Similar questions will be raised and evaluations will be made by IBM customers in response to the company's POWER7 launch. I'll share some of the highlights of today's announcement, but first, a quick point: Considerable work goes on behind the scenes in preparation for these announcement events. Weeks ago I signed a non-disclosure agreement with IBM. In return I received access to relevant training, information and materials. This is standard industry practice -- by allowing, in this case, IBMers and business partners to get up to speed on the POWER7 product line prior to the announcement, these groups can then educate customers and answer questions about the new solutions on day one.
To me, one thing that immediately jumps out is that POWER7 can run in a POWER6 compatibility mode, making this the first generation of POWER hardware that supports previous technology levels. For instance, older operating system versions (AIX and IBM i) can run on the new hardware, and live partition mobility can occur between POWER6 servers and the new POWER7 machines. The binary compatibility between POWER6 and POWER7 allows customers to migrate partitions between their current serves and the just-announced ones. In other words, you'll be able to move your LPARs to and from POWER6 and POWER7 hardware based on your business needs.
Note: Customers that are running POWER5 systems will need to upgrade to POWER6 before upgrading to POWER7.
Here's a breakdown of AIX support on POWER7:
* AIX 5.3 and AIX6 TL2 and TL3 will run only in POWER6 and POWER6+ modes.
* AIX 6 TL4 or later will run in POWER7, POWER6 and POWER6+ modes.
* AIX 6 TL5 will include additional POWER7 performance enhancements with improved memory affinity.
Initially, these AIX Levels will be supported on POWER7:
* AIX 5.3 with the 5300-11 Technology Level SP2 or later.
* AIX 6.1 with the 6100-04 Technology Level SP3 or later.
* IBM i 6.1 with 6.1.1 machine code or later.
* VIOS 126.96.36.199 with Fix Pack 22.1 and Service Pack 2 or later.
HMC V7 R710 is the minimum level for POWER7 support. If an HMC is used to manage a POWER7 processor-based server, it must be either a CR3 or later model rack-mount HMC or a C05 or later model deskside HMC. If IBM Systems Director is used to manage an HMC, or if the HMC manages more than 254 partitions, the HMC should have 3GB of RAM minimum and be either a CR3 or later rack-mount or a C06 or later deskside model.
Active Memory Expansion
Not to be confused with active memory sharing, active memory expansion allows you to effectively trade some CPU cycles for memory size.
Once you've allocated physical memory to an LPAR, the server uses CPU resources to compress memory contents, thus allowing it to function as if it has more memory than it actually does. This reduces the physical memory requirements of existing LPARs and frees up physical memory capacity that can be used to create more LPARs in the same physical memory footprint. The active memory sharing feature is managed by the OS and the hypervisor. The OS compresses and decompresses data based on memory accesses, a process that's transparent to applications.
I'll have much more on active memory expansion in the near future.
General availability for the new Power 750 server is set for Feb. 19. This is a 4U server with four sockets. Customers will be able to select four, six or eight cores running at 3.0 to 3.55 GHz, meaning that a maximum of 32 cores are available on this machine. Maximum memory on the 750 is 512GB.
General availability for the new Power 770 server is set for March 16. This server has a similar form factor to the 570, with up to four 4U nodes. Each node can have 12 or 16 cores running either 3.1 or 3.5 GHz, allowing for a maximum of 64 cores on the 770. Max memory ranges from 512GB with one node to 2TB with all four nodes.
The new Power 780 server is also set for March 16 availability. This machine can also support up to 64 cores, running 3.86 to 4.14GHz. You can have a single enclosure with up to 512GB of memory, or four enclosures with up to 2TB of memory. The 780 features the TurboCore mode -- this allows you to "shut off" four of your original eight cores and aggregate the L3 cache from the unused cores. With more L3 cache available to each core, you should have fewer cache misses and thus, increased performance.
TurboCore chips contain twice as much L3 cache per chip. These cores, when run in TurboCore mode, are twice as fast as POWER6 cores, so your applications can potentially see a doubling of the performance per core. Since software is often sold per core, customers that are sensitive to software pricing will see this option as a great way to go.
Speeds and Feeds
The chips themselves are very interesting. Depending on the machine, four, six or eight cores are available per chip (or socket). An 8-core chip with integrated cache and memory controllers has 12 execution units per core, 4-way SMT per core and 32 threads per chip. POWER7 L1 cache consists of 32 KB I cache/32 KB D cache. L2 cache has 256 KB per core, and L3 cache has 32 MB shared between cores on each chip.
As noted, POWER7 runs at 3 to 4.14 GHz. While clock speed is down compared to POWER6, we should see 20-30 percent higher performance in POWER7, due to increased thread counts per core.
More to Come
As these systems become available and I start to install them at customer sites, I'll continue to let you know about my experiences with POWER7. However, now is the time to start looking at the new features and figuring out how these systems can help your business. IBM and IBM business partners can help you understand how you can benefit from making the move.For more information on the POWER7 announcement, visit IBM Systems Magazine, Power Systems edition.