Editors’ note: Starting today, we are introducing the Deep Throat section. In this section we are going to feature industry insiders who will publish their "no holds barred" opinions. Our first Deep Throat commentator is only to be known as Scali, a software developer knee-deep into the development of current and next gen software millions of people use and enjoy in. This comment is a repost of the original entry, while in the future we’ll both repost selected blog content and publish BSN*-unique articles.
DISCLAIMER: The views presented in Deep Throat sectionrepresent only the personal opinions of the authors. Members of BSN*may or may not agree with the statements expressed in this article.
Deep Throat – Are all AMD fans idiots?
Yea I know, the title sounds a bit like a post on Linux a while ago, doesn’t it? Well, that’s very appropriate, because it’s a similar case of revisionist history by people probably not old enough to ever have witnessed the events at the time.
It’s a pet peeve of mine – the Internet is filled with AMD fanboys, thinking they’re some kind of expert on CPU design and all that. I can understand it up to a certain point? I mean, Intel is that big corporation that people love to hate, and they’re rooting for the underdog, which is AMD of course. I myself have used a long list of non-Intel processors over the years, so I know all about the alternatives to Intel processors, and how they can be better. But I can’t stand people who just don’t bother to check their facts, just like with the nonsense about Linux/Unix and how it was ‘secure by design from the start’, and all that.
I think the whole AMD fanboy movement started with the success of the Athlon. I bet most AMD fans never even heard of AMD before the Athlon, or in fact haven’t even owned a PC before that time. That is the only way in which I can explain their delusional idea that Intel and AMD are somehow each other’s equals in a technological sense, and how they leap-frog over each other, trading the performance crown back and forth.
Clearly, anyone who bothered to study the history since the beginning of Intel’s 8086-range will know that AMD started as an independent seller of x86 processors with the Am386 [after having been a second source for Intel with the 8086 and 80286 for years], and that they did this in 1991. Put this in the proper perspective: Intel released the original 80386 in 1985(!), and released the 80486(!) in 1989. So from the get-go, AMD was about six years behind Intel, with a gap of more than a generation.
One often hears the fairytale that AMD sold much faster 486 derivatives than Intel, so AMD must have had a technological advantage over Intel. While it is true that AMD sold 486 derivatives up to 133 MHz, while Intel’s fastest was only 100 MHz, this has to be put in the proper perspective as well: AMD’s first Am486 was introduced in 1993, actually a month AFTER Intel had introduced the Pentium, which may not have had higher clockspeeds at the time, but the Pentium had far higher performance per cycle, especially the FPU was an incredible deal faster than the outdated design of the 487. In fact, AMD didn’t actually introduce those 100+ MHz 486s until 1995, while Intel released its last 486 in 1994.
So what really happened was that AMD basically was selling overclocked 486 processors as their high-end, while Intel had a much more advanced architecture which delivered much better performance, even at considerably lower clockspeeds. Clearly Intel wasn’t even interested in selling high clockspeed 486 processors, as they would only threaten Pentium sales. And of course AMD was still a generation behind technologically, so the fact that they eventually had a 133 MHz 486 in 1995 doesn’t mean much. Intel offered 133 and 150 MHz versions of the Pentium by then. So not only could Intel match AMD’s clockspeed, but Intel’s processors were MUCH faster at those clocks. In fact, even if we look at the fastest processor that you can put on a 486 socket, it’s not AMD’s, it’s still Intel’s. Intel even offered a Pentium Overdrive processor http://en.wikipedia.org/wiki/Pentium_OverDrive for the 486 socket [although not all motherboards could support it]. It was a true Pentium processor at 83 MHz, complete with the superscalar architecture with the U and V pipelines, the large caches and the massively improved FPU. I’ve actually used it in my home server for a few years, running FreeBSD.
That’s pretty much the story of AMD all around. Usually their CPUs were a generation behind, and also their manufacturing process was usually one node behind that of Intel. In fact, the entire success of the Athlon architecture is partly due to them being a generation behind. While Intel moved on to the Pentium 4 aka Netburst architecture, AMD was still working with an architecture that was closely related to Intel’s P6 architecture, as used in the Pentium Pro, Pentium II and Pentium III.
