Playing 20 Questions (Part 1)
We’ve all played the game, is it bigger than a breadbox? Is it an animal, vegetable or mineral? I decided to try an ambitious exercise, soliciting input directly from folks in IT who have questions about our products and where they are headed.
I received more than 20 in this first round and for the purpose of this post, I’ll focus on five questions related to our upcoming server platforms. Look for more posts that address questions around virtualization/cloud computing and future technologies in the coming days and weeks. Because these questions are around future platforms, some data won’t be available until launch.
Is Socket C32 being designed to be backwards compatible with the existing Socket F?
No. Socket F is a DDR2 design and C32 is DDR3. Having interchangeable processors and memories would mean a massive test matrix that OEMs might find difficult to support. Eventually you’ll have to switch over to the new design. Based on the fact that we expect Socket F to end up with ~5 years of life in it from 2006 to 2011, we feel it has served the market well and we don’t want to be focusing the new generation of processor into a 4 year old platform.
Are you going to allow a C32 socket to accept 8-core or 12-core CPU, or a G34 socket to accept 4-core or 6-core CPU?
We do not have plans to support this. The C32 and G34 sockets are different form factors with different sizes and pin counts and are targeting workloads with different needs in terms of computational capabilities. The C32 and G34 platforms address different form factors and customer workloads. Our goal is to introduce product features (including core counts) that are best aligned to those customer needs.
Does AMD plan to scale into the >100 thread 8P market with G34 systems?
The G34 systems planned for 2010 are expected to offer scalability to 4 sockets each with 12 cores per socket; that represents a total of 48 cores for a 4P platform. The challenge in moving beyond that level for many mainstream business applications can be that the scalability of the software becomes a bigger challenge than the scalability of the hardware. (Specialized HPC-style applications may be another story altogether.) The market for single systems with 96 cores appears to be very limited at the moment, and new technologies like HPC, cloud and developments such as the HyperTransport HNC specification all represent ways of getting greater scalability in a scale out mode vs. scaling the systems up.
Will AMD be the single chipset source for server main boards in the long term?
AMD plans to release the SR5690 chipset later this year. That will be utilized with existing Socket F processors as well as the future “Maranello” and “San Marino” platforms. We can’t speak to the plans of other companies, but we welcome any additional chipset products from other vendors; competition always breeds the best innovation.
AMD has placed a great deal of emphasis on the fact that all of the new Six-Core AMD OpteronTM processors (formerly codenamed “Istanbul”) fit within the same TDP/ACP as the previous Quad-Core AMD OpteronTM processors (formerly codenamed “Shanghai”). Can we expect this trend to continue with “Magny Cours” and “Interlagos” despite the increasing core count?
Because we are moving to a new platform (codenamed “Maranello”), there is not a need to have exactly the same TDP/ACP as past processors. We do recognize that customers are very comfortable with the power/thermal bands that we have established and we plan to continue to use power bands that are very similar to what we use today. In general, we plan to continue the approach of allowing a single platform to be able to support more than one generation of processors (though not mixed in the same system of course), sharing a common power/thermal envelope. Just as the Socket F (1207) allowed us to support Rev F dual -core, “Barcelona,” “Shanghai” and “Istanbul,” we believe that “Maranello” will support both the “Magny Cours” processors as well as the “Interlagos” processors. We expect the “San Marino” platform to support the “Lisbon” processor as well as its follow-on, the “Valencia” processor.
John Fruehe is the Director of Business Development for Server/Workstation products at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.
POSTED IN: AMD Opteron
TAGS: Chipset, Istanbul, Magny Cours, Maranello, Socket C32, Socket G34
SHARE:
COMMENTS: 13
Is AMD has plans to support ‘Torrenza’?
AMD continues to support accelerated computing. I even blogged about it recently:
http://blogs.amd.com/work/2009/04/06/a-million-reasons-why/
Time has passed since 2006 is sufficient and you have not yet been implemented multithreading. Intel, as the same time and wins you through multithreading.
