Not only does the challenge start at an unusual time, but the US just switched to Daylight Savings Time today. Please make sure you've got the starting time correct!
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Interesting. My i3-6100 takes about 2.75 hours.

Edit: Stock CPU clocks, but DDR4-2666.
____________

Luckly I'm near the edge of speed, running the latest gen i5 6600k, OCed to 4.3ghz and with 3200mhz, dual channel, Rank #2 RAM. As a result, tasks are running just under 2h each. Yay! I'm better than the estimate....

... until I get to trusty 2ghz dual core LGA 775 and 1st Gen 2.4ghz APU, both of which estimate around 18h. Mergh. Oh well.


I'd up the challenge estimates to "3h on fast PCs" and "10h+ on slower machines".

i7-4790k 13 hrs.

AMD Interlagos 12 core = 24hrs!! Have two!!! 24 cores[/img]

Interesting. My i3-6100 takes about 2.75 hours.

Edit: Stock CPU clocks, but DDR4-2666.

How stable is your DDR4 @ 2666? I have read reports where anything at 2400 or higher is unstable.

The i3 6100 with LGA 1151 socket motherboard can be picked up for about $139 as a bundle and which is a phenomenal deal. European pricing is about twice that amount. :-(

my i7-6700k running at 4.4ghz is taking 2.1-2.3 hours
my i5-6500 running at 3.3 is taking 2.6-3.07
my q9550s are taking 11.3

HT off, 3 cores loaded.

my mac pro xeons haven't finished a unit yet :(

The Micro Center has the I3-6100 with a Gigabyte GA-H110M-A LGA 1151 mATX mobo for the price I mention above. You are right in that it only supports up to 2133 but it is a great deal for the price. Unfortunately they do not ship overseas or otherwise I would have picked up two such bundles for PG.

How stable is your DDR4 @ 2666? I have read reports where anything at 2400 or higher is unstable.

How stable is your DDR4 @ 2666? I have read reports where anything at 2400 or higher is unstable.

Perfectly stable, at 1.2V. It's on a Z170 board, so that helps. I plan to upgrade to an i7 someday.

Several of the posts in this thread note average times for a CPU type, but these are all going to be slightly less informative than they could be due to the fact that all SR5 units are not created equally. More specifically, the FFT size varies depending on the workunit, and these variations are both significant and variable by CPU.

For example, consider the following CPU:
Intel(R) Xeon(R) CPU E5-1650 v3 @ 3.50GHz (This is stock clocked with 32GB of stock DDR4 running with HT on, but only on 50% cores--i.e., 6 units at a time). For those who are less familiar with Xeon CPUs, this one is very similar to the Haswell-E i7 series. So far, this box has completed SR5 work with four different FFT sizes with the following average times (rounded to the nearest 5 minutes) by FFT...

FFT = 400k Avg. time = 3 hours 45 minutes
FFT = 480k Avg. time = 4 hours 35 minutes
FFT = 512k Avg. time = 4 hours 50 minutes
FFT = 576k Avg. time = 5 hours 15 minutes

As you can see, these times vary quite a bit. More importantly, this variation will itself vary by CPU type given substantial differences in a) CPU cache size and speed and (potentially) b) RAM speed.

But why would a i7-4790k take over 10hrs? Is there a better processor like the i3-6100 for this? I assumed that the i7 were above a i3. Even the e5-2620 v2 takes 10 hrs, with 32gb of memory.

I don't get it.

I thought that i7 were better cpu than i3 or i5. Oh well.

But why would a i7-4790k take over 10hrs? Is there a better processor like the i3-6100 for this? I assumed that the i7 were above a i3. Even the e5-2620 v2 takes 10 hrs, with 32gb of memory.

I don't get it.

Ok, here's the abbreviated guide to Intel CPU performance for bigger LLR tasks, which includes SR5. This does not include smaller tasks, where you just get as much CPU clock as you can and you can't go far wrong. It also doesn't include AMD CPUs, since they took a different design decision early on which means they're highly inefficient at these tasks.

