Q: On-the-Fly [#-CPU-cores] change during MANUAL testing?

[HPmanic]

Hello! First post here, so be gentle...

I am currently working on a HP Z800 that I am close to finishing, and have been tuning / tracking performance (in every step of the upgrade chain) with your nice Passmark tool.

Just two questions:

1. "Database" memory test went down from 70+ to 60+, as soon as second X5660 was added to the Z800 (single cpu-socket RAM was 12GB and dual cpu-socket config. just added another identical 12GB lot, all DDR3, all in Triple-Channel). Why this interesting decrease?

2. How can I switch / change CPU-cores, ON-the-FLY, during MANUAL testing of individual CPU tests, without losing other completed CPU-tests?(my apologies if this has been asked before).

FYI, overall the machine seems doing reasonably well, already above 13,000 and closing in to X5670s and X5680s in database (putting aside other variables, of course).

Thanks, in advance, for your valuable response!

[HPmanic]

UPDATE:

1. I believe I have found answer for my second question, above (so no need to dwelve into this, no need for the answer, itself).

2. As for my first question (#1 above), I still wonder what the reason may be for that drop in Database-test throughput...

All-in-all, machine is now posting 13,500+ (CPU mark) with twin-X5660s, which seems pretty close, already, to its full (non-overclocked) potential. Not sure about RAM, but it is also posting decently (better, for the most part), relative to other samples in its class / performance-level.

Now it is time to enjoy this baby!

[David (PassMark)]

I can't think of any good reason for the drop in the database test after adding a CPU.

I would suggest running the test a few times to see if the result was consistent.

Slim (but unlikely) possibility is that the system is overheating and throttling.

I assume the BIOS hasn't fiddled the memory timings when you added more RAM?

[HPmanic]

Another quick update...

Thanks for your reply...

I do agree that it is particularly strange... before adding second Xeon-5660, machine had 12GB of ram (in triple-channel mode, NUMA node=0), and score came around 72.

After adding second X5660 and adding another 12GB (identical), thus enabling NUMA node=1, best I have been able to extract is ~65.

Can't really explain why (although, in all fairness, ~65 is pretty much in line with all results posted by other twin-5660 Z800s). In other words, it seems that ~65 is normal on this platform / config.

I would like to know what particular variables drive performance or results of Database tests... How much ram it really needs? Extent to which it benefits from multi-core? etc.

Overall, the machine now is posting >13,700 (CPU mark) which is already on X5670-X5680 range, on similarly configured Z800s. Not sure how much else could be extracted (just for joy / curiosity purposes).

[David (PassMark)]

The database tests the speed at which the machine is able to build up a number of commonly used data structures in RAM and store data into them.

Three data structures used.
Queue (linked list)
List (doubly linked list)
Set (self-balancing binary search tree)

Google the C++ STL for more details.

The test is single threaded.

In each case approximately 256MB of RAM is used. So going to 12GB to 24GB should provide no performance advantage in this test. A faster CPU and faster RAM will help of course.

We used the term "database" to describe the test as the underlying structure of a database is often a tree structure and to keep it simple. Most people aren't going to understand all the internal details of STL data structures.