xtremeoverlocking - pushing the 3930K/3960X to 4.5 GHz

[alpha754293]

I'm starting to look into doing a new build that is looking to push the 3930K/3960X to 4.5 GHz.

The system will have at least 16 GB of RAM, but very likely 32 GB, maybe even 64 GB of RAM.

I am looking for recommendations for motherboards, cases, and cooling solutions.

It'll also likely get a cheap PCIe video card (recommendations greatly appreciated for best bang for the buck).

This system is going to be running vehicle/automotive crash simulations and some of the simulations can run for a week or longer - straight so it MUST to be able to run at full load for very long periods of time (say....a month - continuously) at 4.5 GHz.

I would prefer if it can stay < 45 dBA and be the coolest it can be for running such long loads.

Budget is set between $2000-2500 (between board, CPU, case, motherboard, power supply, cooling, video card), but may expand if it can be technically justified.

I already have HDDs and I have a floating optical drive for OS installation, so you don't have to worry about those things.

Here's an example of the type of problems that this computer will be solving: http://www.youtube.com/watch?v=3ppuDedRZDM

Recommendations for this build is greatly appreciated.

I haven't decided on if I'm going to start with the 3930K or the 3960X yet (since from the 3930K, that's a 40% OC, and from the 3960X it's a little more pallatable 35% OC, but the processor alone is an additional $500; so if I can put that towards a better cooler and then push the 3930K to that magical 40% OC, and hopefully that won't kill the processor prematurely because of such a huge OC, and running it continuously for long periods of time)...

Thanks.

[[XC] gomeler]

I'm going to move this to the Intel section as it'll be more appropriate. Informing the OP.

If the simulations you are running can be distributed across multiple computers then it might make more sense to setup a pair of 2600Ks at 4.5 GHz. For roughly the same cost as a 3930K you'll get 16 threads and the ease of expanding out to more threads by putting in the engineering time to setup the distribution system now.

[alpha754293]

I can distribute it, but I'm limited by licensing. More CPUs = more licenses needed.

And also as the number of processors is assigned to a job, the number of inter-processor communication increases (I think it's some kind of exponential/power rule).

The great debate actually is whether having a significantly faster system with fewer cores is going to be better than having four dual-socket Xeon blades (slower clock speed per core, but throwing upwards of 64-threads at it)...

And it used to be that multiprocessing kicked desktop processings' butt, but that's probably a true statement like....10 years ago, and hasn't really been that way for the past oooh...5 years?

My last workstation at my last job (Intel 990X, OC'd to 4 GHz) was able to beat a 48-core AMD Opteron system by a factor of TWO! So...that's why I'm looking into this build. (And it was the exact same input file that was being submitted for the run - same everything - just different hardware.)

P.S. Thanks for moving it. I think that it straddles OC vs. build, but I'll leave that upto you guys, the mods, and the admins. :o)

P.S.#2 four dual-socket Xeon blades - fully decked out will run me between $20k-22k. This system, I'm trying/hoping to build for like...1/10th of that. And probably will be just as capable, if not more so. (Because it would be also DOUBLE in clock speed, and 1/3rd the number of cores, which means something like 1/9th the number of intercore communications. I think. Something like that.)

[Leeghoofd]

I'm starting to look into doing a new build that is looking to push the 3930K/3960X to 4.5 GHz.

The system will have at least 16 GB of RAM, but very likely 32 GB, maybe even 64 GB of RAM.

I am looking for recommendations for motherboards, cases, and cooling solutions.

It'll also likely get a cheap PCIe video card (recommendations greatly appreciated for best bang for the buck).

This system is going to be running vehicle/automotive crash simulations and some of the simulations can run for a week or longer - straight so it MUST to be able to run at full load for very long periods of time (say....a month - continuously) at 4.5 GHz.

I would prefer if it can stay < 45 dBA and be the coolest it can be for running such long loads.

Budget is set between $2000-2500 (between board, CPU, case, motherboard, power supply, cooling, video card), but may expand if it can be technically justified.

I already have HDDs and I have a floating optical drive for OS installation, so you don't have to worry about those things.

Here's an example of the type of problems that this computer will be solving: http://www.youtube.com/watch?v=3ppuDedRZDM

Recommendations for this build is greatly appreciated.

I haven't decided on if I'm going to start with the 3930K or the 3960X yet (since from the 3930K, that's a 40% OC, and from the 3960X it's a little more pallatable 35% OC, but the processor alone is an additional $500; so if I can put that towards a better cooler and then push the 3930K to that magical 40% OC, and hopefully that won't kill the processor prematurely because of such a huge OC, and running it continuously for long periods of time)...

Thanks.

Go for 3930K (slightly less cache then 3960X, but not worth the extra cash) 4500Mhz is a no brainer mate
Corsair Hydro 100 or eg Swiftech kit ( http://www.swiftech.com/H2O-x20-Edge-HD.aspx )
ASUS Sabretooth or ASROCk Extreme 6 or 9 board
2 x G.Skill F3-12800CL9Q-16GBZL (cheap quad channel kit at +/- 120 dollar per kit, 32Gb will set you back trice the price, this at least in europe)
PSU : Corsair AX1200
VGA anything will do as it seems not required for the rendering... GTX560 or 6870 maybe ?


Make sure the case is well ventilated !! so opt for a big tower with eg a sidepanel fan !

[alpha754293]

Go for 3930K (slightly less cache then 3960X, but not worth the extra cash) 4500Mhz is a no brainer mate
Corsair Hydro 100 or eg Swiftech kit ( http://www.swiftech.com/H2O-x20-Edge-HD.aspx )
ASUS Sabretooth or ASROCk Extreme 6 or 9 board
2 x G.Skill F3-12800CL9Q-16GBZL (cheap quad channel kit at +/- 120 dollar per kit, 32Gb will set you back trice the price, this at least in europe)
PSU : Corsair AX1200
VGA anything will do as it seems not required for the rendering... GTX560 or 6870 maybe ?


Make sure the case is well ventilated !! so opt for a big tower with eg a sidepanel fan !

Any recommendations for the case?

I did look at the Hydro 100, but noticed that the radiator is quite large. And I've been looking at the PNY GTX560 as well. I was also looking at a Gigabyte X79-UD5 (8 DIMM slots).

Do I really need a 1.2 kW power supply?

[Leeghoofd]

H100 rad is not that large mate, I can recommend Corsair Carbide 500R or Coolermasters Trooper case, H100 fit's both like a gloove...

AX850 will suit your needs fine too...(but your system running flatout might eat up to 400-500W. The AX1200 PSU is one of the best out there... top quality... hence why I recommended it...

With AX850 and without GPU and ram ( add 500-550) you get this :

http://users.telenet.be/OAP/newegg.jpg

Still way below your budget mate of 2000. I had mucho issues with my UD5 sample, UD3 was far better (believe it or not ) and more stable... But keep in mind that the GB boards still have issues with higher clocks on air or water...

[alpha754293]

I couldn't go with the UD3 because it only has 4 DIMM slots. Well...I could, but then the cost of RAM becomes another $300-500.

Will the system be able to run at full load, at 4.5 GHz for a week (or more) straight?

And it would get decent temps and not sound like a jet taking off?

*edit*
"still way below your budget mate of 2000"

add RAM and GPU, and it'll start getting very close/closer to it.

[Leeghoofd]

Why don't you propose us the sytem you want to go for, seems you have put your eye on some parts... honestly for X79 ASUS is the path to walk...

[BeepBeep2]

+1 to Leeghoofd.
Why come here if you already have your mind up? You seem to be talking the talk, a bit over your head, if you know what I mean.

Unless you need ECC ram, (and I don't have any idea as to why you would need more than 4 slots)...ram is $230

When you keep asking if the system can run full load at 4.5 GHz for a certain amount of time, you need to understand the factors that play into it and come up with that conclusion yourself.

A weak CPU will mean you may need more voltage and/or cooling. If your overclock is right on the edge, you can experience instability. Motherboard needs to have good VRM/PWM components. Your ram needs to be free of any defect, even some that may never show up under a normal/light use, but may show up at full load over several hours/days. When you ask about it getting decent temps without a jet taking off, you need to realize that the fans are what makes that noise, pick lower-midrange RPM fans and you will never have anything sounding like a "jet taking off"...

Custom water cooling components would be best for you in this situation but I really doubt you want to take the time to cut your own tubing, test for leaks, perform maintenance, research the right parts either. That is partially why you are here, to get help with those types of things, no?

[alpha754293]

Why don't you propose us the sytem you want to go for, seems you have put your eye on some parts... honestly for X79 ASUS is the path to walk...

+1 to Leeghoofd.
Why come here if you already have your mind up? You seem to be talking the talk, a bit over your head, if you know what I mean.

Unless you need ECC ram, (and I don't have any idea as to why you would need more than 4 slots)...ram is $230

When you keep asking if the system can run full load at 4.5 GHz for a certain amount of time, you need to understand the factors that play into it and come up with that conclusion yourself.

A weak CPU will mean you may need more voltage and/or cooling. If your overclock is right on the edge, you can experience instability. Motherboard needs to have good VRM/PWM components. Your ram needs to be free of any defect, even some that may never show up under a normal/light use, but may show up at full load over several hours/days. When you ask about it getting decent temps without a jet taking off, you need to realize that the fans are what makes that noise, pick lower-midrange RPM fans and you will never have anything sounding like a "jet taking off"...

Custom water cooling components would be best for you in this situation but I really doubt you want to take the time to cut your own tubing, test for leaks, perform maintenance, research the right parts either. That is partially why you are here, to get help with those types of things, no?

Because that's based on my initial research. Obviously, if ther'es a better solution out there, I'll go with that. (I ended up pricing the system based on what Leeghoofd suggested, so how's THAT for "why come here if you already have your mind up?"??? :P

2) I've never pushed a system +40%, so being XS...it's a point of reference/resource (cuz I would think that some people here have done that any more. The question then becomes have they ever actually ran it like that, at full load, for a month - straight? And have it still be stable, and no deterioration on the life of the CPU).