In general, a higher CPU clock can get more work done up to a point where you're memory bandwidth limited. More cores also means more memory bandwidth demand. You have a balancing act here. Too high a CPU demand means you're memory limited. Too much memory bandwidth, and you're CPU limited. In many cases the CPU will be limited by ram, and it pays to optimise memory further.

CPU clock: faster is generally better, but you will get diminishing returns if there isn't enough memory bandwidth to feed it. This is particularly a problem with the quad core higher clocked CPUs which only have dual channel ram.

CPU cache: doesn't make too much of a difference. If the task is too big to fit in the CPU, you'll be limited by the ram performance. There seems to be a slight gain if tasks are only a little bigger than the cache, so having a bigger cache may help in those situations. I wouldn't worry about picking an i7 over an i5 to get more cache. Note if you run multiple cores, typically you get 1.5MB/core on i5, 2.0MB/core for mainstream i7 CPUs. Enthusiast and server equivalents can go higher. Exception are some Broadwell CPUs with 128MB of L4 cache, which makes memory performance irrelevant as that's big enough to run anything current.

Hyper-threading: generally speaking this doesn't provide any benefit to LLR. Disable it and you get the full performance running all cores. If you want to leave it enabled for other uses, run only as many tasks as real cores. Advanced tip: you get even better performance if you manually set affinity so that boinc tasks can never occupy the HT cores e.g. use only odd or even numbered cores under Windows. This seems to eliminate some inefficiency compared to not doing this.

Instruction set: Look up the CPU on intel's database. Look at the "Instruction Set Extension" line. If it says AVX, that's good. If it says AVX 2.0, even better! If neither, not so good. AVX was introduced with Sandy Bridge and gave about a doubling of performance relative to before. AVX 2.0 was introduced with Haswell. Strictly speaking it isn't interesting, but introduced with it was FMA which gives about 50% more again. So if you get offered a 1st generation i7 cheap, decline, as it will suck at LLR. Also beware the Pentiums and the likes have these features removed.

Ram channels: most systems support dual channel. If you care at all about performance, you want to run dual channel. It is as simple as 2 channels = 2x the bandwidth of 1 channel. Pre-built systems might only run single channel to allow more upgrade capability, but this will be a significant performance hit. If you have a system that supports quad channel ram, I don't think you need me to tell you to fill it up!

Ram speed: generally faster is better. Beware that really high end ram e.g. DDR4 above 3200 or so will start getting disproportionately expensive and compatibility starts becoming a big problem. Certainly aim for 3000 or so since it doesn't cost much more over standard 2133.

Ram rank: this isn't something that is commonly discussed and I only found out after extensive testing to figure out why two similar systems ran with a massive performance difference. I don't fully understand the technical reasons myself, but having more than 1 rank per channel allows for more bandwidth, which as you may have guessed by now, is generally a good thing. The tricky part is this isn't often specified, so you might have to research reviews to see if it is mentioned. 4GB DDR4 modules will likely be single rank. 8GB is a mixture depending. Higher capacity is more likely to be dual rank, but I haven't tested it. Running two single rank modules per channel works like a single dual rank. I also saw this effect in DDR3 but I did less testing on that.

Ram timing: generally tighter is better, but this is a minor effect. Don't worry about this unless you've done everything else first.

So, given all the above, you can't simply say that an i7 will perform better than an i5. My slowest system is actually based on an i5-4570S, running 4 cores at 3.2 GHz with 1600 dual channel ram. It is really ram bandwidth starved. I have two i3-4150T dual core 3.0 GHz systems with dual channel 1600 ram, which run units much faster. So overall, the two i3's work much faster than the single i5.

In another thread I proposed some pure cruncher builds. The latest i3-6100 actually came out highly competitive against i5 and i7 processors in both bang for buck, and performance per watt measures. Basically you have two fast cores and two channels of ram. Since overclocking isn't on the cards you can make do with a basic mobo and low cost ram. Win all around. More in depth investigation into what affects performance is also posted in other threads in this forum.

