Info we all want and need...folding benchmarks done right

[jeepman464]

If anyone is interested I'm going to be doing some comparison benchmarking of different folding systems under the new rules. The following systems will all be tested using the same work units (wasteful insofar as points go, but useful insofar as fair benchmarking.) All system will benchmark two different WU of each of the three types (gromacs, double gromacs, and Tinkers...that coresponds to FAH core 78 79 and 65 respectively) for a total of 6 benchmarks per configuration.

P4 2.8Ghz 800mhz HT=on
P4 2.8GHz 800mhz HT=off
P4 Celeron 2.4GHz 400mhz
P4 1.6Ghz 400mhz
AMD Athlon XP 2400+ 266mhz (2.0GHz)

All of these systems will be tested at stock and at overclocking speeds (for overclocking I will first do a 10% FSB increase, then I will do a second run as fast as the system can remain stable) All the parts should be here this week so the results should be done by friday. Hopefully I can get some interesting info and help all of you future farm builders out there to make the right choices.

Keep Folding

Red_Silver
jeepman464@aol.com

[SavageBasher]

Welcome to the forums, but what is a P4 Celeron? :D

[Tatuya]

O cmon, give him a break what he meant was a Pentium Celeron. He typed P4 too many times before that and involutarily put P4 instead of P or Pentium. Welcome to the team!.

[jeepman464]

The latest revision of Celerons is based on the P4 Northwood architecture. Its really kinda an interesting chip. All the same functions as the Pentium 4 with the exception of Hyperthreading. Other than that, the celeron has a reduced L2 cache, which as evidenced by the increase in performance of the big P4 extreme's with 2mb cache, is going to hurt performance when compared to the regular P4.

Check out the following links for more info on the P4 celeron

http://www.intel.com/products/desktop/processors/celeron/
http://www.tomshardware.com/cpu/20020903/p4_celeron-01.html
http://www.tomshardware.com/cpu/20021016/index.html

[Papsomax]

Well unfortunately, the only cpu that will do the double gromes is the P4 2.8 (I am assuming you mean a P4C or E) As the celeron, the p4 1.8 are not 90nm technology so they do not do sse3 which supports dgroms, these cpus only have sse2. The Athlon also will not do dgroms. An Athlon 64 will (it has sse3), an Athlon XP will not as it only has sse2 capabilities.

MY bad on the sse3 vs sse2.....I meant sse2 vs sse, but i got caught reading about sse3 and made a transcription booboo...long night up unitl 4am working on my farm..so no caffeine!!!!!

Also comparisons like these have been done before in this forum and at the folding at home.org forums too. But feel free to report your data.

[jeepman464]

Very true on the dgroms point. Guess we'll just throw that out or take it into account stochasitcally or something (weight the overall P4 ratings by a percentage of how likely you are the get dgroms or something).
As for the benchmarking, I hate to complain but i have yet to find anything thats been a straight comparison of the same WU's being done by different processors, and especially haven't seen anything that's been done since the introduction of the new points system.

And as for my P4 "typo" its actually more of a convention than a formal nomenclature. The P4 Celeron title was used early on to differentiate between celerons based on PIII architecture and the newer P4 based celerons. Interestingly enough, the PIII 1300mhz and the P4 1700mhz performed about on par with each other. So much for Intel making more efficient clock cycles....more mhz is better, right?

[SavageBasher]

So much for Intel making more efficient clock cycles....more mhz is better, right?

Anyone who thinks different must be an AMD fan. (Insert sarcasm here)

Savagebasher himself is an AMD fan and knows that higher frequency does not necessarily mean more work... ;)

[Fast420A]

Papsomax. That's not right. The double gromacs USE SSE2, not SSE3 and the Athlon does NOT have SSE2. Also, CPU's will fold the double gromacs, just REALLY slowly without the SSE2 boost turned on. All the WU's are benchmarked on a P4 with the optimizations turned off.

[Gasoline]

Welcome to the forums, but what is a P4 Celeron? :D
A Socket-478 Celeron based upon the original P4's Willamette core, not the Northwood core, but with half it's L2 cache missing (only has 128k)


EDIT: Oops! My bad..... had old data.

