PCbanter

PCbanter (http://www.pcbanter.net/index.php)
-   Windows 10 Help Forum (http://www.pcbanter.net/forumdisplay.php?f=52)
-   -   Benchmarking memory settings (http://www.pcbanter.net/showthread.php?t=1104182)

Ed Cryer May 17th 18 06:22 PM

Benchmarking memory settings
 
My PC runs quite well with 16GB RAM at 1600MHz;and also 32GB at 1066MHz.
They feel the same, but there must be some difference.
I want to compare the two, and googling gives me lots of options.
But there's "benchmarking" and there's "benchmarking".
Obviously the extra RAM will help in some apps.

So then, I want an across the board benchmarking.

What do you recommend?

Ed

Paul[_32_] May 17th 18 09:50 PM

Benchmarking memory settings
 
Ed Cryer wrote:
My PC runs quite well with 16GB RAM at 1600MHz;and also 32GB at 1066MHz.
They feel the same, but there must be some difference.
I want to compare the two, and googling gives me lots of options.
But there's "benchmarking" and there's "benchmarking".
Obviously the extra RAM will help in some apps.

So then, I want an across the board benchmarking.

What do you recommend?

Ed


The memtest86+ screen actually has a benchmark, in the upper left.

The various levels of cache are benchmarked, as
to how fast in GB/sec they are. The last level of cache
itself, might be considered to be the RAM sticks.

Your results should bear some proportionality to speed.
In terms of the two numbers you might get. The 1600MHz setting
has more bandwidth than the 1066MHz setting.

This is how I discovered that my RAM with its theoretical
76.8GB/sec, was actually doing pretty poorly at ~17GB/sec
measured in memtest86+. Normally memory might be 60%
efficient, and I could take 0.6 of 76.8GB/sec. But the
actual measurement wasn't nearly that good.

That's an example of a Synthetic test. Similar tests
you can run from Windows, might have the keyword "Stream"
benchmark in the name. Maybe Sisoftware Sandra has
such a bench in it.

https://www.cs.virginia.edu/stream/

*******

An example of an application which is actually "sensitive"
to memory speed, is 7ZIP compression. The interval taken to
compress your goods, will be shorter on the 1600MHz RAM than
the 1066MHz RAM. While 7ZIP should have a menu item
entitled "benchmark", you can just as easily take
a standard file of some sort on a hard drive, and
run the same compression settings on it, with the
RAM at one of the two settings, and compare.

Depending on how shabby a memory implementation a platform
has, 7ZIP could be 50% slower.

Other applications hit in the caches a lot, which
requires fewer actual memory accesses per second. Intel
is especially good at hiding the details of RAM,
with AMD being slightly less good (you have an incentive
to tweak the AMD systems a bit more).

Whereas 7ZIP is a bit of a "cache buster", so the cache
is less effective at hiding the details of the RAM. And
cranking the RAM, will help a bit with 7ZIP compression.
In 7Z "Ultra" mode, the dictionary for each core running
compression, is 600MB. The 600MB figure doesn't fit in
the CPU cache (6MB to 12MB say, in size). Some parts of
each dictionary are consulted more frequently than
others. If you had 8 cores, the entire footprint is
4.8GB of memory, and definitely doesn't fit in any
cache. It's with that sort of compression job, that
the memory conditions make a difference.

So if I was the Geek Squad guy, trying to convince you
what a difference such tuning makes, I would use 7ZIP
for the "before" and "after" results. There are a *ton*
of other things, that laugh at memory optimization,
and don't budge an inch.

I had a Core2, with DDR2-533 RAM in it, and most of the
time, I couldn't tell how hellishly slow it was compared
to the system with the better DDR2 RAM in it. Only
7ZIP let out my secret :-)

https://www.7-zip.org/

Paul

Ed Cryer May 17th 18 10:38 PM

Benchmarking memory settings
 
Paul wrote:
Ed Cryer wrote:
My PC runs quite well with 16GB RAM at 1600MHz;and also 32GB at
1066MHz. They feel the same, but there must be some difference.
I want to compare the two, and googling gives me lots of options.
But there's "benchmarking" and there's "benchmarking".
Obviously the extra RAM will help in some apps.

