I've had an opportunity to run the same code (a big SPSS macro of mine extensively using MATRIX syntax and random number generation) on SPSS 29 (Subscription), SPSS 26 and SPSS 22. On the same machine (Intel Core i9, 64 Gb RAM, Windows 11). While versions 26 and 22 did not differ any considerably in time spent, version 29 showed to be 1.69 times slower! This is an astonishing difference; imagine - it is almost as if I'm back on processor i5 instead of i9. What's wrong with version 29. Can it be speeded up? Or am I mistaken somehow?

------------------------------
Kirill Orlov
------------------------------

Hi @Kirill Orlov

Please open a support case, and let's get your replication scenario in front of a developer.

I've had an opportunity to run the same code (a big SPSS macro of mine extensively using MATRIX syntax and random number generation) on SPSS 29 (Subscription), SPSS 26 and SPSS 22. On the same machine (Intel Core i9, 64 Gb RAM, Windows 11). While versions 26 and 22 did not differ any considerably in time spent, version 29 showed to be 1.69 times slower! This is an astonishing difference; imagine - it is almost as if I'm back on processor i5 instead of i9. What's wrong with version 29. Can it be speeded up? Or am I mistaken somehow?

------------------------------
Kirill Orlov
------------------------------

Hi @Kirill Orlov

Please open a support case, and let's get your replication scenario in front of a developer.

------------------------------
David Dwyer
SPSS Technical Support
IBM Software

Original Message:
Sent: Mon December 18, 2023 01:31 PM
From: Kirill Orlov
Subject: Why SPSS 29 computes so much slower than older versions?

I've had an opportunity to run the same code (a big SPSS macro of mine extensively using MATRIX syntax and random number generation) on SPSS 29 (Subscription), SPSS 26 and SPSS 22. On the same machine (Intel Core i9, 64 Gb RAM, Windows 11). While versions 26 and 22 did not differ any considerably in time spent, version 29 showed to be 1.69 times slower! This is an astonishing difference; imagine - it is almost as if I'm back on processor i5 instead of i9. What's wrong with version 29. Can it be speeded up? Or am I mistaken somehow?

------------------------------
Kirill Orlov
------------------------------

Here is what I've tracked so far. The slow down - at least one of - of version 29 [29.0.2.0] (compared to version 26 [26.0.0.1]) shows up in this example.

set mxloops 1E9.
SET RNG=MT MTINDEX= 123456.

*Compute projections of data x points onto the beams (from origin) of data y points.
MATRIX.
LOOP try= 1 to 300.
*Generate two normal data clouds.
comp x= sqrt(-2*ln(uniform(2000,5)))&*cos(6.28318530717959*uniform(2000,5)).
comp y= sqrt(-2*ln(uniform(500,5)))&*cos(6.28318530717959*uniform(500,5)).
*Compute angle cosins b\w carrier beams of every x point and y point.
comp hx= sqrt(rssq(x)). /*Deviation of every x point
comp hy= t(sqrt(rssq(y))). /*Deviation of every y point
comp cosxy= x*t(y)/(hx*hy). /*The 2000*500 matrix of cosines
comp flag= cosxy>=.001. /*Flag cosines above some arbitrary threshold
*--I detect the slow down (about 1.38 times) emerges with this:.
comp proj= hx*make(1,nrow(y),1) &* flag &* cosxy. /*For each flagged cosine,
     /*multiply it by the deviation of each x point and get
     /*the length of the projection of the x point onto the beam of the y point
*--.
END LOOP.
print msum(proj).
END MATRIX.

@David Dwyer, can you pass it to the development?

------------------------------
Kirill Orlov

Original Message:
Sent: Tue December 19, 2023 04:41 PM
From: David Dwyer
Subject: Why SPSS 29 computes so much slower than older versions?

Hi @Kirill Orlov

Please open a support case, and let's get your replication scenario in front of a developer.

------------------------------
David Dwyer
SPSS Technical Support
IBM Software

Original Message:
Sent: Mon December 18, 2023 01:31 PM
From: Kirill Orlov
Subject: Why SPSS 29 computes so much slower than older versions?

I've had an opportunity to run the same code (a big SPSS macro of mine extensively using MATRIX syntax and random number generation) on SPSS 29 (Subscription), SPSS 26 and SPSS 22. On the same machine (Intel Core i9, 64 Gb RAM, Windows 11). While versions 26 and 22 did not differ any considerably in time spent, version 29 showed to be 1.69 times slower! This is an astonishing difference; imagine - it is almost as if I'm back on processor i5 instead of i9. What's wrong with version 29. Can it be speeded up? Or am I mistaken somehow?

------------------------------
Kirill Orlov
------------------------------

Hi @Kirill Orlov

I've reported this issue to IBM SPSS Statistics Development as Known Issue DT257763
Please follow this Known Issue.

Here is what I've tracked so far. The slow down - at least one of - of version 29 [29.0.2.0] (compared to version 26 [26.0.0.1]) shows up in this example.

set mxloops 1E9.
SET RNG=MT MTINDEX= 123456.

