Event Coverage

AMD's 45nm Opteron Launch

By Vincent Chang - 19 Nov 2008

Goodbye Barcelona, Hello Shanghai

Goodbye Barcelona, Hello Shanghai

After the troubled introduction of AMD's previous CPU microarchitecture, the native quad-core Barcelona that is used in the current Phenom and Opteron processors, the ongoing financial difficulties faced by the company and the recent announcement that AMD was spinning off its foundries into a separate entity (albeit with a remaining 44% stake), the last thing we were expecting was news that AMD would launch its 45nm shrink and refresh of its Opteron processors ahead of schedule.

 A suitably 'Shanghainese' location for the media briefing for AMD's latest Opteron, codenamed Shanghai of course.

Yes, it's goodbye to Barcelona and a warm welcome for the latest iteration of the Opteron quad-core processor, codenamed Shanghai. AMD has plenty of reasons to sell to its customers that they badly need a Shanghai Opteron now. And what better way to do that then to invite the regional South Asian media to the Forbidden City by IndoChine in Singapore for a briefing on the new processor.

 The planned server roadmap for AMD up to 2011. As AMD emphasized constantly, cost is reduced since the Opteron is staying on the current Socket F for the next two to three years at least.

Unlike the recent Financial Analyst Day held at AMD's campus, the regional media briefing was focused on the Shanghai processor. This meant that there were no details about upcoming desktop, mobile or graphics developments from AMD. The only roadmap in the presentation was for the server platform and that too was rather brief. Therefore, if you're hoping for interesting glimpses of future AMD products, we recommend getting the complete coverage through the webcast on AMD's website or your favorite online tech news source for the summary.
 

 IBM's Roadrunner supercomputer captured this year's top spot on the Top 500 supercomputer list and guess what, it was built using Opteron chips.

 According to the Top 500 supercomputer list, there are now more supercomputers built by HP than rival IBM.

What formed the bulk of the briefing instead were presentations from AMD's biggest partners in the server space, including Dell, HP, IBM and Sun Microsystems about their latest products and how the new processor will be the perfect fit for both existing and new customers. The general summary from these OEM presentations: you can just drop-in the new Opterons to existing systems and they should work, requiring a BIOS update at most probably. This platform stability extends to future 6-core 'Istanbul' processors from AMD and it should be reassuring for customers with the fairly depressed economic outlook currently. Obviously, each company were touting their own selling points and strengths but we're not going into those here.

Instead, what really interested us were the details about the new chip and here are some of the major points that AMD made:-

 A higher clock speed and more L3 cache will see the Shanghai be more competitive against Intel's Xeon.

Firstly, the shift to 45nm with a new immersion lithography process allows AMD to ramp up the top speed for the Shanghai Opterons to 2.7GHz. This also probably allowed the company to increase the amount of L3 cache from 2MB on Barcelona to 6MB, which should lead to a small but noticeable performance boost. DDR2-800 memory support has also been added, while the 128-bit integrated memory controller can now be divided into 2 independent 64-bit memory channels for more efficient access since twice as much can be done provided you don't need the full 128-bit bandwidth.

More importantly, the die shrink enables the company to maintain its thermal envelope while enhancing the processor, crucial for power consumption sensitive server chips like the new Opterons.

 AMD has been hammering on the power efficiency angle and there are more such gains to be made with the new 45nm CPUs. Of course, we can't say how important this power issue is for companies, especially those with giant server farms to maintain.

Power efficiency is another card that AMD has played in the past, with its performance per watt comparisons and the 45nm Shanghai arguably extends this approach, with a new Smart Fetch feature that shuts down an idle core without affecting performance by saving the data from that core's L1 and L2 caches into the shared L3 cache.

Other cores are hence able to access those data as required. Other power management technologies, like CoolCore and Dual Dynamic power management are all related to adjusting the power usage of the cores dynamically and quickly so as to reduce consumption. As it is now, all these new Opterons are rated at 75W each.

 AMD's advantages in floating point performance and memory bandwidth is evident here but the question is whether this will remain so against Intel's Nehalem based server chips next year.

Finally, while we don't actually have any numbers to back up the performance of the Shanghai besides AMD's own, early reviews we have seen point to a definite improvement over Barcelona, with AMD holding strong in the power efficiency department against the competition. Although Shanghai may fare well against Intel's Harpertown at the moment, there's no telling what things would be like when Nehalem enters the scene next year, with the brute power of the Core 2 matched with a microarchitecture that's similar to AMD's.

AMD says that the new Opterons are shipping in volume now, with the 2.7GHz model going for US$989 for 1Ku and US$277 for the 2.3GHz Opteron 2376, which are comparable and probably a better deal than Barcelona.

Despite the focus on the Shanghai, you may also have noticed that AMD has switched to a 'Fusion' theme in terms of its branding, which is why our thumbnail is using that Fusion logo. Unfortunately, while the marketing is out and we have been hearing about Fusion for a while now (there's even a white paper on AMD's site about the whole idea of fusion), it's not here yet. Short story: it's about being more efficient by having general purpose CPUs enhanced with integrated, specialized accelerators that are better at certain tasks, like GPUs. However, we won't be seeing this yet, since we believe that AMD will only go full steam on this approach (if ever) when they get to 32nm.

Overall, the earlier than expected launch is positive news for the company. It's about time AMD revived its CPU competitiveness just like it did for its GPU business and we are all looking forward to the 45nm Phenom II next year. Meanwhile, we'll have more in-depth updates, especially benchmark figures from the new Shanghai processor if and when we get a review sample from AMD. So stay tuned to more updates here on www.hardwarezone.com.

Join HWZ's Telegram channel here and catch all the latest tech news!
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.