Intel vs AMD: which does your engine prefer?

[Albert Silver]

...
In other words, on your i7 and Phenom II, using Houdini as a standard parameter, Stockfish improves its relative NPS performance on a phenom II by over 11%. Also, I have trouble believing that Komodo, on a single core, thus no hyperthreading, performs 20% worse on an i7 compared to an i5. Remember, we are not talking absolute performance. We are saying that if the speeds are normalized, it will perform 20% worse on an i7, compared to an i5, which again, is hard to believe.

We are not talking about god, there is nothing to believe!

The subtext is that there is a problem with the values, and you either had a problem that affected the scores, or something else. Since the processors are twin brothers, there is no reason for Komodo to perform so much worse on one as opposed to the other.

[IWB]

Hello Albert,

...
In other words, on your i7 and Phenom II, using Houdini as a standard parameter, Stockfish improves its relative NPS performance on a phenom II by over 11%. Also, I have trouble believing that Komodo, on a single core, thus no hyperthreading, performs 20% worse on an i7 compared to an i5. Remember, we are not talking absolute performance. We are saying that if the speeds are normalized, it will perform 20% worse on an i7, compared to an i5, which again, is hard to believe.

We are not talking about god, there is nothing to believe!

The subtext is that there is a problem with the values, and you either had a problem that affected the scores, or something else. Since the processors are twin brothers, there is no reason for Komodo to perform so much worse on one as opposed to the other.

I am really not sure what you are talking about. The difference on my i920 vs the i5m with Komoodo is 2% (T9/V9) and not 20% and Komodo is 2% better on the 920 and not worse. Stockfish is not performing 11% better but is havin about 10% less nodes when assuming the same clock speed on the i920 and the Phenom2. That are real numbers and not assumed factors (verticaly H to K in my list)!
(Maybe this is because you insist of keeping Houdini as 1000 on every CPU while in reality it itself performes different and sometimes more different than the engine you compare it with. In these case you might get data which look wired)
The overall difference of the 20 UCI engines is 7/8% which is in line compared to i5m to Ph2/C2. One possible explenation would be the cache. On the i5m it is only 3MB while the i920 has 8MB ...
The worst engine difference between those two CPUs is Rybka (not Komodo), which is 27% or 36% (depending from where you have a look.

I will check my numbers again, but I have only access to the i5m notebook on monday. Maybe I have less problems with the difference than you as I dont see them as twin brothers but more as stepsisters

Bye
Ingo

PS: I discovered that my Chiron data on the i5m might be flawed. That is actually something I think I made a mistake! Monday ...

[IWB]

Hi

I am really not sure what you are talking about. The difference on my i920 vs the i5m with Komoodo is 2% (T9/V9) and not 20% and Komodo is 2% better on the 920 and not worse. Stockfish is not performing 11% better but is havin about 10% less nodes when assuming the same clock speed on the i920 and the Phenom2. That are real numbers and not assumed factors (verticaly H to K in my list)!
(Maybe this is because you insist of keeping Houdini as 1000 on every CPU while in reality it itself performes different and sometimes more different than the engine you compare it with. In these case you might get data which look wired)

Some more details:
Assuming the same clock speed (1GHz in my list H to K) and giving the fastest Houdini 1000 (920) it should have 955.3 on the i5m, 838,38 on the C2 and 813,95 on the Phenom2. That is based on hard numbers. If you set all these on 1000 and compare with other engines you add the above relative differences to the other engines. Frankly you can't compare CPUs anymore as you ignore the differences on different hardware of your reference engine.
Just one example: If you set Houdini to 1000 everywhere you have to add (or subtract?) roughly 19% to every other engine on the Phenom2 ... all this seems to be much too complicated for me. I stick with the percentages!
But I am not sure if I got your idea correctly.

Bye
Ingo

[IWB]

Hi,

...

Actually you cannot express it as 100%, because it is misleading. Suppose you say Houdini gets 100%, does that mean Rybka is scoring 6%? And 6% of what? As I explained, the absolute scores are utterly meaningless. What matters are the relative scores compared to other processors....

Ahh, here is the missunderstanding. Of course you can not compare absolut numbers, and I dont do that!

Example: Take the M row. The 1.19 on M6 mean that Houdini is 19% faster on a i920 compared to a C2 when assuming the same clock speed! Now you can do that for every other engine in my list and have an avverage speed gain in percent at the end of the row (14%). This means that the 920 is in average 14% faster per cycle than the Core 2. The other pairings are analog to that example. Unfortunately I dont know how to make that clearer in my list!

Bye
Ingo