1970 with the am2501
AMD beat Intel to 1ghz and above, Intel released and recalled its 1.13ghz model for stability issues. Its probably why Intel went with the netburst high clock speed architecture. Higher clock speeds are better, right? /s
Then they did it again for ryzen. And now they think that they don’t need to add cores to keep up with Intel anymore. They actually think selling a 6 core for the price of a 10 core is a good idea.
"You can't ship desktop CPUs at 150W", the i9 12900k => 250W
i9 13900k => 250W, up to 350W without power limit...
And now i9 14900KS that consumes just shy of 400W without power limits...
@@ismaelsoto9507 I cracked 500W on liquid nitrogen a little bit ago. The VRMs need water cooling if I go much further.
There was a canceled i9-10990XE with a rumored TDP of 380 watts
@@loona5530 Well the 14900ks consumes over 400W.
The IHS is most likely soldered. You can try the lga 771 to 775 mod & use a p4 and/or c2 boards. 771 server boards are so finicky. Try to get a very early bios, asus boards usually seem to do things that they werent ment to.
second that Asus board. I used to have a P5Q Pro that I ran Xeons in and I never did anything to the bios
I mean singlecore sucking up 150W would be very impressive talk about efficiency
ya imagine an 8 core drawing 1200 wattts
As someone who owns a Pentium EE 965 (the last Netburst architecture chip) this is fascinating. These Tejas chips are mythical.
Netburst was bad for gaming, only DX titles, Fury 3D ?
only some Pentium 775 chips, did 4 mb cash on every core, only needing the fancy new DDR III
everyone bought crap, he got that same crap, DDR III was to fancy, but just open, Rambus fakeness, overcoming that ???
This boy is too young to understand it, sorry....long time......
I also got one of those, bought on eBay a year or 2 ago for something like $30 or so. The "TV" indication means it is a Thermal Validation sample, very small chance the silicon on these is actually functional.
But why so many revisions of a thermal sample?
Having done all sorts of overclocking on those 771/775 CPUs, I'd guess if regular Prescott was rated for 115 watts thermal power, and still drew up to 150 W electric power, then to double the clocks you'd have to at least double power draw, but as heat builds up, it becomes less energy efficient. All that under assumption that vcore is constant, I was able to push cedar mill celerons on air to 6 GHz with 1.55-1.6 vcore (from default 1.325) for screenshots or Super Pi 1m, but whole rig idle eats like 280-300 watts (from 120 default). I'd say that TDP on Tejas would have to be in 250-300 W range and it would have easily required 400+ W of power. You might be able to feed and cool that beast with what's available now, but in 2004 there wasn't yet an 8-pin EPS connector introduced, PSU delivering over 500 W were scarce, not to mention you'd need a special motherboard with VRM exceeding ones of that era. This would be something like extreme LGA 3647 stuff done 20 years ago, impressive, mind-blowing while useless and meaningless.
Also you forgot to mention that the old P4's motherboards using 5V rail as a power supply. That's why the old PSU's nowadays sucks because you don't really need 80A at 5V, that's ridiculous. But when you start think about it, when intel would come up with P5, there could be new 24pin connector definicion, like 28 or 30 for powering that hungry as hell Cpu 🤷 who knows, but i'm glad that intel gave up on this project because pipelining the cpu to the sky was big nono and it will not had a future. With P5 7GHz cpu, in CB R15 single core perf. Will be in level with fastest Core 2 duo clocked at 4GHz but the c2d will draw like 130W and the P5 atleast 300
@@samohraje2433 Pentium III and Athlon XP were the last CPUs requiring a high-amp 5V rail my man.
@@samohraje2433 I am not sure if P4 needed that much of 5 V, but I remember having Enermax 465 W PSU back then, that had like 30 or 35 A capacity on 5 V rail. Nowadays you get 20-25 and that's still a lot, since most of the power in PCs goes from 12 V rail. BTW that's also a challenge if you want to overclock an old CPU with new PSU, you might not have enough 3.3 and 5 V rail.
@@samohraje2433 Pentium 4 doesn't need 5V rail. They are mostly 12V.
Only early motherboards from 2001-2002 was built on 5V rail for CPU.
All later motherboards using 12V 4-pin additional supply.
