The Bulldozer Blog
Welcome to the Bulldozer Blog. Our next generation core architecture, a complete new design from the ground up, is called “Bulldozer.” This new core design is planned as the basis for our next generation AMD Opteron™ processors, as well as our high end client products.
The primary focus of the Bulldozer Blog will be commercial systems, and most specifically server, but there will be some client-focused blogs periodically as well. The server business tends to have much longer sales cycles and more architectural discussions, so you will see more focus from us in those areas.
So, what can you expect to see from this blog? Here is a snapshot of some of the things that we will be covering in the near future:
- Hot Chips 22 Disclosures: Each year, the Hot Chips conference brings in-depth discussion of next generation technologies. On August 24th, Mike Butler will be presenting the next generation Bulldozer architecture at the conference. (The same day we’ll also disclose new details on “Bobcat” our other new core architecture scheduled to hit the market in 2011.) We’ll have a series of updates after this conference to give you some of the details, bringing the Bulldozer technology out to a wider audience.
- 20 Questions Blog: One of our most widely read (and participated) blogs for our Magny Cours product was the “20 Questions Blog.” We are going to do this again with Bulldozer this time. We’ll have an email address to send your questions to; we’ll choose the best questions and answer them in the blog.
- Video Interview Blogs: Expect to see several video blogs from AMD as we bring you insight from some of the key engineers, executives and partners behind the product.
- Product Video Demos: As we get closer to the launch, there will inevitably be some public demonstrations of the product. We’ll be sure to videotape those and bring them to you, as quickly as possible.
Just to make sure that everyone is up to speed on what Bulldozer is — a brand new design featuring up to 8 cores for client products and up to 16 cores for server products. Bulldozer will feature a new floating point unit that can support up to 256-bit floating point execution, which will boost the performance for technical applications that rely on floating point math. There will be some new software instructions that will be supported, allowing for greater performance and flexibility, but, it will be backwards compatible so you won’t need to change anything to start using the processor. We will be introducing this processor in 2011, and as we get closer we’ll get more granular on the actual availability.
Just so that we are all on the same page, some things you won’t see us discuss detailed benchmark results, pricing, launch date or any of our partners’ platforms. But, none of that should be a surprise to you because that is standard operating procedure. As we get closer to launch, I would expect to see some partners guest blogging about their platforms, but we’ll leave their disclosures to them.
Be sure to stay tuned and follow this blog; this will be the most interesting place to find all of the information about our next generation “Bulldozer” technology.
John Fruehe is the Director of Product Marketing 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, Bulldozer
TAGS: AMD Opteron, Bulldozer
SHARE:
COMMENTS: 227
Pingback: AMD Bulldozer | GPU Design
Any plans on doing an integrated flash controller? It would be nice to have 32 or 64GB of flash DIMMs on the motherboard right next to the RAM. It would be insanely fast, and cheaper than a SSD.
Several parners today have integrated SD readers for booting and virtualization. Dell and HP both have dual SD cards.
I’m liking what I’m seeing. Wondering mainly about the granularity of Bulldozer’s boost on less threaded workloads compared to the rather binary static 400/500MHz increase on current Zosma/Thuban desktop models. As much as I’d love everything to take advantage of 8 cores, some more mundane things won’t be, and I’d like to know how well it’ll take advantage of any spare TDP it has lying around.
We aren’t releasing any detail on that feature at this point, but it will be improved from today’s implementation.
i nedd a 16 cores cpu for my pc but i still need performance per core of Bulldozer better then core i7 at least
i don’t want my 16 cores cpu get a less benchmark socre than intel’s 12 cores or even 8 cores cpu that make me fell bad
good luck amd and Bulldozer
if performance per core is not as good as core i7, i have to choose intel “sandy bridge”
Your 2 statements do not sync. If you need 16 cores then you are running multi-threaded applications and total throughput is what you should be measuring.
What do you guys expect? Better than Nehalem? Better than SB? That will be disappointed! I have heard some AMD guys reveal, no miracle will happen. The Bulldozer may almost equal to k10.5 per core.
Think, Bulldozer will better than SB to LGA1155 (average, not in all). Than highend SB? I dont know…will see.
