mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Conjectures 'R Us (https://www.mersenneforum.org/forumdisplay.php?f=81)
-   -   This is why you have to vacuum your computer (https://www.mersenneforum.org/showthread.php?t=18763)

KEP 2013-10-27 14:37

This is why you have to vacuum your computer
 
Hi everyone

This is a clear example of why you have to vacuum your computer regularly and as often as you can:

I just vacuumed my I5 and it jumped from sieving my base2 reservation at a rate of 1.53M p/sec to sieve at a rate of 2.74M p/sec after I had vacuumed my computer :smile:

So at least on sieving it can mean a difference in sieveproductivity of 77% on an I5. I'm not sure if the case is the same when using LLR/PFGW, but I really don't see why the speedadvantages shouldn't be somewhat similar or the same.

Now a sad news:

My Q6600 2.4GHz has apparently died, so IF I want a strong CPU that is both a decent siever but also the most efficient when it comes to primality testing, then what kind of RAM do you suggest to what kind of CPU? I'm thinking an I7 is starting to be the most efficient when it comes to productivity per dollar spent, but let me hear your 2 cents :smile:

Regards

Kenneth

retina 2013-10-27 14:51

My thought is that the I5 is the best value. I7 is too expensive to justify the extra gain, and I3 is too weak to justify the strong price.

IMO of course.

TheMawn 2013-10-27 17:45

Completely agree with Retina.

I have a first gen i3 that I've put on small jobs to at least feel like I am accomplishing something. On my stats page, my third gen i5 piece of pie is about a tenth or something of the size of my GTX 670 + GTX 660Ti slice, and the first gen i3 is a tenth of the i5.

Currently the i7-3770k (a buddy of mine has one) costs $370 and is a bit harder to overclock because the extra threads generate some heat while producing very little extra work in high-CPU jobs.

The i5-3570k costs $250 (I have this one) and clocks higher and is identical to the 3770k clock-per-clock assuming hyperthreading does nothing.

The i3-3240 costs $130, has half as many cores and does not overclock, period. Clock per clock, it might even be a touch slower.

So, for any highly CPU intensive job where hyperthreading becomes irrelevant, the i5 costs $62.50 per core, the i3 costs $65.00, and the i7 costs $85.00 per core. The i5's frequency can easily be 10% higher than the i7 and as much as 33% higher than the i3.

From this, it isn't too tough to see that the i5 has the best bang for the buck.

TheMawn 2013-10-27 17:48

As for the RAM, I would look into something like this: [url]http://www.newegg.ca/Product/Product.aspx?Item=N82E16820231587[/url]

The high frequency gives you the bandwidth you need to come anywhere near saturating four workers. I lose 20% productivity per core when I set up my fourth worker, so I can more or less saturate 3.75 cores @ 4.6 GHz with 2400 MHz memory. If you're going for full saturation, you'll have to shell out $200 more for a 2800 MHz kit or $300 more for a 2933 MHz kit.

KEP 2013-10-27 19:21

Thanks for your advices :smile:

I'm currently looking at a computer with an [B]I5 4670 3.4GHz[/B] CPU and [B]Crucial - 8GB DDR3 1600MHz PC3-12800 CL11[/B] as memory. I'm not sure if it is $87 well spent to get [B]2 x Crucial - 8GB DDR3 1600MHz PC3-12800 CL11[/B] as memory. But this option seems to be within the pricerange that I can afford and desires to afford, plus the computer that I'm looking at is offered only with 1600 MHz RAM as the fastest RAM and with the fastest I5 beeing the one selected.

Due to your advices I've decided not to spent the 100-150 USD extra that the I7 costs, compared to the price of the fastest I5, since I risk that it will have less output than the I5. Maybe in the future I can upgrade the RAM to 2900 MHz as suggested, but for now, where it retails at about the same as for an entire new machine, I'm not going to consider it a great business :smile:

Regards

Kenneth

TheMawn 2013-10-27 20:10

Be careful. If you're buying a pre-built computer then the impact of changing parts becomes a lot lesser. There is a big difference between i5-4670k and i5-4670. They both start at the same settings, but you [B]cannot modify the settings for the non-k chip[/B]. The settings are locked and you will not be able to increase the speed.

A discussion has been had earlier about whether or not you can adjust the memory speed on a non-k chip. Because you'll be operating at 3.4 GHz instead of something like 4.2 GHz - 4.6 GHz, you will definitely not be needing 2933 MHz memory. Even 2400 MHz will be beyond what you need.

You'll need to look into your overclocking options before making a decision. Any memory you buy may default to whatever settings the motherboard likes and it's possible you won't be able to set them.

KEP 2013-10-27 20:26

[QUOTE=TheMawn;357630]Be careful. If you're buying a pre-built computer then the impact of changing parts becomes a lot lesser. There is a big difference between i5-4670k and i5-4670. They both start at the same settings, but you [B]cannot modify the settings for the non-k chip[/B]. The settings are locked and you will not be able to increase the speed.

A discussion has been had earlier about whether or not you can adjust the memory speed on a non-k chip. Because you'll be operating at 3.4 GHz instead of something like 4.2 GHz - 4.6 GHz, you will definitely not be needing 2933 MHz memory. Even 2400 MHz will be beyond what you need.

You'll need to look into your overclocking options before making a decision. Any memory you buy may default to whatever settings the motherboard likes and it's possible you won't be able to set them.[/QUOTE]

Well I read what you're writing and I understand your point of view. However my understanding has always bin that there is to some degree a given uncertainty that a result is actually correct, when the result is computed by an overclocked machine. So given that statement and the fact that I doesn't have the skills to overclock my machines, overclocking really never plays much of a role. So if we rule out overclocking, am I looking at a good or a bad choice in terms of productivity per dollar?

TheMawn 2013-10-28 00:16

I'd say factory settings 3.4 GHz i5 with 1600 MHz memory will do quite nicely. The other thing to keep in mind of course is Intel has turbo boost technology on even the non-k series chip so it's rated for 3.4 GHz but will go a bit higher yet. You're still going to be memory limited a bit but you'll be in good shape with that setup.

My two cents, anyway. If you have adjustable settings for memory (memory profiles or even the raw settings) you can look into buying your own 1866, 2000 or 2133 MHz later. In the mean-time, though, you'll be rocking a good machine.

KEP 2013-10-28 09:45

[QUOTE=TheMawn;357663]I'd say factory settings 3.4 GHz i5 with 1600 MHz memory will do quite nicely. The other thing to keep in mind of course is Intel has turbo boost technology on even the non-k series chip so it's rated for 3.4 GHz but will go a bit higher yet. You're still going to be memory limited a bit but you'll be in good shape with that setup.

My two cents, anyway. If you have adjustable settings for memory (memory profiles or even the raw settings) you can look into buying your own 1866, 2000 or 2133 MHz later. In the mean-time, though, you'll be rocking a good machine.[/QUOTE]

Thanks, it's nice to know :smile:

I'm glad that you all took your time to help me out. Due to your advices and my fiscal policy :smile: I guess that my 1 I5 is going to be 2 I5's from this upcoming monday :smile:

Take care everyone

Kenneth

KEP 2014-04-21 10:39

Though it is not as substantial as when sieving, then it's worth mentioning, that on my I5 2300, the speed when running LLR (using AVX), increases by 10% when I have vacuumed my computer, compared to tests done when my computer was dusty. So if you wanna be the most productive, then vacuum your computer and you'll most likely on the AVX machines see a productivity gain of 10% when running LLR and about 77% when running 64-bit sieving :smile:

danaj 2014-04-21 15:58

TheMawn gives good advice.

My US$0.02. The 4670 is good, though with the stock cooler at stock speeds it runs very hot on highly CPU intensive tasks. A simple 120mm aftermarket cooler doesn't hurt, or at least make sure there is good airflow. It's also possible it is just my bad luck -- I didn't reseat or try other thermal paste. Since vacuuming came up, this is within a few days of installation, so no dust issues at that time (it will just get worse).

Strangely newegg is selling the 4670K for less than the 4670 today. Much as I like the 4770K and 4930K, they're not a good value option vs. the 4670/4670K.

TheMawn 2014-04-21 20:54

4930k is a bit less value for your dollar but it's the cheapest (and best value) six-core option, so that gives it some extra attraction right there.


But yeah, stock coolers require good clean cases to function even remotely well at high CPU loads. They blow air inward and it exits the bottom of the heatsink (preferably) and expands radially from there. If you were to put a piece of paper under the bottom of the heatsink (think of dust as a similar obstruction) you would probably lose half of your airflow over the fins right there.

The 10% speed increase you see is due to the fact that you're not hitting the thermal limits so the CPU allows itself to clock slightly higher.


All times are UTC. The time now is 10:07.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.