Latency Revisited - - Again!! - Post Results Here

[bum_jcrules]

There are a lot of different benchmarking tools out there for testing and tweaking memory. Here are some that I have used.


Latency - - <A HREF="http://www.aceshardware.com/articles/reviews/atb/files/cachmem-bin.zip" target="_new">Cachmem 2.6</A> Latency of L1, L2, L3, and the main memory.

(((If it does not work go <A HREF="http://www.aceshardware.com/Spades/read.php?article_id=157" target="_new">here</A> and go to the bottom of the page. Click the link and it will come up.)))


Bandwidth - - <A HREF="http://www.aceshardware.com/files/benchmarks/linpack.zip" target="_new">Linpack</A>


Bandwidth - - <A HREF="http://www.cs.virginia.edu/stream/ref.html#runrules" target="_new">Stream</A>


Utilization - - <A HREF="http://www.outertech.com/downloads.php?product=3&download=30&PHPSESSID=8b51c1d496f8165f5aecda173d17c049" target="_new">Cachman 5.11</A> Also has tweaks for Cache and Main Memory



What programs do you use to test and benchmark you latency, bandwith, and utilization?

Post them here...

<b>"I put instant coffee in the microwave and almost went back in time" - Steven Wright</b> :lol:

 

[jclw]

My most frequently used benchmark is Clibench, but it's not really a memory benchmark.

- JW

 

[bum_jcrules]

Is that <A HREF="http://clibench.daemonware.ch/index.php?seite=download" target="_new">Clibench 0.7.1.5</A>?

<b>"I put instant coffee in the microwave and almost went back in time" - Steven Wright</b> :lol:

 

[bum_jcrules]

Back to the latency issue...

Fatburger posted a reference to an article over at <A HREF="http://www.aceshardware.com/read.jsp?id=45000279" target="_new">Ace's Hardware</A> a few months back involving the P4 2.0GHz and the AXP 2000+. He also had a post reguarding latency and using the benchmark "<A HREF="http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&p=40932#40932" target="_new">Latency 2</A>."

Using the benhemarks above, I was wondering what type of results others are getting with their hardware setups. I wanted to see for myself and for others what the "realworld" performance of RDRAM vs SDRAM. (DDR and SDR)

I am really interested with the Cachemem and STREAM results.


Here is my Cachmem 2.6 results for my laptop...

PIII Mobile 733MHz
PC100 CL2 2-2-2
FSB 100MHz
Chipset is i440BX
MCH is 82443BX

Cache size/Memory speed info tool 2.6MMX - (c) 1999-2001, LRMS - DJGPP compiled
** Warning! Results are unreliable under Windows! **
CPUID support detected... 'GenuineIntel' with FPU TSC MMX
Family=6 Model=8 Step=3 Type=0 Chipset (Vendor/Device ID(Rev)): Intel/7190(03)
CPU clock: 647.2 MHz
Using 32MB physical memory block (alignment = 32)
Bandwidth - MMX linear access test... Read/Write/Copy (MB/s)
Block of 1KB: 4109.9 / 3179.4 / 5774.4
Block of 2KB: 4383.0 / 3293.6 / 5915.0
Block of 4KB: 4432.9 / 3225.5 / 5990.8
Block of 8KB: 4658.2 / 3384.5 / 6024.9
Block of 16KB: 4715.4 / 3394.0 / 3033.3
Block of 32KB: 929.8 / 2301.9 / 2886.5
Block of 64KB: 2597.2 / 2224.0 / 2825.1
Block of 128KB: 2593.8 / 2277.7 / 997.5
Block of 256KB: 1167.4 / 876.8 / 402.8
Block of 512KB: 733.9 / 211.3 / 335.3
Block of 1024KB: 731.4 / 203.1 / 304.2
Block of 2048KB: 733.7 / 189.4 / 287.4
Block of 4096KB: 708.2 / 3.5 / 268.4
Block of 8192KB: 729.9 / 184.1 / 252.8
Block of 16384KB: 735.9 / 179.7 / 252.9
Block of 32768KB: 733.8 / 180.3
Latency - Memory walk tests... ("pointer chasing")