Did I say this was the short version? :)

The i3 6100 with LGA 1151 socket motherboard can be picked up for about $139 as a bundle and which is a phenomenal deal. European pricing is about twice that amount. :-(

Ram rank: this isn't something that is commonly discussed and I only found out after extensive testing to figure out why two similar systems ran with a massive performance difference. I don't fully understand the technical reasons myself, but having more than 1 rank per channel allows for more bandwidth, which as you may have guessed by now, is generally a good thing. The tricky part is this isn't often specified, so you might have to research reviews to see if it is mentioned. 4GB DDR4 modules will likely be single rank. 8GB is a mixture depending. Higher capacity is more likely to be dual rank, but I haven't tested it. Running two single rank modules per channel works like a single dual rank. I also saw this effect in DDR3 but I did less testing on that.

Wouldn't be on overclockers.com would it?

4GB DDR4 being single rank I can well believe. Modern DDR3 is likely but I'm not 100% there. I have two 2x4GB DDR3 1600 kits, one bought as recently as early last year, which are both dual rank. Since then I got the cheap Kingston DDR3 1866 2x4GB kits and they're single rank.

I really wish they did 2GB DDR4 modules as I really don't want to get 16GB of ram to perk up a cruncher!

Wouldn't be on overclockers.com would it?

4GB DDR4 being single rank I can well believe. Modern DDR3 is likely but I'm not 100% there. I have two 2x4GB DDR3 1600 kits, one bought as recently as early last year, which are both dual rank. Since then I got the cheap Kingston DDR3 1866 2x4GB kits and they're single rank.

I really wish they did 2GB DDR4 modules as I really don't want to get 16GB of ram to perk up a cruncher!

In short, yes. I included that consideration in my low cost cruncher thread. I don't strictly know how it scales, but I was assuming single rank ram is -20% the performance of dual rank, all else being equal. That's what I saw on the overclocked Skylake CPUs and it wasn't much different on limited Haswell testing.

Been looking at this setup:

i3-6100
Asrock z170a-x1/3.1
Kingston Hyper X Fury DDR4-2666 dual rank. 8gb (x2)

The Asrock accepts up to 3000mhz DDR4.

All for $302. I think its a good deal.

Any other recommendations as for a better setup?

I think I'll sell the dual AMD setup. Going with Intel.

Speaking of forthcoming I7's, Intel has delayed the release of Cannonlake not only into next year but also the second half of next year.

Has anyone heard what new or extra features Cannonlake may offer and particularly as it pertains to PG?

http://www.superbiiz.com/desc.php?name=HX426FB_8

The above i3 proposed system, ram might be worth having assuming it doesn't actually cost much more than bargain basement stuff. I don't recall much price difference between 2133 and the slightly higher grades but I haven't look too carefully. As for the mobo, again overclocking features isn't really needed, but if the mobo has other features you want, fair enough. Asrock have been perusing bus overclocking of otherwise non-overclocking CPUs, but I'm not familiar enough with that supports what in that department.

I think we could go back a step here. Is this a pure cruncher, or will be be a general system? If general system, there is still sense going higher end.

On AMD Zen, that's another discussion totally... hope we're not already putting too much hardware into the challenge thread!

The above i3 proposed system, ram might be worth having assuming it doesn't actually cost much more than bargain basement stuff. I don't recall much price difference between 2133 and the slightly higher grades but I haven't look too carefully. As for the mobo, again overclocking features isn't really needed, but if the mobo has other features you want, fair enough. Asrock have been perusing bus overclocking of otherwise non-overclocking CPUs, but I'm not familiar enough with that supports what in that department.

I think we could go back a step here. Is this a pure cruncher, or will be be a general system? If general system, there is still sense going higher end.

On AMD Zen, that's another discussion totally... hope we're not already putting too much hardware into the challenge thread!

