Haswell Rig
 

[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16819116899"]Haswell i5-4670K 3.4GHz[/URL] CPU
[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16835181031"]CORSAIR H80i[/URL] for cooling
[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16820231589"]2x8GB DDR3 2400 G.SKILL Trident X Series[/URL] memory
[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16813157369"]ASRock Z87 Extreme4[/URL] motherboard

I'm looking to upgrade my i5-750 rig, primarily due to all of the higher speeds (and new instructions) I can get from a Haswell CPU. I'll be using the CPU for Prime95/LLR crunching, primarily. Maybe some ECM, QS, and NFS sieving and post-processing, too.
I'll overclock it, but not more than I feel is completely stable (4.0-4.2GHz, maybe? Most overclockers seem to go ~4.4GHz, but aren't running their CPUs at 100% all the time). I also have a GTX 560 running TF (~227 GHz-days/day at peak).
Here are the parts where I'd like advice:[LIST][*]Is the memory speed too low, just right, or excessive? I currently have 2x4GB DDR3-1600 and 2x2GB DDR3-1333, but I'm thinking even the 1600 is too slow to want to use with a Haswell.[*]Would a more expensive motherboard be better? What stats or chipset should I look for there?[*]Is there any major Next Big Thing™ in this price range coming up in the near future that I should wait for instead of upgrading now? I've heard [URL="http://www.mersenneforum.org/showthread.php?t=19214"]a bit[/URL] of the 8-core Intel with DDR4, but I'm expecting that as an "Extreme" CPU with the brand-new DDR4, it'll be way outside of my target price range.[*]Is the liquid cooling a waste of money for my purposes, or is it useful?[*]Does anyone have benchmarks for a similar Haswell system on small FFTs, with all 4 cores used? Can I expect that it will be a similar difference to the [URL="http://www.mersenne.ca/throughput.php?cpu1=Intel%28R%29+Core%28TM%29+i5-4430+CPU+%40+3.00GHz%7C256%7C6144&mhz1=3000&cpu2=Intel%28R%29+Core%28TM%29+i5+CPU+750+%40+2.67GHz%7C256%7C8192&mhz2=2700"]FFTs listed here[/URL]? Currently, I get about 1.16 ms per iteration at 84K FFT in CLLR 3.8.9, with all 4 cores running (with ~1.32 million iterations per number, that's 225/day theoretically, actually closer to [URL="http://www.noprimeleftbehind.net/stats/index.php?content=port&server=GB&port=9000"]215-220[/URL] since I use the machine).[*]My current PSU is 500W, and has no apparent problems powering my computer. Can I expect that this is sufficient for and compatible with the new rig? Besides the aforementioned tech, it'll power 1 SSD (new), 1 HDD (existing), and 1 DVD or Blu-ray drive (SATA to replace old IDE DVD drive).[*]Anything stupid I'm missing?[/LIST]

Nope, looks quite good.

I have the same CPU and RAM(G.Skill) and still heavily memory bottlenecked.

EDIT: Also, that i5 4430 is mine, with [I]slow [/I]dual 1600 memory. [URL="http://mersenneforum.org/showpost.php?p=365476&postcount=622"]4670k benchmark[/URL] with dual 2400

Here are benchmarks for a [URL="http://mersenneforum.org/showpost.php?p=363660&postcount=608"]4770k 4.3GHz DDR3-2133[/URL] that may be of interest I got a H100i cooler for it --reasonably quiet and effective. It looks like the 2400 memory wouldn't be too much.

I got a Seasonic S12G-450 PSU for it, and it is running just fine. I'd intended to use a SSD, but ended up using a WD Red drive I had lying around instead.

[QUOTE=kracker;372081]I have the same CPU and RAM(G.Skill) and still heavily memory bottlenecked.[/QUOTE]

Just putting this out there...

I have found, with a great deal of experimentation, that bringing more (real) cores of a single CPU into a LL test can reduce the memory bandwidth issues and improve performance (thanks to the caches), so long as the affinity is done correctly.

I more than doubled my LL throughput by assigning a single LL to multiple (real) cores of a single CPU, thus lessening the memory bandwidth requirements. (This did, however, require I run two instances of mprime in a dual CPU environment -- mprime was unable to manage the situation appropriately itself.)

The H80i is going to be very good for cooling. It performs similarly to a custom solution at a much lesser price (and a much lesser level of fun involved in assembly :razz:)

The H80i fans are garbage. Both of mine were obscenely loud. Buy a pair of [url]http://www.newegg.ca/Product/Product.aspx?Item=N82E16835103061[/url]. You can even get some that light up!

2400 MHz RAM is good. It's gotten more expensive since the Hynix fire and hasn't gone back down, and a lot of skeptics think that memory will never be that cheap ever again. On the other hand, they're wrong. Still, if you need the memory now, it's worth paying an extra $50 than to wait for a few months and use some other kit in the meantime (which most people would have to pay for anyway)

i5-4670k is a must. Don't get it confused with i5-4570k. The 4670k is a 4570k that's binned higher because the batch had a better performing sample. However, there isn't a 4670k so good they can bin it as something else so a 4670k has a much better chance of being a good performing chip.

[QUOTE=TheMawn;372087]The H80i is going to be very good for cooling. It performs similarly to a custom solution at a much lesser price (and a much lesser level of fun involved in assembly :razz:)

The H80i fans are garbage. Both of mine were obscenely loud. Buy a pair of [url]http://www.newegg.ca/Product/Product.aspx?Item=N82E16835103061[/url]. You can even get some that light up!

