Slashdot Mirror
The Chip That Changed the World: AMD's 64-bit FX-51, Ten Years Later
Dputiger writes "It's been a decade since AMD's Athlon 64 FX-51 debuted — and launched the 64-bit x86 extensions that power the desktop and laptop world today. After a year of being bludgeoned by the P4, AMD roared back with a vengeance, kicking off a brief golden age for its own products, and seizing significant market share in desktops and servers."
Although the Opteron was around before, it cost a pretty penny. I'm not sure it's fair to say that the P4 was really bludgeoning the Athlon XP though (higher clock speeds, but NetBurst is everyone's favorite Intel microarchitecture to hate). Check out the Athlon 64 FX review roundup from 2003.
Those were the good old days. How I miss when it took me one day at most to learn about all options I had to build a gaming computer, with enough detail to make an informed decision about what bits and pieces to built it with.
Nowadays just piercing the veil of lies, half truths, false reports and bought reviews, makes the entire process incredibly boring and frustrating.
They swooped in when Intel was being stupid, made the best chips in the world... then committed suicide and haven't built a competitive chip in 3 years. Sad times...
It seems we've lost a lot of quality in the comment fields in the last 10 years. Lots of expertise modded up carefully; now we seemingly have opinion-pieces moderated up by whichever group happens to be awake at the time, and the real expertise is hidden in the +2 or below.
I only just replaced my Athlon 64 motherboard and processor this spring. It was a good product, but not quite up to running Windows 8 IMHO.
W..w..W - Willy Waterloo washes Warren Wiggins who is washing Waldo Woo.
10 years later and we're still running games and applications that are 32bit that only use a single core.
Too bad AMD was just sitting on their laurels after that. Incidentally, in 2 more years, you can start making your own Pentium Pro compatible processor without violating any patents (assuming you're using the same patents that went into the Pentium Pro).
Wow. Ten years. And here I am still dealing with 64 bit incompatability issues every six months or so.
Out of curiosity, how long did 16bit library problems linger after the 32 bit move?
May the Maths Be with you!
for gaming, the GPU took over most of the work which is the way it should have happened
for applications, most don't really need 2 cores. even running multiple apps at the same time you don't really need 2 cores. i was playing MP3's on a computer in the 1990's with minimal CPU usage. there is no way you need to dedicate a whole core to music while surfing the internet. or some of the other idiotic use cases people make up
Fair point and the follow up is WHY?
AMD still makes a better chip for many FP intensive applications, and the price is still superior to Intel to boot. Intel always made a big deal about clock speed, while AMD worked on actual performance. It is really a shame that people pay more attention to marketing than real performance.
-The wise argue that there are few absolutes, the fool argues that there are no probabilities.
Must we be so violent? This is CPU sales not some barbaric conflict.
I was hoping to find a current review of the processor against current CPUs....
However, in AnandTech bench you can compare an AMD Athlon X2 4450e (2.3GHz - 1MB L2) with current CPUs. If you compare this to an Intel Core i7 4770K (3.5GHz - 1MB L2 - 8MB L3, one of the best CPUs right now), you can find that the Intel CPU is between 3 times faster and 9 times faster. Most of the times is about 6-7 times faster.
See http://www.anandtech.com/bench/product/37?vs=836
However, if you could compare an AMD FX-51 with an Pentium 66 Mhz (best CPU in September 1993), I think that the difference would be way greater.
CPU process is currently focused on efficiency and lower power. However, in the ARM field, you can still find progress in CPU performance.
You mean modifying a game outside of its intended scope makes it work in unexpected ways?
That's so weird.
10 years later and we're still running games and applications that are 32bit that only use a single core.
At least 64-bit OSes are widespread now.
Almost ten years after the 80386 was introduced, most people were still running "OSes" which were little more than GUI shells running on 16-bit DOS.
And, of course, AMD is a much cheaper option to get ECC on the desktop. Which is less and less a luxury and more and more a necessity. Simply because the memory densities keep going up, yet the MTTF doesn't go down as much (if at all). Same problem with hard drives, actually, so RAID doesn't really cut it any longer. Need parity files on the disk itself as well as at least mirrored disks.
Still and all, this sort of reminiscing makes me long for alpha and parisc. ARM isn't quite there yet, and mips isn't available in the top performing brackets any longer either. And POWER? Well, few people can take that much gouging.
As the author of the article:
In 2000 - 2001, the Athlon / Athlon XP were far ahead of the P4. But from Jan 2002 to March 2003, Intel increased the P4's clock speed by 60% and introduced Hyper-Threading. SSE2 became more popular during the same time. As a result, the P4 was far ahead of Athlon XP by the spring of the year in most content creation, business, and definitely 3D rendering workloads. Now it's true that an awful lot of benchmark shenanigans were going on at the same time, and the difference between the two cores was much smaller in high-end gaming. But if you wanted the best 'all around' CPU, the P4 Northwood + HT at 2.8 - 3.2GHz was the way to go. Northwoods were also good overclockers -- it was common to pick up a 2.4GHz P4 and clock it to 3 - 3.2GHz with HT.
Athlon 64 kicked off the process of changing that, but what really did the trick was 1). Prescott's slide backwards as far as IPC and thermals and 2) The introduction of dual-core. It really was a one-two punch -- Intel couldn't put two Pentium 4 3.8GHz chips on a die together, so the 820 Prescott was just 3.2GHz. AMD, meanwhile, *could* put a pair of 2.4GHz Athlon 64's on a single chip. Combine that with Prescott's terrible efficiency, and suddenly the Athlon 64 was hammering into the P4 in every workload.
On high end games, the CPU gets hit hard. AI, physics, etc, all need a lot of power. Battlefield 3 will hit pretty hard on a quad core CPU, while hitting hard on a high end GPU at the same time.
AMD, forgotten by most of you, purchased a CPU design company not long after it lost the right to clone Intel CPU designs. The people from this company gave AMD a world beating x86 architecture that became the Athlon XP and then Athlon 64 (and true first x86 dual core), thrashing Intel even though AMD was spending less than ONE-HUNDREDTH of Intel's R&D spend.
What happened? AMD top management sabotaged ALL future progress on new AMD CPUs, in order to maximise salaries, bonuses and pensions. A tiny clique of cynical self-serving scumbags ruined every advantage AMD had gained over Intel for more than 5 years afterwards. Eventually AMD replaced its top management, but by that time it was too late for the CPU. Obviously, AMD had far more success on the GPU side after buying ATI. (PS note that ATI had an identical rise to success, when that company also bought a GPU design team that became responsible for ALL of ATI's world-beating GPU designs. Neither AMD nor ATI initially had in-house talent good enough to produce first rate designs.)
Today, AMD is ALMOST back on track. It's Kaveri chip (2014) will be the most compelling part for all mains powered PCs below high-end/serious gaming. In the mobile space, Intel seems likely to have the power-consumption advantage (for x86) across the next 1.5 years at least. However, even this is complicated by the fact that Nvidia is ARM, and AMD is following Nvidia, and is soon to combine its world beating GPU with ARM CPU cores.
At this exact moment, AMD can only compete on price in the CPU market. Loaded, its chips use TWICE the power of Intel parts. In heavy gaming, average Intel i5 chips (4-core) usually wallop AMD's best 8-cores. In other heavy apps, AMD at best draws equal, but just as commonly lags Intel.
Where AMD currently exterminates Intel is with SoC designs. AMD won total control of the console market, providing the chips for Nintendo, Sony and Microsoft. Intel (and Nvidia) were literally NOT in the running for these contracts, having nothing usable to offer, even at higher prices or lower performance.
AMD is currently improving the 'bulldozer' CPU architecture once again for the Kaveri 4-core (+ massive integrated GPU and 256-bit bus) parts of 2014. There is every reason to think this new CPU design will be at rough parity with Intel's Sandybridge, in which case Intel will be in serious trouble in the mains-powered desktop market.
Intel is in a slow but fatal decline. Intel is currently selling its new 'atom' chips below cost (illegal, but Intel just swallows the court fines) in an attempt to take on ARM, but even though Intel's 'atom' chips are actually Sandybridge class, and have a process advantage, they are slaughtered by Apple's new A7 ARM chip found in the latest iPhones. A7 uses the latest ARM-64 bit design known as ARMv8, making the A7 and excellent point of comparison with the original Athlon 64 from years back.
Again, AMD is now x86 *and* ARM. AMD has two completely distinct and good x86 architectures ('stars-class' and 'bulldozer-class'. Intel is only x86, and now with the latest 'Atom' has only ONE x86 architecture in its worthwhile future lineup. Intel has other x86 architectures, but they are complete no-hopers like the original Atom family, the hilariously awful Larabee family, and the putrid new micro-controller family. Only Intel's current sandybridge/ivybridge/haswell/new-atom architecture has any value.
Damnit people, do not give Adobe's Flash developers any ideas!!
Well, there's spam egg sausage and spam, that's not got much spam in it.
so i was already feeling stoked about finally getting around to finding a matched pair of the fastest cpus that I can put into this board that's been sitting in a box for SIX YEARS, they boot and now I read this. /me - does the peacock strut happy dance thing.
i was always a fan of AMD going back to the 8x300 bit slice stuff. they're clever boys.
Need Mercedes parts ?
Apple just released a 64bit processor, and now AMD is copying it TEN YEARS ago?!?
Can the industry please do something original and quit just following wherever Apple leads it?
-Lod
The misleading thing about benchmarks is that they're generally prebaked - there's no chance for "surprise" physics interactions or various pipeline-stalling things that tended to trip up the Pentium 4. From personal experience I'll tell you that my old 2.8 GHz Pentium 4 generally didn't do as well as my Athlon XP 2400+ in Doom 3, Bioshock, or Unreal Tournament 3. The latter two should have been poster children for the Netburst chip by comparison. Also, the Pentium D 820 was a 2.8 GHz chip: it was the miserably hot 130W TDP 840 that ran at 3.2 GHz. But you're correct on the other counts - the higher IPC and integrated memory controller were both HUGE advantages over a latency-crippled, deeply pipelined architecture. The Pentium D was itself a flailing, mostly failed response to the surge in mindshare the Athlon 64 X2 created until the Core architecture could be prepared.
Comment removed based on user account deletion
The need for computers that can run multiple programs concurrently with a total of > 4GB RAM is more than the need for any single program to consume multiple cores or > 4 GB RAM.
Not for much longer. DICE has already announced that some of their new Frostbite 2 games will be 64-bit only, due to memory requirements beyond 2 GB.
Not that it mattered to the P4-area contest(where desktop OSes and workloads would remain 32 bit for quite some time to come, and RAM fairly expensive); but the A64 was a real smack in the face for IA64...
Intel has their grand, big-iron-class, future-of-enterprise-computing 64-bit architecture, then AMD pops up "Hi guys, who wants a 64-bit CPU, fully backwards compatible with your 32-bit x86 code and pretty damn fast at that, for only slightly more than the price of a nice desktop CPU?"
Boom. Headshot. Game over man, game over.
The overclockable 2.4-2.8 Northwoods kept up on the 32 bit side at the time, and Intel has since swallowed their pride and put out some genuinely brutal 'EMT64' parts; but IA64 was buried beyond hope.
When I'm waiting for an application to do whatever that application is doing, and that application is only using one core, then yes, I really do need it to use more than one core.
To suggest otherwise is also to suggest that computers are fast enough, and that general-purpose computing is a solved problem.
I don't think we're anywhere near that point just yet.
