Interestingly, the person, who managed the copper interconnect project in IBM, that later gave considerable edge to Athlon was actually Dr. Lisa Su. Just imagine, that the person, who helped to develop Athlon manufacturing (albeit indirectly) is the same one, who later served as an AMD CEO during the time, when Zen was developed and is being produced, crushing Intel in HPC. Also, before Athlon and AMD64, Intel and Compaq had DEC Alpha future development cancelled (to ensure, that Itanium woudn't have to deal with RISC compatition), disbanding the development team and leaving some of it's technologies shelved. Some of the DEC engineers then ended up in AMD, developing AMD64 architecture Athlon and Opteron. One of those engineers was Jim Keller, who also much later participated in Zen development. Also AMD licensed EV6 bus from Alpha, using it in Athon.
Now that you think about it, the Athlon name may have have a deeper meaning after all. It was supposed to come from the word decathlon, but with so much of DECs works being used for it, it was almost a "DEC - Athlon" chip.
And as I recall, Itanium seemed like a brain fart by Intel. The X86 instruction set getting extended to 64 bit by AMD made that all too obvious; instruction set compatibility was more important than Intel realized.
@@stevebabiak6997 Itanium was an interesting concept but required too much research to get it done, and by the time it came out x86 caught up in performance and marketshare. More importantly the mere announcement of Itanium made a lot of companies stop developing alternative ISAs (DEC Alpha, PA-RISC, etc), which drove a large chunk of the market to x86. So Itanium basically destroyed its own place on the market with its own announcement. Which is really quite an achievement! AMD64 was just a nail in the coffin; Itanium was already on life support paid for by HP from day one.
@@stevebabiak6997Itanium inherently was a very "riscy" choice (pun intended), because it's at a very extreme end of architecture choices, with its VLIW instruction set. AFAIK the only area where that style has been used for a long time is signal processing (i.e. where you have very specialized, optimized software), where simplifying the chip has huge relative gains on power usage and chip vost. VLIW basically means you have huge instructions that explicitly have fields for each specific ALU of the CPU, so the CPU only needs a very primitive scheduler and no out-of-order handling. In general purpose computing, the difference just wasn't significant enough to warrant the effort to switch over. Also the huge instructions complicate compiler programming, and increase memory usage significantly (besides the extra memory requirement for the 64bit addressing), and that's probably also why Intel always segregated Itanium to be just for servers. And then AMD64 killed off Itanium (and forced Intel to "cannibalize" Itanium with its own 64bit x86 chips). Architecture wise, Itanium probably was great for some types of scientific and Supercomputing applications, but it just doesn't make any sense for e.g. a webserver.
Ah, a stroll down the nightmare memory lane. It was such an adventure watching AMD's struggles against Intel, and now they've finally caught and surpassed them.
@@mddunlap03 Intel might be a bit behind those days but they still got some design strengths and they have an excellent software team and they acquired quite a solid reputation. That's why despite very good products, AMD have difficulties to push through a lot of markets. This time this is not because or coercion. Also Intel have deep pockets and the amount of field they cover is way larger than AMD so they can perfectly manage losing a few battles here and there.
Among Intel's big flops was the new Willamette core of year 2000. (a HUGE project I worked on) It was an ultra complicated, super pipelined (12 stage) generation that required even more stages (18) for the Prescott followup, thus dooming the series to cancellation, as they could not tame the Prescott beast as it fell behind schedule. In hindsight it exposed a major design/tech_path screw up by top management. The 'right hand turn' of multi-cored P6's helped retrieve the situation .. for a while.
I remember in the late 90’s, around 98/99 if I recall correctly, when the Hammer architecture was first unveiled even before the chips came out. I downloaded the PDF and read all about the 64 bit extensions to the x86 architecture. The Clawhammer, Sledgehammer and Jackhammer names were bandied around, what later became Thunderbird and Athlon a few years later. The proposed changes were shared with the world years before the first chip using them debuted. The 64 bit extensions were embraced by Intel without any accolades that they were invented by AMD. When AMD reached 1Ghz before Intel everyone became AMD fanboys overnight. Especially with how easy they were to overclock.