And also, you have fallen behind schedule the implementation of new rules.
Can you send more work to multithreading??? All 4 of the kernel, must share data, and do not forward in kesh. 4 core, 4 scheduler, 4 flow simultaneously processed, the kernel should be as brothers, sharing information on the stages of processing. You need konveer. Split information processing in stages, to make bridges between the stages of processing. Here’s to you and crystal sizes smaller, and finished 8 flows. I would like to know what you are?? Matters is speed, and few people are interested in what will be the nuclei.
Actually, since 2006, I believe AMD has done a better job of building scalable platforms. The key has always been the removal of the front side bus bottleneck, something that we have accomplished years before.
We are currently shipping 6-core processors, for 12-core 2P systems vs. our competitor’s 8 core 2P systems. In 4P, we have 24-core systems, just like our competitor, but in most cases their 24-core systems lag behind not only our 24-core systems but also our 16-core systems.
There was initial speculation that AMD may support SMT in on later…can this G3x support that in future.
Support for SMT is currently not in our plans. As a technology, there is an opportunity for it to increase performance, but depending on the workload there are also cases where it actually slows down the workload. It is far from the “free performance boost” that others have claimed it is.
Instead you will see AMD deliver processors with greater numbers of physical cores, which delivers a much more stable and consistent scalability.
We will continue to monitor SMT and if the technology develops to the point where the negative factors are outweighed, it might be worth considering, but for now, it is not in the roadmap.
what is AMD’s roadmap for the desktop beyong PhenomII? Can we expect six-core phenoms in the line-up for 2010 and desktop versions of Socket G34/”magny-cours?”
Thank you.
I’m in the server business, not the desktop business, so I can’t really comment on their products. When you increase the number of cores, the clock frequencies generally go down. Server applications are far more threaded, so six-core designs are much more appealing to a server market, where more cores (even at lower frequencies) can deliver greater performance. In desktop environments where applications are less threaded, having fewer cores with higher frequencies can generally deliver greater performance.
AMD plans to launch next-generation desktop platforms for the enthusiast and mainstream markets in 2010. In terms of new technology, we believe in bringing the right platform technology to market at the right time. As further platform enhancements become available, we intend to offer them at a price that satisfies customer needs.
Why don’t AMD launch the next-generation desktop and server at the same time for the new platform? Will the price going to be a lot higher than the current processor?
With the introduction of Istanbul, the development efforts between client and server products has split. Server customers buy on a much slower cadence, preferring a longer lifecycle tied to technology transitions.
Client customers tend to buy around seasonal buying cycles (holiday season, back to school, etc.) and refresh technology more often.
Our strategy allows us to better align with the needs of the customers.
We never discuss price of products prior to their introduction, but rest assured that the price performance always gets better for customers with each generation of products.
I am very pleased with this blog i have a question. Will amd’s next generation processor be able to handle multiple threads on a single core?
The current implementations of handling multiple threads on a single core present challenges.
There are instances where the technology can increase performance by 10-15%, but there are also instances where the technology can actually create a bottleneck, leading to overall performance losses (Google the phrase “hyperthreading reduces performance” for more information).
In simple math terms, 2 threads on a single core should be 110% – 115% of a single thread on a single core. But two threads on two physical cores would be 200% of the performance of a single thread on a single core.
Part of the challenge is that to effectively support 2 threads on one core, you need to implement software changes to take advantage of it. Changing the software model is not something customers are generally anxious to do.
Until that technology matures to the point where it can be handled efficiently without the potential for slowing down the system, I don’t see AMD adding it to the processors. Our Magny Cours processor will have up to 12 cores per processor, leading to 2P systems with 24 cores and 4P systems with 48 cores. True actual cores are a better alternative to running more threads on a server because real cores will always deliver more performance without bottlenecks or software changes.
Pingback: What we know about AMD’s next-generation processors - PC SHOW 2010 Pricelist | COMPUTEX 2010 | VRForums - Singapore IT & Lifestyle