3) re: long term system stability - while yes - what you have said is true that it does depend on the luck of the draw for those components. But if no one's been able to be stable say...oh...I dunno...I'm gonna pick a number outta thin air - 10 GHz and run it at that speed for any period of time - they could have had a processor that'll do 7, or even 8 GHz while others won't. And it's impossible to be able to cherry pick a proc like that; which means that if you were to test and OC all of the CPUs available to you in the world, it'll eventually follow some kind of statistical distribution. x1 number will be able to do y1 speed. x2 number will be able to do y2 speed and so on and so forth. (It's very simple statistics. And at the end of the day, you should be able to say "okay...based on x samples, the average or trend tends to be that they're good up to y speed, but above and beyond that, the number of processors that can get it starts diminishing dramatically.)

Can I be guaranteed to be able to hit that speed? No. Am I looking for such guarantee? Definitely not. And yes, I realize that I can help it become more stable with a smidge more voltage, which means (as you've said) more cooling. But if I'm going to be dropping 100 CFM Delta fans on it, I might as well go for the twin 1U blade that's got the 15krpm 40 mm Deltas.

Which brings me to 4) noise - coming from the server/HPC environment, I'm very familiar with the 8.9 B(A) Delta fans and having 6 of them pushing a total of 600 CFM through two passive heatsinks, guided by an air duct.

If it's going to take me 8.9 B to get to 4.5 GHz, then the answer is "no, it cannot be done" based on the parameters specified. If it's 5.5 B, then...well...it's a subjective call once the system is in place and up and running. If it's 4 B, then it's a "you're good to go".

5) Yes, I actually started off the research on the Corsair H80, but I was reading some reviews (and also plotted and X-Y plot between temperature and noise at load) and found that for what they were testing with, even an air cooled solution would have worked. Unfortunately, the air cooled solution that they tested isn't compatible with Socket R, so it became a toss up between the H80 and H100. And I don't trust the people at the local computer store enough to be able to recommend a proper case for it, so...here we are.

And that other people might have experiences with either the H100 or something else that might be better (both in terms of cost and performance). And I was also looking for feedback to see if pushing a 3930K would be better than pushing a 3960X, but putting the difference in the cost towards a better case/cooling solution.

I don't actually need ECC, as pretty much all of my workstations and servers have ECC disabled.

And you are correct, I am looking for more commercially available solutions.

And yes, I did mention the whole thing about being able to run at 4.5 GHz for certain amount of time because in my experience, a lot of people who do OC (40%+) may not be pushing their computer to full load for weeks on end. Yes, some might be running F@H and other similiar distributed computing projects, but unless you're running LINPACK, I have NEVER found anything that taxes a CPU more than that. (considering that the Gauss elimination solver is ALL FPADD/FPMUL operations...) and a good portion of running the vehicle crash simulations is like that. (Not quite as taxing as a pure LINPACK run...) And I know that people use to say, from yore years, that CPUs will die prematurely when OC'd. I don't know if it's the case now or if that has improved dramatically.

Here's the cart that I ended up with:
cart.jpg

[Leeghoofd]

Wow guys lighten up, nothing to get wound up about.

I'll state a small explenation on why I advised some components :

Carbide 500R : big sidefan (speed adjustable ) 2 front fans ( best hooked up to the mobo for balanced speed/noise ratio ) and H100 fit's perfectly
Corsair H100 : best performing and quietest of all the "all in one liquid cooling solutions" H80 and co will be too noisy with the output heatload of a s2011 6core CPU
ASUS Sabretooth : very stable board, biosses have been top notch since day one + install the two little fans + 5 year warranty
3930K iso 3960X : will OC both similar on air, 4500Mhz is doable around 1.33Vcore

You found a nice kit of ram, is that CL10 ? timings don't matter much, but 1600Mhz is a good balanced speed, 1333Mhz is too slow for s2011

Keep us advised how it runs man !! good luck !!

[alpha754293]

Wow guys lighten up, nothing to get wound up about.

I'll state a small explenation on why I advised some components :

Carbide 500R : big sidefan (speed adjustable ) 2 front fans ( best hooked up to the mobo for balanced speed/noise ratio ) and H100 fit's perfectly
Corsair H100 : best performing and quietest of all the "all in one liquid cooling solutions" H80 and co will be too noisy with the output heatload of a s2011 6core CPU
ASUS Sabretooth : very stable board, biosses have been top notch since day one + install the two little fans + 5 year warranty
3930K iso 3960X : will OC both similar on air, 4500Mhz is doable around 1.33Vcore

You found a nice kit of ram, is that CL10 ? timings don't matter much, but 1600Mhz is a good balanced speed, 1333Mhz is too slow for s2011

Keep us advised how it runs man !! good luck !!