Netburst didn’t quite work out, and as a result, Intel had never killed off their P6 architecture completely. They used it in the Pentium M line for mobile devices, as the Pentium 4 just drew too much power [even the Pentium 4M derivative was useful for desktop replacements at best]. Overclockers already knew it, and Intel must have known as well: If you overclock the Pentium M (or later the Core Duo), you get performance very similar to that of the high-end Athlons and Pentium 4s.
So, basically Netburst was just an anomaly. If it was ?business as usual’ with Intel, then AMD would never have been able to touch Intel’s high-end, as their CPU architecture was a generation ahead. And if Intel had stuck with a P6-like architecture, as AMD did with the K7 and K8, then Intel would have had performance and power consumption much closer to AMD’s than they did with Netburst.
And that’s where Core 2 microarchitecture comes in. Intel took a bit of P6, a bit of Pentium 4, and a bit of ‘new’, and they went right back to where they always were: a generation ahead of AMD, and AMD unable to compete with Intel’s high-end performance. The most surprising part here is that Intel didn’t even make use of an onboard memory controller yet. They didn’t need that to outperform AMD’s processors, because the entire architecture was so good [except for multi-CPU systems].
As AMD struggled to get their quad-core answer out in the form of Phenom, Intel worked on a new architecture which finally did leverage an onboard memory controller, and also recycled some of the remaining Pentium 4-technology in the form of Hyper-Threading [as lackluster as NetBurst may have been in many aspects, Hyper-Threading and SSE2 were very nice technologies and will likely be with us for a long time]. As a result, Intel maintains its lead of an entire generation over AMD, and the performance gap became ever larger. Now the multi-CPU problem is also solved. This leaves AMD competing with 6 cores against Intel processors with ‘only’ four cores in the server/workstation market, because the combination of Intel’s more advanced architecture and the reintroduction of Hyper-Threading just deliver more performance per cycle per core. Business as usual.
AMD having higher clocked or better performing CPUs? Outside the Athlon/PIII/P4 era this pretty much never happened, the whole leap-frog thing is a myth. Today AMD is pretty much where they’ve always been, except for that one anomaly. At this point it seems more likely that AMD goes bankrupt than that AMD will once again compete head-to-head with Intel in the high-end market. AMD was always about bargains, bang-for-the-buck, low-end to mainstream systems.
Makes me wonder, do those AMD fans even realize that AMD wasn’t the first, the only, or even the best Intel alternative most of the time? Me, I have nothing against AMD, I’ve used their processors from time to time. I actually had an early Am486DX2-66 back in 1994, and I’ve never had a Pentium 4 myself, I used Athlons through that era, before going back to Intel with Core 2, for obvious reasons. But I’ve also had an IIT 387 coprocessor, which had some very nifty tricks over a regular Intel one, like having 4 stacks of registers rather than 1, which allowed you to fit whole matrix*vector operations on2 stack, for example. I’ve also had a Cyrix 6×86 for a short while, which was a better Pentium alternative than AMD’s K5 at the time. In the end I have to admit I went back to the real Intel Pentium though, because the Cyrix and AMD both were comparable to a Pentium only with integer operations. The FPU on those things was about as weak as a 486, nowhere near a real Pentium. I also had an IBM/Cyrix Blue Lightning 486 at some point. And of course there was the IDT’s WinChip, which VIA eventually bought and reworked into own C-series architecture [Centaur] – just like the AMD bought NexGen and created K6 and Athlon being partly reworked Digital Alpha technology [co-architect of DEC Alpha 2100 is now CEO of AMD, Mr. Dirk Meyer]. And what about the NEC V20/V30? Some of the earliest x86 alternatives, way back in the 8088-era. There’s more, but the point is that AMD wasn’t even Intel’s main competitor, until after all other competitors had given up. I have to give AMD credit for being so persistent, though.
Don?t take my word for it though, it?s all on Wikipedia.