The first Socket 478 Celerons up to and including 2.0GHz were Willamette core, but they switched to a Northwood-based core after that (2.0A and up) and gave it 256k cache.

http://www.upgradingandrepairingpcs.com/articles/upgrade12_02_03.asp


We have a bunch of those 1.8 and 2.0 GHz Willamette-based Celerons at work and they are really lackluster performers.

[Tebore]

Man I wanna see how this goes. Good to see more ppl hating those stupid P4 based Cellys, they give cellys a bad name.

Papsomax as stated by someone else your info on those chips are incorrect and the cores as well. Dgrom uses SSE2 found only on P4s & A64s and grom uses SSE only which is on AthlonXPs, morgan Core Durons and P4 and P3s. So far only the Prescott has SSE3 and no core from F@H uses those instructions.

[nut581]

but i think there's some sort of distribution inequality for these dgromac WUs .. i have seen NONE of A64 users ever got this WU ever since it was circulated .. all the happy folders who got the dgroms were running P4 :temper: meh want meh dgroms for meh a64..... :(

[Karbon]

i cant wait to see the benchmarks- everyone always says that amd farms are the best yada yada yada, their so cheap etc. but i think that a few p4s could blow away 10 amd xp's. a p4 3.0c with http://www.newegg.com/app/ViewProductDesc.asp?description=13-138-018R&type=Refurbish w/ cheap ram and an old vga card totals less than $300, not only would this be a very very fast setup, but u could also run 2 instances with the HT

BTW, im anti-AMD for life

[jeepman464]

glad to see some enthusiasm going on here. i'll let you know how things go. the rest of the parts should be in from zipzoom by wednesday so we'll be up and running by then

[Papsomax]

Yes my BAD on the dgromes, don't know where I came up with sse3?????? (long night working on the farm).....sse2 yes for the dgromes.....but XP Athlons yes NO sse2 just sse....even though an XP can crunch dgromes, I've never got any..only my P4C 2.8 has even gotten a dgrom and a friends A65 3000 and his Opteron 260.....and my 2.7 celey oc'd to 3.3ghz......

i need espresso!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! :eek:

[SavageBasher]

My 1.3 celeron got a dgromac a while back. Don't know why, but it did. :D Took a long time to complete though.

[Karbon]

my 467 celeron takes 110 hours per 100 frame gromac (23 points)

[Papsomax]

cant wait to see the benchmarks- everyone always says that amd farms are the best yada yada yada, their so cheap etc. but i think that a few p4s could blow away 10 amd xp's. a p4 3.0c with http://www.newegg.com/app/ViewProdu...&type=Refurbish w/ cheap ram and an old vga card totals less than $300, not only would this be a very very fast setup, but u could also run 2 instances with the HT

This isn't the best place to bash AMD vs Intel as this is old discussion. It's pure GHz that wins the folding race, period. The higher the GHz the quicker the cpu proceeses fah. SSE, SSE2, help too. But Stanford did their benchmarking with a P4C without any flags on. So an AMD 2200 oc'd to 2400+GHz will beat out a P4A 2.4. A P4C 2.8 running one instance will beat an Athlon XP 2800 hands down. You run both cpu's with 2 instances the results are very close. Most AMD folders run two instances of fah on the XP's. Since I began running 2 instances of fah on my AMD's, I've seen a 15-20% increase in ppw over running just one instance. These are dedicated folding boxes and if you look at the cpu usage one fah instance takes up 48% of the cpu, the other instance takes up 48%.

Here is a link of my pc's frame rates: Frame rates (http://www.ocforums.com/showthread.php?t=284961)

A G5 2.0Ghz plain ars kicks most AMDs and Intels butt. So, my P4C 2.8 @3,220MHz with HT churns away around 1200ppw-1500ppw(protein dependent), my Athlon 2500@2,412MHz with 2 instances churns out around 1000ppw. AMD[Duron vs Athlon vs Athlon XP vs A64 vs Opteron vs FX series), Intel (P4A vs C vs E vs EE), PPC (G5), sse vs sse2[someday sse3] linux vs windows, it's all relevent. I got tired of battling AMD vs P4 vs MAC vs linux vs Windows. I have them all in my boxen farm. My diskless layer will have Athlon XP, Duron, P4E and A64. My wife's getting a dual G5 2.0 soon. My borgs are P4 1.8's and 2.26's.

