SPARCVII+ vs. T5 comparision

UNIX Solaris
1

Hello, I need migrate our applications from M-series to T-series. I would like compare two different SPARC processors - SPARCVII+ from Fujitsu and T5 from Oracle. I found only next benchmarks:

M8000:

_http://c970058.r58.cf2.rackcdn.com/individual_results/Oracle/Oracle_M8000-16_1TB_TPCH_ES_060311_V3.pdf
16 SPARC64 VII+ 3000 MHz Processors, 64 cores, 128 threads, 512GB
209,533.6 QphH@1000GB

SPARC T5-4 Server:

_http://c970058.r58.cf2.rackcdn.com/individual_results/Oracle/Oracle_T5-4_3TB_TPCH_ES_060713.pdf
4 SPARC T5 3.6GHz Processors, 64 cores, 512 threads, 2TB
409,721.8 QphH@3000GB

For comparing I choosed T5-2 which has 1/2 of T5-4 performance (410k/2 QphH) and same QphH performance as M8000:

-------------------------------------------------------------------------
|			|	M8000		|	T5-2		|
-------------------------------------------------------------------------
| QphH			|	~210k		|  ~205k (1/2 of T5-4)	|
| CPUs			|	16		|	4		|
| cores per CPU		|	4		|	16		|
| threads per CPU	|	8		|	128		|
| total cores		|	64		|	32		|
| total threads		|	128		|	256		|
-------------------------------------------------------------------------

So question is, if I will have full filled our M8000 box with 16 CPUs and migrate it to T5-2 Server, it will be enough for CPU performance?

2

Just setup a meeting with your Oracle rep. They have some internal tools to evaluate your current needs and estimate the hardware you need to replace it.

3

Note that the T (Niagara) architecture runs the CPU cores with half the given clock speed i.e. 1.8 GHz.
Normal single-thread applications run slower than on the M series.

4

Could you elaborate on that? http://www.oracle.com/us/products/servers-storage/servers/sparc/oracle-sparc/t5-2/sparc-t5-2-ds-1922871.pdf specifies the CPU frequency as 3.6GHz...

5

I have run single-thread integer and FPU benchmarks on Niagara T1 and T2 processors,
and compared with various other Sparc systems (M series and older Sparc Fire systems).
The factor (execution speed versus clock rate) was about 0.5.

6

Assuming that newest SPARC processors (T4 and T5) behave like T1, which is 8 years old, is unwise in my opinion. T4 is known for it's boost for single threaded processing (which T5 inherited).

7

Agreed... And the statement that the server runs the cpu at half the post speed isn't correct. Rather, we could say that the observed workload resulted in half the performance per cycle on the T series tested versus the M series tested.

The older T series had a strong single threaded performance penalty. The newer stuff is a lot more flexible and can deliver better single threaded performance than a comparable M series.

8

So I need to know if I compare both systems in my table - full M8000 rack with 16 CPU's and T5-2 Server, if they have same computing performance. Accord tpc.org they have, but how it's in reality? Is it possible replace this M8000 by T5-2 Server?

If yes, it will be good for total cost, because I'll save 1/2 of costs for Oracle software licences for databases and middleware.

SPARC64 VII+ and T5 have same core factor 0.5

9

I understand what you're saying in theory. I find out if it would work in the real world, you should work closely with Oracle on a comprehensive review of your needs. The t5-2 is very powerful, but I doubt you could replace a fully loaded m8000.

10

I'll give you a hand in looking at real workload system solutions... Checkout the
Oracle.com/optimizedsolutions pages. From there your local Oracle rep, or Oracle resell partner can get you the best architecture for you needs.

Cheers!
Sean

Full disclosure, yes, I work for Oracle. But I'm not a sales rep.
My job is to help you get the best hardware for your Oracle DB/App/Web software environment.
Although I'd prefer you get Oracle, but if you get from our competition, I'd appreciate if you would let me know why & how we can make Oracle systems better to win you back.