8 core Bulldozer just wins 4 core SB (1155)? I believe it can, but this is a gap! Just like 3 people fighting 1 people, the winness will not be praised by others. John said 16 core Bulldozer just wins 6 core Westmere by 50% in Spec result, that proves the archietecture of Bulldozer is a failure, to others. Maybe the Spec is not a good tool to compare the performance, because Spec has a big demand of cache and bandwidth, especially in parallel compute and multi socket connection. The Xeon 7560 has the worlds records in Spec results, even Power 7 is not the component. But, by Anandtech tests, in other sever tests, XEON 7560 actually loses to Magnycours. You can see, spec has its limitations. I will still expect, in other benchmarks, Bulldozer may outperform component greaterly. I hope this will not be a far future, far hope.
Hi, thanks for the blog, I heard that there was some problem (although I don’t understand the technicalities) that Bulldozer was using FMA4 while Intel was sticking with FMA3 and that the two were not compatible.
Has that been resolved? If not how will it affect Bulldozer performance assuming that most developers will stick with Intels standard simply because of their larger market share.
I think you have it wrong, we are going to write a blog about this in the future. All of the major OS vendors will be supporting all of the features of our processors, this is a manufactred issue, not a real concern.
Will it support SSE5?How about the SSE4, will it be fully compatible now? How about AES-NI?
We support SSE4 today already. We will support AVX and AES-NI. SSE5 is not being supported by anyone that I am aware of.
Will the inter-core latency and L3 cache be reduced sharply?
We are not disclosing any of that detail until launch, but customers should be very happy with the performance.
Will it support more then 512GB Ram?
It will support more than 512GB of RAM, we have not disclosed the full amount, that typically comes at launch.
Pingback: AMD, termina la arquitectura Bulldozer para sus nuevos chips Interlagos | tuexpertoIT.com
Will there be any possibility for Bulldozer to support Anti-Hyper Threading tech? Will it be 28nm with the released version? Will there be a Fusion version based on Bulldozer architecture? Thx.
First, I have no idea what “anti hyperthreading” technology is. If you are referring to technology that makes two threads perform with less throughput than a single thread, intel has already cornered the market on that because hyperthreading has clearly defined cases where 2 threads provide less throughput than one thread today. We are going with cores because cores scale well and will always give you more performance.
Bulldozer will be 32nm SOI.
Can’t speak to fusion around Bulldozer. Currently there are no plans that I can talk about.
Anti-hyperthreading was a crazy rumor mill theory capability for next gen AMD cpus at some point (I think some time between Intel introducting Hyperthreading and dual cores becoming mainstream). I think the idea was that you would interleave two cores clocked half a cycle off from each other to work on the same thread or something of that nature, before people actually understood multicore’s limitations. My head hurts just thinking about trying to do that, getting the timing right wouldn’t be easy with delays, let alone synchronizing the execution state or avoiding race conditions with instructions that take more than half a cycle to settle.
I didn’t expect it to still be making the rounds but rumors never die, they just mutate.
Curious about the cases where hyperthreading decreases performance though. I thought most of those issues were P4-specific. I was under the impression that all hyperthreading was was doubling the state resources so a core can keep track of two execution states at once, and using the two threads to keep the pipeline filled. Only problem I can see issues with is cache thrashing with two threads repeatedly evicting each other’s data from L1/l2 cache.
Hyperthreading causes slowdowns when there is cache thrash (i.e. too much moving of data in and out of the cache vs. executing.) Thing like LINPACK are highly efficient, so they actually see a slowdown because there are the “pipeline gaps” for HT to take advantage of. If you have 2 threads and only one set of pipelines, HT allows the second thread to steal executions in the pipeline when there is a stall.
As applications continue to become more efficient, you have fewer gaps in the pipeline and less opportunity for upside. In server workloads, the increase in performance is ~15-20% and there are some workloads where it actually slows things down or creates unpredictable results.
http://blogs.amd.com/work/2010/01/21/it%e2%80%99s-all-about-the-cores/
As you said the HT would slow things down when the applications continued becoming more efficient. Why will Bulldozer have HT alike function as it is advertised?
A. I did not say HT would slow things down, I said HT will slow down some workloads. There is a difference.
B. Bulldozer will not have an HT function, that has been said 100 different times in 100 different places.
Bulldozer will not have an HT function. We have said that many times in many places; we have never said it would have it.
We will have cores because cores scale and as workloads become more efficient, cores actually scale better, not worse.