About X86 server architectures with 16 channels and above

INTELX86 architecture CPU, very few more than 16 way. HP had 32 way servers based on Itanium (non-x86) cpus, but itanium software was not supported by HP itself. Then Huawei, inspur also made in the national project based on itanium 32 way server. Since INTEL and HP have given up their support for Itanium, this national project is just some accumulation of technology, and it is difficult to have a good growth in the ecosystem.

Now, as Xeon CPUS (X86) continue to improve, HP can use its chip interconnection technology to transform cpus that can only be composed of 2 channels into 16 or more channels, and the performance of expansion also presents a good linearity.

SGI is even more different. SGA was born to be a supercomputer.

Welcome to reprint, reprint please indicate the source: blog.csdn.net/notbaron/ar… \

1. Some SPEC performance metrics

1.1             speccpu

SPEC is short for Standard Performance Evaluation Corporation. SPEC is a non-profit organization composed of computer manufacturers, system integrators, universities, research institutes, and consulting companies whose goal is to establish and maintain a set of standards for evaluating computer systems.

SPEC CPU 2006 is the latest version of the CPU subsystem evaluation software released by the SPEC organization. In earlier years, SPEC CPU 2000 was used in the industry. Like SPEC CPU 2000, SPEC CPU 2006 includes CINT2006 for measuring and comparing integer performance, and C FP2006 for measuring and comparing floating point performance. SPEC CPU 2006 upgraded some of the tests in SPEC CPU 2000 and dropped/added some tests, so the scores of the two versions are not comparable.

In SPEC CPU testing, the processor, memory subsystem, and compiler used by the test system (SPEC CPU provides the source code and allows the test user to perform some compilation optimizations) all affect the final test performance. The I/O (disk), network, operating system, and graphics subsystems have very little impact on SPEC CPU2006.

An ounce of honest data is worth a pound ofmarketing hype is SPEC’s motto. To keep data fair, credible, and valid, The SPEC CPU tests use real-world applications instead of benchmarking with circular arithmetic operations. SPEC CPU 2006 includes 12 integer operations and 17 floating point operations. In addition, there are two random number generation tests, 998.sperand (integer) and 999.specrand (floating point), which are included in the suite and run, but are not timed to score. These two tests are mainly used to verify the correctness of PRNG random number generation functions used in other components. Each test component is basically written in C and Fortran languages. Seven test projects use C++ language, and Fortran language is used to write floating point parts.

CINT2006 includes C compiler, quantum computer simulation, chess program, etc. CFP2006 includes finite element model structured mesh method, molecular dynamics particle method, fluid dynamics sparse linear algebra, etc. To simplify test results, SPEC decided to use a single number to summarize all 12 integer benchmarks. The specific method is to standardize the execution time of the computer under test by dividing the execution time of the computer under test by the execution time of a reference processor. The result is called SPECratio. A higher SPECratio value indicates faster performance (since SPECratio is the inverse of the execution time). The comprehensive test result of CINT2006 or CFP2006 is the geometric mean of SPECratio.

 

2.1 the query

Let’s take a quick look at the CPU performance metric SPEC CPU2006

Spec.org/cpu2006/res…

Open figure 1 below

支那

支那

Click below:

The following figure 2

支那

支那

Figure 3 below appears

支那

支那

One 16-channel and one 32-channel performance indicator.

SGI’s 16-way 16X 8890V2SGI UV 300, SPEC CPU INTRATES value is 9470

SGI 32-way 32X 8890V2SGI UV 300, SPEC CPU INTRATES value is 18600

It was found that the linearity of its performance reached 1.96 when it changed from 16 channels to 32 channels

The SGI UV 300 is really advanced.

Toad looked at HP, which also has related servers, as shown in Figure 4 below

支那

支那

HP uses the 2890 V2, but the performance of the 2890 and 8890 is the same, but the price and QPI quantity are different.

In general, 2890V2 can only be made into a 2-way CPU, but HP used its own chip to make a 32-way CPU from a 2-way CPU. It’s also awesome.

HP 16 x 2890V2 Superdome X SPEC CPU INT RATES 9400

Almost as good as SGI.

These two currently set a new benchmark in cpus based on the X86 architecture.

Do not know the follow-up will have new manufacturers come in to play ~~

Many of us know HP, but not SGI, so let’s take a look at what SGA does.

2.  SGI

SGI is the industry’s leading provider of high-performance computing systems, complex data management and visualization products. It offers the world’s best server family as well as visual workstations with supercomputing power. SGI is one of Fortune magazine’s 500 largest companies/manufacturers in the United States, with over $4 billion in annual revenue. With an average growth rate of 40-50% over the past five years, the company is one of the fastest growing computer companies in the world. In the industry, the company takes the lead in integrating RISC technology, balanced multiple processing technology, digital media technology, computer graphics technology, UMA and CCNUMA architecture and other core technologies in the computer field, forming its own unique style, creating a new direction of visual computing technology and information processing. SGI currently accounts for more than half of the world’s 500 largest supercomputers in supercomputing.