Kid-proof tablet..
I mentioned that Caldera actually sued MS based on the fact that Win9x was still based on DOS in my blog article on the OS/2 2.0 fiasco, because OS/2 never depended on DOS.
Wow. Ten years. And here I am still dealing with 64 bit incompatability issues every six months or so.
Out of curiosity, how long did 16bit library problems linger after the 32 bit move?
16 to 32 was a much more radical change. Segments to flat. In the Unix line, this happened in the early 80's (late 70's?) when few systems were deployed.
In the Wintel line, it was also cooperative to preemptive. Very painful. Very manual. It took 10 years just to let go of 16 bit device drivers and many were never ported.
Classic Mac, Amiga, and Atari ST had an easier time since their "16-bit" systems were already 32-bit internally. Even then you had a few years of dealing with geniuses who stored data the upper 8 bits of pointers and some significant software (like AmigaBasic) were never ported.
Today, I do not see any apps in general use that need or require access to memory beyond a contiguous block of data beyond 4GB (frankly far far less).
Then you're not looking very hard. There are plenty of games that fall over as soon as you install enough mods that you go over 2GB (the maximum amount of RAM the majority of 32-bit apps are allowed to access in Windows). There are also games which install crappy versions of textures in 32-bit mode because they would otherwise hit the 2GB limit.
PAE is a disgusting kludge. There is simply no reason not to run 64-bit apps on a 64-bit x86 OS... unless you're stuck with a proprietary antique like Windows.
Ok, I'm a Java developer, working on Linux. 99% of benchmarks you see on-line are benchmarks of some stuff on Windows, which is immediately skewed towards intel because lots of Windows DLLs are built with Intel compiler which disables optimizations and advanced features on non-intel CPUs, even if those features are supported.
Also, most benchmarks are some synthetic benchmarks (compiled with Intel compiler), or some 3D games, or some video transcoding. I do none of that, and what I do I do not do it on Windows. I am yet to see a site that benchmarks Java compilers, Java IDEs, databases, application or Java web servers, etc. on Intel vs AMD vs ARM platforms on Linux. The only site that comes close is Phoronix. And if you look at their Linux benchmarks, difference between AMD and Intel CPUs is much less than on Anandtech or Toms Hardware. Intel is still making faster CPUs, but not that much faster.
--Coder
Today, I do not see any apps in general use that need or require access to memory beyond a contiguous block of data beyond 4GB (frankly far far less).
Lacking inside information, I could be wrong, but I believe you'll find Photoshop, Premier Pro, After Effects, Solidworks, PTC Creo, various FEA packages, and CAM applications all use blocks of RAM beyond 4GB. And that just a sample of MY machine.
More knowledgeable folks (i.e. coder's for the above apps) can correct me if I've got it wrong. But there is clearly a need for >32bit memory address space.
Place nail here >+
Comment removed based on user account deletion
Sorry, I do not have the time to educate you here on the word "contiguous" in context to OS memory management. Also, I preambled my comment re: memory access not instructions.
Maybe you should spend more time reading than 'educating'. My post was specifically about memory usage, and 'contiguous' is irrelevant to software, because the MMU makes it look contiguous. All Windows gives a 32-bit app is 2GB of RAM, unless the app and every single DLL it loads has the 4GB-aware flag (good luck with that).
Many, many 32-bit Windows apps are hammering hard against that 2GB barrier right now.
Oracle's Enterprise database costs $47,500 per processor core. There is no way in heck that I'd choose AMD over Intel when I have to run more cores to get the same performance.
Microsoft SQL Server Enterprise costs $6,874 per processor core.
AMD has a heavy investment in the server space. They should negotiate lower per-core license costs in these cases; license parity with Intel is throwing them out of the data center.
As the developers of x86-64, they should have a patent portfolio to do serious damage to 64-bit x86 systems vendors. Use it.
Subject pretty much says it all. Oh well, moving on.
