Re: Disappointing upgrade: Sempron 2200+ --> Phenom II 4X 840

DK wrote:
> Wanting more raw computing power, I went from Sempron 2200+
> in ASUS Asus A7V400-MX (socket A) to Phenom II 4X 840 in Asus
> M4N68T-M V2 (socket AM3).
>
> I must say I am quite disappointed. I was hoping for about 10X speed
> boost even with applications that can only use a single core. After
> all, more than six years separate the two (expected boost suggested
> by Moore's law would be 64X). The real life tests show nothing
> of this sort! Not even close to 10X.
>

Is your hard drive 10x as fast?

Bob F wrote:
> DK wrote:
>> Wanting more raw computing power, I went from Sempron 2200+
>> in ASUS Asus A7V400-MX (socket A) to Phenom II 4X 840 in Asus
>> M4N68T-M V2 (socket AM3).
>>
>> I must say I am quite disappointed. I was hoping for about 10X speed
>> boost even with applications that can only use a single core. After
>> all, more than six years separate the two (expected boost suggested
>> by Moore's law would be 64X). The real life tests show nothing
>> of this sort! Not even close to 10X.
>>
>
> Is your hard drive 10x as fast?

Benchmarking, after a hardware upgrade, is all part of the fun.

You want to verify your system is tuned up properly, and also
that you got your money's worth.

All it takes is a benchmark which isn't disk limited, and you
can get some idea how much the new gear is helping.

An example is Cinebench, a synthetic benchmark that scales
perfectly (so doesn't really stress much except the processor
cores). That's better than nothing.

http://www.maxon.net/downloads/cinebench.html

Another test you can use, is 7ZIP in ultra mode compression. My
puny machine can only do about 5MB/sec compression, which means
the hard drive isn't stressed at all. 7ZIP or WinRAR testing
is good for including memory subsystem performance as a component
of the test. The Phenom without an L3 may be marginally slower
than a Phenom with 6MB L3 (at the same clock speed). In many other
situations, the missing L3 isn't a factor. While those benchmarks
are running, you want to pop open Task Manager and make sure the
graphs are well filled (in case your test case is only using one core
by accident).

Paul

DK wrote:
> In article <je5ht2$3oa$1@dont-email.me>, Paul <nospam@needed.com> wrote:
>
>> Another test you can use, is 7ZIP in ultra mode compression. My
>> puny machine can only do about 5MB/sec compression, which means
>> the hard drive isn't stressed at all.
>
> Just tried 7ZIP ultra compression. The version I have seems to only use
> a single core for the ~1.3 MB/sec on average (Task Manager shows no
> more than 30% CPU utilization and that is with email, browser and
> newsreader). Is your 5 Mb/sec for a single core as well? If yes, what
> CPU do you have?
>
> Dima

To do a fair comparative test, we'd have to start with
the same data file. My processor for the test was
the E8400 dual core.

7ZIP is a bit picky about using multiple cores. In some cases,
the interface claims it will use multiple cores, when at runtime
it is obviously not doing so. That's why it is important to
use Task Manager, to make sure it's working.

The slower compression options, like Ultra, are the ones that
may trigger multi-core operation.

If you wanted another benchmark, there is "pigz" for Linux,
which is a multithreaded GZIP. If I need to compress entire
disk drive images, that's my tool of choice. It's a good
tradeoff between compression speed, and amount of compression
achieved. I don't know if there is a Windows port. In Linux,
you get this from your package manager and install it there.

http://www.zlib.net/pigz/

OK, there's a Windows port here, just found it. I won't
be testing this until later.

http://web.archive.org/web/201001070...downloads.html

http://web.archive.org/web/201001070....1.4_win32.zip

*******

To have a fair test, we can create files with similar entropy.

Download a copy of "dd".

http://www.chrysocome.net/download

http://www.chrysocome.net/downloads/dd-0.5.zip

Unzip the file. Open an MSDOS (command prompt) window and "cd"
(change directories) to the directory holding the dd.exe file.
Then try

dd if=/dev/random of=test.bin bs=1048576 count=1024

That will create a 1GB file of pseudo-random numbers. Since the seed used
isn't likely to be the same, we won't have exactly the same file.
On my machine, it takes 24 seconds to make the test file.