I'm sure, that in late 2004, they would plan it on 12V rail, probably with double 4-pin connector, which was introduced anyway on most highend motherboards, for overclocking, when you go above 150W.
If this project would be successful, they would just need to build it few years earlier this way, and introduce 2x 4-pin 12V rail as basic.
socket T was built for large heatsinks from beginning, it was also made of steel, not plastic, and it came handy later, when multicore extreme CPU's appeared, even on core 2 architecture, they start to draw over 125W, particulary after overclock.
Just check Asus P5Q Deluxe motherboard, you'll find there 2x4-pin 12V rail additional connector for CPU.
If this behemoth would be release, they would just need to build motherboards like this. With 8pin 12V connector for CPU instead 4pin, and 8-16 power phase design, (Asus P5Q Deluxe has 16-phases, so it handles over 150W CPU with stability even after extreme overclock... this scenario of course happens only during extreme overclocking, it wasn't meant to run like that, it's basicaly only for records for overclockers, that trying to hit absolute maximum and run 1-2 benchmarks... but Pentium 5 would probably needed it for normal run)
A couple of ES Skulltrails boards were flying in public before final production. There is a little chance if anybody has BIOS dump and try it in your MB.
Oh I didn’t know that - thanks for the tip!
Why would mobo that premiered in 2008 had support for prototype CPU from 2004 in it's BIOS o_O?
Obscure PC hardware is one of my favorite things to find on youtube, because you get to see such interesting oddities. Like how Intel had been banking almost entirely on operating systems continuing to utilize single-threaded workflows and, thus, they were very keen to just continue pushing the power and thermal envelope with increasingly faster clock speeds on a single-core processor architecture. Luckily they've got some wicked smart engineers and were able to pivot quickly to a multi-core design with the Core and especially the Core2 platforms. And, well, it took until Zen 2 for AMD to claw back the performance crown, albeit briefly and under specific circumstances. Zen 3 gave them the title right up until Intel decided to release a special binning of their i9 that is basically uncoolable when fully turbo-ed.
a 7 ghz cpu would be crazy insane I would have loved to see it for the giggles factor alone shame it never came to market
The "A4" stamped on there isn't a stepping/revision code. It's the plant code for where it was manufactured. A4 is the code intel uses for USA, as opposed to Philippines, Malaysia, Costa Rica, etc.
admit it you want a pentium 5 just for bregging rights no other reason
Amazing video !
As for Tejas legacy : SSSE3 instructions (that AMD's Phenom CPUs lack), are direct link between Conroe (a.k.a. Core 2) and Tejas. Additionally (IIRC), some of Tejas branch prediction unit was also incorporated into Core 2 architecture. FSB speeds of 1333MHz and above were also mentioned for it, but that's more of a rumour than actual spec.
I bet you can't make them to boot, because you lack Tejas ucode in BIOS (you can add it manually if you find a source for it).
Thanks for the comment! Yeah finding ucode for these is going to be more than a challenge to say the least.
It's ironic that intel managed to fall into the same trap many years later. Instead of trying to increase IPC they still raised clock speed with newer process nodes and didn't care about power consumption. After the P4 disaster they based their next architecture on the Pentium M, a low clock but high IPC design that was only meant to be used in laptops but turned out to be much better for desktops than P4. This intel "core" architecture was very successful for many years, even persuading Apple to switch from PPC to intel. But intel failed to focus on efficiency and lost Apple as a customer. Fun fact: Apples inhouse architecture is based on a mobile chip developed for smartphones😉. History repeating..
History is also repeating for AMD, they just used Jim Keller in order to stay afloat, their Zen iterations are still based on Zen architecture, with minor changes, even like 3D Cache, it's still on band-aids, and not a new architecture. AMD has no new architecture for years to come.
I don't think it was Intel's plan this time around. They just thought backporting upcoming architectures to their 14nm node would be a waste of resources as they surely would be able to leave that size soon enough. Then their fabs just didn't make enough progress and they were stuck with just one choice which was to crank up the clock frequencies.
Then by the end they did bother making one backport release, namely 11th gen for desktop. Kind of an awkward release.
@@Wokiis Breaking News: The Fab node has lost its meaning once it goes below 12nm.
Heck, it has lost its meaning once everyone go to 3D or trigate transistors instead of single planar transistors.
