Slashdot Mirror
What Went Wrong for AMD's AM2?
An anonymous reader writes "When AM2 was first announced it seemed like it was going to be a guaranteed hit. After all, this platform would be moving the tremendously successful socket 939 into the future with its use of DDR2 memory, a greatly increased memory bandwidth, hardware virtualization, and a number of exciting new CPUs. Despite everything AM2 had going for it, this includes a dedicated enthusiast base and a tremendous amount of pro-AMD spirit at the time, the new platform has largely been dismissed by consumers. The question now is, what happened? How did AMD go from record growth and being the darling of enthusiasts to having a new platform which failed to impress?"
Before it gets slashdotted, or if you don't want 3 pages with ads -- here's most of the text.
--
Before we get started it should be made clearly that despite what people may say, AM2 does make for a capable computer. We took a look at an AM2 build based on an Asus M2N32 SLI motherboard not too long ago and were happy with the system. The disappointment in AM2 is not a result of its failure to perform, but rather the failure to match the performance gains seen in the move to the K8 platform. Our testing has confirmed what the industry at large has found to be true- the move to AM2 should bring performance gains of about 3-10% when compared to socket 939, with an average increase below 5%. This is what we would comfortably call an "incremental" performance boost, but nothing more.
So what happened to AM2? Where did things go wrong for AMD, a company that was on a legendary upswing, during which it could seemingly do no wrong. Even with reasonable pricing, a well-timed release date, and high availability AM2 was unable to take off in a way that was commensurate with its potential.
1. Conroe
An appreciably part of the success of sockets 754 and 939 were due to a colossal blunder on the part of Intel: Netburst. This architecture was kept around since 2001 and was always being improved in piecemeal, rather than simply being replaced. The whole episode was capped off by an unimpressive dual core architecture that was kept alive practically on price alone. During this time (754 came out in fall 2003 and 939 came in early summer 2004) AMD did their homework and put out the impressive but short-lived socket 754 and then 939.
But the landscape was changing by the time AM2's release date was announced. Intel had released its Core architecture and the word had begun to spread about Conroe, what would come to be known as Core 2 Duo. Early benchmarking by a number of hardware sites not only let consumers know that AM2 would be a slight performance increase, but that Conroe would be a dramatic one. By the time AM2 was available Core 2 Duo was one of the most highly anticipated processors of all time and AM2 was the "also ran". There was no way that AMD could compete with Intel's marketing clout, regardless of the performance or previous successes.
2. AM2 is setting up AMD for the future
As good as 939 was, it could only last for so long. AMD had to start to look towards the future, which meant moving to DDR2 memory, increasing the availability of memory bandwidth, launching a platform for improved chipsets and the like. Improvements must be done in stages: Socket 754 brought 64-bit, 939 brought dual core, dual channel memory, and mass acceptance of PCI Express video, and AM2 would bring us DDR2. AM2 may not be terribly exciting, but it is paving the way for K8L, AM3, and AMD's 4x4.
3. AM2 is confusing
Unless you follow the processor market closely, AM2 can be confusing. The naming convention "AM2" or "M2" is much different from 754 or 939 and a little investigation reveals that AM2's socket uses 940 pins. As you may recall AMD has already has a socket 940, it came out along with 754 and was used for Opteron and high-end FX systems. Despite the numerical similarity AM2 and 940 are extremely different and are not compatible with one another. Once consumers get past that they will have to figure out the processor they want, more than a few of which have the same name as their 939 counterpart.
4. 939 was too great
OK, a platform can't perform too well, but the success of 939 meant that in order to top it AMD would have to do bring something really innovative. They were clearly unable to do so (or did not intend to) so most 939 owners were never inclined to upgrade. The strong performance of 939, the availability of cheap processors and great motherboards, and the overclockability of most systems meant that convincing people to upgrade has been difficult. A new system would require a new motherboard, memory, and a CPU in the very least, possibly more if the user was upgrading from a
I think there are two reasons why AM2 isn't enjoying the same popularity as 939 systems. 1) It doesn't offer a large performance increase over 939, so those with decent 939 setups aren't encouraged to upgrade (and those that are are prepared to spend the extra for C2D). However, I do believe that will be changed a little once AMD release their 65nm core (I think it's called Brisbane), and I do believe they'll tweak the memory controller for extra performance (advantage of having it on the die). 2) Conroe. Let's face it, for a high end system, it's virtually a no-brainer. I do think that for the low end machines, the AMD product is still superior to Netburst (I built an AM2 system for my parents a couple of months back), but Conroe has pretty much wrapped up the medium-high end desktop market. Will AMD get it back? I think it depends on whether AMD can release their 65nm product before Intel releases budget Conroe-based CPUs. Once Intel release Core-based CPUs for the low end, AMD will be in a bit more trouble, IMO.