So then, I want an across the board benchmarking.

What do you recommend?

Ed


The memtest86+ screen actually has a benchmark, in the upper left.

The various levels of cache are benchmarked, as
to how fast in GB/sec they are. The last level of cache
itself, might be considered to be the RAM sticks.

Your results should bear some proportionality to speed.
In terms of the two numbers you might get. The 1600MHz setting
has more bandwidth than the 1066MHz setting.

This is how I discovered that my RAM with its theoretical
76.8GB/sec, was actually doing pretty poorly at ~17GB/sec
measured in memtest86+. Normally memory might be 60%
efficient, and I could take 0.6 of 76.8GB/sec. But the
actual measurement wasn't nearly that good.

That's an example of a Synthetic test. Similar tests
you can run from Windows, might have the keyword "Stream"
benchmark in the name. Maybe Sisoftware Sandra has
such a bench in it.

https://www.cs.virginia.edu/stream/

*******

An example of an application which is actually "sensitive"
to memory speed, is 7ZIP compression. The interval taken to
compress your goods, will be shorter on the 1600MHz RAM than
the 1066MHz RAM. While 7ZIP should have a menu item
entitled "benchmark", you can just as easily take
a standard file of some sort on a hard drive, and
run the same compression settings on it, with the
RAM at one of the two settings, and compare.

Depending on how shabby a memory implementation a platform
has, 7ZIP could be 50% slower.

Other applications hit in the caches a lot, which
requires fewer actual memory accesses per second. Intel
is especially good at hiding the details of RAM,
with AMD being slightly less good (you have an incentive
to tweak the AMD systems a bit more).

Whereas 7ZIP is a bit of a "cache buster", so the cache
is less effective at hiding the details of the RAM. And
cranking the RAM, will help a bit with 7ZIP compression.
In 7Z "Ultra" mode, the dictionary for each core running
compression, is 600MB. The 600MB figure doesn't fit in
the CPU cache (6MB to 12MB say, in size). Some parts of
each dictionary are consulted more frequently than
others. If you had 8 cores, the entire footprint is
4.8GB of memory, and definitely doesn't fit in any
cache. It's with that sort of compression job, that
the memory conditions make a difference.

So if I was the Geek Squad guy, trying to convince you
what a difference such tuning makes, I would use 7ZIP
for the "before" and "after" results. There are a *ton*
of other things, that laugh at memory optimization,
and don't budge an inch.

I had a Core2, with DDR2-533 RAM in it, and most of the
time, I couldn't tell how hellishly slow it was compared
to the system with the better DDR2 RAM in it. Only
7ZIP let out my secret :-)

https://www.7-zip.org/

Â*Â* Paul


My memory sticks have 1600Mhz in the name. They seem to do quite well,
so I was wondering if I might try upping the Mhz to the next level.
I guess I can only try. I've been tinkering with memory for some days
now and I haven't fried anything. When Windows has hit a problem it just
shows a message saying that it's got an error, is taking a system dump,
and will then restart.

Do you think your idea of zipping up a large bundle of files would be
better than, say, ripping a large DVD or using Handbrake to convert
something large?

Ed


Paul[_32_] May 17th 18 10:56 PM

Benchmarking memory settings
 
Ed Cryer wrote:
Paul wrote:
Ed Cryer wrote:
My PC runs quite well with 16GB RAM at 1600MHz;and also 32GB at
1066MHz. They feel the same, but there must be some difference.
I want to compare the two, and googling gives me lots of options.
But there's "benchmarking" and there's "benchmarking".
Obviously the extra RAM will help in some apps.

So then, I want an across the board benchmarking.

What do you recommend?