*Compute projections of data x points onto the beams (from origin) of data y points.
MATRIX.
LOOP try= 1 to 300.
*Generate two normal data clouds.
comp x= sqrt(-2*ln(uniform(2000,5)))&*cos(6.28318530717959*uniform(2000,5)).
comp y= sqrt(-2*ln(uniform(500,5)))&*cos(6.28318530717959*uniform(500,5)).
*Compute angle cosins b\w carrier beams of every x point and y point.
comp hx= sqrt(rssq(x)). /*Deviation of every x point
comp hy= t(sqrt(rssq(y))). /*Deviation of every y point
comp cosxy= x*t(y)/(hx*hy). /*The 2000*500 matrix of cosines
comp flag= cosxy>=.001. /*Flag cosines above some arbitrary threshold
*--I detect the slow down (about 1.38 times) emerges with this:.
comp proj= hx*make(1,nrow(y),1) &* flag &* cosxy. /*For each flagged cosine,
     /*multiply it by the deviation of each x point and get
     /*the length of the projection of the x point onto the beam of the y point
*--.
END LOOP.
print msum(proj).
END MATRIX.

@David Dwyer, can you pass it to the development?

------------------------------
Kirill Orlov

Original Message:
Sent: Tue December 19, 2023 04:41 PM
From: David Dwyer
Subject: Why SPSS 29 computes so much slower than older versions?

Hi @Kirill Orlov

Please open a support case, and let's get your replication scenario in front of a developer.

------------------------------
David Dwyer
SPSS Technical Support
IBM Software

Original Message:
Sent: Mon December 18, 2023 01:31 PM
From: Kirill Orlov
Subject: Why SPSS 29 computes so much slower than older versions?

I've had an opportunity to run the same code (a big SPSS macro of mine extensively using MATRIX syntax and random number generation) on SPSS 29 (Subscription), SPSS 26 and SPSS 22. On the same machine (Intel Core i9, 64 Gb RAM, Windows 11). While versions 26 and 22 did not differ any considerably in time spent, version 29 showed to be 1.69 times slower! This is an astonishing difference; imagine - it is almost as if I'm back on processor i5 instead of i9. What's wrong with version 29. Can it be speeded up? Or am I mistaken somehow?

------------------------------
Kirill Orlov
------------------------------

Hi @Kirill Orlov

I've reported this issue to IBM SPSS Statistics Development as Known Issue DT257763
Please follow this Known Issue.

Here is what I've tracked so far. The slow down - at least one of - of version 29 [29.0.2.0] (compared to version 26 [26.0.0.1]) shows up in this example.

set mxloops 1E9.
SET RNG=MT MTINDEX= 123456.

*Compute projections of data x points onto the beams (from origin) of data y points.
MATRIX.
LOOP try= 1 to 300.
*Generate two normal data clouds.
comp x= sqrt(-2*ln(uniform(2000,5)))&*cos(6.28318530717959*uniform(2000,5)).
comp y= sqrt(-2*ln(uniform(500,5)))&*cos(6.28318530717959*uniform(500,5)).
*Compute angle cosins b\w carrier beams of every x point and y point.
comp hx= sqrt(rssq(x)). /*Deviation of every x point
comp hy= t(sqrt(rssq(y))). /*Deviation of every y point
comp cosxy= x*t(y)/(hx*hy). /*The 2000*500 matrix of cosines
comp flag= cosxy>=.001. /*Flag cosines above some arbitrary threshold
*--I detect the slow down (about 1.38 times) emerges with this:.
comp proj= hx*make(1,nrow(y),1) &* flag &* cosxy. /*For each flagged cosine,
     /*multiply it by the deviation of each x point and get
     /*the length of the projection of the x point onto the beam of the y point
*--.
END LOOP.
print msum(proj).
END MATRIX.

@David Dwyer, can you pass it to the development?

------------------------------
Kirill Orlov

Original Message:
Sent: Tue December 19, 2023 04:41 PM
From: David Dwyer
Subject: Why SPSS 29 computes so much slower than older versions?

Hi @Kirill Orlov

Please open a support case, and let's get your replication scenario in front of a developer.

------------------------------
David Dwyer
SPSS Technical Support
IBM Software

Original Message:
Sent: Mon December 18, 2023 01:31 PM
From: Kirill Orlov
Subject: Why SPSS 29 computes so much slower than older versions?

I've had an opportunity to run the same code (a big SPSS macro of mine extensively using MATRIX syntax and random number generation) on SPSS 29 (Subscription), SPSS 26 and SPSS 22. On the same machine (Intel Core i9, 64 Gb RAM, Windows 11). While versions 26 and 22 did not differ any considerably in time spent, version 29 showed to be 1.69 times slower! This is an astonishing difference; imagine - it is almost as if I'm back on processor i5 instead of i9. What's wrong with version 29. Can it be speeded up? Or am I mistaken somehow?

------------------------------
Kirill Orlov
------------------------------