SAN FRANCISCO: CHIPMAKER Intel is slowly winding up its Haswell hype machine ahead of what is expected to be a splashy launch in June.
Intel has been revealing high level details of its upcoming Haswell processors for a long time, such as its statement at IDF 2012 that a Haswell 10W TDP part will be pitched at laptops. The firm not only shown off running Haswell silicon at the recent Game Developer Conference but also announced that its HD Graphics 15.31 driver released this week is designed specifically for Haswell processors.
There's no doubt that Intel's Haswell chip will be the big semiconductor product launch of 2013, but it is rare to see Intel give away so many details of its upcoming processor line, given that AMD is doing relatively little to push the firm's third generation Ivy Bridge Core processors to the limit. However Intel's prelaunch announcements show a company that wants to stress the efficiency of its processors above everything else.
Back in September, Intel's announcement that its standard mobile Haswell chip will have a TDP of 10W was a statement that it can design its full Core architecture down to a point where ARM's opportunity in the mainstream laptop market might be limited. After all, the various Google Chromebook vendors are hardly seeing their units fly off the shelves.
Intel's prebriefing for the HD Graphics 15.31 driver claimed it is "architected for low power" by trying to balance the workload between the CPU and the GPU. The firm even touted OpenCL 1.2 support, which is something you expect to see from AMD in its driver releases rather than a firm that only introduced full profile OpenCL support in its GPU less than a year ago.
There's no doubt that Intel is taking its GPU development more seriously than before. That doesn't mean to say that the firm will suddenly start to threaten AMD and Nvidia in the GPU performance stakes, but with the GPU being used for much more than gaming, Intel has realised that a powerful GPU isn't so much a graphics processing unit but rather a GPGPU that will affect its ability to win the important benchmarks that result in production wins from OEMs.
Chip designers from Intel's rivals almost without fail paint a picture of Intel as a company that is extremely inefficient when it comes to resources, whether those resources be transistors or bang for its research buck. The problem for Intel is that while it has resources that its rivals, aside from Samsung, can only dream of, it knows that even it needs to tighten its belt because the cost of chip design is falling while the cost of chip manufacturing - Intel's unique advantage - is increasing.
Andrew Feldman, best known for leading Seamicro and who is currently corporate VP and GM of AMD's Server Business Unit, told The INQUIRER that the cost of developing ARM based processors is so low that chip vendors' biggest customers will be able to subsidise chip development cost and buy the products on a cost-plus basis. Feldman's future might sound extreme, and it is in the present day semiconductor industry, but ARM based processors are considerably quicker and easier to develop and do not require leading edge process nodes to have the energy efficiency that Intel gains with its advanced manufacturing capabilities.
Feldman's claims actually become pretty obvious when one looks beyond the silicon and considers what high volume server customers run on their equipment. Firms such as Amazon, Facebook, Google, Twitter and anyone else that heavily customises both the operating system kernel and the userland to meet their specific business demands can gain a significant performance advantage by not having to use a generic x86 or ARM core that's available to anyone with a chequebook.
Ask Amazon, Google or Facebook whether it would spend $50m co-developing a processor that its competitors couldn't get hold of, and even offer a one or two percent increase in performance on its current software stack, and it is highly likely they will bite your - forgive the pun - arm off. Feldman said not only can firms easily develop ARM chips within 18 months, some experienced chip vendors can bring that down to a year, something with which Intel might find extremely hard to compete.
Intel is no stranger to designing chips for specific customers. It made Apple's Macbook Air possible by cutting a deal with the firm to develop and supply a custom Merom chip that allowed Apple to all but invent the ultra-portable laptop market. The irony there for Intel is that the Macbook Air that it helped Apple invent is the same laptop design that offers the most competition to its thin and light ultrabooks today.
The question is whether Intel can afford to design more Meroms for customers both in terms of cost and how long the first design to final product will take. After all, there presently are only three x86 chip vendors, but one could easily think of a dozen ARM chip vendors that will drive competition in that market.
Intel will not have a problem matching the inherent power efficiency of the ARM architecture and Haswell will likely do enough to keep its head above anything AMD can pump out in the x86 market, but it might well have a problem matching the low costs and short product cycles that could lure high volume server vendors to ARM. µ