Recently, a customer asked to review the value proposition of Sun's UltraSPARC T2 servers vs x86 servers. I started off by providing some benchmark results. This page on Sun's web site is a good starting point to access those benchmark results:
For my customer, I went through the benchmark and produced this summary with some of my own comments. When ever there is a performance statement, I have included links to the benchmark results that back up that statement
Integer performance (throughput)
SPARC Enterprise T5220 32% faster than Sun Blade x6520 with two 3-GHz Intel Xeon Processors
The Sun CMT servers are great for running many tasks simultaneously. The SPEC CPU benchmark that measures that capability is the CINT2006 Rate. 1 chip T2 system Sun Microsystems SPARC Enterprise T5220 (gccfss) delivered a CINT2006 Rate of 83.2 which was about 32% faster than system Sun Microsystems Sun Blade X6250 (2 chip Xeon) delivered a CINT2006 Rate of 65.0 or 2.5 times faster than a single chip Xeon system Dell Precision 690 (Intel 5160, 3.00 GHz) which delivered a CINT2006 Rate of 33.5.
Sun SPARC Enterprise T5220 95% faster than 3-GHz Intel Xeon Processor
Sun made major improvements in the T2 compared to the T1 chip in the area of floating point performance. T2 chip 2.5 time the performance of a Dell Precision 690 (Intel Xeon 5160, 3.00 GHz - 1 chip 2 core)
Sun SPARC Enterprise T5220 obtained a world record SPECweb2005 result 37,001 SPECweb2005 with one UltraSPARC T2 running Solaris 10 with Sun Java System Web Server
T5220 server delivers 22% greater performance than the four-socket HP ProLiant DL580 G5 with 2.9 GHz Quad-Core Xeon processors
There are no IBM POWER6 results on the SPECweb benchmark
3.1x Dell PE2900: 2 x Dual Core Xeon Windows Server
SPECjAppServer2004 World Record Single-Application server Sun SPARC Enterprise T5220 and a Sun SPARC Enterprise T5120 (Oracle database server)
This result used a Sun SPARC Enterprise T5220 Application Server (single UltraSPARC T2 1.4GHz) and a Sun T5120 Server Database System (single UltraSPARC T2 1.2GHz)
Delivered better results that two Rackable servers, each with 2 sockets with quad core Xeon chips
Delivered better results that two IBM POWER servers
World Record Lotus Domino R6iNotes Sun SPARC Enterprise 5220
The Sun SPARC Enterprise T5220 (one UltraSPARC T2), using Lotus Domino 7.0.1 mail server delivered industry-leading results with best per-socket performance and best power-performance for the Lotus[R] R6iNotes on Domino mail server benchmark
Sun SE T5220 has similar results to an 8 process UltraSPARC IV+ 1.8GHZ V890 at less than half the cost per user
The Sun SPARC Enterprise T5220 performed 2.3x faster than the 4x dual-core HP Proliant DL580 and achieved 30% less price/performance
World Record ERP SAP-SD 2-Tier ECC 6.0 Sun SPARC Enterprise T5120
Best 1 processor, two SAP SD Standard Application Benchmark
The Sun SPARC Enterprise T5120 server (1.4GHz UltraSPARC T2) outperformed the 4-core IBM System p570 (4.7 GHz POWER6) by 7%. The 4RU IBM p570 POWER6 is 4 times larger than the 1RU Sun SPARC Enterprise T5120 system
The Sun SPARC Enterprise T5120 server (1.4GHz UltraSPARC T2) outperformed the HP BL460C with two 3 GHz Xeon quad-core processors by 5%
"a consolidation benchmark to see how an open-source stack performs against the proprietary stack from Microsoft. Solaris, MySQL, and Sun Web Server running on the open-source UltraSPARC T2 processor were pitted against a Microsoft SW stack running on a 4-socket QC Xeon server. This benchmark highlights the continued trend to incorporate MySQL open-source databases and how it works under virtualization (Solaris Zones)"
That benchmark is not an audited benchmark like SPEC . It uses an "Ad-Hoc OLTP workload, called iGEN OLTP 1.6, which was developed from a realistic customer workload. iGen OLTP avoids problems that plague other OTLP benchmarks like TPC-C".
The results are quite interesting and are documented here:
That is actually an excellent and interesting blog. Some of "pro-Sun" stance and railing against bogus or incomplete competitive benchmarks might seem a bit opinionated but those opinions are well substantiated. I love the "Dodge Charger" comparison that was posted recently in that blog:
Look what you can do with not mentioning important information, let's look at the following claims about a car...
Dodge Charger gets 18/26 mpg due to official testing.
Dodge Charger top speed over 244 mph.
Sounds great I'll get one because both statements are true... or is this being misleading? One might assume they are basically the same car, but unfortunately only the nameplate is the same. It is very misleading unless you know that one is the street car and one is the NASCAR "Car of Tomorrow" (COT). It would have been very clear to the consumer if it was written as follows:
Dodge Charger (2.7L V6 190hp) gets 18/26 mpg due to official testing.
Dodge Charger (5.86L Pushrod V8 ~865hp) top speed over 244 mph.
SPEC there are only two more things to add to a disclosure statement (GHz & amount of memory)! By the way...
Dodge Charger (2.7L V6 190HP) gets 18/26 mpg due to official testing, MSRP: $21,320 - $36,355.
Dodge Charger (5.86L Pushrod V8 ~865hp) top speed over 244 mph. cost?
The Sun UltraSPARC T2 CMT servers fit certain applications and workloads very well and others, such as single threaded apps, not so well. That is true for other CPUs. For example, the quad-core Xeons, which Sun sells in some of its systems, have good memory allocation and transfer for a small number of threads, but not so good for a large number of threads. Sun used the open-source sysbench memory benchmarking tool to show up-to 17GB/s for Niagara-2, while only up-to 4GB/s for an Intel 16-core box:
Picking the right CPU and System to meet your requirements can be a challenge, as each has its strengths and weaknesses. Benchmarks can help you pick the right system, Just make sure the benchmarks include all the factors that are really important to you:
