Sign in with
Sign up | Sign in
Your question
Closed

5.0 Ghz Intel Gulftown, 6 cores, 6 GB RAM, $$$X?

Last response: in Systems
Share
October 15, 2010 5:52:30 PM

I need to ask some expert advice from this section of the forum.

I've been staring at this overclocked 5.0 GHz Intel Gulftown, 6 cores, 6 GB RAM, 1 TB Hard drive, 128 GB SSD.

Before I show a link to it and the current asking price, I am wondering what do you guys think a system like this is worth?

Keep in mind this was built by a company, they have to ship it and offer a warranty, stuff like that.

So, what would you guys be willing to pay to own a system like that?

October 15, 2010 8:26:05 PM

Well, if you're in the market for a gulftown that's 5GHz, I'd probably just build your own system.

Are you buying it and reselling it? If so, I'd probably just use common sense and sell it for more than I paid for lol
Related resources
October 15, 2010 10:32:03 PM

Are you selling it or thinking of buying it?
October 15, 2010 11:26:27 PM

I am trying to get people's input before my company does buy one. I am the "resident expert" (unfortunately for them, ha!) so I figure I would ask the highly reputable people on this site before I do something stupid, like recommend to them to buy a $6300 system if someone on here says: "You can build one for $2000"

It seems like everybody on here can do things cheaper than anyone else who has a price in mind or an estimate for a system build, so this is the perfect place to ask the experts!

October 15, 2010 11:35:52 PM

Rather than posting what looks like an advert here, you might be better off pricing a system up over in the homebuilt section.
October 16, 2010 12:56:56 AM

Oh, sorry, was not meaning it to be!
October 16, 2010 1:02:14 AM

I almost deleted as an advert when I first saw it.
October 16, 2010 3:49:38 AM

Well I am glad you let it stick around.

Does anyone here know how to overclock such a box to 5.0 GHz?

In your opinion, is the (very fast already) Gulftown capable or reaching that speed without melting through the core of the earth? It must get hot as blazes at 5.0 GHz, so the cooling solution they have rigged up must be something else! Could that account for the high price?

I'd like to know peoples' opinions of this. If it is "easy" to do, as some of you might say, I'd sure be interested in how. If not, does the price seem reasonable, given the extreme speed?

Sorry for so many questions. You just don't see 5.0 Ghz being offered anywhere!
October 16, 2010 5:45:37 AM

There is literally no real application for this. Most programs use 2 cores at most, and there's not a significant performance gain once you get past the 4.3ghz threshold in a system regardless of what anyone else tells you. This is a complete waste of cash made for someone who wants a bigger 'e-peen'.

You can get a great Xeon 6 core processor for 1k and change, or if you get a dual CPU motherboard such as the SR-2, get two decent Xeons for 1.3k and change and get much better performance. Overclocking any processor above 3.9 ghz requires tons of testing and planning, plus a good liquid cooling solution, in my opinion (some say you can stably and safely overclock to 4.2 on i7 and Xeons on air, but I wouldn't risk it).

For the price this computer is, you can save a lot of money building it yourself, plus get much more out of it. Only 1TB for 6300 dollars? A waste. If you really want a custom built six core with a fair price, I reccomend DigitalStorm.

Also, since you asked, it's a vapor cooling system, which is a giant pain in the butt, and since the 5ghz overclock is really useless in real life application, for a business, it's completely unnecessary.

Any other questions, feel free to ask.
October 16, 2010 1:26:10 PM

Wow.

I can't believe somebody could be so wrong megazeroexe!

Once you get past 4.3 GHz, there's not a significant performance gain?

How about the one hundred quadrillion data points my company analyzes in a simulation of a tall building they wish to implode safely? These simulations run for about 36 hours in a program called LS-DYNA, even on the 12 threads of a stock Gulftown.

Are you going to tell me that at 5.0 Ghz x 6 cores it won't be any faster than 4.3 GHz?

The time savings over what we have in place would be 36 - [(3.3/5.0) x 36] = about 12 and a half hours.

12.5 hours x the number of architects waiting on the data x the number of managers involved in the project x the number of demolition experts involved at about $2000/hour total salaries = $25,000 in labor.

That's 5 times the price of the system, and that's just one project we're talking about!

A vapor cooling system is a "giant pain in the butt"?? Sounds like you have no experience with them.

Our engineers talked to the company selling them. They've seen videos of it during startup, and during operation. They run for weeks at a time, no problem, and they are covered under an 18 month warranty should anything go wrong.

If there was a way to vote for WORST ANSWER EVER PROVIDED, I would click that button for your reply.
October 16, 2010 1:37:24 PM

1) In order to overclock to 5.0 you will need a multiple rad water cooler, or phase change cooling.

2) Except for bragging rights, you absolutely do not need a gulftown chip at 5.0ghz. The stress your putting that chip through at that speed just isn't worth it. You will most definitely be over recommended specification for voltage with that overclock. Not very practical for business or home use, for 24/7 operation. A much more modest overclock of around 3.6-4.0, will crush anything you throw at.

3) And yes you can build a comparable computer for probably less than 2000.

4) Lots of people here can help you with a build like that if you ask nicely :) 
October 16, 2010 1:52:50 PM

I agree with point 1) -- they are using Vapor Phase Change Cooling in their rig.