Motherboards vary by about $10 for comparable features. Seems like the AMD fanboy is just grasping for straws here...
Core 2 motherboards start at $46 (Newegg; VIA chipset) and Athlon 64 FX AM2 motherboards start at $47 (Newegg; SiS chipset).
A motherboard with an Intel chipset can be found at $66, while a AM2 motherboard with the nForce 410 can be found for $57.
The cheapest SLI board for Intel costs $78 (rebate). The cheapest SLI board for AMD costs $85 (sale). Their original prices were $97 and $95 resepectively.
LGA775-compatible CPUs start at $45 (Celeron D 326). Dual core CPUs start at $90 (Pentium D 805). Core 2 Duo CPUs start at $180 (Core 2 Duo E6300).
AMD AM2 compatible CPUs start at $41 (Semprom 64 2800+). Dual core CPUs start at $153.
Summary -- Intel motherboards are usually within a few dollars of an AMD equivalent. Budget CPUs start within a few dollars of each other. Intel dual core is cheaper. Core 2 Duo is $27 more expensive than the cheapest AM2 Athlon 64 X2, but faster.
Meaning that that Core 2 Duo E6600 still crushes that FX-62.
The enthusiast market.
AMD AM2
Can I use my DDR memory on AM2 - NO
Can I use my AGP video card - Yes (Asrock do have an AM2 board with AGP bit still needs DDR II)
INTEL P4 805 Dualcore 2.66GHz with Asrock Dual-VSTA (Cheap board circa £45)
Can I use my DDR memory - YES
Can I use my AGP video card - YES
Can I switch to DDR2 later without changing the motherboard - YES
Can I switch to a PCIE video card without changing the motherboard - YES
Can I run both DDR and DDR II at the same time - NO
Can I run both PCIE and AGP at the same time - YES (With supported cards)
Worse for AMD the Intel solution is CHEAPER and overclocks like a mad beast..
Go on try and find an X2 AMD processor for less than £70 inculding tax and delivery.
You have (miracle) well find a board that will take that processor and my 1Gb of DDR and my AGP Radeon 9550 for less than £45 will you ?
They had the oppourtinity when deciding on the spec of the AM2 interface to include a more flexible memory controller setup and allow the motherboard manufaturers decide what style of memory to equip their boards with but they didn't.
Way to loose the enthusiast market - make things too expensive AND FORCE you to upgrade your memory at the same time.
WAY TO GO AMD
I actually bought myself an AM2 4200, after deciding I needed a new system (I used to have a 2.4 XP) - and seeing it was going to be the new thing, and that it wasn't much price difference. Perhaps at least I can offer some thoughts on actually having used it everyday and built it myself. Its a wonderful system imho. So what if it's not 5% faster than the previous model? Its not any more expensive. At the very least the thought and design gone into the CPU Mounting is great, no more fiddling with stupid heat-sink clips, its a nice clamping system that feels solid. It runs Windows XP, Vista and Gentoo fantastically smoothly (Yep, I've tried all three! I use Gentoo normally.) - and seems to do it better than my Intel 939 at work, which is meant to be faster. It also overclocks like an absolute dream! I can squeeze 8% overclocking on it without a problem. Not just that, but my nice AM2 Motherboard will support an AM3 processor. Hows that for upgrading? C'mon guys, just cos its the new thing - and especially after all the chipset problems with 64 bit systems, this is a nice system. I'd never go back to Intel after using it, personally I can't stand the 'Duo Core', even if it is 1% faster or whatever. Whats going to happen when Intel bring out the next big thing? Dug
Admittedly, 4x4 is dumb. I would consider that to be a quad-processor machine with 4 cores each. But they consider it a dual-core dual-processor machine.
AM2 is, I believe, the socket type. You know, instead of A,754,939,LGA775 (Intel).
x.xGHz+ is something you made up. They had xx00+, which was used for marketing so people knew what it compared to in an Intel processor, since their processors ran at lower clockspeeds than the Intel competition.