Thanks. I'll be honest, I'm not a big fan of cases with side fans because I've actually ran the computational fluid dynamics (CFD) simulation on it and most of them - the air stalls by the time it hits the CPU. They're big, they're quiet, but they don't move a lot of air given it's size. Sure, they might say that they're pushing 100 CFM, but given it's size, that's actually NOT a lot at all. If I have a small, really loud fan pushing 100 CFM and a bigger, but much quieter fan pushing 100 CFM, it would just mean that for the bigger fan, the air has to be moving VERY slowly. Which also means that once it hits any of the installed components, the air just loses all kinetic energy and stalls. And then all that hot air stays in the case. BUT...like you said, it's a case that'll fit the H100.

Now I'm curious, are there other better performing, but probably (slightly) more expensive closed-kit liquid cooling systems out there?

The RAM has a CAS latency of 9 clock cycles, but from what I've been told, the program that I am going to be running ISN'T sensitive to memory speed (or bandwidth) at all.

The system is also likely going to run completely headless after the initial install.

Also, this might be a bit of a silly question - but does anybody know if anybody's going to be coming out with a LGA2011 board that has onboard video that can also OC?

[Leeghoofd]

I've seen 3-5°C drops on GPU, mobo and PWM temps with the sidefan activated or not... have a fun build plz ignore all I have mentioned above... There's more to it then just simulation...

[BeepBeep2]

1. Thanks. I'll be honest, I'm not a big fan of cases with side fans because I've actually ran the computational fluid dynamics (CFD) simulation on it and most of them - the air stalls by the time it hits the CPU. They're big, they're quiet, but they don't move a lot of air given it's size. Sure, they might say that they're pushing 100 CFM, but given it's size, that's actually NOT a lot at all. If I have a small, really loud fan pushing 100 CFM and a bigger, but much quieter fan pushing 100 CFM, it would just mean that for the bigger fan, the air has to be moving VERY slowly. Which also means that once it hits any of the installed components, the air just loses all kinetic energy and stalls. And then all that hot air stays in the case. BUT...like you said, it's a case that will fit the H100.

2. Now I'm curious, are there other better performing, but probably (slightly) more expensive closed-kit liquid cooling systems out there?

3. The RAM has a CAS latency of 9 clock cycles, but from what I've been told, the program that I am going to be running ISN'T sensitive to memory speed (or bandwidth) at all.

4. The system is also likely going to run completely headless after the initial install.

5. Also, this might be a bit of a silly question - but does anybody know if anybody's going to be coming out with a LGA2011 board that has onboard video that can also OC?

1. Sigh. Now that Leeghoofd has given up, I'll say this about airflow. You really want to have directional airflow in an ATX case. I really hate side fans as they disrupt the direction of airflow from front to back, or front to back and top (depending on your case). You really need to look at the real world instead of a simulation to grasp this, but I think that is what you are trying to explain. It has nothing to do with air "stalling", if you draw air in the side and expel it out the top without a front intake and rear exhaust it would work a lot better. However, that isn't how most cases are set up. You are partially correct with your statement about CFM, however when a 120 millimeter fan is used in application pushing 100CFM (lets say a Sanyo Denki fan) it is usually attached to a part or heatsink that can use up that surface area. I don't know if you know this or not, but a ~115CFM 120mm fan at full speed can almost levitate.

I would suggest a different case for you in my personal tastes, but that case does fit the all-in-one joke of a liquid cooling solution you want to use.

2. No. However there are open kits that contain all (or most of, excluding some useful accessories), the parts. For these though, you'd need to mount parts, then measure and cut appropriate lengths of tubing and attach them in the loop using barbs and clamps or compression fittings. After that, you would fill the loop with water as you ran the pump (carefully, because it is self lubricated with the water, it is easier with three hands) by jump-starting your power supply in small bursts with no components attached but your water cooling pump. After it is filled, you can leave the power supply on with the only component attached being the liquid cooling pump. That way you can leak test without worry that your components will be ruined with water. You can simply dry them if you have leaks. This type of loop would need to be drained, taken apart, and cleaned about once a year. If you would like to learn more about this, people here can teach you.

3. Yes, that is an adequate kit. 1600 CL9 will be far from breaking any records, but then again you will see no faster in a Dell or HP machine. Some DDR3 will run 2200 CL7.

4. I strongly advise that not happen, at least not right away, because it will take you a day or two to run Linpack and test your overclock. Overclocking isn't something you can do in a matter of minutes and it is not predictable unless you know your hardware really well. What I was trying to say in my other post, was that you will need to find out if you can run 100% load for a week yourself.

5. As far as the onboard video, that is on the CPU now. I don't know why you would want to overclock that though, especially if you will be using a discrete GPU. In computing performance comparison, it would be like pitting a 64-thread server with an Intel Atom.

[alpha754293]

1. Sigh. Now that Leeghoofd has given up, I'll say this about airflow. You really want to have directional airflow in an ATX case. I really hate side fans as they disrupt the direction of airflow from front to back, or front to back and top (depending on your case). You really need to look at the real world instead of a simulation to grasp this...

If you can find a way to measure the velocity where the instrumentation device won't cause a distruption to the flow, please, by all means - enlighten me.