2400 MHz RAM is good. It's gotten more expensive since the Hynix fire and hasn't gone back down, and a lot of skeptics think that memory will never be that cheap ever again. On the other hand, they're wrong. Still, if you need the memory now, it's worth paying an extra $50 than to wait for a few months and use some other kit in the meantime (which most people would have to pay for anyway)

i5-4670k is a must. Don't get it confused with i5-4570k. The 4670k is a 4570k that's binned higher because the batch had a better performing sample. However, there isn't a 4670k so good they can bin it as something else so a 4670k has a much better chance of being a good performing chip.[/QUOTE]

There is no 4570k.

[QUOTE=kracker;372090]There is no 4570k.[/QUOTE]

Perhaps it's just me, but this seems a bit like the last scene from William Shakespeare's Hamlet....

[QUOTE=kracker;372090]There is no 4570k.[/QUOTE]

Correct. I saw threads with people talking about it, but I guess they were equally misled.

There is a 4570, 4570S, but no 4570K.

[QUOTE=TheMawn;372087]
i5-4670k is a must. Don't get it confused with i5-4570k. The 4670k is a 4570k that's binned higher because the batch had a better performing sample. However, there isn't a 4670k so good they can bin it as something else so a 4670k has a much better chance of being a good performing chip.[/QUOTE]

[QUOTE=TheMawn;372128]Correct. I saw threads with people talking about it, but I guess they were equally misled.

There is a 4570, 4570S, but no 4570K.[/QUOTE]

:unsure:

[url]http://ark.intel.com/products/family/75024/4th-Generation-Intel-Core-i5-Processors#@All[/url]

[QUOTE=TheMawn;372128]Correct. I saw threads with people talking about it, but I guess they were equally misled.[/QUOTE]

Misling people is frowned upon around here, but as in this case it appears to have been unintentional, we'll let it slide.

[QUOTE=kracker;372129]:unsure:[/QUOTE]

The post you're confused about is me figuring out I was wrong after you pointed it out. I don't know what your uncertainty is about...

[QUOTE=ewmayer;372144]Misling people is frowned upon around here, but as in this case it appears to have been unintentional, we'll let it slide.[/QUOTE]

What I meant was I've seen threads in other forums with people talking about the i5-4570k, and I drew the conclusion is that the 4570k was a thing, when in fact they were wrong as well.

[QUOTE=TheMawn;372157]The post you're confused about is me figuring out I was wrong after you pointed it out. I don't know what your uncertainty is about...



What I meant was I've seen threads in other forums with people talking about the i5-4570k, and I drew the conclusion is that the 4570k was a thing, when in fact they were wrong as well.[/QUOTE]

Typos happen, too. We've seen a complaint from Xyzzy that GPU nomenclature is too confusing. I suspect that when you write about such stuff all the time, the alphanumeric soup sloshing around one's brain can generate some brain methane, aka brain farts.

Lol brain methane

Brain methane is mostly harmless.
It is brain methanethiol that is dangerous (to others).

I think the problem is the 25xx and 35xx were both i5 and 26xx, 27xx and 37xx were i7. Now they make 46xx an i5 and that confuses everyone.

Well, there were a few on-topic replies. :rolleyes:
[QUOTE=kracker;372081]Nope, looks quite good.

I have the same CPU and RAM(G.Skill) and still heavily memory bottlenecked.

EDIT: Also, that i5 4430 is mine, with [I]slow [/I]dual 1600 memory. [URL="http://mersenneforum.org/showpost.php?p=365476&postcount=622"]4670k benchmark[/URL] with dual 2400[/QUOTE]
Ok, 2400 sounds good then. :smile:

[QUOTE=TheMawn;372087]The H80i is going to be very good for cooling. It performs similarly to a custom solution at a much lesser price (and a much lesser level of fun involved in assembly :razz:)

The H80i fans are garbage. Both of mine were obscenely loud. Buy a pair of [url]http://www.newegg.ca/Product/Product.aspx?Item=N82E16835103061[/url]. You can even get some that light up![/QUOTE]
Thanks for the suggestion, I wasn't even aware you could change out the fans.

[QUOTE=TheMawn;372087]2400 MHz RAM is good. It's gotten more expensive since the Hynix fire and hasn't gone back down, and a lot of skeptics think that memory will never be that cheap ever again. On the other hand, they're wrong. Still, if you need the memory now, it's worth paying an extra $50 than to wait for a few months and use some other kit in the meantime (which most people would have to pay for anyway)[/QUOTE]
Eh, when you're upgrading the better portion of the whole computer (I have other upgrades that I didn't list here because they are simpler choices), $50 to make sure the RAM doesn't bottleneck me (much) isn't a big deal.

[QUOTE=chalsall;372084]Just putting this out there...

I have found, with a great deal of experimentation, that bringing more (real) cores of a single CPU into a LL test can reduce the memory bandwidth issues and improve performance (thanks to the caches), so long as the affinity is done correctly.

I more than doubled my LL throughput by assigning a single LL to multiple (real) cores of a single CPU, thus lessening the memory bandwidth requirements. (This did, however, require I run two instances of mprime in a dual CPU environment -- mprime was unable to manage the situation appropriately itself.)[/QUOTE]
I'll be sure to set it up for best performance. The benchmarks in the newest versions of Prime95 seem to be very good at trying different things to tell you what'd be best. Takes much longer than the old days where you just had one test to run, but it's much more useful.

[QUOTE=TheMawn;372157]What I meant was I've seen threads in other forums with people talking about the i5-4570k, and I drew the conclusion is that the 4570k was a thing, when in fact they were wrong as well.[/QUOTE]

I think I know you well enough from your posts around here to be sure you would never misle anyone deliberately ... was just making a mod-centric comment on forum etiquette in that regard.

No man is a mis-land, and all that.

[QUOTE=Mini-Geek;372078][URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16819116899"]Haswell i5-4670K 3.4GHz[/URL] CPU
[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16835181031"]CORSAIR H80i[/URL] for cooling
[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16820231589"]2x8GB DDR3 2400 G.SKILL Trident X Series[/URL] memory
[URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16813157369"]ASRock Z87 Extreme4[/URL] motherboard[/QUOTE]

Due to a last-minute price hike on the H80i (+$15 overnight, on both Newegg and Amazon! :yucky: they must've colluded just for me...), I went with the [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16835181058"]H75[/URL], which is largely the same but without Corsair Link (to monitor/change things via USB control). Also, [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16820231672"]G.SKILL's Ripjaws[/URL] series was significantly cheaper ($143 after 10% off vs $189) with nearly the same stats, so I went with that.

I also got a pair of the fans that TheMawn suggested. [URL="http://www.anandtech.com/show/6177/choosing-the-best-120mm-radiator-fan-testing-eight-fans-with-corsairs-h80/5"]One review[/URL] I read showed they had terrible cooling performance, but I think that must've been due to faulty fans or installation, since others have reported [URL="http://www.vortez.net/articles_pages/120mm_and_140mm_fan_comparison,17.html"]better results[/URL].

Thanks to all who read and/or answered this. Now to wait for my parts and hope I don't need to return any of them...

[QUOTE=Mini-Geek;372277]Due to a last-minute price hike on the H80i (+$15 overnight, on both Newegg and Amazon! :yucky: they must've colluded just for me...), I went with the [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16835181058"]H75[/URL], which is largely the same but without Corsair Link (to monitor/change things via USB control). Also, [URL="http://www.newegg.com/Product/Product.aspx?Item=N82E16820231672"]G.SKILL's Ripjaws[/URL] series was significantly cheaper ($143 after 10% off vs $189) with nearly the same stats, so I went with that.

I also got a pair of the fans that TheMawn suggested. [URL="http://www.anandtech.com/show/6177/choosing-the-best-120mm-radiator-fan-testing-eight-fans-with-corsairs-h80/5"]One review[/URL] I read showed they had terrible cooling performance, but I think that must've been due to faulty fans or installation, since others have reported [URL="http://www.vortez.net/articles_pages/120mm_and_140mm_fan_comparison,17.html"]better results[/URL].

Thanks to all who read and/or answered this. Now to wait for my parts and hope I don't need to return any of them...[/QUOTE]

Good luck! :smile:

Looks good! The H75 has a thinner radiator, which may end up playing in your favour depending on your case layout. A thick radiator plus two fans takes up a surprising amount of space.

The good news is that it will take a lot less pressure to get the air to flow through the thinner radiator so you will get more air flow for the same noise level, at the expense of some cooling performance. More flow over less area vs less flow over more area.

My experience with the Sickle Flow fans is pretty good. I'm impressed with the flow through my own thick radiators. The "terrible" cooling performance is probably on the user's end.

I have actually seen a self-proclaimed 5/5 super expert complaining about a fan being crap because two of them overheated his CPU within seconds. He said he knew more about fans than anyone else because he had tested dozens of different pairs of fans on this heatsink. As it turned out, the know-it-all couldn't tell which direction the air flow went on a fan and because this one wasn't labelled, he ended up having the fans pointed at each other.

(Hint: fold your hand to make a cup shape like the fan blades; air flows in the direction outward from your palm)


I personally have eight Sickle Flows. Two of them have a whining sound above 8 volts or so but the noise from the other six is purely air whooshing.

I got everything running over the weekend, and it's looking very good so far! The CPU is running at 4GHz (at 4.2GHz I got a BSOD after a few minutes of Prime95's stress test), using the motherboard's built-in OC settings, and the RAM is at 2400Mhz with the default XMP profile.
[CODE]Compare your results to other computers at http://www.mersenne.org/report_benchmarks
Intel(R) Core(TM) i5-4670K CPU @ 3.40GHz
CPU speed: 3988.43 MHz, 4 cores
CPU features: Prefetch, SSE, SSE2, SSE4, AVX, AVX2, FMA
L1 cache size: 32 KB
L2 cache size: 256 KB, L3 cache size: 6 MB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
TLBS: 64
Prime95 64-bit version 28.5, RdtscTiming=1
Best time for 1024K FFT length: 3.823 ms., avg: 4.109 ms.
Best time for 1280K FFT length: 4.808 ms., avg: 5.130 ms.
Best time for 1536K FFT length: 5.860 ms., avg: 5.946 ms.
Best time for 1792K FFT length: 6.991 ms., avg: 8.769 ms.
Best time for 2048K FFT length: 7.901 ms., avg: 8.053 ms.
Best time for 2560K FFT length: 10.125 ms., avg: 10.228 ms.
Best time for 3072K FFT length: 12.196 ms., avg: 12.271 ms.
Best time for 3584K FFT length: 14.447 ms., avg: 14.589 ms.
Best time for 4096K FFT length: 16.517 ms., avg: 16.658 ms.
Best time for 5120K FFT length: 20.992 ms., avg: 21.127 ms.
Best time for 6144K FFT length: 25.359 ms., avg: 25.502 ms.
Best time for 7168K FFT length: 30.196 ms., avg: 30.334 ms.
Best time for 8192K FFT length: 34.975 ms., avg: 35.111 ms.[/CODE]
With my 84K FFT LLR work, the speed dropped from 1.16 ms (225/day) to 0.27 ms (965/day). Impressive! :smile:

At 4GHz, the cooler keeps the CPU at about 80C (ambient is 25C) on 84K FFTs or about 70C on DC-sized (2M) FFTs.