To test the seed, I created two files with the tool, and the
checksums don't match. So it uses a new seed for each file.

Now, for the test. (If I wanted to do this right, I'd compute
the md5sum of a 7GB movie.iso file, to flush the file cache. But
that's not necessary as this test is CPU bound.)

I did an "Add to archive" in 7ZIP version 4.60 beta (from 2008).

Archive format 7z
Compression level Ultra
Compression method LZMA
Dictionary size 64MB
Word size 64
Solid Block size 4GB
Number of CPU threads 2 (E8400 3GHz Dual Core 6MB L2 cache)

The test uses about 700MB of memory (of 4GB of DDR2-800 CAS5 memory).

Basically, the file is incompressible (high entropy), and the
output file "test.7z" is bigger than the original. The
1048576 KB input file "test.bin", becomes 1062758 KB on output as "test.7z".

The test completes on my machine in 7 minutes 13 seconds.
1,073,741,824 bytes in 433 seconds is 2,479,773 bytes/sec
compression rate. During the test, the CPU runs at only
75%, meaning there is 25% of the dual core not used. (And that
means this test does not scale well, apparently.)

You should be able to beat me, with a quad core, by a bit.
Assuming 7ZIP scales to four threads. I don't know if it
does or not. If you see "Number of CPU threads 4",
the contest should be interesting.

*******

I'm going to run it again now with one core instead of two.
I'm seeing the Task Manager at 50%, which means running with
one thread is much more efficient for this version of software.
Two threads should have run at 100%, but only manage 75%, which
means the threads aren't independent of one another.

That wouldn't happen with the Cinebench CPU test, which scales
perfectly, and doesn't really measure bottlenecks in the processor
quite as well. 7ZIP or WinRAR put more pressure on the cache levels.

Archive format 7z
Compression level Ultra
Compression method LZMA
Dictionary size 64MB
Word size 64
Solid Block size 4GB
Number of CPU threads 1 (E8400 3GHz Dual Core 6MB L2 cache)

In this case, I feel I should be able to beat a single core on
your processor. My time on this one is 11 minutes 17 seconds.
1,073,741,824 bytes in 677 seconds is 1,586,029 bytes/sec

Paul

"Paul" <nospam@needed.com> wrote in message news:je5ht2$3oa$1@dont-email.me...
> Bob F wrote:
>> DK wrote:
>>> Wanting more raw computing power, I went from Sempron 2200+
>>> in ASUS Asus A7V400-MX (socket A) to Phenom II 4X 840 in Asus
>>> M4N68T-M V2 (socket AM3).
>>>
>>> I must say I am quite disappointed. I was hoping for about 10X speed
>>> boost even with applications that can only use a single core. After
>>> all, more than six years separate the two (expected boost suggested
>>> by Moore's law would be 64X). The real life tests show nothing
>>> of this sort! Not even close to 10X.
>>>
>>
>> Is your hard drive 10x as fast?
>
> Benchmarking, after a hardware upgrade, is all part of the fun.
>
> You want to verify your system is tuned up properly, and also
> that you got your money's worth.
>
> All it takes is a benchmark which isn't disk limited, and you
> can get some idea how much the new gear is helping.
>
> An example is Cinebench, a synthetic benchmark that scales
> perfectly (so doesn't really stress much except the processor
> cores). That's better than nothing.
>
> http://www.maxon.net/downloads/cinebench.html
>
> Another test you can use, is 7ZIP in ultra mode compression. My
> puny machine can only do about 5MB/sec compression, which means
> the hard drive isn't stressed at all. 7ZIP or WinRAR testing
> is good for including memory subsystem performance as a component
> of the test. The Phenom without an L3 may be marginally slower
> than a Phenom with 6MB L3 (at the same clock speed). In many other
> situations, the missing L3 isn't a factor. While those benchmarks
> are running, you want to pop open Task Manager and make sure the
> graphs are well filled (in case your test case is only using one core
> by accident).
>

Mine has the 6MB L3, and the 7Zip score is 8.6MB/sec. using 4/4 CPU threads. I watch all 4 cores in Task Manager and
they are like synchronized swimmers at the Olympics.
I love the decompression rate of 135MB/sec.
--
SC Tom

