BM Seer Unofficial thoughts from an anonymous Sun employee

« Lack of proper infor... | Main | SugarCRM v4.5.1i:... »

IBM core confusion

Wednesday Mar 26, 2008

IBM only seems to pin all of their marketing on core count (metric/core). This is disingenuous.

IBM does not compare benchmark results on cost of systems, #RU, hardware cost, TCO, $/perf, watt/performance, etc. I guess IBM would just lose if they use metrics that are important to customers?

Any one can do the math, just ask IBM for price quotes on 4.7GHz with a reasonable 8 GB/core of memory, then divide price by number of cores. OUCH. IBM has by far the most expensive cores industry.

But are are those expensive IBM cores fast enough to give you better performance or $/perf? Nope, Sun's CMT servers cleans their clocks - server to server. Just look at the many benchmarks where a Sun T5220/T5120 beats a 4-core, 4RU 4.7GHZ IBM p570 (with better perf, $/perf, watt/perf, SWaP, TCO, etc). http://www.sun.com/servers/coolthreads/benchmarks/index.jsp

humor with a point to follow: Why do IBM cores cost so much? maybe because to support an IBM core requires so much expensive stuff.

IBM p570 (4RU, 2 CPU, 4 cores) weight 140lb, or 35.0lb/core
Sun T5220(2RU, 1 CPU, 8 cores) weight 52lbs, or 6.5lb/core

[8] Comments

Like this post?

Posted at 06:00AM Mar 26, 2008 in IBM Follies
Tags: cmt ibm performance

Comments:

Christ, your like a stuck record !

Stop comparing enterprise class machines like p570 against entry class T2000 class web servers.

Posted by Alex on March 28, 2008 at 01:59 AM PDT #

So if the 2-chip p570 has
* worse performance than a single UltraSPARC T2,
* worse power-performance than a single UltraSPARC T2,
* worse $/perf than a single UltraSPARC T2,
* worse TCO than a single UltraSPARC T2,

...I guess the only thing IBM can do is say best per core performance, or enterprise class.

Please tell me Alex what do you mean by the nebulous term I also see in IBM marketing hype called enterprise class.

Does it mean virtualisation capabilities? (UltraSPARC T2 has it).
Does it mean redundancy? (UltraSPARC T2 has it).
Does it mean rock-solid production OS? (UltraSPARC T2 has Solaris, which has been scaling to 64 CPUs for over 15 years, more so than AIX).
Does it mean Open-source OS? (UltraSPARC T2 has Solaris)
Does it mean Platinum service? (UltraSPARC T2 has Solaris)
Does it mean it support mission critical apps? (UltraSPARC T2 is used at many Fortune 50 companies right now for mission critical Apps)
Does it mean better reliability? (UltraSPARC T2 has better MTBF)
Does it mean that you can use it for Web, application, Oracle database? (UltraSPARC T2 beats power6 in every tier on perf, $/perf, TCO/perf...)
Does it mean more advanced technology (UltraSPARC T2 has lead the industry on the multi-core and security HW features)

or does it mean since p570 has been beaten, so you just want to "name call" the victor? :)

Posted by BM Seer on March 28, 2008 at 02:15 PM PDT #

Does the T2 system scale to 16 cores?
Does the T2 system scale to 768GB RAM?
Does the T2 system scale to >1.6M TPC-C?
Does the T2 system have hardware retry on error in the CPU?
Does the T2 system have all the virtualization capabilies on the p570?

Answer to all those is of course NO.

And why do you want to compare to the T2000 to the P570. It is at best comparable to the p550 which is 8 cores in 4RU

Posted by Joe on March 29, 2008 at 12:16 AM PDT #

>* worse performance than a single UltraSPARC T2,
At sap-sd the same performance, as 2xPower6

>* worse $/perf than a single UltraSPARC T2,
T2 has worse $/perf because of Oracle licenses on 8 cores (vs 4 core Power6)

>* worse TCO than a single UltraSPARC T2,
Why ?

Posted by Triffids on March 30, 2008 at 02:59 AM PDT #

4-core (2-chip) 4.7GHz IBM p570 with 64GB costs $300K which is 4x to 6x more that a US T2 server. TCO = HW plus SW. A few more Oracle licenses still means Sun is less cost. Less cost and better performance means Sun has better $/perf even if the server is only running Oracle. Sun beats IBM on every count.

Sun beats the IBM 570 4.7GHz 4-core on so many benchmarks, don't try to make the price of IBM look less by saying it is more of a competitor to the p560.

You have to do your own math on SAP-SD prices, as member companies are not allowed to comment on $/perf of SAP-SD benchmarks. As a reminder if you are running SAP app and database together, then you only pay for those cores that are running Oracle. Most cores are running SAP

Posted by BM Seer on March 31, 2008 at 10:56 AM PDT #

This is funny. Quite hilarious how BM Seer is like a broken record always trying to compare an underpopulated p570 prices to the T2 server. Why don't your compare the price of a 4 core p520 or 4 core p550?

Of course the p570 is going to be more expensive. Customers are paying for future upgrade ability of the server.

If its the other way around, you would be crying foul.

i.e. compare a 4 or 8 core Sun SPARC Enterprise M8000 Server or M9000 to the p520 or p550. Those prices won't look pretty for Sun.

Posted by Joe on March 31, 2008 at 06:52 PM PDT #

Isn't price per core an instance of "metric per core?"

WRT comparing prices on a p570 instead of a p520 or p550, have IBM published results using those two systems? Or would comparing to p520 or p550 pricing be assumeing the performance matched the p570?

Posted by rick jones on March 31, 2008 at 08:21 PM PDT #

You can't say T5120/T5220 are not competitive with the 4-core 4.7GHz IBM p 570 - see April 2nd posting on benchmark results. Sun is faster. http://blogs.sun.com/bmseer/entry/sun_t5220_t5120_faster_than
<p>
Price/core points out that metric per core is not a good comparison! Looking at cores and look at system performance and system price-performance. The truth hurts IBM so they avoid the truth and put lots of sugar on top of confusion.

Posted by BM Seer on April 02, 2008 at 11:14 AM PDT #