The Server/Workstation Market: AMD's Position
Question: Over the past 10 years, Intel and AMD have gone back and forth in the server market. Most recently, Intel demonstrated the benefits of its Nehalem microarchitecture with Nehalem-EX in the MP space. Before that, Nehalem-EP and Westmere-EP showcased the architecture's benefits in 1P and 2P configurations. AMD recently told us it was no longer focusing on the workstation space. Has AMD lost the momentum it generated with Opteron back when Intel relied on NetBurst for driving Xeon, or are motherboard vendors still seeing new AMD Opteron offerings as a competitive (profitable) architecture in the server space?
- AMD's new 8-core and 12-core Opteron processors are very robust, and compete well with Xeon. At the platform level, AMD server platforms now have features and performance that rival Intel.
- They have lost their momentum. The Nehalem-based Xeons have taken a huge gain this year, and we don't see AMD regaining the server space anytime soon.
- While it is true that AMD missed out on a large server buy-up recently, they still have a very price-conscious product, and that will do well in this economy.
- There is no reason to suspect this is anything but normal. AMD and Intel have always been at war. Even though we see Intel as rather successful in the [1P and 2P server] market, they have made mistakes (i.e. IA64), and so it is good to have a competitor to help drive prices and technology.
Looking back over the past four years, an increasing number of businesses had to deal with slashed budgets. IT departments were no exception. Server purchases--particularly in the 1P and 2P space--languished, along with a slowdown in general business PC sales. The former were, for the most part, simply budget-constrained. The demand was still there, but purchases were delayed, rather than outright diminished. Earlier this year, this proved to be prophetic, as we saw a heavy uptick in low-end server sales ($25 000 and under) with a revenue increase just shy of 33%*. Mid-range and high-end server sales at the margin (profit per unit) have seen a decline, as we see more scalability- and performance-driven systems enter the market, thanks to new technological growth. We have to imagine that the budget cutbacks help reinforce some of this "migration."
Particularly in Q2, we saw a proverbial micro-flood of sales for volume servers. Intel was in a great position here with its new Xeon 5600-series. In retrospect, rather than being the direct cause of its success, Intel enjoyed good sales due to an inevitable IT server refresh. AMD lost some of its usual server momentum by being late to the game, which seems almost reminiscent of Nvidia's low-end to mid-range Fermi solutions. If AMD's July earnings conference call is any indication of where things need to go, CEO Dirk Meyers is well aware the company is behind Intel and just starting to nip at its competitor's heels.
The blame, according to the company, lies with OEMs, who were slow to bring new designs to the market. Frankly, the blame can't be offloaded completely. It's likely an even split between AMD and its partners. The company knew that Intel was reaffirming its market share by building on the success of Xeon 5500, and it chose to scrap backward compatibility (Socket G3 and G3MX) in order to achieve better-scaled performance through redesigning and reorganizing its Opteron brand, implementing two multi-core dies. The cost of implementing a new socket (G34) no doubt slowed down the speed of introducing the company's new server processor lineup. With the recent economic landscape, OEMs were weary to bring a new solution to market at such a fast pace, particularly since this meant nearly scrapping older AMD server platforms.
|Opteron 6100-series, Eight-core "Magny-Cours" (45 nm)|
|Opteron 6128||2.0 GHz||8 x 512KB||2 x 6 MB||3.2 GHz||10x||1.3 V||80 W||115 W||G34|
|Opteron 6134||2.3 GHz||8 x 512KB||2 x 6 MB||3.2 GHz||11.5x||1.3 V||80 W||115 W||G34|
|Opteron 6136||2.4 GHz||8 x 512KB||2 x 6 MB||3.2 GHz||12x||1.3 V||80 W||115 W||G34|
|Opteron 6100-series, Eight-core "Magny-Cours" (45 nm), High Efficiency|
|Opteron 6124 HE||1.8 GHz||8 x 512KB||2 x 6 MB||3.2 GHz||9x||1.2 V||65 W||Row 5 - Cell 8||G34|
|Opteron 6128 HE||2.0 GHz||8 x 512KB||2 x 6 MB||3.2 GHz||10x||1.2 V||65 W||Row 6 - Cell 8||G34|
|Opteron 6100-series, Twelve-core "Magny-Cours" (45 nm)|
|Opteron 6168||1.9 GHz||12 x 512 KB||2 x 6 MB||3.2 GHz||9.5x||1.1875 V||80 W||115 W||G34|
|Opteron 6172||2.1 GHz||12 x 512 KB||2 x 6 MB||3.2 GHz||10x||1.1875 V||80 W||115 W||G34|
|Opteron 6174||2.2 GHz||12 x 512 KB||2 x 6 MB||3.2 GHz||11x||1.1875 V||80 W||115 W||G34|
|Opteron 6176 SE||2.3 GHz||12 x 512 KB||2 x 6 MB||3.2 GHz||11.5x||1.25 V||105 W||137 W||G34|
|Opteron 6100-series, Twelve-core "Magny-Cours" (45 nm), High Efficiency|
|Opteron 6164 HE||1.7 GHz||12 x 512 KB||2 x 6 MB||3.2 GHz||8.5x||1.075 V||65 W||Row 13 - Cell 8||G34|
But how does AMD fare on the cost front? Given that this is usually AMD's strong suite, it does well. We currently see market prices on the 6174 right around $1150, which is decent considering it falls between the X5660 (~$1220) and the X5650 (sub $1000). But the scale of the Opteron 6100-series tops out with the 6176, just one model up from the 6174. Meanwhile, Intel has four processors faster than the X5660 ($1300 and up).