DK wrote:
> In article <je5pmf$3pn$1@dont-email.me>, Paul <nospam@needed.com> wrote:
>> I did an "Add to archive" in 7ZIP version 4.60 beta (from 2008).
>
> I had version 4.65 installed but just to be sure downloaded and installed
> version 4.60 beta. And for some strange reason I do not see quite the
> same GUI options as you list:
>
>> Solid Block size 4GB
>
> Only option is "Create solid archive" - I checked it.
>
>> Number of CPU threads 2 (E8400 3GHz Dual Core 6MB L2 cache)
>
> Only option I see is "multi-threading" - I checked it.
>
>> The test uses about 700MB of memory (of 4GB of DDR2-800 CAS5 memory).
>
> Using 670 Mb here.
>
>> The test completes on my machine in 7 minutes 13 seconds.
>> 1,073,741,824 bytes in 433 seconds is 2,479,773 bytes/sec
>> compression rate. During the test, the CPU runs at only
>> 75%, meaning there is 25% of the dual core not used. (And that
>> means this test does not scale well, apparently.)
>>
>> You should be able to beat me, with a quad core, by a bit.
>> Assuming 7ZIP scales to four threads. I don't know if it
>> does or not. If you see "Number of CPU threads 4",
>> the contest should be interesting.
>
> Alas, I am not even close, confirming my first impression that
> this system is considerably slower than it "should" be:
>
> The test took 18 min 36 sec with the overall CPU usage between
> 30 and 33%.
>
> Without multi-threading, CPU usage is 25% and it takes
> correspoindingly slightly longer (canceled before it finished).
>
> CPUZ show stock bus speed (200) and multiplier (16), so
> nothing fishy in the CPU setup. Same for memory - no OC,
> expected frequency.
>
> Hmm, what going on?
>
> - Dima

That motherboard seems a bit weird. Check out this suggestion.

http://vip.asus.com/forum/view.aspx?...Language=en-us

*******

If that doesn't help, I'd try setting a few BIOS things to manual. I
have a couple Asus motherboards, where "manual" exposes the
most settings, and some of those settings can be interesting.

For example, I'd set "CPU Overclock" to [Manual] rather than [Auto].
It may allow you to observe clock frequencies, while in the BIOS.
And check if something isn't right.

Your memory "DCT Unganged Mode" is set to [Always] by default, which
is best for a processor like a quad core. I don't even see an option
to disable it, which might be beneficial if you were using
a single core processor. Maybe there aren't any single core
processors that work in the motherboard ?

Do you have two sticks of RAM, or only one ? Using one stick,
might drag down the multithreaded efficiency.

I was toying with the idea of suggesting a "cache and memory" type test,
which gives bandwidth curves. But the problem is finding a program
that does a nice job. My copy of SiSoftware Sandra didn't do a good
job. It took me forever to figure out how to get the displayed
results to appear (copy of Sandra from 2009).

Paul

On Fri, 06 Jan 2012 01:38:13 -0500, Paul wrote:

>I was toying with the idea of suggesting a "cache and memory" type test,
>which gives bandwidth curves. But the problem is finding a program
>that does a nice job. My copy of SiSoftware Sandra didn't do a good
>job. It took me forever to figure out how to get the displayed
>results to appear (copy of Sandra from 2009).

Memtest86+ shows the speed of the memory and the three levels of cache
(if present).

--
Regards - Rodney Pont
The from address exists but is mostly dumped,
please send any emails to the address below
e-mail rpont (at) gmail (dot) com

Rodney Pont wrote:
> On Fri, 06 Jan 2012 01:38:13 -0500, Paul wrote:
>
>> I was toying with the idea of suggesting a "cache and memory" type test,
>> which gives bandwidth curves. But the problem is finding a program
>> that does a nice job. My copy of SiSoftware Sandra didn't do a good
>> job. It took me forever to figure out how to get the displayed
>> results to appear (copy of Sandra from 2009).
>
> Memtest86+ shows the speed of the memory and the three levels of cache
> (if present).
>

I was looking for something with graphs, but you're right,
memtest86+ will give the answer. I was thinking maybe the
shape of the graph would be important. I'm not sure whether
memtest86+ determines the numbers by actually disabling cache levels,
or uses an inflection point method.