Fold on, hooah....whatever :argue: :rolleyes: :bang head :temper:

[Karbon]

my point wasnt to bash amd, but rather to show my opinion. It seems that about 90% of people make AMD farms, why doesnt anyone use a p4 farm? it would cost more per layer but it would probably be more productive for the same price

[Papsomax]

Because it is cheaper to make an AMD layer vs P4C or E layer, even at newegg refurb price. I'm building a diskless layer now and here are the approximate costs:

AMD server layer:
Athlon XP 2400 with OEM HSF (heat sink/fan): $32
PC Chips KM400 mobo: $30
PC2100 512MB: $70
2nd WOL Nic: $4
maxtor 40G HDD: $40
Samsung 52X CDROM: $22
Leadman 400W PSU: $25
Total for server layer: $223

AMD client layer:
Athlon XP 2400 with OEF HSF: $32
PC Chips KM400 mobo: $30
PC2100 128MB: $25
1/2 cost Sparkle 300W psu with splitter: $15
Total for AMD client layer: $102

Intel 2.8GHz P4E client layer:
P4E with OEM HSF: $139
ASUS P4R-800 VM mobo: $40
256MB PC2700: $25
1/2 cost Antec Trupower 550 with splitter: $45
Total for P4E client: $249

So let's say we are going to do a 5 layer farm. *It will have an Athlon XP 2400 server layer, 2 Athlon XP 2400 client layers and 2 P4E client layers.
Server layer cost: $223
2xAthlon client layers cost: $204
2xP4E's 2.8GHz client layers cost: $498
8 Port Linksys switch: $17
17" monitor: $35
KB: $10
Mouse: $5
10 case fans: $50
Cat5 cables: 5x3ft $15
Cat5 cable: 1x10ft $10
shelving: $40
misc (screws, washers, fan grids, etc): $25
UPS: 1000 PowerCom BNT-800: $57
Total for 5 layer diskless farm: *$1189

As you can see it's almost double the cost to build a P4E layer. Even if you got a 2.4C at $114 at newegg....it's still double the cost. You can almost buy 3 XP 2000's with mobo, ram, etc for the cost of one P4C/E layer. AND with those 3 cpus oc'd to 2200MHz and each running 2 instances of fah = 6 instances of fah vs only 2 for the P4C/E layer. One day P4C and E will eventually come down further in price, but AMD will also go down. So it's a personal preference. I just bought an A64 2800 and mobo, where I could have purchased 3x2000 XP,s etc. I just don't have an A64 so tis why I went that route.

paps

[Karbon]

hmmmm, you may have a point there. where did u find "Athlon XP 2400 with OEF HSF: $32" ? That's a pretty low price I'd say. For a performance point of view, I'd never consider using AMD for anything, but for folding layers it looks like the way to go. schools out in 3 weeks, and I'll make about $300 or so a week. I might be able to put up a farm. If i did, i would definitely use Windows 2000 pro, linux seems rather annoying to me. right now i have 3 pcs on a table, i need to find some shelving somewhere

[AmigoThree]

Maybe he meant 2000+, because that was what he talked about in the last paragraph. Either way, that is still really cheap.

[jeepman464]

one other point to mention.
Its not exactley a GHz race here. The amd's do a bit better on a ghz to ghz compariosn because of the fact that AMD have 3 FPU units and the Intel's only have 2 FPU's. That and the shorter pipeline of the AMD architecture are how they justify their rating system (i.e. 2.0ghz => 2400+ XP). But what I'm interested in is how exactley does this translate and how exactley should we be setting up farms to be the most efficient under the new scoring sytem.

[Papsomax]

Actually if you watch, newegg had a 2400+ oem (they said NO hsf for that price...it came with a hsf....just like the oem P4E 2.8's I got for $139 with shipping...they said NO hsf..but they both came with oem hsf......