I think that the only way it would work is if the case was made out of acrylic or polycarbonate and then you used DPIV. Then you can prove that the velocity goes to zero as it reaches the components.

If you want to pay $20 grand for the test, I'll run it. (The lasers I believe were about $3k each if I remember correctly. I forget how much the optics were and the control module and the camera, but I think that it works out to be close to $20k.) The program I was using at the time, was licensed (legitimately) for $1k and I could run as many tests as I can for a year.

And how do you think that the simulation softwares are developed???

... but I think that is what you are trying to explain. It has nothing to do with air "stalling", if you draw air in the side and expel it out the top without a front intake and rear exhaust it would work a lot better. However, that isn't how most cases are set up. You are partially correct with your statement about CFM, however when a 120 millimeter fan is used in application pushing 100CFM (lets say a Sanyo Denki fan) it is usually attached to a part or heatsink that can use up that surface area. I don't know if you know this or not, but a ~115CFM 120mm fan at full speed can almost levitate.

volumetric flow rate is determined by flow cross section area * velocity.

If you have an 80 mm fan (and yes, I know, the actual flow cross section area isn't ACTUALLY 80 mm in diameter), to make the calculations simple (it really doesn't matter that much to prove the point), if you start out with 100 CFM, that 0.04719 m^3/s.

80 mm (0.08 m) diameter has an circular flow cross section area of 0.005027 m^2. That works out to 9.39 m/s exit velocity.

For a fan that's 120 mm in diameter, but that has the same volumetric flow rate, that exit velocity drops down to 4.172 m/s.

For a fan that's 200 mm in diameter, again same volumetric flow rate, that exit velocity drops down to a puny 1.5 m/s. It doesn't take a freakin' rocket scientist to be able to feel (in the real world) that it doesn't take much for the air to come to a standstill because it's NOT moving very fast at all when it leaves the fan!

The difference is that with the CFD (which, you were all about the real world - why would Formula One teams have SUPERCOMPUTERS doing this???) - I can have very complicated inlet and outlet configurations that you can't tell me how it's going to perform before you cut a check for half-a-million-dollars-worth of tooling. ;o) How's that for real world?

I would suggest a different case for you in my personal tastes, but that case does fit the all-in-one joke of a liquid cooling solution you want to use.

2. No. However there are open kits that contain all (or most of, excluding some useful accessories), the parts. For these though, you'd need to mount parts, then measure and cut appropriate lengths of tubing and attach them in the loop using barbs and clamps or compression fittings. After that, you would fill the loop with water as you ran the pump (carefully, because it is self lubricated with the water, it is easier with three hands) by jump-starting your power supply in small bursts with no components attached but your water cooling pump. After it is filled, you can leave the power supply on with the only component attached being the liquid cooling pump. That way you can leak test without worry that your components will be ruined with water. You can simply dry them if you have leaks. This type of loop would need to be drained, taken apart, and cleaned about once a year. If you would like to learn more about this, people here can teach you.

3. Yes, that is an adequate kit. 1600 CL9 will be far from breaking any records, but then again you will see no faster in a Dell or HP machine. Some DDR3 will run 2200 CL7.

4. I strongly advise that not happen, because it will take you a day or two to run Linpack and test your overclock. Overclocking isn't something you can do in a matter of minutes and it is not predictable unless you know your hardware really well. What I was trying to say in my other post, was that you will need to find out if you can run 100% load for a week yourself.

5. As far as the onboard video, that is on the CPU now. I don't know why you would want to overclock that though, especially if you will be using a discrete GPU. In computing performance comparison, it would be like pitting a 64-thread server with an Intel Atom.

I've actually ran the CFD simulations for some of the other people here and elsewhere as well. I think one guy - Jake - goes by the handle Cpt. Planet - when he started with his dry ice cooling and he was machining his tubes, he actually had me simulate the thermal performance and we actually spent a little bit of time optimizing his design with CFD before he ran. I think at one point, he might have held a record or two. I forget.

I've also CFD'd waterblocks too. So, CAN I engineer a custom system? I probably can. One of my friends actually builds CNC kits too, for like $2k, so I can CNC my own blocks as well. Course, that would be time spent on getting the system up and running instead of actually having it up and running and doing something useful with it.

3. I was told that going from DDR3-1333 to DDR3-1600 only made like a few percent difference despite the huge speed increase.

4. I meant that once the system has settled in on an OC, and it has passed all of the stability tests, then the system is going to be running completely headless.

5. From what I've been able to find out after posting that, Socket R doesn't even have ANY on-board GPU available. Why would I want to do that? Because there's no point of putting in a discrete GPU when the system is going to be ultimately going headless. And I'll only need that if stuff goes wrong.

[OC Nub]

Its a waste of time and energy to try and help someone that knows everything.

[l0ud_sil3nc3]

Its a waste of time and energy to try and help someone that knows everything.

yup, very true.

[BeepBeep2]

1. If you can find a way to measure the velocity where the instrumentation device won't cause a distruption to the flow, please, by all means - enlighten me.

