Hi All,

I have a workstation which crunch for me all the time, i'm however puzzled of the relative poor numbers htese pretty hefty cpus are getting from the cpu benchmark;

------------------
CPU type GenuineIntel Intel(R) Xeon(TM) CPU 3.06GHz
Number of CPUs 4 (thats 2 hyperthreaded cpus)

Measured floating point speed 1363.35 million ops/sec
Measured integer speed 1123.04 million ops/sec
------------------

I think thats pretty low numbers... heres what a non ht p4 is getting;
-----------------
CPU type GenuineIntel Intel(R) Pentium(R) 4 CPU 2.53GHz
Number of CPUs 1

Measured floating point speed 1328.72 million ops/sec
Measured integer speed 2703.29 million ops/sec
-----------------

Whats happening - is boinc setin not ht compliant / optimized - or does it just not runn well on xeon cpus ??

do i need to disable the ht feature of my xeon processors ??

/Henrik

---

hello henrik,

I do not know for sure, but it's possible that the benchmarks are done only for one of the virtual CPUs.
The crunching definetly would be one WU per virtual CPU.
I myself do not have these types, so I can only give you my impression of the various posts I read about this on this and other boards.

---

Gruesse vom Saenger

For questions about Boinc look in the BOINC-Wiki

Well,

Yes I think so too - but i've also seen other xeon class machines with much higher benchmarks - (i still find the numbers low)..


see this example;


--------------------
http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=36273

CPU type GenuineIntel Intel(R) Xeon(TM) CPU 3.06GHz
Number of CPUs 4

Measured floating point speed 2998.48 million ops/sec
Measured integer speed 8998.8 million ops/sec
--------------------

--------------------
http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=138525

CPU type GenuineIntel Intel(R) Xeon(TM) CPU 3.06GHz
Number of CPUs 4

Measured floating point speed 2979.01 million ops/sec
Measured integer speed 8978.74 million ops/sec
--------------------

I really wonder.... maybe i should just try to turn off HT and see...

/Henrik

---

I agree with OhmegaStar - I have a HT CPU at home and the calculation IS per virtual CPU. To get the best out of your CPU's make sure your boinc configuration shows all the CPUS that your task manager shows you and set it to use all the CPU's. Watch out for the CPU temperature then! I have found that if I run mine this way, I will have to modify the hardware for better cooling efficiency.

When your boinc is set as I suggested you will crunch multiple WU's simultaneously.

---

The client runs one benchmark thread per CPU then divides the total result by the number of CPUs. This causes the benchmarks on a HT machine to be about half of what the same computer would do with HT disabled. It all comes out in the wash, the benchmarks are reasonably close for what they are used for. They do still need some work though.

---

BOINC WIKI

BOINCing since 2002/12/8

You might not be comparing apples to apples. Yes, I know you are checking other XEON HT machines benchmarks, but before 4.25 (or was it 4.19) the benchmarks were flawed (Windows cheated). You need to find a machine that is on a current client, or maybe try to find ones on a Linux, which didn't skew the benchmarks.

I am sure there could be other factors.

---

My movie https://vimeo.com/manage/videos/502242