There is also a place here in town that you can get refurbs and system pulls for dirt cheap.....they were selling system pulls 2.4A for $40!!!!!!!!......they are a liquidating company on a first come first serve basis...I missed out on that one :bang head

I have a P4C 2.8 and 2 x P4E 2.8's and one (on order) A64 2800+ along with an XP 2400+, 2200+, 2400-M, 2500+, 2800+, celey 2.7 and 1.7 and a Duron 1.3 and 1.6 in my personal farms. Also have on old G3 333 and PII 333 as well. And my wife is getting a new dual G5 2.0Ghz (I was hoping she'd wait until the new model came out) for her business. My borgs are P4 1.8's and 2.26's. My gaming system is my unlocked XP 2500 that I have oc'd up to 2.5GHz, but with 512MB of HyperX ram and a GeForce FX5950 and the oc backed down to 2412, I will put it up against any system to play games.

I use Win XP Pro/Home, Mandrake 10.0 Linux, OSX, and Overclockix, I use Oracle 9iAS and I am trying SQL Server 2000 Enterprise Edition.....so I am well rounded in all my systems. I used to be MAC only, then Intel only, then AMD only...........so there's my spiel. If you have unlimited funds go for Intel C/E farms..if you have limited funds then go with AMD......

EDIT

. The amd's do a bit better on a ghz to ghz compariosn because of the fact that AMD have 3 FPU units and the Intel's only have 2 FPU's. That and the shorter pipeline of the AMD architecture are how they justify their rating system (i.e. 2.0ghz => 2400+ XP).

I didn't want to get into that but yes that is why AMD's do so well folding and the same for PPC (G5's). Thanks for mentioning that! :beer:

If you notice the link of my various cpus that data is with the -advmethods and -forcesse flags.....but if you are going to do all you said with and without flags with and without ht and with and without dual fah on AMD's then KOOL!

[walaka7]

well when it comes to gromacs then yes mhz is the main deciding factor. but you throw tinkers into the mix and once again AMD's own just like they did in the early days, and even more so now that tinkers are benched on a P4 thus yielding 2 times the point value as before. Not saying that everyone should go out and build amd rigs, just stating an old and perhaps forgotten fact.

[Agent_Mull]

Papsomax, where did you hear that people run 2 instances on XPs?

[Papsomax]

I read about assigning 2 instances of fah to XP's here in the posts and threads. I'd have to serach the exact thread (but search is not working yet). If you run 2 instances of fah on XP's and non HT P4's you'll see 12-20% increase in ppw vs only running one instance. Your OS basically assigns one fah to 50% cpu usage and the 2nd fah for the remainng 50%.....this is on dedicated folding only cpu's...my actually OS assignment is 48/48.....with the remaining 4% to handle other os assignments...now if you try this on a non-dedicated box (ie gaming, etc_ your fah assignments might be 49/32 for example.


paps

[Agent_Mull]

K, ill try that when i get my farm. Ill just keep one on my rig.

[jeepman464]

adding a 1.8 duron to the mix too, although they're almost too slow to even consider these days....

[Papsomax]

adding a 1.8 duron to the mix too, although they're almost too slow to even consider these days....

Good cause I added A Duron 1.3 and 1.6 to my farm....

[jeepman464]

well shipping was delayed b/c of memorial day so parts get in tomorrow....expect results monday. oh and by the way does anyone know why i turned in 5 WU's this morning and got 2 points according to this webiste:
http://folding.extremeoverclocking.com/user_summary.php?s=&u=77567

seems that some new WU's don't give you any points....p1064,p1058,p1059

any ideas?

[Fast420A]

well shipping was delayed b/c of memorial day so parts get in tomorrow....expect results monday. oh and by the way does anyone know why i turned in 5 WU's this morning and got 2 points according to this webiste:
http://folding.extremeoverclocking.com/user_summary.php?s=&u=77567

seems that some new WU's don't give you any points....p1064,p1058,p1059

any ideas?


See your other post.

[jeepman464]

well the first bit of good news:
my $70 celeron 2.4 will stabley overclock to almost 3.5ghz. its currently installing windows xp at 145mhz FSB
24x145=3480mhz stable.....

first bit of bad news:
the amd board was DOA so will be delayed a bit

[jeepman464]

first bit of bad news: on protein 543 these are the following results (gromac core)
Duron 1.8 @ 2.0ghz= 60ppd
XP 2400 @ 2.08 ghz=105ppd
Celeron 2.4 at 3.4gz=72ppd


So L2 cache matters mucho much.