• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

any difference in performance dual channel memory vs single?

B

Bill Clo

Gawd
Joined
Jan 24, 2002
Messages
959
I'm contemplating expanding my boxes again, and I notice that the Socket 754 processors are a lot cheaper than socket 939. I am wondering if there is any advantage in Folding with dual channel memory (socket 939) vs single (socket 754)? I know that socket 939 would allow me to upgrade to the dual processor chips later on.

I figure by the time the dual processor chips come down in price to more reasonable levels, that at least one of my boxes will be ready for retirement...
 
I haven't done any testing lately, but in the past we've never seen any significant difference in folding performance with anything besides processor speed. SDR, DDR, single channel, dual channel, fast hard drives, slow hard drives, FSB speed - they've all had little or no effect on folding speed. It's all about the processor!
 
Since you are going AMD, dual channel vs single channel does not matter yet. However, I hear the new QMD core proteins that are Intel only apparently benefit a lot from dual channel but no word on whether they will be the same for AMD, unless you really need the comp right now, I suggest you wait a month or two until the QMDs come out for AMD.
 
From an intel point of view it makes a big differance. I've got 2 systems both at 3.0/3.06 gzh. One is 533 FSB single channel and one is 800 FSB dual channel. On QMD's the one does ~273 PPD and the dual channel does a freakish ~475 PPD on the 450 pointer QMD's.
 
With my 754 I got a 10-20 sec speedup per frame when I upped the speed of the memory. 200-230mhhz Not all that important, though it does add up over time. Whatever you decide on you need to make sure you tweak the system to make it as fast as possible while still making sure it's doing calculations correctly.

I'm not sure how important QMD proteins are to a folding only box because it requires so much memory. Unless people are ok with buying 512mb-1gb+ of ram for the box.

-MN Scout
 
marty9876 said:
From an intel point of view it makes a big differance. I've got 2 systems both at 3.0/3.06 gzh. One is 533 FSB single channel and one is 800 FSB dual channel. On QMD's the one does ~273 PPD and the dual channel does a freakish ~475 PPD on the 450 pointer QMD's.
Click to expand...
Are they both Prescott cores or are they Northwoods?
 
Mattman said:
Are they both Prescott cores or are they Northwoods?
Click to expand...

I forget. The 800 FSB was bought from newegg just a month or so ago- Prescott I think then. The other is a Northwood I think (it's from my notebook....don't ask) it is a HT chip thou.
 
If it's a Prescott, then it has SSE3, where the Northwood wouldn't. That could be part of the reason why it folds faster, because QMDs use SSE3.


Oldbenwa
 
Oldbenwa said:
If it's a Prescott, then it has SSE3, where the Northwood wouldn't. That could be part of the reason why it folds faster, because QMDs use SSE3.


Oldbenwa
Click to expand...

U sure on that, I did not think anythign was using SSE3 from a folding point of view.
 
As far as I can tell they just require SSE2 also, although I wish they would start allowing A64s to get them.
 
marty9876 said:
U sure on that, I did not think anythign was using SSE3 from a folding point of view.
Click to expand...


From one of my logs:

[14:28:04] Folding@Home QMD Core
[14:28:04] Version 1.04 (Apr 7, 2005)
[14:28:04]
[14:28:04] Preparing to commence simulation
[14:28:04] - Assembly optimizations manually forced on.
[14:28:04] - Not checking prior termination.
[14:28:05] - Expanded 260843 -> 826210 (decompressed 316.7 percent)
[14:28:05]
[14:28:05] Project: 1912 (Run 63, Clone 6, Gen 8)
[14:28:05]
[14:28:05] Writing local files
[14:28:05] Extra SSE2/SSE3 boost OK.
[14:28:05] Entering QMD...
[14:28:49] System: p1912_ALA-dipeptide_umb2
[14:28:49]
[14:28:49] Verifying checksum
[14:28:49] Finished
[14:29:40] Completed 560 out of 2072 steps (27)
[14:43:54] Timered checkpoint triggered.
[14:47:18] Completed 581 out of 2072 steps (28)

So yeah, I'm sure. :p

Oldbenwa
 
howd u get to use sse3???? I have a venice and it only says sse boost

I would like to see some boostage!

Thanks

Edit: They dont have it for A64 yet..... :-(
 
Depends on the core. You probably don't have a QMD core since you are running an A64. They haven't started sending them to any A64s to the best of my knowledge.
 
With the on-die memory controller, I'm not sure sure an A64 would need dual channel as much as a P4 would. It is well known that P4's are slowed down a lot by single channel in just about everything while A64's aren't hindered as much. I remember with some of the first A64's that dual channel didn't help a whole lot versus single channel.

 
Oldbenwa said:
From one of my logs:

So yeah, I'm sure. :p

Oldbenwa
Click to expand...

Geez, you made me walk over and check mine. Ok, I'll check the home box this evening see if it says this also.

I recall reading on the offical forums no need for the precision SSE3 offers, and I'm too lazy to go find the thread. Honestly some of the mods over there don't know much more than we do.

Ok, back on topic. I dunno.... :)
 
not too sure how this impacts folding since it's mostly dependant on CPU speed, but, from what i've heard on G[H], dual channel nets you about a 5% performance increase in AMD systems, with like, a huge (10%+) increase in Intel systems (since they're more dependant on memory bandwidth)

a note about Socket 754: if you go with the Sempron, rememember that's a crippled A64 so it only has a single memory controller on die -- in other words, it will benefit more from dual channel than a real A64 would.