Ed


The memtest86+ screen actually has a benchmark, in the upper left.

The various levels of cache are benchmarked, as
to how fast in GB/sec they are. The last level of cache
itself, might be considered to be the RAM sticks.

Your results should bear some proportionality to speed.
In terms of the two numbers you might get. The 1600MHz setting
has more bandwidth than the 1066MHz setting.

This is how I discovered that my RAM with its theoretical
76.8GB/sec, was actually doing pretty poorly at ~17GB/sec
measured in memtest86+. Normally memory might be 60%
efficient, and I could take 0.6 of 76.8GB/sec. But the
actual measurement wasn't nearly that good.

That's an example of a Synthetic test. Similar tests
you can run from Windows, might have the keyword "Stream"
benchmark in the name. Maybe Sisoftware Sandra has
such a bench in it.

https://www.cs.virginia.edu/stream/

*******

An example of an application which is actually "sensitive"
to memory speed, is 7ZIP compression. The interval taken to
compress your goods, will be shorter on the 1600MHz RAM than
the 1066MHz RAM. While 7ZIP should have a menu item
entitled "benchmark", you can just as easily take
a standard file of some sort on a hard drive, and
run the same compression settings on it, with the
RAM at one of the two settings, and compare.

Depending on how shabby a memory implementation a platform
has, 7ZIP could be 50% slower.

Other applications hit in the caches a lot, which
requires fewer actual memory accesses per second. Intel
is especially good at hiding the details of RAM,
with AMD being slightly less good (you have an incentive
to tweak the AMD systems a bit more).

Whereas 7ZIP is a bit of a "cache buster", so the cache
is less effective at hiding the details of the RAM. And
cranking the RAM, will help a bit with 7ZIP compression.
In 7Z "Ultra" mode, the dictionary for each core running
compression, is 600MB. The 600MB figure doesn't fit in
the CPU cache (6MB to 12MB say, in size). Some parts of
each dictionary are consulted more frequently than
others. If you had 8 cores, the entire footprint is
4.8GB of memory, and definitely doesn't fit in any
cache. It's with that sort of compression job, that
the memory conditions make a difference.

So if I was the Geek Squad guy, trying to convince you
what a difference such tuning makes, I would use 7ZIP
for the "before" and "after" results. There are a *ton*
of other things, that laugh at memory optimization,
and don't budge an inch.

I had a Core2, with DDR2-533 RAM in it, and most of the
time, I couldn't tell how hellishly slow it was compared
to the system with the better DDR2 RAM in it. Only
7ZIP let out my secret :-)

https://www.7-zip.org/

Paul


My memory sticks have 1600Mhz in the name. They seem to do quite well,
so I was wondering if I might try upping the Mhz to the next level.
I guess I can only try. I've been tinkering with memory for some days
now and I haven't fried anything. When Windows has hit a problem it just
shows a message saying that it's got an error, is taking a system dump,
and will then restart.

Do you think your idea of zipping up a large bundle of files would be
better than, say, ripping a large DVD or using Handbrake to convert
something large?

Ed


The benchmarks should be the most sensitive, with 7ZIP.

Handbrake will have a dependency too, but at a guess,
not quite as severe.

Your progression of "non-windows" tests, were supposed
to prevent Windows throwing an error. (Like running
Prime95 under Linux, before booting back to Windows.)

The worst case corruption in Windows, comes from carrying the
Registry portions in RAM and writing them back at
shutdown. That's a "magnet" for errors, and you don't
really want to run Windows if the system is unstable,
for fear of borking the Registry.

A System Restore point keeps a copy of the Registry,
and you can use that as a fallback if you don't have
any other backups handy.

Paul


All times are GMT +1. The time now is 10:28 AM.

Powered by vBulletin® Version 3.6.4
Copyright ©2000 - 2024, Jelsoft Enterprises Ltd.
Copyright © 2004 - 2006 PCbanter
Comments are property of their posters