Intel Xeon E5-2600 v2 Series Processors (Ivy Bridge-EP) in Q3 2013

More details about the upcoming 22nm Intel Xeon E5-2600 series processors (aka Ivy Bridge-EP) are starting to leak out. Initially, we will see the single-socket Xeon E5-1600 v2 series and the two-socket Xeon E5-2600 v2 series processors, that share the 22nm Ivy Bridge microarchitecture. They use the same LGA2011 socket as the current 32nm E5-1600 and E5-2600 series (Sandy Bridge-EP), so that should mean that they will show up quickly in existing server models, such as the Dell PowerEdge R720 and the HP ProLiant DL380p Gen 8.

The chart that TechPowerUp found does not show the physical core count for each processor model, but we can infer that based on the total L3 cache size, since Intel has been using 2.5MB of shared L3 cache for each physical core in the Sandy Bridge and Ivy Bridge microarchitectures. That means you can divide the L3 cache size (in MB) by 2.5 to get the physical core count. Based on this calculation, it looks like the Intel Xeon E5-2600 v2 processors will have from eight to twelve physical cores. The Xeon E5-2697 v2 and E5-2695 v2 have twelve physical cores and a 30MB shared L3 cache, which means that you give up quick a bit of base clock speed to get twelve cores. If you want to minimize your physical core counts for SQL Server 2012 licensing purposes, the eight-core, 2.6GHz Xeon E5-2650 v2 would be a good choice. I am a little surprised that there is not a four or six-core model in the lineup, but we may not have all of the available models yet. Table 1 shows the known specifications for the Xeon E5-2600 v2 series processors.

ModelPhysical CoresBase Clock SpeedL3 Cache Size
Xeon E5-2640 v282.0GHz20MB
Xeon E5-2643 v210 (??)3.5GHz25MB
Xeon E5-2650 v282.6GHz20MB
Xeon E5-2650L v2101.7GHz25MB
Xeon E5-2660 v2102.2GHz25MB
Xeon E5-2667 v2103.3GHz25MB
Xeon E5-2670 v2102.5GHz25MB
Xeon E5-2680 v2102.8GHz25MB
Xeon E5-2687W v2103.4GHz25MB
Xeon E5-2690 v2103.0GHz25MB
Xeon E5-2695 v2122.4GHz30MB
Xeon E5-2697 v2122.7GHz30MB

Table 1: Intel E5-2600 v2 Series Model Information

I am a little suspicious of the core count and L3 cache size for the Xeon E5-2643 v2. I think it is more likely to have eight physical cores and a 20MB L3 cache based on it’s base clock speed. Even though the Ivy Bridge-EP core counts are up to 50% higher than the 32nm Xeon E5-2600 series, the TDP ratings are about the same, which shows the benefits of moving to a smaller 22nm manufacturing process. The only low-power SKU in the lineup is Xeon E5-2650L v2. It has ten CPU cores, a 1.7 GHz base clock speed, and a 70 Watt TDP rating. It would not be a good choice for SQL Server, since you give up a significant amount of performance to save perhaps 20-30 watts of electrical usage per processor. The Xeon E5-2600 v2 processors are supposed to have two QPI 1.1 links, up to 40 lanes of PCI Express 3.0 interface, and a 4-channel DDR3 memory controller that supports DDR3-1866 memory.

You should be able to get an existing model server with these processors sometime in July. Getting a two-socket system with two Intel Xeon E5-2690 v2 processors will be a pretty compelling choice for a lot of SQL Server 2012/2014 workloads. I predict we will see TPC-E scores in the 2500-2600 range for that type of system, which compares quite well to a four-socket, Intel E7-4870 system that would cost twice as much for SQL Server 2012 Enterprise core licenses.

Hardware 201: Selecting and Sizing Database Hardware

I had the opportunity to present Hardware 201 in Boulder on June 18 and then do it again in Denver on June 20. It was fun presenting to both user’s groups nearly back to back. Here is the abstract:

Hardware 201: Selecting and Sizing Database Hardware for OLTP Performance

The foundation of database performance is the underlying server hardware and storage subsystem. Even the best designed and optimized database application can be crippled by an inadequate hardware and storage infrastructure. Recent advances in new processors and chipsets, along with improvements in magnetic and SSD storage have dramatically changed the evaluation and selection process compared to the past. Many database professionals struggle to keep up with new technology and often simply let someone else make their hardware selection and sizing decisions. Unfortunately, the DBA usually gets the blame for any performance issues that crop up later. Don’t let this happen to you! This session covers current and upcoming hardware from both Intel and AMD and gives you the tools and resources to make better hardware selection decisions to support SQL Server OLTP workloads.

New AMD-based TPC-E Benchmark Result

HP recently submitted a TPC-E benchmark result for a two-socket HP ProLiant DL385p Gen 8 system that is using two 2.8GHz AMD Opteron 6386SE, 16-core processors. This system had a score of 1416.37 TpsE, which does not sound too bad until you compare it to a very similar two-socket HP ProLiant DL380p Gen 8 system that is using two 2.9GHz Intel Xeon E5-2690, eight-core processors. The Intel system had a score of 1881.76 TpsE, which is 32.8% higher than the AMD-based system. That is pretty significant just from a pure OLTP performance perspective, but the story gets even worse when you look at the per-physical core performance and the SQL Server 2012 Enterprise Edition licensing cost differences for the two systems, as shown in Figure 1.

ProcessorTPC-E ScoreScore/Physical CoreSQL 2012 License Cost
Intel Xeon E5-26901881.76117.61$109,952
AMD Opteron 6386SE1416.3744.26$164,928

Figure 1: Comparative TPC-E Metrics and SQL Server 2012 License Costs

The DL385p Gen 8 system has 32 physical cores (that all must be licensed with $6872.00 SQL Server 2012 Enterprise Edition core licenses). The SQL Server 2012 Core Factor Table gives a 25% discount for modern AMD processors that have more than six physical cores. You can download it here (PDF warning), although Microsoft has not updated it to include the newer Opteron 6300 series of processors, which is probably just an oversight by Microsoft.

The DL 380p Gen 8 system only has 16 physical cores that must be licensed (hyper-threading is not taken into account for licensing on physical servers), so the SQL Server 2012 Enterprise Edition licensing cost is significantly less. Paying about 50% more to get about 33% less performance does not seem like a compelling value proposition!

I really wish that AMD was able to present some viable competition in this area for Intel, but I just don’t see that happening based on the currently available roadmaps from both companies. Lacking any real competition on the performance front, there is less incentive for Intel to meet their release schedules for new processor microarchitectures.