 
Carnival Forces said:
a note about Socket 754: if you go with the Sempron, rememember that's a crippled A64 so it only has a single memory controller on die -- in other words, it will benefit more from dual channel than a real A64 would.

Click to expand...

Also be sure to get the 3100+ and OC the bejesus out of it!
 
Carnival Forces said:
not too sure how this impacts folding since it's mostly dependant on CPU speed, but, from what i've heard on G[H], dual channel nets you about a 5% performance increase in AMD systems, with like, a huge (10%+) increase in Intel systems (since they're more dependant on memory bandwidth)

a note about Socket 754: if you go with the Sempron, rememember that's a crippled A64 so it only has a single memory controller on die -- in other words, it will benefit more from dual channel than a real A64 would.

Click to expand...


I had been contemplating one of the socket 754 3400 models. I definitely won't be getting the Semperon. I looked the cost of socket 754 vs 939, and it doesn't make sense for a maybe 5% speed increase with dual channel. Socket 939 3500 (2.2ghz) is $262, versus $156 for tthe Socket 754 equivalent (3200, 2.2ghz). Socket 939 3800 (2.4ghz) is $359, versus $175 for the Socket 754 equivalent (3400, 2.4ghz)...big difference there.

Works out to be $254 for a 2.4ghz socket 754 (CPU, mobo, memory), versus $464 for the Socket 939 equivalent. Whew.

I don't want to go the OC route again, so this is more cost effective for me.

Thanks for the info, guys.
 
wha? so you'll be getting a different socket Sempron? or an A64? i couldn't understand what you said...

 
There are Socket 754 non-Semperon models (Newcastle and Hammer). I'd be getting the Newcastle. Not half cache like the Semperons.
 
Bill Clo said:
There are Socket 754 non-Semperon models (Newcastle and Hammer). I'd be getting the Newcastle. Not half cache like the Semperons.
Click to expand...

Why L2 cache? Cache is unimportant with F@H otherwise Celreon's wouldn't kick so much ass at folding.
 
I hate to say it guys...

But the old school part of me is comming out here...



This thread is way the hell way tooooo much on topic.


I like cookies..


And im trying to convince myself buying 5x new cases is a bad thing.
 
Spectre said:
Why L2 cache? Cache is unimportant with F@H otherwise Celreon's wouldn't kick so much ass at folding.
Click to expand...

I was under the impression that the Celerons suffered doing Folding due to the small cache. I was also under the impression that while Folding isn't highly cache sensitive (ie, 1Mb cache made no difference), that too small of one can be a problem.

In any case, that CPU is ordered now, so...2.4ghz of AMD 3400 (512kb cache) for $173 is schweet.
 
KillerMob and I were comparing folding times on 64 bit rigs.

He was at 2.6 with a 754 and I was at 2.5 with a 939. Both cpu’s had the same cache size, my box is slightly faster per frame then his. About the only thing that makes sense is the memory bandwidth, which is about double on the 939.

I then set up to machines here at home, one 754 and one 939, same amount of ram, same clock speeds. Again, the 939 is faster. Not by a ton, but several seconds per frame faster.
 
Spectre said:
Also be sure to get the 3100+ and OC the bejesus out of it!
Click to expand...
Make sure to get extra bejesus, those OC better. The more you start with, the more can come out, right?
 
Bill Clo said:
I was under the impression that the Celerons suffered doing Folding due to the small cache. I was also under the impression that while Folding isn't highly cache sensitive (ie, 1Mb cache made no difference), that too small of one can be a problem.

In any case, that CPU is ordered now, so...2.4ghz of AMD 3400 (512kb cache) for $173 is schweet.
Click to expand...

Ok just saying...that 256k cache versus 512 shouldn't be an issue.
 
marty9876 said:
From an intel point of view it makes a big differance. I've got 2 systems both at 3.0/3.06 gzh. One is 533 FSB single channel and one is 800 FSB dual channel. On QMD's the one does ~273 PPD and the dual channel does a freakish ~475 PPD on the 450 pointer QMD's.
Click to expand...


is that all? my P4 @ 3.9 1114 FSB is pulling 600+ PPD


The [H]orde needs You!
 
jfb9301 said:
is that all? my P4 @ 3.9 1114 FSB is pulling 600+ PPD


The [H]orde needs You!
Click to expand...


you have been oficaly added to the bastard list along with unhappy_mage, oldbenwan and Kodiak Star.

Hey, it's a server doing stuff too ya know. That is pretty PPD action you got there.
 
BillR said:
KillerMob and I were comparing folding times on 64 bit rigs.

He was at 2.6 with a 754 and I was at 2.5 with a 939. Both cpu’s had the same cache size, my box is slightly faster per frame then his. About the only thing that makes sense is the memory bandwidth, which is about double on the 939.

I then set up to machines here at home, one 754 and one 939, same amount of ram, same clock speeds. Again, the 939 is faster. Not by a ton, but several seconds per frame faster.
Click to expand...

That's good to know. But the much higher cost of the socket 939 system isn't worthwhile for a dedicated folding system (IMHO). Totally different story for a gaming system though.
 
You must log in or register to reply here.
Back
Top