[QUOTE=Mini-Geek;372657]I got everything running over the weekend, and it's looking very good so far! The CPU is running at 4GHz (at 4.2GHz I got a BSOD after a few minutes of Prime95's stress test), using the motherboard's built-in OC settings, and the RAM is at 2400Mhz with the default XMP profile.

At 4GHz, the cooler keeps the CPU at about 80C (ambient is 25C) on 84K FFTs or about 70C on DC-sized (2M) FFTs.[/QUOTE]

If you find the time, it might be of interest to you to play with the voltage a bit longer. My board has an overclock feature but it was way to aggressive with the voltage. If memory serves, it put me at 1.38 V and 4.6 GHz. Manually I was able to achieve 4.6 GHz at 1.28 V and 4.5 GHz at 1.20 V.

Take it down a few notches at a time and see where it stops being stable, and go a few notches back up from there.

Alternatively, leave everything where it is and enjoy your new rig!

[QUOTE=TheMawn;372666]If you find the time, it might be of interest to you to play with the voltage a bit longer. My board has an overclock feature but it was way to aggressive with the voltage. If memory serves, it put me at 1.38 V and 4.6 GHz. Manually I was able to achieve 4.6 GHz at 1.28 V and 4.5 GHz at 1.20 V.

Take it down a few notches at a time and see where it stops being stable, and go a few notches back up from there.

Alternatively, leave everything where it is and enjoy your new rig![/QUOTE]

Haswell has a thing called adaptive voltage, is a damn pain to have if you use FMA3... It just really takes a lot longer and a bit of creative thinking if you want it on.

Well, my Ivy Bridge has this offset voltage thing that I've gotten back into. It takes a base voltage and applies a +/- value to it under load. I believe it's a function of the multiplier in some regard.

I'll look up this adaptive voltage some other day, to see if it's the same. The offset voltage was enough of a curve to learn. Hopefully this one isn't worse.

What I'm reading here is encouraging and I'm due to upgrade (last time was 2006).

This is interesting: i7-4770K, 16 GB DDR3 - 2400, boart to be determined but will be a server type. Last boart was Asus P5MT, not much in bios settings but did last running prime 24/7 for nearly 8 years

Can I get away with stock cooling? I may have to buy the parts overseas and DHL them here. A heavy copper block cooler would add to the shipping (140 $ / kg). It is not hot here in Rwanda, daytime 25 / 27 degrees and 16 / 18 at night, uniform all year round. I may overclock modestly but no pushing things as seems to be done around here.

What I like to do is P-1 and ECM. Perhaps 16 GB ram is not enough - 32 sounds better. Any feedback on this particular is appreciated.

[QUOTE=PageFault;373687]What I'm reading here is encouraging and I'm due to upgrade (last time was 2006).

This is interesting: i7-4770K, 16 GB DDR3 - 2400, boart to be determined but will be a server type. Last boart was Asus P5MT, not much in bios settings but did last running prime 24/7 for nearly 8 years

Can I get away with stock cooling? I may have to buy the parts overseas and DHL them here. A heavy copper block cooler would add to the shipping (140 $ / kg). It is not hot here in Rwanda, daytime 25 / 27 degrees and 16 / 18 at night, uniform all year round. I may overclock modestly but no pushing things as seems to be done around here.

What I like to do is P-1 and ECM. Perhaps 16 GB ram is not enough - 32 sounds better. Any feedback on this particular is appreciated.[/QUOTE]

In my experience with a 4670K, stock clocks are at borderline throttle with the stock cooler. A Hyper 212 EVO is not bad, not sure on how heavy though. Get a 4670K, not the 4770K, performance is almost same.

[QUOTE=PageFault;373687]
This is interesting: i7-4770K, 16 GB DDR3

What I like to do is P-1 and ECM. Perhaps 16 GB ram is not enough - 32 sounds better. Any feedback on this particular is appreciated.[/QUOTE]

In my experience, ECM responds very well to hyperthreading, so ideally you'd like enough memory to run 8 ECM processes. I would start with 16GB, and buy another 16 in a year or two as B2 and ECM memory requirements rise (are you testing Mersenne numbers, or for other projects?). Testing enormous B2 bounds is fun, and does present the chance to find a near-record factor- but whether that option is worth another $200 on a machine is personal.

If you have 4GB now, you could run a single process with your desired ECM parameters, measure memory use, and see how many you could run in 16GB. Perhaps you'd run 4 large processes and 4 smaller ones (say, 4 P-1 tests and 4 ECM) once you have 16GB.

[QUOTE=PageFault;373687]What I like to do is P-1 and ECM. Perhaps 16 GB ram is not enough - 32 sounds better. Any feedback on this particular is appreciated.[/QUOTE]
An ECM with the following bounds:

ECM2=1,2,9195881,-1,50000,5000000,150

Takes up to 10177 MB of memory in stage 2 (I allowed 12000 MB, that's all it took). Stage 1 took about 50 minutes of CPU time, and stage 2 took around 25-30 minutes. If you have 8 ECM processes running, you'll have approximately 3 in stage 2 at once (at this speed), using up to ~30GB. So having 32 GB of RAM for ECM/P-1 doesn't sound completely wasteful.