It's like sex, except I'm having it!
Why should anything be wrong with the AM2 platform? Nothing. It is just an evolutionary step for the AMD.
I agree, nothing went wrong. The only point that I think that the article was right on was cost. Since the performance difference between the current generation of AM2 processors and 939 processors is so small (or almost negligible), the average consumer is buying based on price. And since 939 processors and systems are still available, though less "desirable" from being "older tech", the prices are usually better on 939.
There isn't anything confusing about the naming conventions of the CPU, at least not any more confusing than they were on 939. The AM2 versus 940 "confusion" statement is a red herring, because they're not both called socket 940, and I have yet to see an AM2 product advertised as 940.
The only real "problem" that AM2 has is that for the average consumer who buys a PC, then throws it out and buys a new one in 3 years, is the fact that AM2 doesn't add anything spectacular in this generation of chips. Sure, hardware support for virtualization is great for those few of us professionals who use virtualization, but the average user doesn't care. And while DDR2 is the new standard, and will undoubtedly have more benefit in the future at higher speeds (and with later revs of the AM2 CPU's memory controller), it wasn't really necessary at today's performance levels. The Athlon 64 went from a a very low latency (due to the integrated memory controller), medium bandwidth memory technology to a medium latency, high bandwidth memory technology. You wouldn't expect much change, except in extremely bandwidth-limited situations. On the other hand, the Intel line of CPUs went from a medium latency medium bandwidth memory techology (using an external memory controller) to a medium latency high bandwidth technology, so you would expect an improvement.
Back in the day if you're shopping for CPUs and come across a P3 933, you instantly had an idea of the chips performance, at least enough to say well that's probably a bit faster then an Athlon 750. I'm sure some nitpicking AMD fanboys will argue and say it wasn't, but lets face it 933 > 750.
Your own example is the very reason that AMD "broke" the naming scheme. It was because idiot consumers like yourself were apparently incapable of making the leap of logic that "clock speed" != "performance." Since Intel was aggressively pushing clockspeed while AMD was pushing the operations per cycle, this would leave AMD at a great marketing disadvantage. So they named their chips with numbers represented the clock speed of the Intel chip they roughly performance-competitive with. In reality, you got what you wanted - numbers that represented performance, not just clock speed.
If you don't know where you are going, you will wind up somewhere else.
There is also HD video playback, HD video editing, currently, people are asking me about this and I keep telling them the technology is coming and there is no reason to update because your pc needs to be hdmi ready
Anything that needs HDMI will not be edittable. HDMI is only necessary to support DRM, not to support any technical requirements of HD and DRM, as its backer's perceive it, precludes editting.
I play back and edit raw HD transport streams on my AGP system all the time and since they have no DRM it works just fine. Even some of the DRM'd stuff (notably MS WMV9 encoded stuff like Terminator 2, Brothers Grimm, and a whole slew of IMAX titles) all play back 'fine' on the current system, at full rez.
When information is power, privacy is freedom.
This was just a rumor and has subsequently been said to be near impossible.
http://arstechnica.com/news.ars/post/20060713-726
peak power use (just okay)
The core 2 duo has a TDP of 65W (75 for the "Extreme"). The X2s had 85, then 65, and now a mere 35W, or basically half of the core 2 duo.
memory bandwidth and latency is still behind X2s, but I know you don't really care about those "facts"
Well, considering I mentioned them, you might not want to assume quite so much...
And yes, I called them "abysmal", because they haven't even caught up to the X2, despite having a year and a half since the X2s came out to play catch-up.
Poor overall performance
Would you like links to a similarly stacked test showing the X2s trouncing the C2D?
Here, how about one that matches a few of your linked benchmarks, yet shows the opposite result? Funny how that works.
Again, Intel has merely "caught up". Performing on par with an 18-month old chip, I consider a poor showing indeed - particularly when you consider that, while Intel has played its hand, AMD (beyond releasing lower power versions) still hasn't fully revealed its next gen yet.
You need to pull AMD's dick out of your mouth and try again.
Did I mumble?
http://www.eecg.toronto.edu/~moshovos/ACA05/read/A kkary.1998.MICRO.pdf T R2208.pdf s calar.pdf
http://www.crhc.uiuc.edu/~mfrank/pubs/Malik-2006-
ftp://ftp.cs.wisc.edu/sohi/papers/1995/isca.multi