@@AlfaPro1337 you just spouted nonsense.
@@Wokiis twice
I bet a very experienced coreboot dev on a properly setup board for low level debugging could rig something working, especially with the recent advancements in decoding and patching Intel microcode.
That assuming it doesn't just work on those 775 patched bioses that boot just about anything after doing the mod to the socket
A silly part of me wonders what a chip taking the Tejas/Jayhawk approach would be like if it were done with current process technologies, and using the increased transistor density to make the pipelines even deeper.
A 50 stage pipeline CPU on TSMC's 5nm process could reach 10ghz I bet, with some water cooling of course.
-_- from my memory bulldozer when they first explained what the architecture was... it made me pull the trigger on the 2600k.... because i was like wtf AMD why are you making your own version of netburst.... i already saw the issues coming
@@wnxdafriz Yeah, bulldozer was a thing... back in day I had a Phenom II X6 system and I remember hearing that the upcoming bulldozer parts would use the same socket, and thinking to myself it would be nice to have that as an upgrade option. Sadly, bulldozer turned out to be a flop, not even a decisive upgrade from what I already had. My next CPU purchase was a 4790k which I'm still using to this day. Still holds up quite respectably for a quad core.
You will absolutely need a modded bios for these to work, you can also try a very early ASUS board with a first bios revision maybe you can have some luck
This is great content. Cool subject, well researched and good presentation. If you keep making content like this your channel will blow up!
Thanks! 😁
50 pipeline stages sounds a bit like a cargo train with 50 fully loaded wagons with the bogies and couplings welded fixed going at top speed. In the mountains.
I built my first gaming PC in 2003 and AMD was the clear winner at that time even doing it at lower clock speeds, this was a marketing nightmare , how do you sell a 1700mhz chip thats faster then intels 2.4ghz chips? thats when they introduced the + naming conventions and called their 1800mhz procesors 2500+ to give you an idea what the speed was compared to intels slow pentuims 4's. But intel took the crown back with their core2duos in the coming decade and really has hung onto the leads until recent ryzens. But at this point im looking forward to Risc architecture so im not interested in more hot big cpu's from eitehr AMD or intel.
i have so much nostalgia for the p4, it was the processor in the first ever desktop computer i owned myself. it was slow as shit on windows 7 but it was mine.
Nice find! Thanks for the video
This is awesome work! Keep them coming!
Thanks!
Even today's chips do not run at 7GHZ out of the box! This would have been insane if possible with stock cooling and considering how much power it would draw from the wall.
I'm not sure if you realize this, but "tejas" is simply the Spanish pronunciation of Texas. Jayhawk is the name of a hunting ranch in Texas, although the name may be from something else as well.
Wow, just discovered your channel in my recommended feed. Very nice work, I like it!
Thank you!
Man, I have been watching your vids for some time now, and for sure you deserve A LOT more subs, but until then DON'T GIVE UP. Your material is really good! From someone in the same growing struggle, keep going! I really enjoyed your vids and on point facts and presentation. If there is anything you need help with (graphics, video graphs, logo, whatever) let me know, I will try to help you out of good will. Take care my friend.
Many thanks for the kind words! Best of luck to you as well :)
Intel in the 2000's "we cant possibly launch a 150w processor" .... Intel 2022 "hold that..." (i9 12900k)
Tejas isn't just a place in Texas. Tejas is an older name for Texas.
Just came across this on my homepage and it's fantastic. Liked and subscribed
wow, never heard Socket T is named after Tejas. Interesting finding, never figured out, why they used T. I just thought it is random letter.
Also dont know, why Socket A was chosen as A.
LGA775 was my favorite socket. In limited scenarios, it's still even useable in 2024 (though not recommended obviously).
It just had such an incredible lifespan. Single core, 1GHz CPUs all the way up to 4 core 3GHz+ behemoths. So much variety, so many price points.
And you'll never see Intel release so many CPUs on a single socket again. The fact that Core 2 was a drop-in upgrade on many P4 boards was amazing.
Amazing work Guido! Your best video yet.
Thanks Mike 💪
What is the chip called and were can I find it also does it require a specific motherboard?
Great video detailing this interesting processor history. I am enjoying your channel :-)
I feel like it could have been possible for Intel to release these cpus as "speciality units" that would have specific requirements for purchase
did you try a good overclocking socket 775 motherboard and 771-775 pin mod you would need to set the clock speed and voltage manually
This stuff reminds me of the FX9590, where AMD tried to squeeze the impossible from an old flawed architeture that could give extremely high clocks, but low IPCs. The story is quite the same, and also the infamous FX chips where characterized by abnormally long pipelines, hence high power consumption and long latencies in case of wrong branch predictions.
Why did AMD decide to make the same mistake Intel made on its previous iterations is still a mystery to me ahaha.
Your take on this is grossly inaccurate. if you compare the FX9590 with any 32nm Intel chip, the FX9590 wins easily. What you lacked over the years is knowing what Intels advantage actually was. It wasnt a better design. Intels designs have been clear verifiable worse for 25 years. Intel hit it big, jumping what was essentially several nodes ahead when going from P4 to Core2, because of both a node shrink AND a new novel 3d trigate design. The Core2 was also based on P3 instead of P4, but still not as good as AMD's designs. Also, that 3D trigate shit was what put Intel in the VERY BAD position it is in today, because they spent 8 YEARS trying to move it from 14nm to 10nm and eventually gave up. Todays Intel chips are using FinFET again like everyone else, and Intel is worse. Yep.
@@styleisaweapon Don't get me wrong, I wasn't considering neither raw performance nor lithographic nodes for my comparison. I was considering concepts (long pipelines) and relative consumption (> 150W). I know that AMD has been superior at designing chips at most times (not always tho). 2nd gen FX was superior to Intel in raw integer calculations and had a way larger cache amount. The way that resources were handled, tho, was poor. Shared caches had a terrible latency and collisions between cores, and FPU cores were halved to cointain costs. So that design wasn't the best anyway. What penalised FX series was their terrible 32nm GlobalFoundries node, which was obsolete even at the time. FX could have gathered much more success, but it had limits due to its structure. Even if they used a better node, pipeline complexity increasing over generations would have caused consumption to become incontrollable. And latencies too. This is why in many applications FX was way less responsive than lower-tier Intel chips of the time (as i3 and i5), despite being more powerful on paper.
More latencies + More consumption = Consumer Market failure. And the reason in the absurdly long pipeline. That's why I can't get why they used this approach.
Also, another inaccuracy you said is that Intel was always (and still is) inferior in design. After AMD K6, which is corresponding of Intel P1, AMD was getting each iteration closer to Intel, and superseeded it in the terrible P4 era. Athlon Classic was more efficient than P4 at the same way Intel Core 3-4th gen was towards FX 1-2nd gen. The Phenom era wasn't the same, tho. Intel was way more efficient at the time, despite being overall slightly less performing.
Also, the first Ryzen processors weren't *better* than Intel ones. They were much more cheaper and had a way better value (on a similar node than Intel's). That's why they succeded. This changed with Ryzen 3000 series, which was both cheaper, better performing and better consuming.
This is also changing again, with Intel Amber Lake being better performing, but more expensive and worse consuming in some categories. Times change.
@@rotten_cyano blah blah wall of text that proves youi still simply dont understand what intels advantage was - everything you said is literally invalidated by it - EVERYTHIHG
@@styleisaweapon I suppose you didn't even read. You just said random things in your previous comment and I made a further explanation.
@@rotten_cyano ah yes random - intel being effectively 2 nodes ahead didnt entirely explain intels performance advantage at the time .. you are so right ... not only doesnt it entirely explain it... it doesnt explain it at all .. how clever of you to realize that having a budget of 4 times as many transistors doesnt mean anything to performance
Cool video! I would try these on a p35 chipset socket 775 motherboard with the socket modded to take 771 cpus (with sticker on the cpu) I have a p35 board that runs a Xeon 5050 which is old old Pentium D era stuff. Good luck! Need to see these running. Any more available? Ive lots of different server boards ect to test on
Well thats a shame it doesn't work. You could also try a 775 chipset board with one of those stickers which swaps the pins round?
I have the Intel black ops CPUs for LGA 2011 and although they are unstable I can boot and run them on Asus z9pe d8 ws and rive.