I think that the only way it would work is if the case was made out of acrylic or polycarbonate and then you used DPIV. Then you can prove that the velocity goes to zero as it reaches the components.

If you want to pay $20 grand for the test, I'll run it. (The lasers I believe were about $3k each if I remember correctly. I forget how much the optics were and the control module and the camera, but I think that it works out to be close to $20k.) The program I was using at the time, was licensed (legitimately) for $1k and I could run as many tests as I can for a year.

And how do you think that the simulation softwares are developed???



2. volumetric flow rate is determined by flow cross section area * velocity.

If you have an 80 mm fan (and yes, I know, the actual flow cross section area isn't ACTUALLY 80 mm in diameter), to make the calculations simple (it really doesn't matter that much to prove the point), if you start out with 100 CFM, that 0.04719 m^3/s.

80 mm (0.08 m) diameter has an circular flow cross section area of 0.005027 m^2. That works out to 9.39 m/s exit velocity.

For a fan that's 120 mm in diameter, but that has the same volumetric flow rate, that exit velocity drops down to 4.172 m/s.

For a fan that's 200 mm in diameter, again same volumetric flow rate, that exit velocity drops down to a puny 1.5 m/s. It doesn't take a freakin' rocket scientist to be able to feel (in the real world) that it doesn't take much for the air to come to a standstill because it's NOT moving very fast at all when it leaves the fan!

The difference is that with the CFD (which, you were all about the real world - why would Formula One teams have SUPERCOMPUTERS doing this???) - I can have very complicated inlet and outlet configurations that you can't tell me how it's going to perform before you cut a check for half-a-million-dollars-worth of tooling. ;o) How's that for real world?




I've actually ran the CFD simulations for some of the other people here and elsewhere as well. I think one guy - Jake - goes by the handle Cpt. Planet - when he started with his dry ice cooling and he was machining his tubes, he actually had me simulate the thermal performance and we actually spent a little bit of time optimizing his design with CFD before he ran. I think at one point, he might have held a record or two. I forget.

3. I've also CFD'd waterblocks too. So, CAN I engineer a custom system? I probably can. One of my friends actually builds CNC kits too, for like $2k, so I can CNC my own blocks as well. Course, that would be time spent on getting the system up and running instead of actually having it up and running and doing something useful with it.

4. I was told that going from DDR3-1333 to DDR3-1600 only made like a few percent difference despite the huge speed increase.

5. I meant that once the system has settled in on an OC, and it has passed all of the stability tests, then the system is going to be running completely headless.

6. From what I've been able to find out after posting that, Socket R doesn't even have ANY on-board GPU available. Why would I want to do that? Because there's no point of putting in a discrete GPU when the system is going to be ultimately going headless. And I'll only need that if stuff goes wrong.

In this post, I have given you and spelled out, spoon-feeding to you everything you should need to make your own deductions about what parts you need and what decisions to make about the build of your new machine. You come off to be very blunt and rude. It is obvious that you don't need any help, because by your current tone, you already know it all and everyone else, especially all the people here, know nothing. I don't even know why you're here anymore, and for the sake of other members, we've all spent way too much time already trying to help you. If I had to look back at this post, and could do it over, I would not have written it as you are not deserving of the information given. It is 4:56 AM, I am a junior in high school, and I am now going to bed. Please though, enjoy the following information while you invalidate everything I say in your head due to what all of us have said previously, and my age.

1. I do not know of any device that will measure velocity without disrupting the flow. I apologize for my "inability to enlighten you" about a product you and I both know is mythical. I am talking about directional flow in a push-pull configuration with the components in the middle. That is how a computer is designed around ATX specification, and how the components are designed to cool themselves efficiently, or is it not? Of course velocity goes to zero as it reaches components if it were in open air and flow was perpendicular to the surface, but that is not what happens here. We are not talking about the side fan (which I already said I disliked) because it is nearly useless other than to help direct flow like I said already but you neglected to heed.

That said, I am obviously not intelligent in your eyes because I am not an engineer, deducted by the way you speak, so I doubt anything I say will have any merit to you.

I will not pay $20,000 for a test. I don't care how much the lasers cost. I don't care how much the optics were, and I don't care how much the control module cost. I don't care how much the camera cost. I don't care how much your program cost either, but thanks for making those smarta** statements about the cost of equipment you already knew I couldn't care less about.

How do I think simulation software is developed? I am not a dumb***, by real-world testing of course. However, you are asking me to pay apparently over $40 grand for a scientific real-world test, one that could be used to help design a simulation software, AND one that I don't care about.

Please watch the video in the link below, to get an idea of how airflow is supposed to be directed in an ATX chassis.
http://www.youtube.com/watch?v=Qe-2ZqmSGug

Alrighty! I bet you didn't learn anything from that video, since you don't learn anything from what anyone tells you here, but when you realize that the components sit on the farthest side of the case, those useless 200mm fans can be used to help direct that flow of air on to the components and help with positive air pressure within the case to provide more velocity (so it isn't completely useless!) instead of bypassing and directly being exhausted outside of the case.

