climateprediction.net (CPDN) home page
Thread 'memory timings'

Thread 'memory timings'

Message boards : Number crunching : memory timings
Message board moderation

To post messages, you must log in.

AuthorMessage
ChrisD

Send message
Joined: 8 Aug 04
Posts: 69
Credit: 1,561,341
RAC: 0
Message 39664 - Posted: 1 May 2010, 20:39:52 UTC

I would like to ask users who run an AMD Phenom II X4 processor, to post in this thread what Memory Modules they are using. DDR2 or DDR3 and memory timings please.

I can find the Phenoms, of course, but the machine specs only tell the Floating Point speed and the Integer speed.

Reason is that I currently use 2 Phenom 9500\'s, the original Agena core, clocked at 2400 and 2600 MHz.

I am using High Speed DDR2-6400 RAM Corsair clocked at 440 and 480 MHz at 5-5-5-15 timings.

Just got enough dole to buy a Phenom II X4 BE Version clocked at 3.4 GHz.

Now I wonder how much extra speed I get if I buy a new AM3 Motherboard and RAM to go with it.

Thank You for Your cooperation.

ChrisD
ID: 39664 · Report as offensive     Reply Quote
Profilegeophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2187
Credit: 64,822,615
RAC: 5,275
Message 39665 - Posted: 1 May 2010, 23:28:48 UTC

ID: 39665 · Report as offensive     Reply Quote
Belfry

Send message
Joined: 19 Apr 08
Posts: 179
Credit: 4,306,992
RAC: 0
Message 39669 - Posted: 2 May 2010, 18:01:19 UTC - in response to Message 39664.  

Phenom II X4 945, 3.0GHz, revision C2; 2 x 2GB Corsair XMS2, DDR2-1066; Ubuntu 9.04 x86_64; BOINC 6.4.5; geophi\'s indispensible patch run against hadsm3_um* (he didn\'t author it, but discovered its applicability to CPDN\'s slab models.)

For energy consumption considerations I normally run at 2.8GHz, DDR2-800 (4-5-5-16, unganged), 1.225 Vcore, 1.90 Vdimm; giving me ~125 watt system power consumption with three hadsm3\'s and one hadam3p running. I observe 0.8225 sec/TS over twenty minutes on one of the slab models (no post-processing occuring on any of the hadsm3\'s.)

3.0 GHz, DDR2-800, 4-5-5-16: ~135 watts, 0.8065 sec/TS (1.275 Vcore)
2.8 GHz, DDR2-1066, 5-6-6-18: ~134 watts, 0.7458 sec/TS (2.10 Vdimm)
3.0 GHz, DDR2-1066, 5-6-6-18: ~144 watts 0.7247 sec/TS

For an extra 10 watts, I get a much bigger performance boost by speeding up the memory than by bumping up the processor. Can this progression hold through DDR3-1600? I\'ll just have to move to DDR3 someday.

Unfortunately there isn\'t a way to fully enable SSE2 with the Windows hadsm3_um* binary. For reference my machine (2.8 GHz, DDR2-800) does about 1.25 to 1.30 sec/TS under a Windows 7, VMware virtual machine.
ID: 39669 · Report as offensive     Reply Quote
Belfry

Send message
Joined: 19 Apr 08
Posts: 179
Credit: 4,306,992
RAC: 0
Message 39670 - Posted: 2 May 2010, 18:23:28 UTC - in response to Message 39664.  

One other thing, if you plan on running with integrated graphics, get a board with sideport memory and disable the shared memory in BIOS. 3D performance will drop, but CPDN and other memory-intensive applications will benifit significantly. I forget my exact numbers from when I tested it a while back, but I think it\'s around 20% improvement in sec/TS for slab models.
ID: 39670 · Report as offensive     Reply Quote
Belfry

Send message
Joined: 19 Apr 08
Posts: 179
Credit: 4,306,992
RAC: 0
Message 39671 - Posted: 2 May 2010, 18:32:17 UTC

Oops, I\'m out of order. The third post in this thread should be someone who joined the project on Aug 6, 2004.

Sorry!
ID: 39671 · Report as offensive     Reply Quote
old_user140164

Send message
Joined: 18 Dec 05
Posts: 1
Credit: 538,436
RAC: 0
Message 39676 - Posted: 3 May 2010, 16:40:41 UTC - in response to Message 39671.  

:-D

Oops, I\'m out of order. The third post in this thread should be someone who joined the project on Aug 6, 2004.

Sorry!


ID: 39676 · Report as offensive     Reply Quote
Profilegeophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 2187
Credit: 64,822,615
RAC: 5,275
Message 39677 - Posted: 3 May 2010, 17:29:32 UTC - in response to Message 39671.  

Oops, I\'m out of order. The third post in this thread should be someone who joined the project on Aug 6, 2004.

I couldn\'t figure that out until valle posted the smiley and I went back and looked again. Now I get it, I\'m just a little slow.
ID: 39677 · Report as offensive     Reply Quote
Belfry

Send message
Joined: 19 Apr 08
Posts: 179
Credit: 4,306,992
RAC: 0
Message 39678 - Posted: 4 May 2010, 1:23:38 UTC
Last modified: 4 May 2010, 2:11:52 UTC

First there was nothing: link. Then the universe started to cool: link, link, link, link, link.

Raise a glass to the progenitors of boinc-based CPDN!

Now back to the thread:

But I realize just increasing the memory speed won\'t boost performance without end. At some point you have to increase the clock-rate to accomodate the higher bandwidth. Maybe someone with a Phenom II 955/965 and some overclockable DDR3 can run some tests to determine how this happens. Like maybe four slab models run with the following settings:

3.0 GHz, DDR3-1066
3.2 GHz, DDR3-1066
3.4 GHz, DDR3-1066

3.0 GHz, DDR3-1333
3.2 GHz, DDR3-1333
3.4 GHz, DDR3-1333

3.0 GHz, DDR3-1601 (240 HT, 12.5 multi, maybe with a 2.4 GHz northbridge)
3.23 GHz, DDR3-1594 (239 HT, 13.5 multi)
3.45 GHz, DDR3-1587 (238 HT, 14.5 multi)

Oh, and it\'d be really great if you\'ve got Linux with the SSE2 patch applied, as it might be the only way to push the limits of the AMD architecture.
ID: 39678 · Report as offensive     Reply Quote
ChrisD

Send message
Joined: 8 Aug 04
Posts: 69
Credit: 1,561,341
RAC: 0
Message 39681 - Posted: 6 May 2010, 10:30:59 UTC
Last modified: 6 May 2010, 10:33:18 UTC

Belfry wrote:

Phenom II X4 945, 3.0GHz, revision C2; 2 x 2GB Corsair XMS2, DDR2-1066; Ubuntu 9.04 x86_64; BOINC 6.4.5; geophi\'s indispensible patch run against hadsm3_um* (he didn\'t author it, but discovered its applicability to CPDN\'s slab models.)


Seems that patching against this \'Intel does not like AMD\' BUG-- is still very much needed. When will they ever stop that crap??

Look here

INTEL CPU: 2971 FLOPS, 6228 INT

AMD CPU..: 3558 FLOPS, 8689 INT (FLOPS + 19.75%, INT + 39,5%)

Guess what, These 2 CPU\'s crunch current WU\'s at exactly the same speed???

AMD\'s are still forced to execute less efficient code.

We must find a way persuade AMD to make an AMD Compiler.

ChrisD
ID: 39681 · Report as offensive     Reply Quote

Message boards : Number crunching : memory timings

©2024 cpdn.org