And this is one of the few times Microsoft actually did the right thing: they were ON it like a whiz. They had caught wind that Intel would partner with HP and their HPUX as the server operating system of choice for Itanium. Microsoft didn't like that one bit, and put all hands on deck to make a good x86_64 compiler and port Windows to the new instruction set extension.
No. Athlon K7 was the first major effort in 1999 and was still 32-bit. Thunderbird were later Athlon models reaching over 1Ghz,. Sledgehammer and Clawhammer were Athlon64 and had no competition because the pentium 4 was shit and only clueless idiots got those.
I actually worked at TI in the 1990s (1995 & 96 I think) on a Pentium class clone processor. Internally it was called it the Amazon project. TI had effectively stolen Cyrix CPUs by signing a manufacturing deal with Cyrix that allowed TI to sell the chip directly to OEMs and to undercut Cyrix. TI pursued the Amazon project for a few brief years in the hopes of coming up with their own successor chip, but we were several months behind schedule when we began getting silicon and TI's bean counters decided to can the entire project and sell off the IP rather than try to be an also-ran in the x86 CPU business. The Amazon chip had several innovated design features that would have allowed it to run rings around a typical Pentium processor, but the it lacked out of order execution, which was the next new thing with the up and coming Pentium Pro. Apparently TI management didn't want to spend the resources to go chasing that target. Awesome video. Brings back lots of memories.
Forget the other statements about AMD kicking Intel's ass because Intel will move to TSMC before that happens and would probably push AMD to the side a little with TSMC as they already have, buying up the remaining run of TSMC N3 for this year and I think into next which then keeps AMD from being able to use N3 for most their product lines that will come out next year. So, a future look? How much do you know about CPUs? AMD will have to redo their current MCM architecture and probably move to what Intel is putting out in 2024 to at least laptops (Meteor Lake, MCM architecture that is tiled, so one chiplet pushes up against the next and there are direct connects between them). Many insiders feel this is going to be better than AMD's Infinity Fabric where processor cores have to send data to/from another die called the IOD which has a multiplexer for getting data to where it's supposed to be, and this design was fast enough for Zen 2/3 but is giving them problems for Zen 4. So AMD will probably move to a direct connect MCM architecture just like Intel and I believe they have the rights to do so as I think that connection was licensed by a few companies so CPUs could be made with a wide variety of chiplets or tiles. So, imagine a world where a CPU can be made with lots of different chips from different companies. That's the next 5 years. CPUs will have hardware accelerators, AI cores similar to what AMD is already putting in some laptop products. You can have a graphics processor be a separate die, etc..... You could even have a chiplet/tile simply be cache, similar to how AMD is stacking L3 cache on their X3D parts put no need to stack it any more. Yeah, that's the next 5 years and it's going to be exciting and both Intel and AMD are ready for this.
@@johndoh5182 Sorry, but I didn't bother reading that blurb as you started out with blatant speculation that nobody but TCMC, AMD and Intel knows anything about.
A64 was incredible - in terms of price AND performance. A bit confusing with the different socket types, going 754 locked you down to a 3.4Ghz maximum but they were much harder to find and more expensive than the 939's. I spent a fortune on a DFI Lanparty 754 board and regretted it pretty soon after. But hey .... Core 2 Duo was round the corner so it was a no brainer what to buy next.
One reason Abu Dhabi investmsment looked at AMD was intel getting Core and other innovations out of the Intel branch in Israel. The two CPU giants are proxy warriors for a very different kind of fight.
I remember the time my small amount of AMD stock was worth practically nothing. it was in spitting distance of being de-listed. I.... still have it though. ;-)
Ohhh The K6-2 My first PC, i learned a lot in that PC for my motherboard had a sketchy L2 cache and gives a lot of BSOD's LOL.. and the Thunderbird Athlons.. those where good and proud times. So Proud of AMD's achievements! keep going!!!
I built myself computers in 99, 04, and 06 or 07, and used AMD for all three. Those were the last new computers I've had, and finding used/refurbished AMD stuff is not easy. I'll likely build another in a year or two, and it'll be back to AMD.
This is great content, the late 90`s and early 00`s were a great time for competition in desktop processors, coming along with the discovery of what graphics card could (and would) do. Thanks for your work.
HOLY SHIT, are you telling me there was a chance, that Jensen huang would had managed AMD and NVidia. OMG can you imagine that? GPUs would cost 5 times more and CPUs would cost the double.
@@cv990a4 Why would intel do that, only AMD needs to buy themself in, sell nintendo gear ! intel needs proprietary developing, not China steel it ! Do it yourself, own x86 and ARC !
I will always have a deep love for my Opteron 165. Using 1A-Cooling (a German company) watercooling parts in 2005, 3.2/3.3GHz+ was my daily driver frequency. The waterblock itself was a heavy chunk of metal. The contact point/cold plate was two strips of copper and one of brass in the middle magically joined seamlessly. It mounted with a single thumbscrew dead in the middle to provide pressure directly down on the silicon which would cause the block to be sometimes comically askew with the torque from the tubing with no reduction in performance. The whole block was seemingly one unbreakable monolith, and I always wondered what it looked like inside. 4x 1GB DDR sticks, GeForce 6800GT, NB and GPU were also waterblocked (wildly thin blocks, same company, same single point of pressure mounting) with a super thick 240MM radiator. Those were the days... The Eheim pump was 110V and required the case to have an AC plug routed out the back. Thanks for the walk down memory lane here :)
I really love your video analysis and your newsletter, I am sincerely impressed by your work as it has the right blend of thorough analysis and the right dose of sprinkled in humor! Keep it going, my guy
My PC history was Pentum MMX (fine), K6-2 upgrade (fantastic), AMD K8 / Athlon 64 (great speed and efficiency). Then an Intel i7 2600K that never made me happy for some reason. Problems with suspend etc. Last year I bought a not-so-leading-edge Ryzen 7 5800g. Laptop chip in desktop package. Love it again.
Much nostalgia for the chips featured in this video. Still blows my mind that AMD shipped Athlons with exposed die and users could damage it when applying heatsink lol
great CPUs for mid/high gaming solutions and mobile ones. But their desktop parts are actually lacking in applicative, not that great in idle/light task power consumption and their low-end offer is not competitive at all. Same for High-end workstation, the new Threadripper is not the game changer it was before. Finally, their recent platforms seem very unfinished/unreliable, more than typical AMD style.
I had the first Athlon XP based computer back in 2001, there were some problems with it. The chipset wasn't very stable. I did have AMD products before this, like a AMD K6-2 350 MHz, and AMD K6-3 450 MHz, but those products didn't offer the performance of a Pentium 2 or 3. I changed to a Pentium 2, and Pentium 3 in 1998 and 1999. I tried a Athlon XP, but it wasn't very stable and crashed a lot for some reason. Then, I changed to Penitum 4 1.8 GHz . The last AMD product I used was in 2009, and it was AMD Phenom II 945 Denab, and I switched to a Intel I7-960. After that I just used Intel products ever since. A lot of people are saying AMD stuff is better now, and when I have to get a new computer next time it will be a AMD product.
I’m literally reading how Eagle Pass crossing is one of two border closures at the exact time you mention Ruiz crossing the border into Eagle Pass. Timing
In years 1995 to 2002, I suggested AMD computers to gamers and Intel computers to engineers. AMD traditionally had better in-processor networks, and Intel had better raw calculation capabilities. And, I indicated ATI Radeon video cards in both Intel and AMD; the Sound blaster AWE32 or similar made for a great all-rounder PC...
rayoflight62 AMD for gaming, engineers need the same system, why use less compatible systems ! EU laws, muhahahahahaha, legal Clone IBM Bios, replace the intel for AMD in dresden... GAMING PC only, SoundBlaster was only needed for games ! way more better cards on the market, compatible in GMAES you needed ONLY ! U understand it now, why engineers need intel, if one of them does the calculations on AMD, it's not compatible !