2. Again I will say, I am not a dumb***. I already know that volumetric flow rate = v * A. I've also passed high school math.

What's the point in your calculation? 100 CFM from an 80mm fan is not a reasonable number for a desktop /or headed workstation environment, where we don't rape our ears and typically choose fans consistent with the SPL of all the other fans in the system, so lets use 40 CFM instead which is normal for an 80mm fan running 3000 RPM at a rated 35 dBa and a circular cross section area of .005027 m^2:
40 CFM = ~.019 m^3/s.
Which works out to be ~3.78 m/s velocity.


Now lets use 80 CFM for a 120mm fan, running 2000 RPM at a rated 35 dBa and a circular cross section area of .007539 m^2:
80 CFM = ~.038 m^3/s.
This works out to be ~3.36 m/s velocity.

These calculations are still way, WAY off because you and I both did not take into account the size of the motor hub in the center which is variable, though roughly twice the area of the motor hub for 120mm fans common to 80mm fans. This is worthless still, even if you were going to look at fans to be put on a radiator for several reasons.

The motors make most of the noise, ratings are inaccurate, and simulations won't save you when it comes to perception. Second, radiators and water temperature scale logistically. The folds in between of the flat tubes that make up the water channels in a radiator are measured by fins per inch or "FPI". The more folds you have, the more restrictive the radiator is on airflow, and the worse it does with low speed fans. Radiators with more FPI do much better with high speed 100CFM+ fans. Radiators with low FPI do very, very well with low-speed (quiet) fans, compared to their high FPI counterparts.

What just made me laugh about your creation of this thread is how you mentioned this:

I haven't decided on if I'm going to start with the 3930K or the 3960X yet (since from the 3930K, that's a 40% OC, and from the 3960X it's a little more pallatable 35% OC, but the processor alone is an additional $500; so if I can put that towards a better cooler and then push the 3930K to that magical 40% OC, and hopefully that won't kill the processor prematurely because of such a huge OC, and running it continuously for long periods of time)...

The 35% OC on the 3960X might as well be the same as overclocking the 3930K by 40%. Cooling doesn't even matter. One 3960X may only overclock reasonably 30%, then another 3930K will overclock by 45%. It is hilarious to think you are gaining the extra 5% on the 3960X by using better cooling on it. You are gaining 5% on both processors. This is also why I told you that the people here couldn't just tell you that a specific cooling combination and overclock will be able to run. Every CPU is not the same, some even run 10 degrees Celsius higher than an identical model at the same frequency and voltage. Also, you run a risk of killing the processor prematurely when you don't know what you're doing. I feel safe saying that you don't know what you're doing. Skilled overclockers know the tolerances of their parts.

I'm not going to waste my time calculating a 200mm fan. Everyone already knows those are useless, especially by themselves at the speeds most run, and the results that Leeghoofd mention are because of the way the 200mm fan are because of a diversion in the direction of the flow throughout the case as it runs over the heatsinks that are optimized for directional flows and low restriction.

3. Did I say a damned thing about running CFD simulations on waterblocks? I don't care if you can design your own blocks, OR manufacture them. What I said has absolutely NOTHING to do with that. It has to do with you putting a few extra days of effort to buy separate parts, including but not limited to a waterblock, radiator(s), barb fittings and clamps, or compression fittings, bulk tubing, pump and reservoir, also cutting the tubing to correct length, installing all the components, meticulously filling the finished closed loop and testing it for leaks. You commented on the subject and had no idea what you were talking about. In fact, you completely avoided the practical aspect, and only talked about the engineering aspect, before a part(s) become(s) a product in use.

Also, I am sorry, but there is much more than fluid flow and restriction to take into account with a water block. There are waterblocks currently on the market with a low pressure drop that perform exactly the same in a loop as a waterblock with 3x the pressure drop when combined with the same radiators and pumps.

4. You are right, but you don't understand why. That would be because the memory does not do any of the computation. It only stores what is needed at any given time. You can also buy a hard drive twice as fast as the one before and only see a 5% benefit, because the CPU, memory, hard drive, software reliance, other system hardware bus may carry equal or a biased weight in the final result. DDR3-1600 CL9 will be roughly 20% faster than DDR3-1333 CL9 in both effective latency and read/write speeds provided the interconnect is not a "bottleneck". That simply means the CPU can access whatever it has stored or store information in memory 20% faster, and some programs will be insensitive to memory speed for the simple fact that they will not read and write to RAM often. If your program makes frequent read/writes to RAM, one could definitely see a realistic 10-15% improvement in performance.

5. Good for you. You weren't very clear, and it didn't seem like you were going to put much effort into the overclock on your new CPU. Overclocking is not necessarily an easy thing to do, even when you know the basics of what it is. By some standards, overclocking is an art. Also, the NVIDIA GTX 560 is worthless for you if your program does not benefit from GPGPU computational speed. Programs that take advantage of the parallel computational power of GPUs will run 1 to 20x faster than they do on a CPU. If you know for sure that your program doesn't, then simply pick a low end PCI-Express GPU. I'm sure you can figure all of that out on your own, you are obviously a capable man and more capable than the users here.

