Fritz 11, XP in Parallels, Vista in Boot camp, ...surprises

[mikebore]

I have Fritz 11 running in XP in a Parallels virtual machine. I also have it running in Vista on a Boot Camp partition.

The machine is a MacBook Pro original Core Duo 1.83 with 2 Gb RAM. All the RAM is available to Vista in Boot Camp of course, and the XP virtual machine has 768Mb allocated.

The Fritz Performance Benchmark test results are surprising to me:

Kilonodes/sec:

Vista with two processors 2230
Vista with one processor 1164
XP with one processor 9759

These scores are reflected in normal (infinite analysis) usage, compared analysing the same games, single processor only of course.

I would expect Vista to be a bit slower than XP, all else being equal. But Vista has all the RAM and the whole machine at it's disposal, where XP has less than half the RAM and is sharing the CPU with OSX.

Do these results make any sense to anyone? Is it possible I have some settings wrong, either in Fritz or Vista which are slowing it down.

Thanks,

Mike

[pedrox]

The values of Vista appear to be real, but the image of XP does not seem real value, I believe it is impossible to get almost 10000 kN with a core duo processor at 1,83 (1 thread)

[mikebore]

Thanks for quick reply!

Is there another way of comparing, perhaps depth level in a given position after a defined time? I am interested in knowing how much performance penalty I am taking if I start Fritz up XP/Parallels instead of Vista/Boot Camp which is less convenient. I can assess subjectively but like measuring things where possible.

What would you expect the real difference between Vista and XP to be (leaving out any Parallels/Boot Camp issues) ?

Thanks,

Mike

[pedrox]

The values that you have for Vista in your Mac I think it could be very similar to those reached a PC with the same processor. I have no experience with Mac but if I tested other programs of virtual machines for the PC.

The difference in the number of nodes between Vista and XP on the same computer has to be minimal. In your case as XP runs in a virtual machine will run a little slower, perhaps to 80% of its speed (arround 900 kN?)

Try to solve a mate, for example as follows and you can see the difference in speed between the virtual machine with XP and in the real mode of Vista.

Mate in 11. Calculate the time it takes to make the mate and compared the times

[d]r4r2/1q2ppk1/b1np2p1/6P1/4P3/1BNnBP2/PP1K4/R2Q3R w - - 0 1

If the mate is easy, then probe with another more complicated.

You can also configure a position that is not mate and calculate for example the time it takes to reach deep 18.

[mikebore]

Thanks for that. can't try at the moment, but will report back later.

[mikebore]

I have spent some time trying to solve for mate the postion you posted.
The first problem was that if I use the mate search option it doesn't show any engine in the window presented. Help talks about special engines. Would be grateful for any tips how to do this.

So what I did instead was run it in "infinite analysis" mode, single line and wait for it to calculate mate. I was surprised by the variability of different runs (restarting prog between each). Sometimes it would find a mate in 12 or 15 or 17 then stop analysing. And the times varied too. There was a typical time which each option took, but occasionally it would take much longer.

Nevertheless, I got a reasonable feel for the difference. Typically Fritz in XP in the Parallels VM took 3 mins 30 secs mins to find a mate, and in Vista direct (Boot Camp) it took 1 min 15 secs. XP hash was 256 MB and Vista 384 Mb. Larger didn't seem to impove anything.

I have also recently got Hiarcs 11.2 MP running in SigmaChess in OSX. This typically took 15 secs to find mate, usually in 11, once in 10, but it occasionally took very much longer, up to 2.5 mins. It was noticeable that Hiarcs seemed to find the first move Rh7+ very quickly and concentrate on that, while Fritz looked at a lot of other first move options before Rh7+. and when it found Rh7 it was quick to finish. One versus two processors with Hiarcs didn't make as much difference as I expected. Some single processor runs were 15 secs.

The Fritz interface is much better than the rather basic Sigma.

Thanks for your help. I understand much better now, but would be grateful for any more input you have.

Mike

[pedrox]

I have spent some time trying to solve for mate the postion you posted.
The first problem was that if I use the mate search option it doesn't show any engine in the window presented. Help talks about special engines. Would be grateful for any tips how to do this.

So what I did instead was run it in "infinite analysis" mode, single line and wait for it to calculate mate. I was surprised by the variability of different runs (restarting prog between each). Sometimes it would find a mate in 12 or 15 or 17 then stop analysing. And the times varied too. There was a typical time which each option took, but occasionally it would take much longer.

Nevertheless, I got a reasonable feel for the difference. Typically Fritz in XP in the Parallels VM took 3 mins 30 secs mins to find a mate, and in Vista direct (Boot Camp) it took 1 min 15 secs. XP hash was 256 MB and Vista 384 Mb. Larger didn't seem to impove anything.

I have also recently got Hiarcs 11.2 MP running in SigmaChess in OSX. This typically took 15 secs to find mate, usually in 11, once in 10, but it occasionally took very much longer, up to 2.5 mins. It was noticeable that Hiarcs seemed to find the first move Rh7+ very quickly and concentrate on that, while Fritz looked at a lot of other first move options before Rh7+. and when it found Rh7 it was quick to finish. One versus two processors with Hiarcs didn't make as much difference as I expected. Some single processor runs were 15 secs.

The Fritz interface is much better than the rather basic Sigma.

Thanks for your help. I understand much better now, but would be grateful for any more input you have.

Mike

You should always get the same times and the same principal variant pv, if you get different values it is possible that the program is not clearing hash tables and because you have different solutions.

Every time you want to do an analysis, go to the menu File menu, and go to New and setup position, but you do not change anything about the configuration of position then OK this will clear the hash tables and you can always get the same amount of time.

I am not familiar with the engines to mate, so I can not say anything about this.

On the results out, it seems that the virtual machine has a speed of 50%, which might be reasonable.

[mikebore]

Thanks. Yes I get more consistent times by clearing the hash before timing.

I have now put the Hiarcs 11.2 MP engine in Fritz 11 in Vista, and this seems to be the fastest and most consistent combination. It always seems to find mate in 11 (not 12 or 14 etc) and does it in around 10 seconds.

So this seems to be the best of both at present....fastest engine (Hiarcs Multi Proc) and preferred UI (Fritz)

Now I will stop measuring and start using!

Thanks for your help.

Mike