IBM POWER6™ processor-based servers are designed to allow you to get the most out of your IT investment. The combination of leadership performance, energy efficiency, flexible virtualization features and RAS features designed to maximize application availability can enable you to take back control of your IT infrastructure.
Power = Performance + Control
IBM Power Systems has a history of leadership performance on a platform with a solid track record of innovation delivered to clients as promised. That’s why statistics show a steady migration from competitive UNIX® platforms to Power Systems. So, from time to time the competition chooses to release information that can be misleading.
Clients around the world are taking advantage of the New Power Equation with POWER™ processor-based performance, plus the efficiency, availability and manageability enabled through IBM-unique Power Systems Software designed to enhance the advanced capabilities of the POWER architecture.
Check out these facts that have the competition concerned:
- IBM is the only major vendor to gain revenue share in the UNIX segment for the past five years (+11.2 points) while both Sun (-1.9 points) and HP (-5.7 points) lost share, according to IDC.1
- IBM has helped nearly 1,200 customers migrate from Sun, HP and other UNIX platforms to IBM's AIX or Linux on Power.2
Claim: Sun says that the Sun SPARC Enterprise T5440 “blows away” the IBM Power 550 Express and Power 570 with up to 4 times faster performance and up to 2 times higher performance per WATT.
FACT: Many of the more than 500 clients who have migrated to Power Systems from Sun SPARC systems noted that outstanding POWER performance was a major factor in their decision to move to IBM.2
|Performance3||32-core Power 570||16-core Power 570||16-core Power 560||8-core Power 550||4-core JS22||32-core T5440|
|Cores / Threads / Processor Chips||32/64/16||16 /32 /8||16 /32 /8||8 /16 /4||4 /8 /2||32 / 256 / 4|
|SAP SD 2-tier users||14,432||8,000||3,104||7,520|
Power = Performance + Scalability:
The 5.00 GHz 16-core IBM Power 570 uses the industry’s fastest processor and performs at levels of up to 2.3 times better than the T5440 -- with ½ the cores. Compare the 32-core 4.20 GHz Power 570 to the 32-core T5440 and again the winner is clearly the 570 with 1.9 to 2.7 times the performance.
In fact, the scalability of the 570 is only one of the reasons to question why Sun would even try to compare its T5440 to this top-selling system with the following enterprise-class features:
The T5440 does not have any of these features. Therefore, it is more appropriate to compare it to the IBM Power Express product line. Get the facts: Dare to compare Sun’s M-Series to the Power 570 and 595 enterprise systems. (US)
Power = Performance - Costs:
Most software today recognizes cores and threads. However, Sun focuses on performance per processor chip. Why would Sun do this when major software venders, like Oracle and IBM, license database and application serving software based on the number of cores?
IBM concentrates on providing the most performance per core and the most performance per thread to minimize software cost. And, when you compare the Power Express line to the T5440, even the IBM BladeCenter JS22 has at least 2.25 times better performance per core.
|Performance per core4||16-core Power 560 Express||8-core Power 550 Express||4-core JS22 Express||32-core T5440|
|Cores / Threads / Processor Chips||16 /32 /8||8 /16 /4||4 /8 /2||32 / 256 / 4|
|SAP SD 2-tier users||388||235|
It is possible that having fewer chips could lead to lower energy usage, as Sun claims. But, Table 3 demonstrates that selecting the right form factor optimized for efficiency makes better ‘green’ sense. The BladeCenter JS22 has a greater than 70% advantage over the T5440. For the other IBM systems noted, the range is from Sun with a 6% advantage to Power with a 32% advantage.
|Performance per WATT5||16-core Power 560||8-core Power 550||4-core JS22||32-core T5440|
This analysis is based on the Maximum WATTs required for the systems, not the actual energy used in the benchmark. This is an important metric because the Maximum WATTs in site planning documentation defines the energy infrastructure that must be built into the datacenter. Power distribution, uninterruptible power supplies or motor generators, and other infrastructure are based on the maximum requirements. Actual energy usage will always be less than the maximum requirement.
Sun’s energy claim compares the T5440 ‘actual’ usage of 58% of maximum (as measured during a benchmark) to an ‘estimated’ usage of 80% for the IBM Power 560 Express. This comparison is not based on fact since IBM did not measure actual energy usage when running the benchmarks.