6. You're correct about LGA2011 (and the majority do not call it Socket R, even intel officially calls this socket FCLGA2011 in specification, FCLGA standing for "Flip-Chip Land Grid Array") being absent of an integrated GPU. I apologize for my assumption. The mid-range intel socket that is current right now, LGA1155, has a CPU line-up that all contain integrated graphics. There are no longer integrated graphics on the motherboard for any of AMD or intel's consumer class CPU lineups.

I already understand what a "headless" machine is. What struck me and why I replied what I did, was not because you asked if there were motherboards that had integrated graphics but that they would be able to be overclocked. Thanks, for making me look ignorant as to why it would be preferable to use integrated graphics though.
I even said to you:

I don't know why you would want to overclock that though, especially if you will be using a discrete GPU. In computing performance comparison, it would be like pitting a 64-thread server with an Intel Atom.

If the GPU is not being utilized by your simulation/computing program, why would you need to overclock an integrated GPU? Do you think you need to overclock the GPU to get a better display signal or something? A $30 graphics card can do that. You completely misunderstood me (I realize you have no idea about the performance level of a GPU, what you use it for, or what you can do with it as far as compute capabilities), and you tried to make me look stupid while you did so.
In a case that your program did utilize the compute power of your GPU, then you would want to be using discrete as the performance difference would be akin to pitting a 64-thread Intel Xeon server with a server containing a single Intel Atom CPU.

I won't be spending another second giving you advice or input, about anything. I doubt many others will say much either. In fact, I am adding you to my forum ignore list. You should have what you need now, so take it and use it, though I would prefer, and I believe I speak for the majority of members here, just take it and leave. Don't sit and try to prove everyone here wrong with arrogance and spite, when you came here because you weren't sure what you were doing in the first place. You look like an a$$hole.

[l0ud_sil3nc3]

You look like an a$$hole.

the TLDR version

[Leeghoofd]

IF all F1 simulations are that great, really worries me then that some teams are lacking in speed, have a hard time getting a car balanced and set up for a track. That's where the input of the pilot comes in...and engineers can provide : if, but, when and how,... yet that's where they mostly will fail as the pilot has to drive thier simulated setup...

IF AMD's simulations were that reliable then the FX CPU wouldn't be in it's current state for sure... Simulations are simulations, sometimes they work and sometimes they only tell you half of the story...

peace... have a nice stay in your little matrix...

[BeepBeep2]

the TLDR version

ROFL, Did you even read it all? ...of course not. You should read it. XD

IF all F1 simulations are that great, really worries me then that some teams are lacking in speed, have a hard time getting a car balanced and set up for a track. That's where the input of the pilot comes in...and engineers can provide : if, but, when and how,... yet that's where they mostly will fail as the pilot has to drive thier simulated setup...

IF AMD's simulations were that reliable then the FX CPU wouldn't be in it's current state for sure... Simulations are simulations, sometimes they work and sometimes they only tell you half of the story...

peace... have a nice stay in your little matrix...

Exactly, people like this guy engineer, but have no idea how to use the product they can make. (Like him talking about designing and running simulations on a waterblock when it is largely irrelevant in the loop as a whole because it can be less restrictive and give more flow for the radiator or more restrictive and perform better but only if you used a better pump. It is the same principle with the radiator folds per inch and airflow.) ...It is absolutely hilarious, actually.

[SnowyOwl]

Hi !
To help you with your first question... This is the major component of my Catia V5/CFD analysis workstation. I use a corsair H100 on low setting @4.6ghz it's fast stable and quiet. Professional cheap graphic card always a must with these software. High power PSU for low noise & added safety when running 24/7 @100%.



Have a nice day
David

[alpha754293]

oh no. I think that I just lost the whole reply because I clicked the attachment 124846 below. eff eff eff eff efff.....:'(

[alpha754293]

How does that whole auto-save work? Can I restore it? Probably not eh?

Anyways, for those that helped, thank you.

For those that didn't, thanks for nothing.

SnowyOwl
What do you use for your CFD? I'm on mostly Ansys CFX and Fluent, but I'm just about to get started with LS-DYNA.

[Bun-Bun]

I just built a 3930K system using the Asus P9X79 WS mobo cooled by the Noctua NH-D14. No problem overclocking it to 4.5Ghz. In fact I run it 4.6ghz 1.34v 24/7 crunching WCG. It is LinX stable and the Noctua keeps it under 70° while crunching.

I used the corsair 800D case which has lots of room.

For a video card I just picked up a 8400GS for $ 15.

Personally i do not like the all in one liquid cooling kits. Expensive, complicated, and fail to beat the best air coolers.


Oh and after reading this entire thread I must say, as a fellow professional in engineering/technology, I am insulted. You come here asking for help and then rip everyone a new one when you feel it threaten's your intelligence. All that does is prove how not smart you are.

Please, get off your high horse and join us back in reality.