Sierp. base 6 team sieve

[mdettweiler]

Originally Posted by KriZp

yea, I suppose I should take a larger chunk, anon's tiny starting ranges made me think a small range would be appropriate :) Also a small range this early in the sieve would shorten the interval between sr2data.txt updates, wich I suppose would speed the sieve up marginally. I will reserve a slightly larger range now, but split it in the sr2work.txt file so I get more factorsXXX.txt files and mail them as they get done.

125G-200G complete KriZp
reserving 210G-500G KriZp

Actually, at this point in the sieve, the marginal improvement from removing candidates from the sieve file is so marginal that it's often not worth the trouble. I'll still be releasing updates to the sieve file, but I'll probably only do so every 100G or so. But, of course, it certainly won't do any harm to submit factors more often.

Meanwhile, 200G-210G is complete.

[KriZp]

Originally Posted by Anonymous

Actually, at this point in the sieve, the marginal improvement from removing candidates from the sieve file is so marginal that it's often not worth the trouble. I'll still be releasing updates to the sieve file, but I'll probably only do so every 100G or so.

Ah, I thought it would be less marginal this early, since there are thousands of candidates removed just by sieving a few G. I know that removing ks is what really counts regarding speed. In any event, don't burn yourself out on the administrative tasks.

[gd_barnes]

Originally Posted by KriZp

yea, I suppose I should take a larger chunk, anon's tiny starting ranges made me think a small range would be appropriate :) Also a small range this early in the sieve would shorten the interval between sr2data.txt updates, wich I suppose would speed the sieve up marginally. I will reserve a slightly larger range now, but split it in the sr2work.txt file so I get more factorsXXX.txt files and mail them as they get done.

125G-200G complete KriZp
reserving 210G-500G KriZp

And I haven't even started on my range yet. lol It starts tomorrow morning.

At this rate, we'll easily be done sieving before the base has been tested up to n=100K!

[gd_barnes]

Originally Posted by gd_barnes

And I haven't even started on my range yet. lol It starts tomorrow morning.

At this rate, we'll easily be done sieving before the base has been tested up to n=100K!

My range started ~9 AM CDT this morning. ETA is 7 AM CDT Friday morning so this is going to zoom right along.

Therefore, I'd like to go ahead and reserve the range of P=500G-750G. If 75G took 1 day, then 250G should take ~3.5 days on the same Athlon DC, although should be a little faster since higher P-ranges sieve faster, even with the same # of k/n pairs remaining.

Anon, once you get the completed range from KriZp up to P=500G, can you remove the factors and re-post the sieved file? (I'm assuming he'll be done before I'm done with my 2nd reservation of P=500G-750G.) I'll then re-download it and should be able to get a reasonable estimate on the optimum sieve depth for breaking off n=100K-200K using the LLR time of an n=170K k/n pair.

Optionally since the LLR times are much longer here than powers-of-2 bases, we could break off n=100K-150K when the removal rate is about that of an n=135K k/n pair. It'd be somewhat more CPU-efficient in the long run but it'd be a little more upfront work here while sieving.


Gary

[KriZp]

I'm at p=335G atm, doing around 150G/day, so 210-500 should be done in perhaps 30 hours. sr2sieve reports ~1,6 Mp/sec, 27 sec/factor. I'll keep this core on this project untill it is decided that the sieve is deep enough.

[gd_barnes]

Originally Posted by KriZp

I'm at p=335G atm, doing around 150G/day, so 210-500 should be done in perhaps 30 hours. sr2sieve reports ~1,6 Mp/sec, 27 sec/factor. I'll keep this core on this project untill it is decided that the sieve is deep enough.

Very speedy siever. What kind of machine? My 1.6 Ghz Athlon DC laptop is only getting ~450P/sec on each core.

I'm running both cores on it just for my total speed to be a little over half your single-core speed. (lol) Of course I bought it before starting prime-searching so it wasn't intended for speed in the first place.


Gary

[KriZp]

Sounds like you are running 32bit :) I put 64bit linux on my 64bit processors back when sr2sieve came with a 64bit version, IIRC it more or less doubled in speed, very fun.

/proc/cpuinfo says
model name : AMD Athlon(tm) 64 Processor 3000+
cpu MHz : 2430.709
cache size : 512 KB

[mdettweiler]

Originally Posted by gd_barnes

My range started ~9 AM CDT this morning. ETA is 7 AM CDT Friday morning so this is going to zoom right along.