However, I'm pretty sure we're well into diminishing returns, and the extra 16 GB would be a small benefit, not a large one. I'm having trouble finding it, but someone looked at the % chance to find a P-1 factor when assigning different amounts of memory. For ordinary P-1s (around the first-time wave), having over 1GB (or 500MB? something relatively small like that) is practically irrelevant. This isn't a perfect analog, but it does show that extra memory doesn't speed things up dramatically. So I really wouldn't worry about 16 vs 32GB.

Note that I did this test with just one core doing ECM, so it's probably not taxing memory bandwidth, another big factor. For ECM/P-1, I think I'd rather have 8GB of DDR3 2400 memory than 32GB of DDR3 1600 memory. (I could be wrong about that, but that's my guess)

Thanks for the feedback guys - I really appreciate it.

A bit more money on ram is nothing - I can do it, if justified. And this seems to be the consensus - 32 GB minimum. I hear about cooling, in this case I want the best - cost inconsequential (it must last another 8 years). Big copper block recommendations please ...

I'll post back about the boart. I looked at three, one supermicro, two asus, at the same price point.

I will always use the official client - please educamate me on ECM and extreme settings - these interest me, I'm after the bizarre factor possibilities.

If you upgrade that infrequently I might suggest waiting for a DDR4 system. Haswell-E should be available this year probably Q3. Should you require to upgrade your memory at a later point it will then be possible cheaply. I am stuck with a DDR2 system currently(with a Q6600). Originally had 2GB of memory and later upgraded to 4GB. When I first bought the pc DDR3 was out and I should have got a motherboard that supported it in hindsight. I could do with more but by the time I needed it DDR2 was rare and extremely expensive compared with DDR3. You look like you could end up in the same situation. You could also end up with 6-8 cores as part of this if you go for Haswell-E although the initial outlay will be more.

What is your current system?

Not sure exactly on what you mean by "big copper block", custom water cooling?
If you want max life out of your investment, I wouldn't recommend overclocking too much, thus you may not need any exotic cooling.

[QUOTE]I'm having trouble finding it, but someone looked at the % chance to find a P-1 factor when assigning different amounts of memory. For ordinary P-1s (around the first-time wave), having over 1GB (or 500MB? something relatively small like that) is practically irrelevant.[/QUOTE][url]http://www.mersenneforum.org/showpost.php?p=282335&postcount=10[/url]

[QUOTE=PageFault;373716]Thanks for the feedback guys - I really appreciate it.

A bit more money on ram is nothing - I can do it, if justified. And this seems to be the consensus - 32 GB minimum. I hear about cooling, in this case I want the best - cost inconsequential (it must last another 8 years). Big copper block recommendations please ...

I'll post back about the boart. I looked at three, one supermicro, two asus, at the same price point.

I will always use the official client - please educamate me on ECM and extreme settings - these interest me, I'm after the bizarre factor possibilities.[/QUOTE]

Hyper 212 is a good cooler and very inexpensive. It is not terribly heavy. The contact area is heat pipe direct. It will certainly surpass the stock cooler. At the high end there are Noctua, and various closed loop liquid coolers. The high end of Noctua gives many liquid cooler a run for the money.
[url]http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=50012454%2040000574&IsNodeId=1&Manufactory=12454&name=CPU%20Fans%20%26%20Heatsinks&SpeTabStoreType=0&isdeptsrh=1[/url]

Currently I am using my work box - it is i3, a store bought HP box. More than twice as fast as my old build (Pentium D).

What I mean by copper block is a large surface area air colled rig. I will never use water, as it adds additional layers of complexity, ones that are beyond the understanding of overclocking. Doubly so in Africa.

I have to do this, can't wait. I can always add additional boxes later.

Thanks,

PF

[QUOTE=kladner;373739]The high end of Noctua gives many liquid cooler a run for the money.
[/QUOTE]

+1. I ´ve been a happy Noctua user for more than 4 years, on a i5-750. Low noise, very good performance, running 24/7 for most of the time.

The Noctua NH-D15 looks good... not released yet though.. :sad:
[url]http://www.tweaktown.com/reviews/6313/noctua-nh-d15-cpu-cooler-review/index.html[/url]

Arctic cooling rate their coolers based on watts of heat. Does anyone know of any wider comparisons of coolers in a similar manner?
[url]http://www.arctic.ac/eu_en/products/cooling/cpu.html?cpu_cooling_capacity=111[/url]

[url]http://www.tomshardware.com/reviews/h100i-elc240-seidon-240m-lq320,3380-12.html[/url]

This page and the following page give an interesting comparison of several liquid coolers vs the Noctua NH-D14.

Thermal dissipation is always rated in watts (1 watt = 1 joule / sec).

Copper is the preferred metal due to its high thermal conductivity. Gold is better, but too expensive. Some day yes, as I have just discovered a coltan deposit and the company is very happy and my year end bonus is already 1 kg of good delivery metal. Let's see how much this may increase ... that would be an awesome project, a 9999 fine cooler block.

I'll look at the Noctua. Noise doesn't bother me - a constant background like a fan actually helps me to sleep (even the fop 38, remember those? I once had four of those, within a few feet of my sleeping ear!).

We shall discuss power. It is very unreliable here, cutting out several times per day, from a few seconds to several hours. My house is almost finished and I'll use a backup genset. These can be programmed to automatically kick in after a preset time of power loss - main feeds through the unit, and it will disconnect grid / start and connect local user. The resulting spikes are very hard on power supplies, so a UPS is essential. Recommendations?