That’s very cool - I’d love to get a black ops. Unfortunately it seems that without proper microcode support it won’t matter what motherboard I try. Perhaps an engineering sample 771 board, but the chances of getting hold of those is slim.
Thank you, this was very informative
In reality, no large company loses money on failed projects. They simply sell more expensive products that did work.
seeing one of these actually running would be so cool!
That is a well made video. Congratulations!
Thanks!
Awesome video! Neat trivia note about the socket names!
Thanks!
Do you know about NetBurst Replay system?
Very impressive presentation!
Young Hugh Downs in the making,
Subscribed and shared. Cheers from TAIWAN :)
Many thanks :)
good information . we support you .
Thanks!
I did a tour of the Intel facility in Portland during school. At the time, P4 had just been released. PCI Express 1.0 was being debated on. They had a *10ghz* ALU running at the time.
I would have loved such a school trip
Wouldn't need any air-conditioning in winter, a Pentium "5" would heat the place up nicely!
Diep onderzoek, intelligente analyse en prachtige presentatie
Austin engineers were kind of punished for Tejas. Next product that came out of Austin? Atom. They were tasked to develop sub-1W x86 CPU. It was paired with 10W chipset (i945), but it was possible to run first Atom silicon at 1W.
Folsom is a stone throw away from Santa Clara (in fact it is a 'backup' site of sorts, in case of SC goes to hell if San Andreas fault going off would actually wreck Western California), so there was major push from board of directors to continue all development in FM, since they had a chance to personally oversee all the projects this way. I strongly believe that Austin site is a 'talent sink', in a region.
Really cool thx for the share
I have some of these 771s I got in a batch of engineering samples, I never had a board to run or test them. I still have them, still have no Board for them. I thought I was getting 775s. Most are branded Zeon though. Isn’t the socket named for Jayhawk?
They were selling these stickers that swapped two pins on the 771 to run on Intel 775 boards. I never got around to ordering them because at the time I was going to swap them myself and never did.
This is really remarkable! I remember very clearly when Intel announced the cancellation of Tejas, and after that I think most assumed there was either never any silicon produced, or what little there was would never see the light of day. So to see a physical sample, even if it's non-working, is pretty incredible. Even better to get insight from one of its designers. That was brilliant on your part!
It's honestly hard to believe it got this far. I can't even believe it was built on 90nm! That just seems insane, given that as it was, Prescott was already getting Intel in trouble, and the Athlon 64/Opteron was spanking them in almost every market segment. Even Intel diehard OEMs like Dell suddenly started offering the option for AMD during this time. Intel created a performance-per-watt problem, and painted themselves into a corner, while letting their rival soak up all the goodwill. The fact that many reported the 3.8GHz Prescott couldn't run stable in many systems should have been (and maybe was) a wake-up call.
Many thanks for the comment man 😁 I remember the AMD Dell systems - I had a Prescott 530 Dell Dimension back then and it pushed out so much heat for the performance it offered. Intel must have been not amused by that switch lol
@@FullyBuffered Dell was the last tier 1 OEM to adopt AMD at the time and it was a pretty big deal. It was years of reasons why not for this reason and that reason to suddenly full embrace... brings back memories...this is a cool score BTW...
Netburst took "It's better to burn out then fade away" to an extreme and going to potentially 7GHz is proof of this 😂 Im surprised your house didn't burn down from testing those sample chips.
I wonder of Intel would be able to revive such chip for something like 20'th anniversary of last Pentium, just dig out of its archives the "working board" for those, and maybe make on prototyping process like 500 of those for collectors. Not bothering with anything but bootable board, with first commercially available 7Ghz CPU. That would be a nice marketing they really need right now.
I would love to see that. It would be a neat marketing stunt indeed.
People don't realise that increasing frequency by deeper pipeline is quite equivalent to more parallel functional units, because you need the same amount of independent instructions to fully utilise resources in every case. In principle, every stage of pipeline execute different instruction.
To some degree relation speed vs. width is similar for another parts of processor, like cache.
Man your channel is gonna blow up. Trust me... Great documentary
Thank you! I do hope so 😁
Great video brother.
Finding out processors try to anticipate data gave me a glimpse of how little I know about the subject. If our civilization collapses, we will never be able to figure out what all this crap did in just a few decades. Once all the books rotted, there will be heaps of brittle plastic and trinkets with no discernible function. We will assume that it’s all for religious sacrifice or something dumb like that. Just like we think about EVERYTHING from the past.