Maybe my results would have looked better if I had the graph in
linear mode rather than log. That could have been why it
didn't look too good.

http://fileshosts.com/intel/Asus/P5K...che_mem_tn.png

Paul

On Fri, 06 Jan 2012 03:02:13 -0500, Paul wrote:

>I was looking for something with graphs, but you're right,
>memtest86+ will give the answer. I was thinking maybe the
>shape of the graph would be important.

Being colour blind I prefer the AIDA 64 benchmarks:

http://www.aida64.com/product/aida64...ition/overview

The blue section at the left shows the performance, the faster it is
the longer the blue section.

--
Regards - Rodney Pont
The from address exists but is mostly dumped,
please send any emails to the address below
e-mail rpont (at) gmail (dot) com

"DK" <dk@no.email.thankstospam.net> wrote in message news:je74hd$phs$1@dont-email.me...
> In article <je5ps8$4js$1@dont-email.me>, "SC Tom" <sc@tom.net> wrote:
>
>>Mine has the 6MB L3, and the 7Zip score is 8.6MB/sec. using 4/4 CPU threads. I
>> watch all 4 cores in Task Manager and
>>they are like synchronized swimmers at the Olympics.
>>I love the decompression rate of 135MB/sec.
>
> What version of 7Zip do you have? The old 4.6 beta only
> uses 33% of the 4 cores and they are definitely not sybchronized.
>

4.65. It doesn't say Beta, so I assume it's the standard released version. Which reminded me to check for an update.
I'll let you know how 9.20 goes once I install and test with it.
--
SC Tom

"DK" <dk@no.email.thankstospam.net> wrote in message news:je74hd$phs$1@dont-email.me...
> In article <je5ps8$4js$1@dont-email.me>, "SC Tom" <sc@tom.net> wrote:
>
>>Mine has the 6MB L3, and the 7Zip score is 8.6MB/sec. using 4/4 CPU threads. I
>> watch all 4 cores in Task Manager and
>>they are like synchronized swimmers at the Olympics.
>>I love the decompression rate of 135MB/sec.
>
> What version of 7Zip do you have? The old 4.6 beta only
> uses 33% of the 4 cores and they are definitely not sybchronized.
>

I installed 9.20 and ran the benchmark on it. Pretty much the same as before- 8.7MB/s and all 4 cores synchronized
within 3 or 4% of each other. It would start at ~60%, then ramp up to 100% and stay there until the next pass. Memory
used was 851MB (forgot to mention that on the last reply).

I ran the 1GB "create a 7z file" as in one of the other sub-threads. Mine took 3 min. 35 sec. to complete, with the 4
cores fairly balanced from 48-53%. Towards the end of the compression, core number 1 (I think) dropped off to about 10%.
But that may have been because OE6 was checking for new mail. The other cores picked up the slack, increasing usage by
about 7-8%.

I did mistakenly start creating a .zip file instead of a .7z one and it took considerably longer, even though it still
used all 4 cores, fairly balanced across them all. I quit it after 6 or 7 minutes with it being about 50% done. Don't
know why there would be such a disparity in time, unless 7ZIP doesn't create WinZip files that well. Is that possibly
what you did that took 18.5 minutes?
--
SC Tom

"DK" <dk@no.email.thankstospam.net> wrote in message news:je82m6$iku$1@dont-email.me...
> In article <je7nu8$h25$1@dont-email.me>, "SC Tom" <sc@tom.net> wrote:
>>
>>"DK" <dk@no.email.thankstospam.net> wrote in message
>> news:je74hd$phs$1@dont-email.me...
>>> In article <je5ps8$4js$1@dont-email.me>, "SC Tom" <sc@tom.net> wrote:
>>>
>>>>Mine has the 6MB L3, and the 7Zip score is 8.6MB/sec. using 4/4 CPU threads.
>> I
>>>> watch all 4 cores in Task Manager and
>>>>they are like synchronized swimmers at the Olympics.
>>>>I love the decompression rate of 135MB/sec.
>>>
>>> What version of 7Zip do you have? The old 4.6 beta only
>>> uses 33% of the 4 cores and they are definitely not sybchronized.
>>>
>>
>>4.65. It doesn't say Beta, so I assume it's the standard released version.
>> Which reminded me to check for an update.
>>I'll let you know how 9.20 goes once I install and test with it.
>
> Even more weird. 4.65 is what I had before and it was definitely not
> using all cores. I now tried 9.2 and in it I can at least see options that
> Paul said he had in 4.60. In the 9.2m it has "Number of CPU threads"
> but bor some reason only 1 and 2 shows in the pulldown list. The CPU
> utiization with 2 threads is 40% and consequently the compression
> rate is a little faster but nowhere near yours. And we have identical
> MB and RAM, same OS (XP SP3, right?) and very similar albeit
> not identical CPU.

