What are the minimum speeds of CPUs for various DC projects?

[networkman]

Okay, so what's the minimum speed of processor for some of these projects anyways?

I think it would be very helpful to know this so we don't end up putting a machine to use on a certain project only to find out that it's not going to be able to complete the workunit in time for the project.

I'll start with an example:

I just picked up (6) Tyan S1563 Dual Pentium AT motherboards for a little under $5 each. I have all the processors(P166s), ram, ps, etc to get these working, plus I figure this'll give me a chance to build a crackrack like I've always wanted too albeit slower than I'd planned.

Okay, assuming (2) P166 chips per board, that's 333MHz(166*2) * 6 boards = 1,998MHz or roughly 2GHz, almost 10% of MereMortal's stated goal for ECCp additions.

Now, the question is, will these machines be able to complete work in a reasonable enough time to be helpful to this project? Assuming that the client will even run on them? If not, then please suggest alternative projects. Thanks.

 

[Evadman]

SETI would kill those machines, same for folding. I would say ECCp is about your only choice.

 

[Freewolf]

the p166 taks 2 1/2 days more or less per a wu for seti

 

[ProviaFan]

SETI does not have a time limit on the WUs after all. Nevertheless, those processors would be in reality way too slow to practically run SETI (imho, coming from a F@H and ECCp guy ). Folding@Home, unfortunately, requires at least an (approximately) 300MHz CPU to meet the WU deadlines, so that's out (and no, dual processors don't count - F@H is not designed to distribute the processing of one WU over several processors). ECCp will run even on old 486s (albeit very slowly), since there are no time limits on the WUs.

Message edited to remove untruth that might mislead new persons

 

[Spacehead]

Originally posted by: jliechty
SETI does have a time limit on the WUs, but iirc it's something quite manageable (like a month or so).

SETI considers you inactive if you don't return a result in, in 28 days. However, i don't think there is a time limit on returning WU's.

 

[ProviaFan]

Originally posted by: Spacehead

Originally posted by: jliechty
SETI does have a time limit on the WUs, but iirc it's something quite manageable (like a month or so).

SETI considers you inactive if you don't return a result in, in 28 days. However, i don't think there is a time limit on returning WU's.

Oops, my mistake. Sorry.

 

[networkman]

Okay, so S@H would be insanely slow(I kinda figured this) and F@H probably won't even work due to the requirements.

I know that both RC5 and OGR will run, since those can be run on 486s; actually, RC5 can even be run on a 386 - OGR doesn't appear to though.. I tried putting in numbers in the Speed Calculator and even having a 1,000 GHz of 386s won't result in any numbers for OGR.

ECCp would appear to have the same requirements as OGR, namely that of a 486 minimum.

So far we have this:

RC5 ----- 386
OGR ----- 486
ECCp ---- 486
S@H ----- 486
F@H ----- Pentium II(?)

How about some other projects or corrections of info?

Might even something good to add to our FAQs, especially so that people know they can help out our Team with even the most humble of machines.

 

[ViRGE]

Originally posted by: networkman

I know that both RC5 and OGR will run, since those can be run on 486s; actually, RC5 can even be run on a 386 - OGR doesn't appear to though.. I tried putting in numbers in the Speed Calculator and even having a 1,000 GHz of 386s won't result in any numbers for OGR.

I don't have any numbers for OGR on a 386, so there's nothing to derive a multiplier from. That said, OGR is pretty hefty, enough so that I'll bet that a 386 is too slow to run it within reason. As for a Pentium, the Pentium will be far better in terms of speed, but if I remember OGR results correctly, we're only talking a couple of WU's/day on average, maybe less, as it's been a while since I've run OGR on a Petium class machine. RC5 really seems to be the way to go with something like a rack of Pentiums.

 

[Swanny]

I would say either SETI or ECCp for those boards. But you'd probably be better off with ECCp.

 

[Swanny]

Oh and the quote in jliechty's sig is hilarious!

 

[GhettoFob]

You could try Distributed Folding.

 

[Slaughter]

SETI would run, but how well would determine how much ram you have in them. I had a 133 with 64megs and it spit WUs every 2 days or so...and for comparisons sake, I had a 120 with 16 megs crunching and becuase it used swap all the time it took almost 10 days. Go figure. Its up to you! And who knows, maybe you could overclock those 166's a bit and churn them out a wee bit faster.

 

[Assimilator1]

S@H ----- 486

Err ,I wouldn't bother running SETI on a 486 ,too slow ,& would only run with a DX version or an SX with a Co Pro FPU.
Min RAM for SETI should be 32Mb (SETI Takes 15Mb)with no other programs running! ,48Mb would obviously be better ,with 64Mb being safe (as long as you don't run dozens of other programs alongside).
I would say the slowest worthwhile for SETI would be a Pentium 200/166MMX(single) .Having said that if a slower machine is all you have & you want to run SETI then go for it!
Inccidently the Pentium MMX's are considerably faster in SETI than an equivalently clocked Pentium classic (due to L1 cache size & bandwidth mainly)

Btw I had a Pentium 233MMX@266 with CAS2 SDRAM & was averaging 30.5hrs/WU.When that rig had a Pentium 166MMX@188 it was managing .....[edit](damn the newer GUI SETIQ logs are much harder to look through,can't find info)........found it ,the Pentium 166MMX@188 did most WU's in about 39-40hrs

Freewolf
Dual P166 times would be considerably slower due to shared memory bandwidth

 

[Swanny]

Looking at all the stats, it would seem that ECCp needs help the most. Although I don't do the project myself, I think that's where the power should go.

 

[PeterN]

Talking out of experience (and without any paper help ), they could be used excellent for RC5 (can be used for almost every proc).
OGR-25 is too heavy.
OGR-24 can be done, but there are not much of those left in the project.

ECCp runs on almost everything. Even on a 386/12MHz (don't ask how I did it or how fast it went. It wasn't worth the trouble It was fun though )
AMD is most suitable for this project.

Seti requires much more resources. Have tried it on a P166MMX@200 with 96Mb, but times still were over a day per WU.

Prime95 is possible, but only factoring. The big WU's for real prime number crunching are only done in a reasonable time with a 1GHz or higher, since the calculations/WU grow as time and project progresses.

Haven't got enough experience in other projects to help you out further.

 

[networkman]

Okay, this is all good to know.

I'm just thinking that if I/we were to put together some information on the "Minimum Requirements" and "Recommended Equipment" like what is done on most software packages now, that it would be especially helpful to the people looking to get involved in the various projects.

 

[RaySun2Be]

Originally posted by: networkman
Okay, this is all good to know.

I'm just thinking that if I/we were to put together some information on the "Minimum Requirements" and "Recommended Equipment" like what is done on most software packages now, that it would be especially helpful to the people looking to get involved in the various projects.

Excellent idea, NWM!

 

[ken008]

By the number of CPU`s ECCP is the best choice. When two clients produce the same DP`s we will have a solution. ( I think ) Therefore the more clients running the better the chance for a collision. Solving the thing is the best way to keep fifth place. So in the case of ECCP the number of clients running has some bearing on the probability of success regardless of speed. This is what I think, if I am wrong please correct me.

 

[Postman]

Above applies ofcourse only in the dreamland of free electricity . I tend to judge equipment usefullness by TOC/work done. $10 486/66 system versus $100 P350 running for a year 24/7. You know .

 

[Assimilator1]

Something else that could be worth while adding to an FAQ is a link toRojakpot's excellent bios guide.

Seeing as many projects benefit from tuned memory timings then this would help greatly