What a wonderful video, A big thank you for this content... I remember owning a 3.06Ghz P4 Prescott chip back in 2004 it was one of the first LGA 775 jobs and they were real power houses, mine stayed in service up till around 2010, I recall a day when there was a glitch and when i finally got task manager up it said CPU usage was at 12Ghz, it was super laggy and the fan was screaming loud but it does show the potential of the architecture. with a better cooling solution and better ram i think those chips could have easily gone to 5Ghz out the box. Its a shame in a way they decided not to continue it but i'm happy to see some of the R&D went in to future products Hyper threading being one example.
I had a laptop with a Pentium 4 desktop chip clocked at 3.5MHz. It was quite fast but the most notable feature is it melted the plastic on the bottom of the case. It was too hot to put on your lap and would leave burn marks on a wooden desk.
I recall the upheaval at Intel when the Israel labs came up with Pentium M, which would later birth the Intel Core architecture. The desktop division was a 10,000 engineer-strong organization. The Israel lab had some 100 engineers. Imagine Intel Austin being showed up by Intel Israel. I was told that the Austin lab was seriously downsized after that. Dadi Perlmutter rocketed up the Intel hierarchy thanks to this, as his team's Pentium M lead to Intel Core, which effaced all the bad rep Pentium 4 had created.
I had an old netburst celeron, which was supposed to be the lower TDP version and that thing still killed 3 of the measly old heatsinks we used to use back in the day.
What about delidding them to see the actual die? Same process derbauer did with the threadripper (assuming these samples have solder), you just heat em up to 150C and then carefully remove the lid
This reminds me of a long-standing question I have of what we would be on if Intel kept using the x86 name structure. Any thoughts?
I don't think the marketing department at Intel would be able to cope with that haha
I love that CM 690 you have under the desk
crazy thing I remember reading on Anantech in the early 2000s about how Intel was going to make 4 GHz and 5 GHz Pentium 4s. They also talked about the goal of it being 10 GHz. Insane when you look back at it, single core CPU at 10 GHz? I find it fascinating how Intel was much bigger than AMD so they doubled down on single core chips with insane clock speeds but AMD did the opposite. They took the time and money to invest and create CPUs with integrated memory controllers Hypertransport and two cores on the same die and x86 64. The Anandtech writers all sided with Intel and even said the Pentium D would be superior to the AMD Athlon 64X2 because Pentium D had two CPUs in different dies. Insane right, Intel must have been paying them off or something. IBM also followed AMD with their Power 970 architecture. Intel saw the error of their ways and finally course corrected with their Core designs.
Without propper bios u never be able to boot this chips...if its working chips.Are they heat in motherboard or not?
It would be interesting to see what happens if you have a port 80 debug card connected and try the CPU again. That would at least let you know if if it actually executes any instructions after the power-on reset or if they're totally inoperable. If it makes progress but hangs somewhere, knowing what code it hangs at might give you a clue as to how to get past that particular hang,
Ok this is awesome. I really wasnt sure if actual silicon was made that wasn't scrapped - the explanation on why this isn't a thermal or mechanical sample makes sense. Also interesting the 7 GHz target appears to have been on 90nm. I am curious of the engineers opinion on how much Dennard Scaling breaking down influenced the cancellation - did they know 65/90nm (and below) wouldn't scale like prior nodes when designing this chip?
i was excited to see the chip work
Intel thought they were being slick with Prescott and the high clock speed. Basically trying to trick people with the high clock speed. Only problem was that AMD's Athlon64 at 2.4 Ghz was significantly better than the P4 at 4.0 Ghz. Which really opened people's eyes to IPC rather than clock speed.
Great video. Try the 771 775 mod on an Asus motherboard. Asus motherboards seems to boot unsupported CPU's, it just displays an "Unknown CPU" message at startup. (tried a 3.6 p4 cpu on a core2duo Asus motherboard, and got the "Unknown CPU" message but I pressed F1 to continue, and it worked out just fine) It wouldn't hurt to try, but I doubt that it will work with that cpu.
Since CPUs are now using 300-400-500W I wonder what a Pentium 5 using 500W would have looked like.