Those of you who have been around for a while know that our challenges, while fun, tend to be exceptionally effective as prime-repellents. We almost never find any large primes during challenges.

Not this time.

Yesterday Aus_Karlos found the world record base 5 mega prime 92158*5^2145024+1. This prime is 1,499,313 digits long.

This eliminates k=92158 from the base 5 Sierpinski problem, leaving 33 k's remaining. The base 5 Riesel problem has 77 k's remaining.

Congratulations!!!

____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Those of you who have been around for a while know that our challenges, while fun, tend to be exceptionally effective as prime-repellents. We almost never find any large primes during challenges.

Not this time.

Yesterday Aus_Karlos found the world record base 5 mega prime 92158*5^2145024+1. This prime is 1,499,313 digits long.

This eliminates k=92158 from the base 5 Sierpinski problem, leaving 33 k's remaining. The base 5 Riesel problem has 77 k's remaining.

Congratulations!!!

Okay, so, THIS GUY, it's like, 12th task, "imma find a world record prime that helps prove a conjecture".

Why.....

I'd rather have Scott Find it. At least it would make sense, given his army of PCs.

But no, let's do it with a 1st gen i7 that doesn't even have AVX.

#mad.


Still, congrats on the finding!

Okay, so, THIS GUY, it's like, 12th task, "imma find a world record prime that helps prove a conjecture".

Why.....

I'd rather have Scott Find it. At least it would make sense, given his army of PCs.

But no, let's do it with a 1st gen i7 that doesn't even have AVX.

#mad.


Still, congrats on the finding!

Did I ever tell you the story of how I found that big GFN prime in my signature? There's a reason I call it "my lucky number". :)

But no, let's do it with a 1st gen i7 that doesn't even have AVX.

If you're seeing a lot of inconclusive or invalid results, you have a problem that needs to be fixed.

There's a very high number of bad results coming back during this challenge:

Of the tasks that have been returned successfully:

12546 (38%) are pending validation. [12199 (37%) / 0 (0%) / 347 (1%)]
20205 (61%) have been successfully validated. [19635 (60%) / 0 (0%) / 570 (2%)]
50 (0%) were invalid. [50 (0%) / 0 (0%) / 0 (0%)]
120 (0%) are inconclusive. [118 (0%) / 0 (0%) / 2 (0%)]

And we are done! Great challenge everyone. Here's the final statistics:

Challenge: From Pi to Paddy
App: 19 (SR5-LLR)
(As of 2016-03-17 03:29:01 UTC)

65884 tasks have been sent out. [CPU/GPU/anonymous_platform: 64693 (98%) / 0 (0%) / 1191 (2%)]

Of those tasks that have been sent out:

6304 (10%) came back with some kind of an error. [6301 (10%) / 0 (0%) / 3 (0%)]
40480 (61%) have returned a successful result. [39359 (60%) / 0 (0%) / 1121 (2%)]
18952 (29%) are still in progress. [18890 (29%) / 0 (0%) / 62 (0%)]

Of the tasks that have been returned successfully:

12208 (30%) are pending validation. [11880 (29%) / 0 (0%) / 328 (1%)]
28067 (69%) have been successfully validated. [27276 (67%) / 0 (0%) / 791 (2%)]
92 (0%) were invalid. [92 (0%) / 0 (0%) / 0 (0%)]
113 (0%) are inconclusive. [111 (0%) / 0 (0%) / 2 (0%)]

The current leading edge (i.e., latest work unit for which work has actually been sent out to a host) is n=2182813. The leading edge was at n=2145542 at the beginning of the challenge. Since the challenge started, the leading edge has advanced 1.74% as much as it had prior to the challenge!

Of course, the most important statistic is "1", as in 1 prime found!

Now the cleanup begins. With short deadlines, the cleanup should go quickly and will probably be done in about two weeks.

Cleanup Status:

Mar-16: From Pi to Paddy: 12320 tasks outstanding; 12224 affecting individual (296) scoring positions; 9395 affecting team (72) scoring positions.