I have looked at the Noctua and this is more than adequate - only concern is that it is too quiet. I'll DHL it in.

Thanks guys - next help me with the boart.

About ECM - how can I be more aggressive? Can I do more curves by adjusting the last argument in a worktodo line, the one ending in 3? What would be good B1 / B2 bounds, for the 9xxxxxx assignments the server is giving me?

[QUOTE=PageFault;373787]I have looked at the Noctua and this is more than adequate - only concern is that it is too quiet. I'll DHL it in.

Thanks guys - next help me with the boart.

About ECM - how can I be more aggressive? Can I do more curves by adjusting the last argument in a worktodo line, the one ending in 3? What would be good B1 / B2 bounds, for the 9xxxxxx assignments the server is giving me?[/QUOTE]

Yes, Noctua is known for being whisper quiet... :razz:
I don't know much about UPS though, sorry.

[QUOTE=PageFault;373787]

About ECM - how can I be more aggressive? Can I do more curves by adjusting the last argument in a worktodo line, the one ending in 3? What would be good B1 / B2 bounds, for the 9xxxxxx assignments the server is giving me?[/QUOTE]

Yes, you can adjust that parameter.
The "good" B1/B2 bounds are the ones provided by the server, as they take into account how much ECM has previously been done on the exponents.
To look for large factors, you can venture in the lowest exponents regions (say below 500K, where exponents have already been ECM-tried to larger values of B1/B2, so yielding larger factors, if you´re lucky enough to find one).
Check the "Results Queries" ->"ECM progress" page, and then "Exponents with no known factor" to see what I mean. The karger the value of B1, the larger the number of curves to run, and the more difficult to find factors. But if you are looking for large ones, that´s the way to go.
Good luck!

Thanks lycorn

ok I get part of it - choose low exponents and up the B1 / B2 bounds, in proportion to what I am seeing and then register the exponents. What about the last argument? Will the server automatically adjust this at registration, or do I have to do it (i.e. how)? I want to go extreme and have decided on 64 GB of ram. Who knows, perhaps I can find a monster factor in about a year or so.

My bush mansion will be ready by month end - details like plumbing and kitchen remain. Mountain top view is fine for me, Rwanda looks like a mini Switzerland. And that type of tobacco grows wild here ... as does coffee and all sorts of natural food. So this build is going live in early june. I am going to confirm specs here and order parts within a week or so. I check with the company's comp supplier and see what they can do and DHL the rest.

[QUOTE=PageFault;373824]I want to go extreme and have decided on 64 GB of ram. Who knows, perhaps I can find a monster factor in about a year or so..[/QUOTE]
A i7 4770k doesn't support that much ram! Max Memory Size for Haswell is 32GB according to Intel: [URL]http://ark.intel.com/products/75123[/URL]

IvyBridge-E (for instance an i7 4930k) does support up to 64GB: [URL]http://ark.intel.com/products/77780[/URL]

[QUOTE=Mini-Geek;373706]An ECM with the following bounds:

ECM2=1,2,9195881,-1,50000,5000000,150

Takes up to 10177 MB of memory in stage 2 (I allowed 12000 MB, that's all it took). Stage 1 took about 50 minutes of CPU time, and stage 2 took around 25-30 minutes. If you have 8 ECM processes running, you'll have approximately 3 in stage 2 at once (at this speed), using up to ~30GB. So having 32 GB of RAM for ECM/P-1 doesn't sound completely wasteful.

Note that I did this test with just one core doing ECM, so it's probably not taxing memory bandwidth, another big factor. For ECM/P-1, I think I'd rather have 8GB of DDR3 2400 memory than 32GB of DDR3 1600 memory. (I could be wrong about that, but that's my guess)[/QUOTE]

Mini-Geek-
ECM bandwidth is very gentle- I use GMP-ECM as hyper-thread tasks to fill the i7 when LLR or NFS matrix-solving is nearly filling the CPU and bandwidth on the first four threads, and get "free" work because the balance of ECM low-bandwidth and LLR high-need works out. Memory speed shouldn't matter here, provided one chooses 1600 or up.

PF-
Having less memory leads ECM (not P-1!) to use slightly slower settings, rather than overlooking factors. Stage 2 can be broken into pieces to fit into whatever memory is available (within reason). This causes a small hit to stage 2 time only- stage 1 is not memory-intensive, and the two stages do not effect each other's times. So, you don't need 64GB (or even 32!) to chase massive factors, though extra memory may get you a small speedup- say, 30% of stage 2 time for a doubling of memory when using large B2, when stage 2 is less time than stage 1.

You should get a copy of GMP-ECM and read the readme file for information about what B1 and B2 bounds are usually used to find what size factor. Note that Prime95 does not use GMP-ECM default B2, due in part to memory requirements. Also, notice that the readme discusses factors by digit length, where Mersenne folks often refer to bit length. A 30-digit factor is close to 100 bits!

GMP-ECM is not the most useful program for finding Mersenne factors- I am not recommending the program, just the readme.

So, if power use isn't an issue, you're better off getting 16GB instead of 64, and building a cheap i5-8GB system with the $600 savings. You can always add a second pair of 8GB sticks in a year when they're half the price.

[QUOTE=PageFault;373824]

ok I get part of it - choose low exponents and up the B1 / B2 bounds, in proportion to what I am seeing and then register the exponents. What about the last argument? Will the server automatically adjust this at registration, or do I have to do it (i.e. how)? I want to go extreme and have decided on 64 GB of ram. Who knows, perhaps I can find a monster factor in about a year or so.