The point is the new register set. Registers being wider is a happy side-effect, as is greater virtual address space. But the main point of AMD64 is more registers. and it started a sequence of ISA extensions that have dramatically improved compute-bound throughput via SIMD.
Absolutely right. And Intel really wanted the Itanium to succeed, because only Intel could build Itaniums and they really really wanted to be a sole-source for processors. That way they could more easily jack the price up.
I'm very glad AMD was able to cut the legs out from underneath the Itanium.
So the focus on consoles in the industry, plus the lingering zombie life of Windows XP, had basically kept everyone stuck at Direct3D 9/Shader Model 3 and 32-bit for years. With next-generation consoles moving to Direct3D 11/Shader Model 5 and 64-bit, we should see more PC titles that take advantage of the latest generation PC hardware. BTW, for some technical background on 64-bit and gaming, see http://www.gamasutra.com/view/feature/3602/sponsored_feature_ram_vram_and_.php
I don't remember this, and can't find it noted. Can you link to where you saw that? I've been using tom's hardware for years.
Most don't really need two cores, but that's not a reason not to want two cores.
I fell in love with multiple core processors when I first got one, not because my computer in general became faster (I'll bet that all but one of my cores are idling most of the time) but because my computer wouldn't get unresponsive when I was doing computationally heavy tasks (or programs crashed).
While you may have the benchmark you like the best that you think is wonderful, probably because it shows your favoured product in a good light, big numbers aren't what most people care about. They care about wall time of execution, or FPS, or the like of what they run. Well that varies person to person, but it is actual applications they care about.
That's the reason I have an Intel CPU. For the programs I use, it does better. I don't really care why. You can argue that it is lack of optimization, hating AMD, Intel nefariousness, whatever. Doesn't matter, what matters is it works faster for me. That is what I care about.
I don't know about xbox 360, but the ps3 is most definitely using a 64bit cpu, i was running a powerpc64 linux distribution on it.
http://spamdecoy.net - free throwaway anonymous email - avoid spam!
The instruction set itself was an yawner - I was looking forward to 64-bit being the point where all CPUs become RISC, and where Windows NT could go from being Wintel only to NT/RISC.
However, one delicious piece of irony that I love about the Opteron/Athlon 64 is that this was the architecture that sunk the Itanic. If the Itanium sank far worthier chips before it - PA-RISC, DEC Alpha and MIPS V, this architecture brought out the Itanic in Itanium. Originally, the Itanium was supposed to be the 64 bit replacement for x86, but thanks to this gag from AMD, it never happened. Instead, AMD started stealing the market, and to add insult to injury, when Intel tried entering w/ 64-bit extensions of its own, Microsoft forced them to be AMD compatible. So that Intel was ultimately forced to let x64 be the successor to x86, and let Itanium wither on the vine.
Once that happened, Itanium followed the same path as the better CPUs that it killed above. Microsoft dropped support for it after Server 2008 and XP or Vista were never supported, Monterrey collapsed and to add insult to injury, even Linux - the OS that boasts about being ported everywhere - didn't want to remain supported on the Itanic. Today, the Itanic has as many OSs as the DEC Alpha had at its peak - 3: HP/UX, Debian Linux and FreeBSD.
So no, the x64 didn't change the world. But it sure sunk the Itanic!
I can't mod this up, so have a +1 thank you.
EnBoost my friend. Check it out on skyrim nexus. Works wonders if you have a 3d card with a lot of ram, but it helps a lot with stability even on a 1gb card!
Tom's got nailed over the "burning" AMD Athlons. If you dig through the HardOCP forums from back in 2001-2002, Kyle Bennett and others managed to trace some not so savory relationships with Intel and the site.
I've never trusted Tom's since and they have a well established Intel bias. HardOCP is much more transparent in it's testng.
[RIAA] says its concern is artists. That's true, in just the sense that a cattle rancher is concerned about its cattle.