The benchmarks comparing the two competing server solutions make this a bit more interesting. In previous years, Xeons were the power hogs, but clearly beat out Opteron in almost all performance metrics. Today, based on what we have seen, the new Opterons can beat similarly priced Xeon 5600-class processors in some areas, but the power consumption figures are nearly neck and neck. This means that the applications you plan to use dictate the correct hardware platform, be it molecular 3D simulation or Compressible Fluid Dynamics. This shouldn't come as a surprise, as multi-threaded performance can diverge based on the implemented code-paths.
This helps explain why we are getting a 50/50 response from our survey participants, as half of them see AMD as still-profitable in the server market space. No doubt, it remains an uphill battle, but AMD's long-term server position doesn't look as bad as many people make it out to be. Obviously, the late support of HP, IBM, and Dell didn't help the matter, but what's done is done. In our opinion, part of the reoccurring problem is that server model names and product lines don't specifically differentiate between AMD- and Intel-based solutions. AMD needs a way to get its name out there in a more recognizable way. Furthermore, the paramount concern for AMD in the server fight, at the moment, needs to be pricing. Granted, for most of 2011, we don't see AMD hitting any home runs for its server group. But right now, AMD is in endurance game until it can bring the Bulldozer architecture to bear.
As an aside, it's odd that AMD is so readily willing to relinquish its traditionally strong showing in the workstation/high-performance PC space. According to representatives at the company, a lack of modern supporting core logic held back the Opteron's interoperability with high-end graphics solutions. AMD has a PCI Express 2.0-based chipset now, though, so it remains puzzling that there are so few platforms designed for Opteron-based workstations.
Absolutely agree. This is basically where I've been placing my builds for the last 10 years; at the bottom of the price curve where performance hits the sweet spot for price. I alone typify that logic.
Smartphones/PDA? They've existed for a long time now. The problem is the technology wasn't around to give them the power they needed to do everything a Netbook can do. That is rapidly changing. I hate to say it, but I disagree with Netbooks being a long-term investment. The consumer now is driven by convenience. If my smartphone can be my multimedia outlet, document editor, day planner, browser, camera, accessory portal (ear pieces/headsets, printers, scanners, etc...) and telephone, they why would I want to lug around seperate devices for each of those?
Very short-term. At the way things are going, that will be one to two years worth of earnings at the most. Hardly worth the R&D IMHO.
This is the bottom line for everything, basically. This motto can not falter.
Complexity is definitely the direction the industry has taken. However, I would think if a manufacturer wanted to baseline a board with IGPs, they would do so in terms of finding a way to allow additional discrete GPU and/or CPU installments for those that tinker. I know this has been tried in the past, but I'm not talking about simple onboard graphics processors.
The baseline board would be for the general consumer and could handle day-to-day tasks found in every household. While additional GPU and CPU configurations that would work in conjuction with the onboard processors appeals to the specializing category. We just need a manufacturer to take that step to allow them to co-exist in the same environment and provide that extra benefit of accessorizing.
The P55 is a great middle of the road platform, and if one graphics card (even two) is enough to wet your whistle, you could'nt ask for more.
I think brand loyalty comes about by great customer support and innovation, and it doesn't hurt to have a well used and supportive attached forum.
I've hide my eggs with Intel, WD, Corsair, ASUS, and EVGA for many years now, and they are safe.
In an economy where it is more important that things last, what I want to see is a focus on durability. Gigabyte advertises this very well, but I find that ASRock has many of the same elements (e.g. all-solid, Japanese caps, ferrite chokes) and costs a lot less.
Thought it is some software that does what with motherboards ?
And talking heads music band has got what to do with this ?
If not for the weird article name the rest is very nice.
Good to see Toms getting serious about hardware and reality from a users point of view.