Oracle Sparc T5 server is not faster than IBM server, claims IBM

BANGALORE, INDIA: Seems like Oracle, who recently unveiled its new line of Sparc servers, went a little overboard while laying out figures that state that its Spacrc T5 servers are 'faster, cheaper and more efficient' than its rival IBM's servers.

Let us see whether they are indeed as 'faster, cheaper and more efficient' than IBM servers, as being claimed by its developers. And if not, where do they fall short of.

Oracle had claimed that the Sparc T5-8 server is 2.4 times faster per chip compared to IBM Power 780 three-node cluster results for TPC-C tpmC and is 2.5 times less expensive per $/tpmC.

However, here Oracle had not specified that the T5-8 TPC result was obtained on a single node of T5-8 with 128 cores of T5 at 3.6 GHZ and is non clustered. The per core performance of T5 at 3.6 GHZ is around 66817 tpmC. Oracle is here comparing a clustered benchmark on Power 7 against non clustered benchmark on T5.

Our sources say, "The comparison Oracle is making is against Power 780 was a clustered benchmark on 3 numbers of P780 each with 64 cores of P7 at 3.86 GHZ. IBM Power 780 with 8 cores at 4,14 GHZ of Power 7 (http://www.tpc.org/tpcc/results/tpcc_result_detail.asp?id=110041301&layout=) delivers 1200011 tpmC which is a non clustered benchmark. So, IBM POWER 7 has got 2.25 times better per core performance compared to Oracle T5 at 3.6 GHZ. Oracle compares SPARC T5 against P7 at 3.86 GHZ,when IBM has already started shipping POWER7+ based systems from October 2012 ,which on an average delivers 20 per cent better per core performance compared to P7."

Oracle had also claimed that its Sparc T5-8 server delivers 1.4 times better performance than the next best single system result, the 32-processor IBM Power 595, at one-fifth the price/performance. One thing to be noted is that here again Oracle is comparing T5 against Power 6 technology, when today IBM is shipping server portfolio based on Power 7/7+ processors.

"Power 6 core at 5 GHZ performs 1.4 times better than a T5 core at 3.6 GHZ. Compared to P6 core at 5 GHZ (highest Clock) P7 core at 4 GHZ is around 35 per cent better in performance and similarly a P7+ core at 4.42 GHZ performs around 48 per cent better than the P6 core at 5 GHZ. So it is not the right technology comparison, unless Oracle feels T5 at best could afford to compete against P6. Most importantly per core performance is the most important criteria for selecting a hardware platform because better per core performance leads to lower Middleware license and support cost which inturn helps a customer lower his operating expenses," adds the source.

Oracle had also claimed that the T5-8 server has 2.8x times better performance than the 4-processor IBM x3850 X5 system equipped with Intel Xeon processors, also with a lower price/performance. Whereas, the benchmark was done with the Intel Server with 4 Sockets 40 cores, and 80 threads, which was done in 2011.

"Performance per CPU is not an accurate way of measurement, as there is a huge variation in CPUs. The Oracle T5 has 16 cores per CPU, with 8 threads per core, whereas the Intel Xeon E7-8870 has only 10 cores per CPU, and 2 threads per core. Yet, if you look at the performance of each core, the x3850X5 performs 12 per cent faster than the Oracle T5. To make things even more interesting, if you look at the performance per thread, the 3850X5 is 4.5X more powerful than the Oracle T5. And it costs 37 per cent of the price of Oracle system," adds the source.

Oracle. during the announcement, had also claimed that its T5-8 server delivers 1.4 times the performance per chip compared to IBM Flex x240 result. However, here performance is being compared on a 2S blade server, not costing more than 5 Lakhs, vs the Oracle system with 8 sockets, and 128 cores, costing much higher. So, even when you look at the price/tpmc, it is not a fair comparison. Moreover, performance per CPU is not an accurate way of measurement, as there is a huge variation in CPUs.

"The Oracle T5 has 16 Cores per CPU, with 8 threads per core, whereas the Intel Xeon E5-2690 has only 8 cores per CPU, and 2 threads per core. Yet, if you look at the performance of each core, the x240 performs 40% faster than the Oracle T5. To make things even more interesting, if you look at the performance per thread, the 3850X5 is 5.6X more powerful than the Oracle T5. And it costs 16.7 per cent of the price of Oracle system. So, at a performance per core, or even performance per thread - the highest performing server is the Power server," adds the source.