____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Cleanup Status:

Mar-16: From Pi to Paddy: 12320 tasks outstanding; 12224 affecting individual (296) scoring positions; 9395 affecting team (72) scoring positions.
Mar-17: From Pi to Paddy: 10374 tasks outstanding; 9485 affecting individual (291) scoring positions; 7037 affecting team (69) scoring positions.
Mar-18: From Pi to Paddy: 6228 tasks outstanding; 5391 affecting individual (279) scoring positions; 3368 affecting team (58) scoring positions.
Mar-19: From Pi to Paddy: 3385 tasks outstanding; 2534 affecting individual (262) scoring positions; 1547 affecting team (34) scoring positions.
Mat-20: From Pi to Paddy: 1551 tasks outstanding; 945 affecting individual (212) scoring positions; 116 affecting team (14) scoring positions.
Mar-21: From Pi to Paddy: 759 tasks outstanding; 386 affecting individual (143) scoring positions; 17 affecting team (6) scoring positions.
Mar-25: From Pi to Paddy: 44 tasks outstanding; 15 affecting individual (14) scoring positions; 0 affecting team (0) scoring positions.
Mar-27: From Pi to Paddy: 16 tasks outstanding; 4 affecting individual (4) scoring positions; 0 affecting team (0) scoring positions.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Cleanup Status:

Mar-16: From Pi to Paddy: 12320 tasks outstanding; 12224 affecting individual (296) scoring positions; 9395 affecting team (72) scoring positions.
Mar-17: From Pi to Paddy: 10374 tasks outstanding; 9485 affecting individual (291) scoring positions; 7037 affecting team (69) scoring positions.
Mar-18: From Pi to Paddy: 6228 tasks outstanding; 5391 affecting individual (279) scoring positions; 3368 affecting team (58) scoring positions.
Mar-19: From Pi to Paddy: 3385 tasks outstanding; 2534 affecting individual (262) scoring positions; 1547 affecting team (34) scoring positions.
Mat-20: From Pi to Paddy: 1551 tasks outstanding; 945 affecting individual (212) scoring positions; 116 affecting team (14) scoring positions.
Mar-21: From Pi to Paddy: 759 tasks outstanding; 386 affecting individual (143) scoring positions; 17 affecting team (6) scoring positions.
Mar-25: From Pi to Paddy: 44 tasks outstanding; 15 affecting individual (14) scoring positions; 0 affecting team (0) scoring positions.
Mar-27: From Pi to Paddy: 14 tasks outstanding; 2 affecting individual (2) scoring positions; 0 affecting team (0) scoring positions.
Mar-28: From Pi to Paddy: 11 tasks outstanding; 2 affecting individual (2) scoring positions; 0 affecting team (0) scoring positions.
Mar-29: From Pi to Paddy: 7 tasks outstanding; 1 affecting individual (1) scoring positions; 0 affecting team (0) scoring positions.
Mar-30: From Pi to Paddy: 5 tasks outstanding; 1 affecting individual (1) scoring positions; 0 affecting team (0) scoring positions.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

Cleanup Status:

Mar-16: From Pi to Paddy: 12320 tasks outstanding; 12224 affecting individual (296) scoring positions; 9395 affecting team (72) scoring positions.
Mar-30: From Pi to Paddy: 5 tasks outstanding; 1 affecting individual (1) scoring positions; 0 affecting team (0) scoring positions.
Apr-1: From Pi to Paddy: 2 tasks outstanding; 1 affecting individual (1) scoring positions; 0 affecting team (0) scoring positions.
Apr-2: From Pi to Paddy: 2 tasks outstanding; 1 affecting individual (1) scoring positions; 0 affecting team (0) scoring positions.
____________
My lucky number is 75898⁵²⁴²⁸⁸+1

TheDawgz have heard back from Charley - he has been and still is swamped at work ("very looooong days") - he will deal with this as soon as possible - but, it will likely not happen until next week sometime.