Claim: Sun says the T5440 is 1/5 the cost of the IBM Power 570
FACT: Purchase price is not the only important cost factor when choosing the right hardware platform. Total cost of ownership is a more valid consideration. This includes hardware, maintenance, operating systems and other SW costs – as well as operations costs like management, energy and physical space.
Power = Performance + Savings First, compare price/performance by looking at the purchase price of the systems used for the SPECint_rate2006 benchmark. The T5440 costs 25% more than the Power 560 and has 20% lower overall performance -- and much less performance per core. The really remarkable fact is that the 4-core BladeCenter JS22 has price/performance that is more than three times better than the T5440 and costs less than 10% of this 32-core server.
|Price7||16-core Power 560||8-core Power 550||4-core JS22||32-core T5440|
|Price / Performance8||16-core Power 560||8-core Power 550||4-core JS22||32-core T5440|
A complete cost comparison must include maintenance for the server, operating system and storage. The following table provides this detail.
|Maintenance9||16-core Power 560||8-core Power 550||4-core JS22||32-core T5440|
|Three Year HW||$14,400||$6,624||$879||$20,452|
|Three Year OS Maintenance||$19,392||$9,696||$3,228||$0|
|Three year maintenance||$33,792||$16,320||$4,107||$20,452|
Next, it is important to understand that software costs can often be higher than the cost of the associated hardware. Table 7 details Oracle database costs, while Table 8 shows the cost of the clustering option to improve database scalability and data availability.
|Oracle price10||16-core Power 560||8-core Power 550||4-core JS22||32-core T5440|
|EE List Price||$456,000||$228,000||$121,125||$855,000|
|Oracle RAC price11||16-core Power 560||8-core Power 550||4-core JS22||32-core T5440|
|RAC 3-year maintenance||$144,528||$77,418||$40,986||$291,456|
Table 9 shows that on a total cost of owner comparison, Power Systems are less expensive than the T5440.
|16-core Power 560||8-core Power 550||4-core JS22||32-core T5440|
|Total Cost 12||$1,336,253||$702,778||$327,619||$2,408,513|
Now look at the comparison of total cost of ownership divided by performance – a metric analogous to price / performance. POWER6-based systems range from two to more than 2.4 times better than the T5440. Regardless of any claim Sun makes about price, Power delivers twice the performance for the cost.
|16-core Power 560||8-core Power 550||4-core JS22||32-core T5440|
|Total cost / performance 13||$3,681||$3,315||$3,868||$8,002|
While energy, space, and management costs are not included, they are certainly important elements to consider. This is where the performance per WATT advantage will tip the scales in favor of Power systems over the T5440 in a total costs analysis comparison
IBM performed these calculations for SPECint_rate 2006 because this industry standard benchmark allows price / performance calculations. The rules for other such benchmarks prohibit these calculations. IBM encourages the calculation of costs based on actual workload with the confidence analysis will show that Power delivers two times or greater the price / performance of the T5440.
Claim: Sun says that cranking the clock frequency to scale performance is an evolutionary dead-end; while Sun and the industry as a whole move forward with multi-core, multi-thread processors. And, that IBM is the last holdout and is exhausting the old technique with higher power consumption and higher heat generation.
FACT: Sun appears to again ignore that software is licensed by the core, making performance per core a critical element in today’s data center economics.
Power = Performance + Innovation: Performance per core can reduce costs when running Oracle database, DB2, WebSphere Application server, BEA or other similar software. So, the better the performance per core, the lower the total cost of ownership. (Power System performance per core leadership is documented above.)
When it comes to the adoption of multi-core, multi-thread technology, IBM was the first to deliver dual core technology with the introduction of POWER4 in 2001. Dual core technology was not available in a Sun SPARC system until in 2004 And, while Sun supports eight threads per core with the T5440 and other systems using the UltraSPARC T2 Plus processor, only two threads per core can be simultaneously executing – the same threads per core that can be executing on Power systems.
Sun’s power consumption and heat generation claims do not consider the performance per WATT advantage of Power, as noted above. Statistics show that the Power 550 and Power 560 are more efficient than the T5440. And, the JS22 blows has an astounding 70% more efficiency.
