If I may make a request on behalf of 32 bit PCs

[petrw1]

There have been several posters that have noted that older 32 Bit hardware (PIII, Duron, etc) actually perform pretty decent doing TF up to BUT NOT above 64 bits. Doing TF above 64 bits or LL/DC work drops their thruput in half.

Coupled with that the currest published Primenet server default TF-LMH assignment rules assign exponents below 400M up to 64 bits. Any exponent 400M and above get TF-LMH assignments to 65 bits. As defined above the work up to 64 bits completes at thruput "X" and from 64-65 bits at 1/2"X".

Once all exponents are TF'd to 64 bits all these machines drop in thruput by 50%.

So what I am so humbly asking is that for those of you with 64 bit PCs that are assigning your own TF work below 64 bits is to consider taking exponents at 400M and above and leaving those lower to the 32 bit PCs

Thanks for your time.

[Uncwilly]

There are several users that hit hard in that range.
Torchwood Institute
Tydus
linded
monst
GIMPS Visualization
Some may not read the forum. You may be able to contact them directly.

[petrw1]

Originally Posted by Uncwilly

There are several users that hit hard in that range.
Torchwood Institute
Tydus
linded
monst
GIMPS Visualization
Some may not read the forum. You may be able to contact them directly.

Thanks...the biggest by far is GV (aka Chalsall) who is a frequenter of this forum; and I have seen monst on this forum too. The other three are newer and may not have found it yet.

[lindee]

Almost all of linded's TF <= 64 bits is on PIIIs. I have been doing this ever since we noticed the performance issue doing TF above 64. We just happen to have about 60 of them going.

[petrw1]

Originally Posted by lindee

We just happen to have about 60 of them going.

Wow....that's a lot of PIII's.

Thanks for replying.

[chalsall]

Originally Posted by petrw1

So what I am so humbly asking is that for those of you with 64 bit PCs that are assigning your own TF work below 64 bits is to consider taking exponents at 400M and above and leaving those lower to the 32 bit PCs.

OK. Please note that there are still just short of 15 million exponents currently below 64 bits. I would argue it is going to take quite some time before these are exhausted.

However, in the interests of cooperation, I have instructed my cluster to work in the 800M to 900M range, and will stay above 400M.

[petrw1]

Originally Posted by chalsall

OK. Please note that there are still just short of 15 million exponents currently below 64 bits. I would argue it is going to take quite some time before these are exhausted.

Doing a quick calc in my head tells me that based on the effort in the last few months that all exponents will be at 64 bits before year end.

But thanks...

[chalsall]

Originally Posted by petrw1

Doing a quick calc in my head tells me that based on the effort in the last few months that all exponents will be at 64 bits before year end.

In your opinion, is this necessarily a bad thing?

[petrw1]

Originally Posted by chalsall

In your opinion, is this necessarily a bad thing?

I'm really NOT trying to tell people what they should or shouldn't do; it up to them.

I am just suggesting that once there are no more <64 bit TF assignments left then many (probably hundreds of) 32 bit machines thruput will be cut in half and along with that their ability to contribute most effectively to this project. This includes Intel PIII, AMD Duron, etc.

My PIII and Duron both take 4 times as long to complete TF from 64 to 65 bits as from 63 to 64 for only double the credits. This observation has been confirmed several times by others on this forum recently. By the way these same 32 bit machines perform just as poorly at all other work types as they do at TF above 2^64.

On the other hand most (all?) 64 bit PC's effectiveness doing TF does not diminish whether to 58 bits or 68 bits.

So the more 64 Bit PC's that take TF at 2^64 and above then the longer the most useful life of the 32 bit machines will be....without impacting the effectiveness of the 64 bit PC's

IMHO...