Yep, I'm running XP Home SP3. The only "tweaking" I've done to my system is unlocking the other two cores (like I posted
earlier), and changing my RAM timing from 9-9-9-25-34 to 8-8-8-21-31, but I don't see where that would make as much
difference as there seems to be between our two systems. The only other real differences I see are that I'm running an
add-in video card (GT 240) and SATA HDD instead of PATA. Do you have the latest BIOS and other drivers from the ASUS
site?

Here's a greatly snipped text report from CPU-Z for comparison, if you want:

CPU-Z version 1.58

Processors
-------------------------------------------------------------------------
Number of processors 1
Number of threads 4

Processors Information
-------------------------------------------------------------------------

Processor 1 ID = 0
Number of cores 4 (max 4)
Number of threads 4 (max 4)
Name AMD Phenom II X4
Codename Deneb
Specification AMD Phenom(tm) II X4 B55 Processor
Package Socket AM3 (938)
CPUID F.4.3
Extended CPUID 10.4
Brand ID 29
Core Stepping RB-C3
Technology 45 nm
TDP Limit 160 Watts
Core Speed 3215.1 MHz
Multiplier x FSB 16.0 x 200.9 MHz
HT Link speed 1004.7 MHz
Instructions sets MMX (+), 3DNow! (+), SSE, SSE2, SSE3, SSE4A, x86-64, AMD-V
L1 Data cache 4 x 64 KBytes, 2-way set associative, 64-byte line size
L1 Instruction cache 4 x 64 KBytes, 2-way set associative, 64-byte line size
L2 cache 4 x 512 KBytes, 16-way set associative, 64-byte line size
L3 cache 6 MBytes, 48-way set associative, 64-byte line size

Chipset
-------------------------------------------------------------------------
Northbridge NVIDIA GeForce 7025 rev. A3
Southbridge NVIDIA nForce 630a rev. A2
Graphic Interface PCI-Express
PCI-E Link Width x16
PCI-E Max Link Width x16
Memory Type DDR3
Memory Size 4096 MBytes
Channels Dual, (Unganged)
Memory Frequency 669.8 MHz (3:10)
CAS# latency (CL) 8.0
RAS# to CAS# delay (tRCD) 8
RAS# Precharge (tRP) 8
Cycle Time (tRAS) 21
Bank Cycle Time (tRC) 31
Command Rate (CR) 1T
Uncore Frequency 2009.4 MHz

Memory SPD
-------------------------------------------------------------------------
DIMM # 1
SMBus address 0x50
Memory type DDR3
Module format UDIMM
Manufacturer (ID) G.Skill (7F7F7F7FCD000000)
Size 2048 MBytes
Max bandwidth PC3-10700 (667 MHz)
Part number F3-10666CL9-2GBXL
Number of banks 8
Nominal Voltage 1.50 Volts
EPP no
XMP no
JEDEC timings table CL-tRCD-tRP-tRAS-tRC @ frequency
JEDEC #1 6.0-6-6-17-23 @ 457 MHz
JEDEC #2 7.0-7-7-20-27 @ 533 MHz
JEDEC #3 8.0-8-8-22-31 @ 609 MHz
JEDEC #4 9.0-9-9-25-34 @ 685 MHz