The business value of IBM’s Green Data Center (US) is delivered with the Power 570. In fact, the new 32-core Power 570 is nine times better in performance per WATT than its predecessor introduced seven years ago – the p670.14
IBM is synonymous with innovation. But, taking care of clients by providing platform consistency and real business value with the delivery of innovation is what differentiates Power from the competition. This commitment is proven with the introduction of each new POWER processor generation and efficient form factors like modular 570 and blade technology
Claim: Sun says that instead of creating their own technologies, IBM is borrowing competitors' innovations, especially Sun's Solaris OS technologies in their new proprietary AIX 6.1 release.
FACT: Here is a ‘fact-check’ on specific technologies that Sun alleges AIX copied from Solaris.
Why would Sun want to have an ‘innovation duel’ between Solaris and AIX? Here’s a recap of why AIX is an acclaimed UNIX operating system:
Sun has little to compare with Power and AIX in the area of virtualization. These IBM innovations are notable:
Claim: Sun says its products and services portfolio delivers breakthrough performance with dramatic space and power efficiency. And that the free and open Solaris OS is a proven, industry leading operating system with features designed to save your business time and money.
FACT: The metrics in the following table counter this claim with industry standard benchmark-based comparisons between a Power blade server and the T5440. The more efficient system is the BladeCenter JS22, designed for efficiency. Where space or power consumption are key concerns, a Power blades configuration or a highly virtualized environment could be the best solution.
|Performance per rack unit 15||IBM BladeCenter JS22 Express||Sun SPARC Enterprise T5440|
Claim: Sun says that on its platform Linux customers get the no-cost license, open-source applications and community support that attracted so many to Linux, but with all the renowned enterprise-class backing that Sun can offer.
FACT: Instead of providing Linux-like capabilities, all Power servers support Linux. This mean that Power users can take advantage of all open-source applications available on Linux, while Sun users are limited to those available on Solaris.
1. IBM is the only major vendor to gain revenue share in the UNIX segment for the past five years (+11.2 points) while both Sun (-1.9 points) and HP (-5.7 points) lost share. (Source: IDC) IDC Worldwide Quarterly Server Tracker, August 2008.
2. Source: IBM internal numbers.
SPEC® and the benchmark names SPECrate®, SPECint®, and SPECjbb® are registered trademarks of the Standard Performance Evaluation Corporation. For the latest SPEC benchmark results, visit http://www.spec.org. Competitive benchmark results reflect results published as of October 21, 2008. The results are the best IBM POWER systems 32-core and lower results compared to the Sun SPARC Enterprise T5440 results.
|Number of Benchmark Users||Operating System||ERP Release||Central Server||Chips||Cores||Threads||Technology||GHz||Certificate Number|
|8000||AIX||6.0 (2005)||IBM Power 570||8||16||32||POWER6||4.7||2007039|
|14432||AIX||6.0 (2005)||IBM Power 570||16||32||64||POWER6||4.2||2008057|
|3104||Redhat Enterprise Linux 5||6.0 (2005)||IBM System p 550||4||8||16||POWER6||4.2||2008002|
|7520||Solaris 10||6.0 (2005)||Sun SPARC Enterprise T5440||4||32||256||UltraSPARC T2 Plus||1.4||2008058|
For full details on these and other SAP benchmarks, see SAP SD Standard Application Benchmark Results, Two-Tier Internet Configuration at http://www.sap.com/solutions/benchmark/sd2tier.epx
|SPECint_rate2006 results as of 10/21/2008||Cores||Chips||Cores/Chip||Threads/Core||Peak|
|IBM Power 570 (4.20 GHz, 32 core)||32||16||2||2||832|
|IBM Power 570 (5.00 GHz, 16 core)||16||8||2||2||542|
|IBM Power 560||16||8||2||2||363|
|IBM System p 550 (4.20 GHz, 8 core)||8||4||2||2||212|
|IBM BladeCenter JS22 (4.00 GHz, 4 core)||4||2||2||2||84.7|
|Sun SPARC Enterprise T5440||32||4||8||8||301|
|SPECfp_rate2006 results as of 10 /21/2008||Cores||Chips||Cores/Chip||Threads/Core||Peak|
