UKGamer


Content

Latest

News
Articles

Community

Forum

UKGamer

About us
Network

Scan Windows 7

Body
As AMD surprised many by announcing its ARM license will be used to make server chips instead of ones destined for use in tablets, for once the firm is looking like it is in a good position to take advantage of a unique position in the semiconductor market.

Ever since AMD introduced the first Opteron chips back in 2003, the firm has been well equipped to tackle the server market. Not only was the firm's AMD64 architecture significantly better than anything Intel had for a good few years but in using the HyperTransport bus and an on-die memory controller it had performance where servers needed it the most.

Only AMD did what AMD does best, fumble around for years on end and allow Intel to catch up. The firm even got the design wins it so badly needed from vendors such as Dell and HP but all of that has resulted with IDC reporting AMD having a meagre 5.7 percent of the server and workstation market at the end of 2011.

Last year at AMD's Bulldozer Opteron launch, Leslie Sobon, AMD's corporate vice president of Worldwide Product and Outbound Marketing told me that it was simply unacceptable that a small company, one that should be nimble and able to react quickly, couldn't disrupt the server market to get a 10 percent share. Sobon may have marketing at the end of her job title but the comment was blunt and the above all true. Quite simply AMD needs more than X86 and AMD64 to make any serious dent in what is very lucrative market.

ARM offers redemption

Both Dell and HP have been very public in their support of ARM-based servers and both firms have told me that it is 64-bit support that is needed before ARM can truly start to take on Intel in the server market. With ARM's A50-series launch, the firm has finally delivered designs based on an architecture it announced last year and effectively fired the starting pistol for AMD and other more established ARM chip vendors.

ARM's 64-bit ARMv8 architecture is designed to create sub-5W thermal design power (TDP) processors, meaning AMD doesn't have to rely on GlobalFoundries or TSMC to come up with sub-20nm process nodes to help it achieve the same TDP from its X86 designs. Effectively this allows AMD to compete with Intel in a market its Atom S processors in 2014, something that its Opteron chips were looking increasingly unlikely to do.

As for AMD's competitors, Calxeda has been making a name for itself in the past two years, and my friend over at ZDNet Jack Clark has been following the company for a while now and speaks very highly of the firm whenever our schedules match up and we talk over a drink or two. I share Clark's enthusiasm over Calxeda, despite its decision to produce what is a shockingly inaccurate presentation claiming to show power utilisation between its ARM-based server and an Intel-based server.

While Calxeda may be the hot commodity at the moment having managed to secure business as server vendors and businesses assess the potential of ARM kit in their data rooms, the firm and every other ARM vendor looking to get into servers, should be very wary of AMD.

AMD's brand is far stronger than its market share deserves. Calxeda, Qualcomm and even Samsung, which I've heard is looking very seriously at entering the ARM server market, will have a hard time competing, at least in terms of brand recognition, in the server market. And perhaps most importantly, AMD has a good decade or so of being in the datacenter which should not be underestimated.

Then there is AMD's new found position in the channel. In numerous talks with Dell and HP both companies claim its customers want to buy from a single supplier in order to cut down on costs both in capital and operational expenditure. Of course one would expect Dell and HP to make such a statement, after all the two firms supply everything from desktops to multi-socket servers with printers and network equipment in between, but even network infrastructure vendors such as Cisco has been taking this route in recent years by offering servers. While I don't think anyone expects AMD to start selling its own desktops, with its purchase of Seamicro, the firm can become a one-stop server shop and that will tempt some large server buyers.

Seamicro is the ultimate hedging strategy

Back in March John Fruehe, AMD's marketing director for enterprise products admitted the firm had bought Seamicro for its interconnect technology and Lisa Su, AMD's senior vice president and general manager of the firm's Global Business Units confirmed that by stating Seamicro's recently launched Freedom Fabric was the "secret sauce" that would set the firm apart from other ARM vendors.

To further cement AMD's position of reliance on Seamicro, Andrew Feldman, who used to be the CEO of Seamicro before it was bought by AMD told me that AMD would not be licensing the Freedom Fabric to other CPU designers. But Feldman's company brings something else to AMD, a direct sales channel for servers that includes both AMD and Intel processors effectively making AMD's product offering, at the silicon level, on par with server vendors such as HP.

Through Seamicro, AMD is able to sell servers that sport X86 and ARM-based processors effectively meaning the firm is able to hedge its bets on the popularity of ARM servers, without much risk to its established business. Given that most server vendors and ARM itself is expecting ARM-based servers to account for 15 per cent of the market in the next three to five years, AMD has a better than average chance of significantly boosting its current server market share.

As for Seamicro, I think it AMD's decision to purchase the firm has been one of its best decisions in a very long time. I had an interesting chat with Feldman at this year's Intel Developer Forum in San Francisco and it became very apparent that this is a man that knows the technical and business domain of servers and is determined to stick it to established vendors. Quite frankly AMD needs more Feldmans.

I've lost count of how many AMD employees have told me the firm cannot compete with Intel because of its size and wealth; extrapolating that sentiment would lead to question why AMD is bothering to keep the lights on. I only hope Feldman sticks around at AMD longer than any contractual obligations that may exist.

AMD's decision not to rely on X86 for its future is something the firm has been edging to with its CPU and GPU Fusion processors. However the firm's so-called accelerated processing units does not bring TDP down to the sub-10W range for servers and vector processors are not useful for the vast majority of current server workloads.

AMD had to wait for ARM's ARMv8 architecture before stating 2014 will see Opteron chips based on both ARM and X86. It offers the firm a chance to move from simply being a silicon designer reliant on fabrication plants to a chip vendor that can offer its own ARM and X86 servers.