Hello,

Is there a rational explanation for the apparent 'slowing down' of the BOINC core client CPU Benchmarks?
Especially the Floating Point MIPS. In fact the Integer MIPS have recently taken a boost.
Is this me or BOINC?

Here are the stats:

21/05/2005 12:56:32||Starting BOINC client version 4.25 for windows_intelx86
21/05/2005 12:58:34||   Number of CPUs: 1
21/05/2005 12:58:34||   1927 double precision MIPS (Whetstone) per CPU
21/05/2005 12:58:34||   3625 integer MIPS (Dhrystone) per CPU
----------------------------------------------------------------------------
21/05/2007 03:23:15||Starting BOINC client version 5.8.16 for windows_intelx86
21/05/2007 13:10:57||   Number of CPUs: 1
21/05/2007 13:10:57||   1900 floating point MIPS (Whetstone) per CPU
21/05/2007 13:10:57||   3545 integer MIPS (Dhrystone) per CPU
----------------------------------------------------------------------------
08/02/2008 16:29:31||Starting BOINC client version 5.10.30 for windows_x86_64
09/02/2008 01:37:53||   Number of CPUs: 1
09/02/2008 01:37:53||   1821 floating point MIPS (Whetstone) per CPU
09/02/2008 01:37:53||   4693 integer MIPS (Dhrystone) per CPU

Maybe I need to do a fresh OS install - it's been some years...
Thank you for any insight.

Yours curiously,

---

For comparison, what was your CPU doing at the times of those benchmarks?

---

For comparison, what was your CPU doing at the times of those benchmarks?

Right. The CPU Benchmark feature fascinates me. Well, it's free!
Anyway...
I am aware extra processes will slow the CPU down. I am always careful to make sure there's a minimum in memory before I 'record' the results.
The small trend downwards of 50-100 MIPS a year maybe attributable to the Windows registry getting well clogged.
So whats the Integer MIPS boost about? That's HUGE!

---

See 2 different OSses. The latter a 64bit which does have a large integer increase which is normal. Vista? Well don't expect performance miracles with flops.

Indeed :-)
I've been using XP64 before the 64 bit project & client became available. The first set of stats are under XP32, the remainder are all under XP64. It's a single core Athlon64 3200(2GHz).
True to say I've seen a work unit completion time improvement recently, now we have 64 bits. Do projects typically mess around with the mantissa to speed things up?

---

As I understand it do the 64 bit CPU's themselves only have 32 bit floating point ability, which is why you would not see much change on the Whetstone side.

I see. What a can of worms!

Thanks, Sekerob.

Guy

---

Same here on Suse64-10.2 with E6320.

Messured with 64bit-5.10.21:
1804 floating point MIPS
5123 integer MIPS

Messured with Augustines 64bit-5.10.30:
1773 floating MIPS
7820 integer MIPS

---

These benchmarks are pretty much meaningless - as far as I know none of the projects use them for credits anymore, and they're certainly not a realistic guide to the speed of the machine.

The best test of a machine's performance is how well it runs your favourite project's workunits.

---

...Messured with Augustines 64bit-5.10.30:

Different compilers or different compiler settings always cause that, it is quite normal.

The TruXoft clients have an optimized integer speed with an advantage in the same range.


p.s.: the benchmarks are useful - but only for cache management, estimating the amount
of work a host can do. Once a year would probably be sufficient for that ;-)

As I understand it do the 64 bit CPU's themselves only have 32 bit floating point ability, which is why you would not see much change on the Whetstone side.

Actually, both 32-bit and 64-bit machines have 32/64/80-bit floating point ability.

---

...Messured with Augustines 64bit-5.10.30:

Different compilers or different compiler settings always cause that, it is quite normal.

The TruXoft clients have an optimized integer speed with an advantage in the same range.


p.s.: the benchmarks are useful - but only for cache management, estimating the amount
of work a host can do. Once a year would probably be sufficient for that ;-)

I have seen that behavior with every new boinc-version and it is independed from Augustines version. The background for this version is only to be more compatible with older distro's and not cheating!

I think the boinc-dev-team had quirks something because everyone could compile it's own version. Furthermore comes that also to a time where the not excisting problem with to many granted credits on some projects are hot discussed in some project-boards.

Only for cache managment? If the values are higher than your realworld-values boinc would have to many work on the queue. Great!!!

---

As I understand it do the 64 bit CPU's themselves only have 32 bit floating point ability, which is why you would not see much change on the Whetstone side.

Actually, both 32-bit and 64-bit machines have 32/64/80-bit floating point ability.

Then indulge on the ignorant and explain why a 32 bit client on a 32 bit OS shows the near equal whetstone of a 64 bit client on a 64 bit OS.

Exactly for that reason. 32-bit machines already had 64-bit floating point. 64-bit CPUs haven't had any (major) change in how floating point works.

---

As I understand it do the 64 bit CPU's themselves only have 32 bit floating point ability, which is why you would not see much change on the Whetstone side.

Actually, both 32-bit and 64-bit machines have 32/64/80-bit floating point ability.

Then indulge on the ignorant and explain why a 32 bit client on a 32 bit OS shows the near equal whetstone of a 64 bit client on a 64 bit OS.

Exactly for that reason. 32-bit machines already had 64-bit floating point. 64-bit CPUs haven't had any (major) change in how floating point works.

Most of the projects use double (80 bit) floating point. Since this was already available on 32 bit machines, the only change would be the speed of memory access. If most of the data was in cache, this would not show much effect.

---

BOINC WIKI

Hello there . I too have noticed the generous increase in MIPS ratings shown by the CPU benchmarks .I'd like to make the following observation: I have a Hyper-threading Pentium 4B in my HP D530 ...the benchmark numbers are really double what they should be because the values for BOTH threads seem to be added together , even though the results say "Per CPU" . 64 bit mode is an addressing range and has nothing to do with arithmetic speeds at all.I've only seen this increase since "upgrading " to Vista Ultimate .By the way , MIPS really stands for "Meaningless Index of Performance " !