|IBM Power 570 (4.20 GHz, 32 core)||32||16||2||2||602|
|IBM Power 570 (5.00 GHz, 16 core)||16||8||2||2||544|
|IBM Power 560||16||8||2||2||263|
|IBM System p 550 (4.20 GHz, 8 core)||8||4||2||2||178|
|IBM BladeCenter JS22 (4.00 GHz, 4 core)||4||2||2||2||75|
|Sun SPARC Enterprise T5440||32||4||8||8||230|
|SPECjbb2005 results as of October 21, 2008||JVM Instances||Cores||Chips||HW Threading||bops||bops/JVM|
|IBM Power 570 (4.20 GHz, 32 core)||16||32||16||Yes||1,390,087||86,880|
|IBM Power 570 (5.00 GHz, 16 core)||8||16||8||Yes||867,989||108,499|
|IBM Power 560||8||32||16||Yes||593,904||74,238|
|IBM Power 550 (4.20 GHz, 8 core)||4||8||4||Yes||350,642||87,661|
|Sun SPARC Enterprise T5440||32||32||4||Yes||692,736||21,648|
4. Performance per core = the performance shown in Table 1 divided by the number of cores for each benchmark shown in footnote 1
5. Performance per WATT
Benchmark results reflect results published as of October 21, 2008. Each result is the best performing result for the benchmark. The benchmark results and detail can be found in footnote 1.
Performance/WATT is calculated by dividing the performance from the Table 1by the recommended maximum power usage for site planning. This defines the requirement for the power infrastructure. Actual power used by the systems will be less than this value for all of the systems. For Power Systems, this information is available in the respective site planning guides for each system available through www.ibm.com. For Sun systems, this information is available through Site Planning Guides available through www.sun.com. All data is from those documents and is current as of October 21, 2008
|System||Maximum Energy Requirement (WATTs)|
|IBM Power 570||5,600|
|IBM Power 560 Express||2,400|
|IBM Power 550 Express||1,400|
|IBM BladeCenter JS22||430|
|Sun SPARC Enterprise T5440||2629|
6. “IBM Power 560 (2 building blocks) power specifications calculated as 80% of maximum input power (2,400W) reported on 10/5/08.” This quote is in the notes for Benchmark outcome 6 for SPECjbb2005. Sun measured 1525 WATTs for the T5440 which is 58% of the maximum power requirement of 2629 WATTs.
7. All prices are list prices in USD as of October 21, 2008. Prices from resellers may vary. Prices are subject to change without notice. Prices based on configurations used in the SPECint_rate2006 benchmark including processors, memory and disk. All IBM systems used internal storage. The price of the JS22 includes a BladeCenter H Chassis. Since that chassis could be shared by 14 blades, the effective price per blade could be lower than is shown here. Sun SPARC Enterprise used Sun StorEdge 6140 Array for disk. Sun prices from the vendor Website.
8. Price / performance calculate by dividing the prices in Table 4 by the performance in Table 1.
9. Maintenance prices are for three years. Sun SPARC Enterprise T5440 warranty upgrade to level Gold 24 x 7 selected because that is the recommended upgrade from Sun.
10. Oracle Database Enterprise Edition list price in USD from Oracle Store on October 21, 2008 available at http://oraclestore.oracle.com/OA_HTML/ibeCCtpSctDspRte.jsp?section=11365&media=os_g_english_help_
11. Oracle Real Application Cluster list price in USD from Oracle Store on October 21, 2008 available at http://oraclestore.oracle.com/OA_HTML/ibeCCtpSctDspRte.jsp?section=11365&media=os_g_english_help_
12. Total cost is calculated by summing the price in Table 4 with the Maintenance price in Table 6 plus the Oracle Database Enterprise Edition price in Table 7 plus the Oracle Real Application Clusters Option price in Table 8.
13. Total cost / performance calculated by dividing the Total Cost in Table 9 by the SPECint_rate2006 performance from Table 1.
14. Based on comparison to the predecessor product, p670 using rPerf as the performance metric.
15. Performance per rack unit calculated based on dividing the performance in Table 1 by the number of Rack Units (1.75” in a standard 19” rack) that each system requires. The 4 RU height of the T5440 is from the Tech Specs at www.sun.com. The rack units for the JS22 were calculated based on filling a 9u BladeCenter H Chassis with 14 BladeCenter JS22 servers.