DIMM # 2
SMBus address 0x51
Memory type DDR3
Module format UDIMM
Manufacturer (ID) G.Skill (7F7F7F7FCD000000)
Size 2048 MBytes
Max bandwidth PC3-10700 (667 MHz)
Part number F3-10666CL9-2GBXL
Number of banks 8
Nominal Voltage 1.50 Volts
EPP no
XMP no
JEDEC timings table CL-tRCD-tRP-tRAS-tRC @ frequency
JEDEC #1 6.0-6-6-17-23 @ 457 MHz
JEDEC #2 7.0-7-7-20-27 @ 533 MHz
JEDEC #3 8.0-8-8-22-31 @ 609 MHz
JEDEC #4 9.0-9-9-25-34 @ 685 MHz


Display Adapters
-------------------------------------------------------------------------
Display adapter 0
Name NVIDIA GeForce GT 240
Revision A2
Codename GT215
Technology 40 nm
Memory size 512 MB
PCI device bus 2 (0x2), device 0 (0x0), function 0 (0x0)
Vendor ID 0x10DE (0x1043)
Model ID 0x0CA3 (0x8328)
Performance Level Default
Core clock 135.0 MHz
Shader clock 270.0 MHz
Memory clock 135.0 MHz
Performance Level 2D Desktop
Core clock 405.0 MHz
Shader clock 810.0 MHz
Memory clock 324.0 MHz
Performance Level 3D Applications
Core clock 550.0 MHz
Shader clock 1340.0 MHz
Memory clock 1700.0 MHz
--
SC Tom

DK wrote:
> In article <je7nu8$h25$1@dont-email.me>, "SC Tom" <sc@tom.net> wrote:
>> "DK" <dk@no.email.thankstospam.net> wrote in message
>> news:je74hd$phs$1@dont-email.me...
>>> In article <je5ps8$4js$1@dont-email.me>, "SC Tom" <sc@tom.net> wrote:
>>>
>>>> Mine has the 6MB L3, and the 7Zip score is 8.6MB/sec. using 4/4 CPU threads.
>> I
>>>> watch all 4 cores in Task Manager and
>>>> they are like synchronized swimmers at the Olympics.
>>>> I love the decompression rate of 135MB/sec.
>>> What version of 7Zip do you have? The old 4.6 beta only
>>> uses 33% of the 4 cores and they are definitely not sybchronized.
>>>
>> 4.65. It doesn't say Beta, so I assume it's the standard released version.
>> Which reminded me to check for an update.
>> I'll let you know how 9.20 goes once I install and test with it.
>
> Even more weird. 4.65 is what I had before and it was definitely not
> using all cores. I now tried 9.2 and in it I can at least see options that
> Paul said he had in 4.60. In the 9.2m it has "Number of CPU threads"
> but bor some reason only 1 and 2 shows in the pulldown list. The CPU
> utiization with 2 threads is 40% and consequently the compression
> rate is a little faster but nowhere near yours. And we have identical
> MB and RAM, same OS (XP SP3, right?) and very similar albeit
> not identical CPU.

Just out of curiosity, have you looked at your BIOS screen recently ?

First, you should have "full screen logo" disabled, in case the
motherboard presents an image instead of text. (A couple of my
motherboards default to presenting the full screen logo, so this
has to be disabled.)

Next, I'd want to check the BIOS declaration of the processor identity.
Is the processor mis-identified, or is the model information and
frequency right ?

Either 7ZIP is only offering "1" and "2" as options, because the program
can only handle two threads of execution. (Some algorithms can't be
"divide and conquer" indefinitely.) Or, the program might be offering
those options, because it thinks the processor only has two cores.
And it might get that information from the operating system.
I understand as well, from watching Linux boot screens (dmesg), that
the BIOS passes information about the number of cores in some kind
of table. So it might be possible for the BIOS to mis-inform the OS.
This wouldn't be a problem, if the OS also had its own identification
procedures. So it's a matter of whether the OS places all its trust in BIOS
tables, or whether it also does some of its own detection.

For example, when I boot Linux in a virtual machine on my PC, Linux
complains that the BIOS table "reports one core" which is correct,
but "the core number is 1 instead of 0", implying the virtual BIOS
isn't passing "core0" as the identity of the virtual processor. That's
how I know that at least with Linux, Linux is inspecting some info
from a BIOS table, and in that case, did not like what it saw. Linux
didn't crash or anything, and the message was more of a warning than
an error. It didn't actually affect the operation of the OS.

Paul