FritzMark Results

[LiquidNitrogenOverclocker]

Is there a page somewhere showing a list of FritzMark results?

I just ran mine and got the following:

Relative Speed 29.80 (compared to PIII @ 1.0 GHZ = 480 KN/sec)
Kilo-Nodes/sec = 14303

Just wondering how this result stacks up.

My hardware:

8 cores
Intel i7-860 @ 3.91 GHz

[Andre]

here
http://www.jens.tauchclub-krems.at/dive ... marks.html

[LiquidNitrogenOverclocker]

here
http://www.jens.tauchclub-krems.at/dive ... marks.html

Thanks, that is exactly what I was looking for!

[LiquidNitrogenOverclocker]

Looks like I would be 13th from the top. Not bad for about $1650 worth of hardware.

[M ANSARI]

The Fritzmark seems to wrongly give higher scores for hyperthreading on. I don't know if all those tests were with hyperthreading ON or OFF. I do remember overclocking my Skulltrail with hyperthreading OFF and it had a score of about 44 overclocked at 5 Ghz. It is nice to see that this can be even bettered by a single chip system today.

[LiquidNitrogenOverclocker]

The Fritzmark seems to wrongly give higher scores for hyperthreading on.

A node is a node is a node. I understood the test to be a fixed-time benchmark where nodes were merely counted, then averaged at the end.

I don't see how hyperthreading could lead to miscounting nodes.

[Gian-Carlo Pascutto]

It's misleading because it reports the raw NPS and doesn't account for scalability.

This will cause a machine with HT to get a higher result reported even though it won't actually be faster.

The NPS count will be OK. It's just that this number has no meaning for performance.

[LiquidNitrogenOverclocker]

Then why is the benchmark so widely reported I wonder?

Is this a "generally acknowledged" misnomer, or is it just speculation? Has the programmer acknowledged this?

Can a test case be shown?

[CRoberson]

What GCP is trying to tell you is that NPS doesn't always correlate to Elo gains, especially in multiprocessor tests. One can have a NPS improvement, but not gain in Time to Ply. Time to Ply is a measure of speed that more accurately correlates to Elo gains.

From lots of testing I have done, Hyperthreading doesn't produce the same gains across all programs. While it may greatly help one program, it may not help another at all. One might think that if it helps one chess program it will help all. That is not the case,

My tests with lots of different programs using various tree search algorithms indicate that its greatest gains are to the programs that have the least amount of work at the leaf nodes.

If I understand correctly, the version of Fritz used in the benchmark has a very light weight eval, thus fitting with my test results.

[LiquidNitrogenOverclocker]

I was not trying to correlate the FritzMark score to a chess performance expectation (in terms of rating gain).

I am trying to determine if a higher NPS correlates to faster hardware.

I overclocked systems such as the one shown here:

http://www.liquidnitrogenoverclocking.com

...and I want to know if faster NPS means the hardware is technically faster.

From what I read, hyperthreading "skews" the result somehow.

I am not looking for chess gains, just hardware throttling.