Imagine a house burning down yeah just like that
Just imagine the 28 cores of bullshit from 2018, that's what Tejas would have been.
Are they LGA771 or 775 ? Could it be that, Y they didn't turned on?
man if Pentium 5 ever came out, news headlines would read “NEWER INTEL COMPUTERS REPORTED TO MELT DOWN, PRODUCING SMOKE”
150W at 7Ghz? That is probably the optimistic projection. More like 5Ghz at 250W. Let's also add a little thing. While Intel would like to be dismissive of AMD, imagine if AMD would use the same power consumption and how much performance they would have gotten out of it then.
maybe it just need bios update, you can make that by your self if you can found the cpus microcode and then insert them by mmtool to motherboard bios
Any updates on this?
how about trying to run them on 775? Intel server boards usually have some whitelist of supported chips, desktop board shouldn have it.
Could you put a really old card running on that server board? i.e. something from 2004, for example. The evolution of PCIe has lead to some anomalies where some graphics cards on really old CPUs don't display a POST, and you are basically hanging there waiting for something that never happens.
Funny thing, I am using Socket 478 CPUs to this day...
Brother you deserve more followers, you have a calm voice and you are very smart , love from Romania❤️ 🇷🇴 Europe
The motherboards will need a compatible microcode...
Was this prexeon lga 771?
Yes, that's right.
Tejas was the original name of the state while it was still part of Mexico. It was renamed to Texas when the US took it.
Thanks for the comment!
it still is pronounced like that, just for Spanish speakers. Just like Mexico is still spelled Mexico just pronounced as "Mehico".
@@iplyrunescape305 I see. Nice to know thanks. :)
Fantastic content! I was shocked when I saw your sub count, count me as one more for sure.
Thanks!
A brilliant video! I wonder if some Intel engineer saved some engineering motherboards from the trash bin with the needed microcode to run these chips?
here's a thing: there are two x86 µarches gentoo has ever seen that gain a sizeable amount of performance with µarch specific compiler options/flags. those are:
* bulldozer
* netburst
and owing to this, I've concluded that those two µarches have heavy justifications for build arches specific only to each of them. sure, since it's x86, every x86 processor can run it, but obviously they'll make bulldozer/netburst shine like they never got a chance to. you may remember a similar scenario with the original raspberry pi; many distros used an armel kernel and binaries (that is, software floating point), but the bcm2835 had hardware for floating point, so it could also run (and actually benefitted from) armhf kernels and binaries due to not having to emulate floating point instructions. for many people back in the day, picking an armhf distribution/system literally meant free performance.
Oh yes...those times. My first rig was a Pentium 4 Prescott on the Asus P4PE2-X before we started cranking up a Northwood core. But - that's not the story that I am heavily interested in - it would be the Pentium M (this laptop runs the 725 - yes it's about 18 years old now) that later turned into the Core (mobile only) and Core 2 processors. The Netburst architecture wasn't really all for waste...it introduced HT that waited until the Core i-series to reappear...and of course the higher FSB speeds which were all developed for it.
As for making those CPUs work...everyone else already said the deal. Hope the bios modders could also unlock my step-up VAIO to read all the 4 GB of RAM that the chipset (i945GM) is supposed to support and enable Core 2 specific features (Enhanced C-States) because it just got a CPU upgrade
Edit: Intel would have named that the Pentium V for extra flair
my first comp was 486.
I remember being very dismayed at just how much heat was pumped out by the Northwood and Prescott chips. Netburst really was a mis-step for Intel....
You will probably have to find out all the info on this cpu like family, ext family, model, ext model, stepping and revision and add the microcode, containing this info, to the latest bios and flash the board with it in order to have the p5 cpu working. Its only natural that the microcode is missing from every lga 771 and 775 board right now making the cpu not work. Ive modded lga 771 cpus to work with lga 775 boards in the past. Adding the microcode isn't hard but you will need all the info.
You should to add context of the quantum tunneling issues and the success of the high-k effort and extreme power management that led to Core. Last chips of that era I had were some DC Dempsey for 2 socket servers - socket compatible with the Core series Xeon - quickly swapped to the much better performing X5465 because the 5080 with FBDIMMs could eat 180W per socket!