[/QUOTE]

To expand my previous answer a bit more and to be more precise:

If you want to test the low exponents I mentioned, you get them manually (Manual Testing -> Assignments), and then you choose ECM as work type, and enter in the "Optional exponent range" the exponent you have previously chosen. From my experience, the program will hand you the exponent with the B1/B2 already set, according to how far the exponent has been tried before. So don´t fiddle with whatever numbers are there. For these small exponents, the number of curves prescribed by the program usually defaults to 150. If you want to run more (or less) curves, just edit that part of the worktodo.txt line.
As for the memory, note that the FFT size used varies largely with the size of the exponent, and so does the memory needed. For exponents under 500K the memory used, even in stage 2, is not significantly large (I think under 1 Gig).
The running time depends both on the FFT size and the bounds used. As an example, I have done a large number of curves on M5503, with B1 bound 44000000, and each curve was taking roughly the same time than on M2423 with B1=110000000. The smaller FFT effect was nearly "cancelled" by the higher value of B1.
To see what exponent suits you best have a look at the ECM Progress page, and pick one, based on the above considerations and your particular preference. From that page you will be able to tell how "far" the exponents have been tried. Note in the column headers that the corresponding factor size in digits (not bits!) is indicated. So the algorythm is progressing from the smaller to the larger factors.
One last thing: you may reserve several instances of the same exponent, and run each one on a different core. The chance of finding a factor is closely tied to the number of curves run, so ECM tests are ideal candidates to be parallelized. Once one of the workers finds a factor, you may stop them all. Job done, and move somewhere else.
Finding very large exponents with ECM, with the current state of the hardware available, only seems possible for very small exponents, where small FFT sizes are used, allowing us to search through higher B1 bounds within a reasonable time limit. Even a 9M exponent would take a huge amount of time to a complete search up to B1=1000000, as the number of curves required is substantial. You may try this for yourself. That´s why I insisted on small exponents

Thanks guys.

I shall try to digest all of this over the weekend - my brain is a bit dizzy from trying to open two mines simultaneous - it resembles the task of Sisyphus.

Crunch on,

PF

OK - that's helpful.

Is there a difference between the two, except the newer one being unable to support ram (!!!??? WTF)

What is the ideal stick configration - I notice the boarts all have 8 slots? 8 X 8?

Cheers,

PF

[QUOTE=VictordeHolland;373828]A i7 4770k doesn't support that much ram! Max Memory Size for Haswell is 32GB according to Intel: [URL]http://ark.intel.com/products/75123[/URL]

IvyBridge-E (for instance an i7 4930k) does support up to 64GB: [URL]http://ark.intel.com/products/77780[/URL][/QUOTE]

ECM2=F8FC338FC24CE5D099185B29AD7218DD,1,2,501013,-1,250000,25000000,150

OK - now what do I modify?

I got this manually - I don't think I can stick in worktodo without changing setting to manual? Otherwise I have to clear out all existing assignments.

Test=N/A before the ECM2?

No change, just stick it to worktodo. Do a copy file in case something went wrong and P95 deletes it, but for me it looks like a pretty valid assignment. The "test=" part is for LL tests only, you don't need it here.

[QUOTE=PageFault;374166]ECM2=xx,1,2,501013,-1,250000,25000000,150[/QUOTE]

It's not too important for ECM reservations, but FYI: you shouldn't post your assignment IDs publicly (that's the long hex number at the start of worktodo items that PrimeNet gives you). This can let people steal your assignments. With ECM, I could probably get an identical assignment (i.e. same exponent, same bounds) if I wanted to, but the same is not true of LLs.

[QUOTE=PageFault;374166]ECM2=F8FC338FC24CE5D099185B29AD7218DD,1,2,501013,-1,250000,25000000,150
[/QUOTE]

If you don't mind satisfying my curiosity, can you report chip GHz/time per curve/memory used during stage 2?

250,000 is the B1 value, which means these settings are intended to search for 30-digit (nearly 100 bit!) factors. 150 at the end is number of curves it will try for this number. That's less than the number needed to have reason to step up to a higher B1 and digit range, which is why someone else could get the identical reservation without wasting effort. Stepping up to B1 = 1M should use about double the memory in stage 2; that will determine whether you do a lengthy effort at 250,000 vs taking a few exponents into the 1M tests that are best for finding 35-digit factors.

I forgot about the assignment key - I was on the sidelines for a couple years, unable to replace my obsolete junk - an antique copper plated calculator from Taiwan, or electronic abacus. I'll slightly change parameters, so that nobody can poach it (why anyone would is beyond me, not like this kind of work is in shortage).

vbcurtis:

I'll likely build my personal machine and then run the small M tests. I have the 500000M above and also a 50000M. CPU will be an i7 with HT and it will do these ECM tests exclusively. I'll get the DDR3 2400, what is better for this custom build - 8 x 4 GB or 2 x 16? Once the OC is settled (It will do 20 triplechecks per core, small M and they must [B]all[/B] match database) I will post the stage 2 details for you.

[QUOTE=PageFault;374184]vbcurtis:

I'll likely build my personal machine and then run the small M tests. I have the 500000M above and also a 50000M. CPU will be an i7 with HT and it will do these ECM tests exclusively. I'll get the DDR3 2400, what is better for this custom build - 8 x 4 GB or 2 x 16? Once the OC is settled (It will do 20 triplechecks per core, small M and they must [B]all[/B] match database) I will post the stage 2 details for you.[/QUOTE]

Nobody would poach this work- he was ensuring you don't do that for a future LLR reservation.