S is for supercomputer, G is for graphics workstation, and I is for breakthrough insight. With our servers, you have extraordinary computing power to help you solve your toughest problems. Our graphics workstations can view, manipulate, and use data in a more visual way. When you combine with our worldwide service team, you can get the solutions you need to achieve your vision.

 

 

 

3. About the HP SuperDome X

Most X86 high-end servers use INTEL’s official CPU interconnection architecture, which is inefficient.

HP SuperDome X uses its own eXternalNode Controller (XNC2).

The following figure 5

 

His 16P architecture is likely to look like Figure 6:

 

 

 

 

 

 

 

4. About SGI UV300

Based on the original UV 300H device, it is currently used in SGI’s UV for SAPHANA products. It is a single-node system based on the Intel xeon E7 8890 v2 15 core processor with a maximum expansion capacity of 32 slots (i.e. 480 computing cores) and a total of 24TB of unified shared memory from 32 gb dimms.

Toad omitted a zillion thumbs up for this architecture

HP and SGI are somewhat similar in that any memory access between two cpus does not go through a third CPU.

 

 

 

5. About Top 500 servers

top500.org/list/

Check out the top 10 machines of 2015 below:

5.1 Top 10 Server Systems in 2015

RANK SITE SYSTEM CORES RMAX (TFLOP/S) RPEAK (TFLOP/S) POWER (KW)
1 National Super Computer Center in Guangzhou China Tianhe-2 (MilkyWay-2)  – TH-IVB-FEP Cluster, Intel Xeon E5-2692 12C 2.200GHz, TH Express-2, Intel Xeon Phi 31S1P NUDT 3120000 33862.7 The 54902 activity 17808
2 DOE/SC/Oak Ridge National Laboratory United States Titan– Cray XK7, Opteron 6274 16C 2.200GHz, Cray Gemini Interconnect, NVIDIA K20x Cray Inc. 560640 17590.0 versus 27112 8209
3 DOE/NNSA/LLNL United States Sequoia – BlueGene/Q, Power BQC 16C 1.60 GHz, Custom IBM 1572864 17173.2 20132.7 7890
4 RIKEN Advanced Institute for Computational Science (AICS) Japan K computer, SPARC64 VIIIfx 2.0GHz, Tofu interconnect Fujitsu 705024 10510.0 versus The 11280 activity 12660
5 DOE/SC/Argonne National Laboratory United States Mira – BlueGene/Q, Power BQC 16C 1.60GHz, Custom IBM 786432 8586.6 10066.3 3945
6 Swiss National Supercomputing Centre (CSCS) Switzerland Piz Daint– Cray XC30, Xeon E5-2670 8C 2.600GHz, Aries Interconnect, NVIDIA K20x Cray Inc. 115984 6271.0 versus 7788.9 2325
7 King Abdullah University of Science and Technology Saudi Arabia Shaheen II – Cray XC40, Xeon E5-2698v3 16C 2.3GHz, Aries interconnect  Cray Inc. 196608 5537.0 versus 7235.2 2834
8 Texas Advanced Computing Center/Univ. of Texas United States Stampede – PowerEdge C8220, Xeon E5-2680 8C 2.700GHz, Infiniband FDR, Intel Xeon Phi SE10P Dell 462462 5168 us 8520 us 4510
9 Forschungszentrum Juelich (FZJ) Germany JUQUEEN – BlueGene/Q, Power BQC 16C 1.600GHz, Custom Interconnect IBM 458752 5008.9 5872.0 versus 2301
10 DOE/NNSA/LLNL United States Vulcan – BlueGene/Q, Power BQC 16C 1.600GHz, Custom Interconnect IBM 393216 4293.3 5033.2 1972

 

 

5.2 Tianhe-2

Tianhe-2, a supercomputer system developed by the National University of Defense Technology, topped the list with a peak calculation speed of 549 trillion calculations per second and a sustained calculation speed of 336 trillion double-precision floating-point calculations per second, making it the world’s fastest supercomputer.

On November 17, 2014, China’s Tianhe-2 won the top spot for the fourth time in a row, nearly twice as fast as the second-ranked Titan of the United States.

In May 2015, tianhe-2 successfully conducted a cosmological numerical simulation of 3 trillion particles, neutrinos and dark matter, revealing the long evolution of the universe from 16 million years after the Big Bang to about 13.7 billion years ago.

China’s Tianhe-2 topped the list of the world’s top 500 supercomputers for the fifth time with a speed of 33.86 petaflops (petaflops), according to the latest list released at the International SuperComputer Conference 2015 in Germany.