Therefore, I'd like to go ahead and reserve the range of P=500G-750G. If 75G took 1 day, then 250G should take ~3.5 days on the same Athlon DC, although should be a little faster since higher P-ranges sieve faster, even with the same # of k/n pairs remaining.

Anon, once you get the completed range from KriZp up to P=500G, can you remove the factors and re-post the sieved file? (I'm assuming he'll be done before I'm done with my 2nd reservation of P=500G-750G.) I'll then re-download it and should be able to get a reasonable estimate on the optimum sieve depth for breaking off n=100K-200K using the LLR time of an n=170K k/n pair.

Optionally since the LLR times are much longer here than powers-of-2 bases, we could break off n=100K-150K when the removal rate is about that of an n=135K k/n pair. It'd be somewhat more CPU-efficient in the long run but it'd be a little more upfront work here while sieving.


Gary

Okay, I'll be sure to post an updated sieve file when you finish your 70G-125G range.

As for when to break off n=100K-150K--I don't get it, isn't n=135K the standard 70% anyway?

[gd_barnes]

Originally Posted by Anonymous

Okay, I'll be sure to post an updated sieve file when you finish your 70G-125G range.

No, I suggested to post an updated file when KriZp finishes his 210G-500G range, which will be after my small reservation. I'll be done tomorrow morning with my file. He should be done later in the day. It makes more sense to wait a few hours to get more than twice the sieved range factors removed.

As for when to break off n=100K-150K--I don't get it, isn't n=135K the standard 70% anyway?

No, again. If we break off n=100K-200K, then n=170K is the 70% point of the range. If we break off n=100K-150K, THEN n=135K is the 70% point. It's the 70% point of the range you break off.

Effectively, what you're looking for is the approximate average LLR time of only the range that you're breaking off as a reference for the optimum removal rate of the sieve. The 70% point approximately accomplishes that.

If we wanted to be lazy and sieve the entire n=100K-400K range all to exactly the same depth, then n=310K would be the 70% point and the most optimum. It would be very inefficient when LLRing at the higher n-levels to do it that way but as far as the optimal depth if that depth was the SAME for the entire range, then that would be as good as we could do. I've done that for some smaller efforts on base 2, perhaps for n=25K-200K or something when I didn't want to mess with optimizing the sieving and LLRing. The time saved wasn't worth my extra hassle.

Edit: I just looked closer at the timing of of KriZp's post. It would be closer to 1 AM EDT late night tomorrow when he is done. Regardless, go ahead and wait and remove the factors then. I'll then temporarily stop my P=500G-750G range, copy in the file with less candidates remaining in it on Saturday during the day, and restart it.


Gary

[mdettweiler]

Originally Posted by gd_barnes

No, I suggested to post an updated file when KriZp finishes his 210G-500G range, which will be after my small reservation. I'll be done tomorrow morning with my file. He should be done later in the day. It makes more sense to wait a few hours to get more than twice the sieved range factors removed.


No, again. If we break off n=100K-200K, then n=170K is the 70% point of the range. If we break off n=100K-150K, THEN n=135K is the 70% point. It's the 70% point of the range you break off.

Effectively, what you're looking for is the approximate average LLR time of only the range that you're breaking off as a reference for the optimum removal rate of the sieve. The 70% point approximately accomplishes that.

If we wanted to be lazy and sieve the entire n=100K-400K range all to exactly the same depth, then n=310K would be the 70% point and the most optimum. It would be very inefficient when LLRing at the higher n-levels to do it that way but as far as the optimal depth if that depth was the SAME for the entire range, then that would be as good as we could do. I've done that for some smaller efforts on base 2, perhaps for n=25K-200K or something when I didn't want to mess with optimizing the sieving and LLRing. The time saved wasn't worth my extra hassle.

Ah, I see what you mean now. I didn't know you were comparing it with the possibility of breaking off 100K-200K.

Edit: I just looked closer at the timing of of KriZp's post. It would be closer to 1 AM EDT late night tomorrow when he is done. Regardless, go ahead and wait and remove the factors then. I'll then temporarily stop my P=500G-750G range, copy in the file with less candidates remaining in it on Saturday during the day, and restart it.

Okay, sounds good.

[KriZp]

ETA of my range is now 15 hours.