Figure 1: AMD Bulldozer module

A single Bulldozer processor would have eight of these modules. Each module has 2MB of shared L2 cache, with up to 16MB of L3 cache that is shared between all of the modules in the physical CPU. This processor will also have a feature that AMD calls Turbo CORE, which has the ability to boost the clock speed of all of the cores by 500MHz, and of fewer cores by an undisclosed, higher amount (subject to the actual TDP rating of the processor). This is pretty similar to the Turbo Boost feature in recent Intel processors.

These are supposed to go into production in Q2 and to formally launch and be available for sale in Q3 of 2011. They will be socket compatible with the current Opteron 6100 Magny-Cours, using the same G34 socket, which will make it easier for system vendors to start using it. AMD has a short video about the Bulldozer available here.

I have high hopes for this processor, since I really want AMD to be a viable competitor to Intel from a performance perspective. If AMD cannot compete with Intel, we will all lose, since Intel will have little incentive to continue to innovate at a rapid pace.

The AMD Opteron 6100 “Magny Cours” is an eight or twelve-core processor used in two and four socket servers starting in April of 2010. It is built on 45nm process technology and has clock speeds ranging from 1.7GHz to 2.5GHz. Each core has a 512K L2 cache; while there is a 12MB shared L3 cache. It uses DDR3 SDRAM and Socket G34. It is basically two six-core “Istanbul” processor dies combined in a single package, with improved, dual DDR3 memory controllers and improved hyper-transport (HT) connections between the processor dies. There are also eight-core versions of the “Magny Cours”. The AMD Opteron 6100 “Magny Cours” is part of the high-end “Maranello” platform, which supports both two socket and four socket configurations. Each processor die has its own memory controller, so there are four DDR3 memory channels on “Magny Cours”. The “Magny Cours” requires a new socket called Socket G34. The “Magny Cours” is AMD’s best performing processor in the 2010 to mid-2011 timeframe. The “Magny Cours” seems to perform better with multi-threaded workloads because of its high physical core count and overall design. This makes it better suited for non-OLTP usage (such as reporting, data warehouses, etc.).

AMD is different from Intel in that they use the exact same processors in both two socket servers and in four socket (and above) servers. This makes it less confusing when you are looking at processor models, and makes it easier for systems vendors to design and build systems that use AMD processors.

One interesting fact is that there have been only four published TPC-E benchmarks for systems based on AMD processors vs. 36 published TPC-E benchmarks for systems based on Intel processors. Both the Intel Xeon X5680 and the Intel Xeon X7560 seem to do better than the AMD Opteron 6176 on TPC-E.

The AMD Opteron 6180 SE is the best performing CPU you can get from AMD until the AMD “Bulldozer” series is released, presumably in Q3 2011.

The Intel Xeon E7 processors have up to ten cores (plus hyper-threading in most models). They have four Quick Path Interconnect (QPI) links and two memory controllers, which each have two dual-channel interfaces per memory controller. The new memory controllers support 32GB DIMMs and low-power memory modules. This means that a four socket system could support up to 2TB of RAM (which is the current operating system limit for Windows Server 2008 R2).  Of course, you will need pretty deep pockets to do that, because 32GB DIMMs will be very expensive when they are first available. My guess is that 32GB DIMMs will cost around $3000 each, at least initially.

Using the on-board memory buffer, the E7 processors can run DDR3-1333 memory at data rates of 800, 978 and 1066 MHz.  The E7 processors will support AES instructions, Trusted Execution Technology, and Virtualization VT-x, VT-d and VT-c.

Intel claims up to 40% better database performance for the top E7-4870 model in comparison to the previous generation Xeon X7560 model. Performance of the E7-4870 CPU in integer and floating-point applications is better than the X7560 by up to 22% and 19% respectively. The E7 processors will be socket compatible with the current Xeon 7500 processors, which means that existing systems from your favorite vendor should be able to use them as soon as they are available.

Here are some specifications for the four-socket E7 models.

The E7-4850 and above have 10 cores, plus hyper-threading, so a four socket server could have 80 logical cores. Don’t forget that you need to have SQL Server 2008 R2 running on top of Windows Server 2008 R2 in order to use more than 64 logical cores.

The E7-4850 and above will also use Turbo Boost to increase the clock speed of some cores by up to 400MHz, which can give a nice performance boost to single-threaded workloads.

Finally, here is some Intel “marketechure” about this series.

I think Intel might be launching this series in the next week or so.