I was asking about your current machine because I have an old Athlonx2 with 2GB that I might be interested in trying ECM work like this.

Back to cooling

Bad african intarweb is hindering me in finding a Noctua for socket 2011 - have a model number anyone?

edit:

CPU FAN NOCTUA NH-D14 SE2011

How about this? Or this?

CPU FAN NOCTUA NH-D14 SE2011 LP

edit, again

for the build, how's this:

INTEL CORE I7 4820K 3.70G/10M/S2011 NOFAN
CPU FAN NOCTUA NH-D14 SE2011
CORSAIR VENGEANCE PRO 2400MHZ DDR3 16GB KIT CL11 RED X 2
Asus Rampage IV Black Edition socket 2011 boart

Is this adequate, or am I being a piker? Money no problem

[QUOTE=PageFault;374166]ECM2=F8FC338FC24CE5D099185B29AD7218DD,1,2,501013,-1,250000,25000000,150

[/QUOTE]

Choosong that exponent is not in line with your willingness to find "monster factors". If you look at the "ECM Progress" table, you´ll see that it has just started to be tested in the 30 digits range. I don´t really know how you define a "monster", but I assume a 30 digit (~100 bits) factor wouldn´t qualify as such If you want something around, say, 55 digits (~180 bits) you have a long way to go sweeping through a considerable amount of intermediate ranges - note how many thousands of curves to run to get there...
My point is, if you really want a "monster" go straight to where they may be hiding. If you want something laround 55 digits, your first choice would be M2423 (for the smaller factors the prescribed number of curves has already been run), if around 60 digits pick M2273, and for 65 digits, M1277. Check the table named ECM on Mersenne Numbers with no known factors in [URL="http://www.mersenne.org/report_ecm/default.php?txt=0&ecm_lo=1&ecm_hi=20000&ecmnof_lo=1&ecmnof_hi=2500"]http://www.mersenne.org/report_ecm/default.php?txt=0&ecm_lo=1&ecm_hi=20000&ecmnof_lo=1&ecmnof_hi=2500[/URL].

I hear you. I still don't know what I am doing and time to read is scarce - try a Director level position, it keeps one [B]extremely[/B] busy and work doesn't magically stop at four or something.

I also have a 53000M and a 12000M. I can grab the ones you suggested and run the hell out of them. How's the hardware choices? I can't get the intel site because the patriot act has blocked this country's IP. I am not at all lamiliar with modern choices.

The ECM progress table: I don't know how to interpret it. If I go with yout picks, settings / bounds please ...

I suggested those exponents because you seem interested in finding VERY large factors. For the ones suggested, you will either give up after a certain time, or, if you are lucky and a factor pops up, find a large one.
The hardware components you mentioned seem quite right. If possible, go for a Noctua cooler.
The ECM Progress Table shows you how far we have gone in testing the numbers in the leftmost column. For factors up to 25 digits long, the B! bound is 50000, and the prescribed number of curves is 280. This means that the program starts by issuing worktodo lines where B1=50000. When the server has already received results for 280 curves, it ups B1 to 250000, which allows to find factors up to 30 digits long. The prescribed number of curves is now 640, and only after receiveng results for this number of vurves will the server start issuing worktodo lines with B1=1000000. Note that, for the same exponent, the curves run with B1=250000 take longer, each one, than the ones run with B1=50000. As we progress, and B1 gets larger, the longer will the curves take.
So the thing is: don´t worry about setting B1 values. The program will do it for you according to the current status of the exponent. If for example you decide to test M2423, the program will send you the following line for your worktodo.txt file: ECM2=xxxxx(AID),1,2,2423,-1,110000000,11000000000,150, because the number is currently being tested for factors up to 55 digits, which implies to use a B1=110000000. The last parameter is the number of curves, that for these small exponents defaults to 150. You may change this parameter, in case you wish to perform a different number of curves.
Hope this helps. Have fun!

The problem with ECM on small exponents in the 1000 to 10000 area is you need to learn how to run GMP-ECM to do stage 2. This is not a simple task. Given OP's limited time, I'd stick with P-1 using large bounds or ECM on exponents between say 10000 and 100000.

I´m surprised. I´ve already tried ECM on exponents this small, and the program (Prime95) would happily accept them and run both stages. What am I missing here?

[QUOTE=lycorn;374365]-- What am I missing here?[/QUOTE]

Both can do it, but GMP-ECM does stage 2 much better than Prime95.

OK thanks, got it.
Is that the same for stage 1, or may we use one program for stage 1 and the other for stage 2 (in case the savefiles are compatible)?

I guess currently the optimal way to do ECM on special numbers like Mersenne and (generalized) Fermat numbers is to run stage 1 with Prime95 and stage 2 with GMP-ECM. Someone can correct me if there have been advances in software that have changed this.

The Prime95 file after stage 1 can be used in GMP-ECM. See undoc.txt:
[QUOTE]Alexander Kruppa wrote some code that allows the output of ECM stage 1 to
be passed to Paul Zimmermann's more efficient GMP-ECM stage 2. This program
is usually faster in stage 1. You can activate this feature by entering
GmpEcmHook=1
in prime.txt. Then select ECM bound #2 between 1 and bound #1. Results.txt
will contain data that can be fed to GMP-ECM for stage 2.[/QUOTE]

Thanks a lot. I´ll give it a try.