steps: 4 8 16 32 64 128 256 512 1k 2k 4k (bytes)
Block of 1KB: 3 3 3 3 3 3 3 3 - - - cycles
Block of 2KB: 3 3 3 3 3 3 3 3 3 - - cycles
Block of 4KB: 3 3 3 3 3 3 3 3 3 3 - cycles
Block of 8KB: 3 3 3 3 3 3 3 3 3 3 3 cycles
Block of 16KB: 3 3 3 3 3 3 3 3 3 3 3 cycles
Block of 32KB: 3 4 5 7 7 7 7 7 7 7 7 cycles
Block of 64KB: 3 4 5 7 7 7 7 7 7 7 7 cycles
Block of 128KB: 3 4 5 7 7 7 7 7 7 7 7 cycles
Block of 256KB: 3 5 13 47 39 39 38 38 39 42 29 cycles
Block of 512KB: 18 23 42 72 72 72 73 74 76 82 91 cycles
Block of 1024KB: 18 23 42 72 72 85 73 74 76 81 90 cycles
Block of 2048KB: 18 23 42 73 74 72 73 74 76 81 90 cycles
Block of 4096KB: 18 23 42 72 74 110 73 74 76 80 158 cycles
Block of 8192KB: 18 23 42 72 72 72 73 74 76 80 88 cycles
Block of 16384KB: 18 23 42 75 73 72 73 74 76 81 88 cycles
Block of 32768KB: 18 23 42 72 72 72 76 74 76 80 90 cycles
Done.
This system appears to have 3 cache levels (enabled).
L1 cache (16KB) speed (MB/s): Read=4658.2, Write=3384.5
L2 cache (128KB) speed (MB/s): Read=2597.2, Write=2224.0
L3 cache (256KB) speed (MB/s): Read=1167.4, Write=876.8
Main memory speed (MB/s): Read=735.9, Write=179.7

Press any key...

Post what you are getting...


<b>"I put instant coffee in the microwave and almost went back in time" - Steven Wright</b> :lol:

 

[Crashman]

WARNING! Memory benchmarks are the WORST way to gather information about performance gains of using less latency. Why? Simple-did you really build your computer to only run memory benchmarks? I think NOT.

If gaming is the place you most need a performance increase, use gaming benchmarks. If video editing is your greatest performance concern, use those benchmarks. The only thing that matters is what affect it has on system performance, not memory-only performance.

What's the frequency, Kenneth?

 

[jclw]

Agreed.

For a general idea run a benchmark that does memory intensive calculations (like Clibench).

- JW

 

[FatBurger]

Game benchmarks are hard to do between different platforms with different video cards, hard drives, sound cards, etc.

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft

 

[Crashman]

True, but memory benchmarks cannot be interpreted for performance. For example, some programs will run better with faster memory while others will run faster with slower, lower latency memory. Memory benchmarks cannot interpret these implications.

What's the frequency, Kenneth?

 

[FatBurger]

Of course not, but the point is to find out more about memory technologies, not to help people decide what kind of RAM they need for their system.

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft

 

[bum_jcrules]

What would you use to do latency calculations???

I agree the real world performance is the bottom line. However, the original debate was latency within memory architectures. So again I ask the forum, What would you use to measure latencies in main memory?

<b>"I put instant coffee in the microwave and almost went back in time" - Steven Wright</b> :lol:

 

[Crashman]

I'll give you that, but because of the differences between, say, DDR SDRAM and RDRAM, comparing cas latencies could only be accurate if you converted the numbers to real time responses.

What's the frequency, Kenneth?

 

[FatBurger]

I was planning on that, however, relative latency is important too, not just realtime latency. It helps show how latency scales with clock speed.

<font color=blue>If you don't buy Windows, then the terrorists have already won!</font color=blue> - Microsoft

 

[bum_jcrules]

As per Slv_Phoenix,

<A HREF="http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&p=40932#40932" target="_new">"Okay, I've had a few thoughts lately and want to re-visit this thread. Can people help me out by running the stream program (the one that outputs copy, scale, add, triad) and posting their results. Also, post the exact specs regarding memory (chipset, memory type, memory clock used, CL, density, etc., etc.) please. Without at least the memory type and memory clock, the data will be generally useless.

I'm specifically looking for standard settings, such as PC800 actually clocked at PC800, PC100 actually clocked at PC100, and so on. Overclocking of the memory is acceptable if that overclock is at a markettable setting. (Such as PC800 OCed to PC1066.) The exact reason for this request is that the access times in ns are available for standard markettable memory settings, but not for PC100 running at PC112.

Thanks to everyone who helps. When/if enough data is collected, I will post my results."</A>


Let's try to do this again...

<b>"Sometimes you can't hear me because I'm talking in parenthesis" - Steven Wright</b> :lol: