Re: UltraSPARC IV

So, the reason they claim almost twice the throughput is because the ultrasparc iv is a multi-core chip. That is to say, while there is a single processor form factor (one little module you whack into the cpu/memory board), there are two instruction cores internally. This means that the processor can simultaneously execute two processes at the same time, one on each core.

The dodgy parts about it is that while it is two separate instruction cores, there is shared cache, which is a bit of a performance hit, and the cores share access to the memory bus, which can also be a performance hit. The advantage is that its an in-place upgrade for most of the mid- to high-end servers - pull out your single-core chips and put in the multi-core and double your effective processor count.

But - since you're not actually increasing clock speed, you're increasing parallelization, are you actually doubling your performance?   Depends how you define that. Any single thread is going to execute at pretty much the same speed (there is a bit of a clock speed jump from the IIIs to the IVs, to be fair), but you'll be able to run twice as many threads at once, so there's less time waiting in the scheduler queue. But, this is not the same thing as a doubling of execution time, so its a little disingenuous to claim a doubling of performance.

Multi-core processing is different from Intel's HyperThreading technology - in that case, its a single processor core that shows itself as two processors. This is because Intel has built a certain amount of possible parallelization into their processors already that generally wasn't being taken advantage of by compilers and software engineers. So, the idea there was make a processor look like two processors, let the OS throw two simultaneous processes at it, and the processor will attempt to parallelize internally. It works alright, there's a variety of logistical problems with it.

The REALLY important thing, though - Oracle counts every core of a multi-core processor as a processor. This means if you swap 8 Ultrasparc IIIs for 8 Ultrasparc IVs, you have to license a total of 16 processors for Oracle. The US IV is not the normal situation of "get a processor upgrade to improve performance without licensing more processors " - this physical upgrade requires that you buy more Oracle licenses. Needless to say, sun is not particularly thrilled about that, but since Intel and IBM have also said that multi-core CPUs are the way of the future, Oracle has taken a stand early on that every core counts as a processor and must be licensed as such.

Hope this helps,
Matt

--
Matthew Zito
GridApp Systems
Email: mzito_at_gridapp.com
Cell: 646-220-3551
Phone: 212-358-8211 x 359
http://www.gridapp.com


On Jun 4, 2004, at 6:07 PM, Tim Johnston wrote:



> Mladen Gogala wrote:

>

>> How did they compare? Hopefully, that claim is not based on dhrystone 

>> MIPS?

>> You know what MIPS stands for, don't you? (MIPS = Marketing Invention 

>> for Pushing Sales).

>>

> I don't even have that...  Only the marketing statements with no real 

> numbers to back it up...  Things like:

>

> "Developed in concert with the SolarisTM Operating Environment to meet 

> the performance, reliability, and scalability requirements of 

> mission-critical enterprise, HPTC, and other compute intensive

> applications, the UltraSPARC IV processor can almost double the 

> current UltraSPARC III processor-based system throughput."

>

> "Provides up to 2 times the application throughput of UltraSPARC III 

> processor"

>

> "Twelve UltraSPARC IV processors execute 24 simultaneous computing 

> threads."

>

> "This CMT technology nearly doubles current compute densities and 

> reduces overall heat dissipation, resulting in significant end-user 

> cost-benefit savings."

>

> " Up to 12 dual-threaded, near-linear scalability, UltraSPARC IV 

> processors with 16 MB cache."

>

> I'm searching for something with a little more meat...  One of the 

> reasons for this email...

>

> :-)

>

> -- 

> Regards,

> Tim Johnston

> Tel: 978-322-4226

> Fax: 978-322-4100

>

>

>

> ----------------------------------------------------------------

> Please see the official ORACLE-L FAQ: http://www.orafaq.com

> ----------------------------------------------------------------

> To unsubscribe send email to:  oracle-l-request_at_freelists.org

> put 'unsubscribe' in the subject line.

> --

> Archives are at http://www.freelists.org/archives/oracle-l/

> FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html

> -----------------------------------------------------------------


----------------------------------------------------------------
Please see the official ORACLE-L FAQ: http://www.orafaq.com
----------------------------------------------------------------
To unsubscribe send email to:  oracle-l-request_at_freelists.org
put 'unsubscribe' in the subject line.
--
Archives are at http://www.freelists.org/archives/oracle-l/
FAQ is at http://www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------