Interestingly, AMD's original Athlon CPUs used the same bus as the Digital Equipment Corp Alpha 21264 EV6. They licenced this from DEC. This is a DDR bus, and it's why they had much greater memory bandwidth in one of your slides on Intel vs AMD performance in the Athlon vs Pentium 3 era.
Looking at a Pentium with a small number of pins as opposed to what they have now with 1700 pins is pretty funny. It shows how little conductivity these earlier CPUs had.
The K5 wasn't as bad as this video suggests, it was the choice for budget systems at the time as the performance was sound and the price below the intel parts. One might tell the story as: AMD got to use an outdated fab without going bankrupt.
Itanium brings back memories. Itanic! It might be able to run HP PA-risc. But had really really bad IA 32 performance due to emulsion and abysmal compilers. iIA 64 was nearly as bad because compilers. Amd aIA64 could run native or nearly native IA32 code with minimal changes. Plus and easy to access IA64 mode with much better compilers, with all code far easier to adjust. No contest.
AMD saved x86 from Intel’s own shortsightedness. They were focusing on an Itanium based 64 bit future. The AMD64 extensions were a game changer and are still in use today. Whilst Itanium might have been good for HP and Intel no one else in the industry wanted it to be the successor to x86.
To be more specific, The Itanium optimization chain was to simply expose everything and leave micro optimization like pipelining and branch prediction on the compiler. The simplified hypothesis being one one hand that the compiler knows the bigger picture of an application and while slower than hardware it will only need to optimize the binary once, as opposed to the CPU making fast hardware optimizations during every run. And this lack of HW optimizing also frees up substantial silicon area and power budget for more core logical processing units. (Which is largely why GPUs can do so much calculation per dollar for specific tasks). This could actually work well, but as you stated the compiler-optimizers of that time did not have the needed advances and Intel didn't bother to put that horse before the cart, and secondly it requires compiling for very specific models of hardware (less portable binaries, more prone to vender lock-in). End users also get attached to their legacy proprietary binary software so even with a good cooptimizing compiler they can't just recompile to get the advantages, thus the marketing importance of good backward compatibility.
@@PainterVierax Itanium is RISC, not "pretty much". (Paedantry!) But being RISC wasn't the core market failure. It was just a poorly supported implementation. ARM's various designs are RISC and do just fine. Historically the market is much more complicated, but 80x86 being CISC was not the major factor in it's market dominance. It was much more a matter of momentum, contracts, compatibility, and supply competition at a critical time in the Micro-computer market.
@@mytech6779 ARM did a bit better because the embedded nature of the products allowed OS to adapt from it. Performance and compatibility is a reason why Android uses Dalvik then ART instead of a more classic binary packaging. That's also why specific distros like Armbian or Raspbian were formed on top of the Debian work and why Arch or BSD are quite popular on ARM and RV devices. Add to this the huge library incompatibility with the vast x86 environment and you got why Apple took so long to make an ARM transition to their "desktop" environment (not iOS) despite a very secured marketshare.
I still have a box with Pentium 100 CPU's and K5's and K6's, also Slot A and Slot 1 processors and MB's, I bet they still work to be honest. I'm holding onto them as a collector \ hoarder type of thing but they are all in one box marked "Old computer parts" Then of course the Athlon came and performance was way better than Intel for a while until Core 2 Duo came and since then and until recently it's been AMD playing catchup. 8:00 fun fact, upper right corner the L1 and those dots were the pencil trick. with a #2 pencil you could unlock the multiplier, all 4 needed to be connected.. I did this with many Athlons :)
i own 2 identical PIII system HP systems, one on slot and one on Socket, same PIII chip on both. Why thy did this, AMD did second level cash daughter board for it, a better solution. include the memory controller and both cash !
Someone, somewhere, much smarter than me, already combined moore's law, the capital cost of new fabs, and the limits the human user has to calculate where the diminishing returns to all this capital madness will start. Semiconductors will have a "Concorde/SST" moment sooner than we think.
4:00 Not sure if this was intentional or not, but note that the "Pentium" you show here on the left is not one of the Pentiums under discussion, but a much later low-end Intel Skylake chip from 2015 that re-uses the Pentium name for marketing purposes (well into the Core i# era)
I hated both AMD for their AthlonXP and Intel for their P3.. as a former PC shop owner, selling CPUs with open silicon dies was probably the worst idea the industry could ever have. I've lost so much money, time and nerves in fighting customers too dumb to install their cooler and trying to claim warranty on cracked dies. Sure, it made them cheaper to fab, but that just shifted the cost to the last link in the chain.
Back then 1999, I was writing code for biometrics and using Intel Pentium was able to analyze more than 600 fingerprints per seconds. Did not worked well with AMD or Cyrex. Don't know how that's going now with crazy fast processors.
AMD bring a lot of nostalgia during the late 90s throughout to the mid 00s. For me it's kind of weird hearing AMD struggling back then as basically not one but ill informed novices would even consider Intel in the enthusiast space. Like 90% of gaming PCs that all my friends and local gamers in general went the AMD Athlon XP, 64 and FX series plus mostly nvidia combo, ATi being the better and more popular option in the early 00s due to the nvidia FX gen being an absolute flop. We all knew that P3 was too expensive and slower for gaming than the Athlon, and the 64/FX was just light years ahead of the cluster f**k that was the P4. It wasn't until I think about 2006 or 2007 when intel Core was compelling and offered an upgrade, that the AMD preference finally started to wane. I suppose 99% of consumers back then had no idea about the anti competitive shady stuff intel was doing back then, not like the awareness people have now. We all assumed AMD was racking it in back in those glory days of the Athlon era. I also remember the woes of the first Phenom, I bought one very early on and it was good but very much totally limited by clocks due to that design bug. The Phenom II was fantastic though, one of the best OCing AMD chips I ever had, but was too late to the market by that stage. My main desktop is still an FX8320, its one of the best PCs I've had in terms of snappiness in general everyday use even though it really lacks the modern gaming and heavy number crunching power even when it was new. But It's way faster than the i7 2500K I tried for a while in that respect, much for muchness in the other areas, and even is more snappy than the lower end Ryzen laptops I currently use which I did not expect. I am not sure why any of this is the case, I suspect something to do with the latency though the pipeline being better than early Ryzen and Intel of the same gen? But I am no expert on those older architecture. Maybe, one day I'll upgrade to that mid/high end Ryzen desktop I have been talking about for the past 5-6 years 😅
Slots, muhahahaha We have a Celeron Slot 1 CPU, we know what it is. AMD sold cash broad too, level 2 cach expansion slot, that was that. NOT on Die !!!!
Only games If you all work on the same project, all hardware needs to be compatible ! why is that so hard to understand, he is not a coder, a gamer too ?
18:04 just imagine if nvidia+AMD merged way back then (and Huang took over). Ruiz' days were ultimately numbered anyway and Huang is the much stronger CEO. AMD also started slipping come 2006, while the ATI merger was strategic, their lackluster CPUs compared to Intel's Core 2 and then also long time required to get quad core chips out to masses were really their death knell, and then intel integrated the FSB and any benefit from AMD was gone (wasn't until several CEOs later and Ryzen did they finally compete again). nvidia also really wants an x86 license and this would've solved it. But with nvidia blocked from buying ARM these days, I don't think they would've been blocked merging with AMD given the economy and regulatory environment back then. I think under Huang AMD would've had a much better chance all the way up to at least the Ryzen days - depends on how products would've performed.
Pentium 4 taking the crown from Athlon Thunderbird? haha. I beg to differ. P4 was hot, big and expensive. Pentium 3 Coppermine and Tualatin were much better as the time showed. Best wishes.
I would like to add that Globalfoundries, TSMC etc. are pure-play foundries. As such, if you were to included others like Samsung, which are IDM and also takes orders from other players, GF is hardly placed 3rd.
In the developing countries it is still impossible to find AMD powered laptops in retail chains for electronics and local internet shops. The ones available in small nambers are the premium models that cost over 1k and very few people can afford. That's why 1366x768 models with Celerons are still selling like hot cakes,since there is nothing else available in that segment. Intel is still doing shady busines and its a shame that amd is not suing them.
Please, can you tell us about Intel too?, how the company was founded and before it was founded?, and what are Intel's products from when it was founded until now that we haven't heard about and also before Intel was founded?
Any words that would properly convey my sentiment for Hector Ruiz and his time at AMD would certainly get me flagged for excessively foul language. Suffice it to say he was probably the worst thing to happen to AMD besides Raja's tenure.
So, now do one on Intel. Or, even better do two - one on Intel's history, and one on their current plans to retake the industrial technical leadership again. And then do a third one on who you think will actually be the future leaders in CPUs, with all the problems of further die shrinkage, potential for needing to move to non-silicon technologies, as well as the disruption of quantum computers coming online.
AMD is ahead of Intel in market cap right now, but Intel is still the bigger company in many measurements. Intel also has its own foundry which means they have an added security compared to AMD and other fabless companies that are dependent on third party foundries like TSMC. With today's geopolitics, AMD might even turn to Intel for production.
I can tell you are not working in the semiconductor industry and don’t understand the geopolitics between China and Taiwan. I am an engineer and work in many different US fabs. I am from Asia. Intel is virtually no way to compete with TSMC 99.999 confident. You may have Intel stocks. It will be just like IBM.
Hearing about the Cache issues with the Opteron chips, reminds me of the much more recent cache issues with Radeon 7000-series GPUs. (meaning that the top end cards were 25 or 50 % slower than they would otherwise have been.
Why did AMD totally spin off their fab as GloFo when they could've done what Intel recently did, where they spun it out but still own it. so they can take in clients and their money, and then use some of that money to help bail out CPU R&D costs?
AMD had a massive amount of debt because of the ATI buy. Revenue was also going down because Phenom could not compete against Intel CPUs. Once revenue goes down and you have issues paying debt you basically go bankrupt. AMD had to spin out their fabs and get 1 billion dollars plus Global Foundries taking in some of the debt.
I am an engineer and work in many US Fabs. AMD had a smart move. Running a fab is very expensive. Just “one” ASML advanced photo stepper/scanner costs over 300 Mils. The profit margin of the Fabs are not as good as designing the chips as AMD.
Oh, come on. Athlons (XP) slaughtered pentium IV in every possible way except sales. Netburst was an utter technology failure and was rightfully terminated and forgotten...
Paradoxically, ATI seemed like a bad buy at the time, but it was what kept them alive long enough after the bulldozer fiasco. Consoles sales kept them afloat, even with such tiny margins. If Ryzen wasn't a thing, AMD would have probably done for by now.
@jairo8746 Why you say bad buy ? They needed a partner in the GPU market, what other company you would used, Matrox, others ? U don't understand why AMD outsourced all production to TSMC ? Why they failed !
@@lucasrem It seemed like it, because none of the business they were banking on having materialized, they simply didn't get their money's worth right away, and incurred in huge debt. But it was their IP that proved to be a great investment for the future.
Early ARM guy here. Long before is was a low watt CPU, we got a BBC micro, who is able to host and ARM chip, interface for it. In 1995 all other systems failed, X86 was that cheap, it erased all competitors, Sun Digital etc, Sgi too. The IBM chip i own too, Cyrix x86 on 120 hz, no level one cash, why they never went to TSMC to produce it, that code was stolen, unable to outsource. Stable system, got all updates for it, used as a DOS system for argiving needs, in use till 2016, replaced by a Windows system. All mobile ARM developing i did in the 1990 Ish, was on apple, by then the only system we needed, low power, smaller instruction set, Windows CE Intel Strong ARM was the only survivor, but after intel, AMD stopped producing ARM chips too. the ARM community did it all by themselves, all for themselves, open market developing. Rules, intel should own the x86 code, it's theirs ! AMD should pay rights producing X86 chips. Or produce their own system.
Should have bought nvidia but the pride of the CEO (who would have to step down and lose their prestige job) made them take decisions not in the best interests of the company
AMD is welcomed for keeping the price performance pressure on Intel, but I only bought 1 AMD CPU, and 1 ATI 3d accelerator, in the last 43 years. I learned my lesson with both.
I recently bought my first ever AMD processor and I was also struggling with it, until I stopped using the "balanced" power plan in Windows 11. Set to the Performance plan it works just fine. It wouldn't surprise me at all if Microsoft and Intel have some sort of arrangement - they seem to have similar worldviews.
@@TheVanillatech Is Jim still around? It seems like he's taking a smaller role after expanding into a website and podcast. I remember watching his masterpieces like _Nanomatters_ and _Path Tracing_ years ago.
I love these historical flashbacks of these massive companies. I've learned a lot about AMD, especially it's early beginnings from this current trilogy! Will you make a fourth one covering where this video left off until present day?
You missed/misrepresented some things I think... One of the main reasons the second generation of Celeron CPUs was better, was that they had 128k on-chip L2 cache, which ran at the full CPU frequency, as opposed to 0 on the first generation and 512k off-chip L2 cache of the P2 which ran at half the frequency. So equivalently clocked Celerons were actually faster in some workloads (particularly gaming) than P2s (and much cheaper). Celerons also overclocked well, since they were made on the same process, so a 300MHz Celeron 300A (A was the cached version) was known for being able to run at 450MHz. The cache was actually a key reason the second Celeron generation was successful, not the clock speed increase. Iinitially there was only the 300A and 333, (compared to the 266 & 300 of the cacheless parts). The Pentium 4 NetBurst architecture was generally considered a flop. intel went for a very deep pipeline, to try to get high clock speeds, but sacrificing IPC. This didn't work out so well in practice. This partly opened the door for AMD to take the performance crown with the Athlon. Would have probably been worth mentioning for better context. At around the same time as the move towards 64-bit computing occured, intel created a completely new architecture (IA-64, which actually originated at HP) and its Itanium processon (colloquially known as the Itanic, due to its collosal failure), primarily for servers. These were never consumer CPUs though. IA-64 was a "very long word architecture", which relied on compilers to order instructions efficiently, but these were difficult to write and never really materialised, resulting in generally poor performance. In addition, as it was not compatible with x86, x86 code required emulation, further killing performance. The architectural difference is rather important. On the other hand, AMD extended the original 32-bit x86 architecture (later name IA-32), to the currently used x86-64 (aka. amd64), which maintained backwards compatibility (not requiring emulation). So the first consumer 64-bit CPU was the Athlon64. Intel later followed suit and also stopped producing the Itanic. After NetBurst, Intel went back to iterating on the P6 architecture (of the Pentium ii & iii), under the "Core" branding, which performed much better. So missteps from both companies were involved in the switch of performance leadership. The video doesn't really reflect this... (Actually, the situation today feels somewhat akin to the P4/Athlon days, from the consumer product side). Also, you mentioned that the lower clock speed on Barcelona CPUs was due to the cache bug (I guess specifically the TLB bug. Afaict this wasn't a problem "translating between its caches". Note the Translation Lookaside Buffer (TLB) is a very specific cache used by the MMU and the bug occurs on page entry modification - perhaps that's where this description came from?). The actual bug is kind of cache corruption issue, where the same cache line might end up residing in both the L2 & L3 cache, even though the archtecture is such that those are exclusive caches (eg. another core might accesse the data from L3, even though it "also" lives, potentially modified, in a different core's L2). In the video there's also the following statement: "That software workaround compromised performance so Barcelona failed to hit the clock speeds AMD promised..." This doesn't make any sense. It may well be that those chips had both the bug *and* were not able to hit the expected clock speeds, but the lower clock speeds would not be due to the bug (which may itself lower performance, by effectively lowering IPC).
I bought Ahtlon64 and had built lots of winapps with it. Now, knowing its history, I'm even more proud that I had that chip. Thanks for the history lesson!
I had an athalon xp 1900+ in the early 2000s, and my dad had a 2ghz p4. My athalon ran circles around his p4, it wasn't even close. my amd was more prone to overheating, but it was definitely a superior cpu.