I disagree with point 2) -- you are perhaps used to "consumer applications" and not "business applications." The simulations we are running do quadrillions of calculations. 1 quadrillion = 1000 x 1000 x 1000 x 1000 x 1000. There is no way for computers to "crush" this, all of the simulations we do still run for about 36 hours, some much longer. Any speed increase we get has tremendous savings.

As for point 3) -- I don't know how to respond to "probably" based on some of what you wrote prior. If you think a 3.6 GHz machine will "solve the need" then you don't understand what it is we are trying to do, so you "probably" could not build a solution that would work.

This is NOT a gaming machine. This is NOT something we are using to run Microsoft Word faster.

We have a massively-parallel piece of simulation software that can use all 6 cores and task them fully. Even 6 processors running 24x7 take about 36 hours to finish one of our LS-DYNA simulations on a stock Gulftown.

So, I guess my question is still unanswered.

Is there a way to build a 5.0 GHz Gulftown system for under $5000?

If there is, I'd like to hear about it, but so far all I have heard is "I can build a 4.3 GHz system for $2000" with nothing backing it up.

By the way, it's perfectly fine for you guys to say "I can't build a 5.0 GHz system", in which case, you are helping!

If nobody can build one, then we go with the only solution being offered that is out there.

October 16, 2010 2:22:50 PM

1) you didn't fully read or comprehend what I said in point 2. I never said 5.0 isn't faster- What I said was you will be "overvolting" the chip. That means, at 5.0 you will be doing damage to the processor, hence not practical for business, since business computers, especially the one you are proposing, is being used at a pretty high capacity for long periods of time. Your just looking for a failure, and last time I knew, a work machine doesn't work well with a bad processor. Electron migration will be the killer here, you can look that one up. What you should be looking for is a server/workstation class computer, not an overclocked single chip.

And yes again, You can probably, most likely build a comparable, computer for around 2000. That that doesn't mean it comes at 5.0 out of the box. That means you can buy the same parts for around that much.
October 16, 2010 2:47:53 PM

I literally said running two Xeon's at a comparable price would get you much higher performance than a Gulftown; sure, it will be slightly faster at 5.0 compared to 4.3, but two six cores mildly overclocked > an insanely overclocked six core, AND the stress you're putting on the 5.0 chip is ridiculous. Like I said, this absolutely isn't worth it.

If you gave me advice in my other thread saying how overpriced and underpowered it was, you should be able to realize this is the same situation.
October 16, 2010 2:56:57 PM

Well here is your 2000 dollar computer with 12gb of ram capable of reaching 5.0, with the right overclocking know how.



October 16, 2010 3:01:10 PM

Forgot the hard drives, but you can save money if you subtract 6 gigs of ram, i made the build with 12gb. Or for around 2300, with 12gb an ssd and a 1tb drive
October 16, 2010 3:23:50 PM

Depending on what you're doing, a better graphics card for GPU computing can help. I don't know much about business application.

By the way, I didn't mean to sound snarky or anything earlier, no hard feelings :p 
October 16, 2010 3:58:21 PM

It sounds like he's going to be using the cpu for crunching exclusively, but not a 100 percent on that.

October 16, 2010 4:33:55 PM

Quote:
Forgot the hard drives, but you can save money if you subtract 6 gigs of ram, i made the build with 12gb. Or for around 2300, with 12gb an ssd and a 1tb drive


Great post!

RE: 12 GB - I was told getting all 12 GB to work with such a high overclock is nettlesome. It is relatively easy to get 6 GB to behave comparatively speaking.

RE: Overvoltage at 5.0 GHz - Not a problem, the voltage delta is not as severe as you might think. All well within spec, and they have systems installs with several months longevity at 24x7 so far. Nothing longer than that because the technology is relatively new.

RE: complete parts list - I am checking it all out now, THANKS 1000 x 1000!

I'm sure there's quite a bit more to it than that, but I'll let you know how it turns out, whatever the company ultimately decides.
October 16, 2010 4:46:40 PM

Forgot to add, you will have to do some insulating on the mother board for that phase change setup, or you will have condensation problems. You don't want to fry any components. There are tons of guides online to help you out in that regard. I'm sure there are some guru's here too, but I'm not one of them.
October 16, 2010 4:55:15 PM

Thanks for the tip, and yes, I was vaguely aware of stuff like the condensation issues. I have been reading everything I can on here.

Our company may need to buy several of these things, so I'm a bit nervous if they ask me to build them all!

I might request that they buy the one online already made, and spend some money on the parts you listed, and I'll try an build one. If they're happy with the one they buy, I won't try too hard to upstage it. If it turns out to be a dog like everyone is supposing, or not worth the money, then time will tell and I will report this too.

Thanks again for all of your help and good advice.

It's a long shot, but one the company will have to take, either way.
October 16, 2010 4:57:51 PM

LOL another thing... Most cases aren't setup to receive a phase change cooler. You might have to cut a hole in the bottom or the side of the case, with a metal hole saw.
October 16, 2010 6:36:00 PM

I'm good at that, I punch holes in walls with a sledgehammer for a living :wahoo: 
October 16, 2010 9:14:06 PM

Good deal, I would do some research on a good case for your build. The one I pocked was just a good quality case for the money I literally found in two minutes
October 16, 2010 11:07:17 PM

Sorry to break this news to you, but according to the company that has an actual working system at 5.0 GHz, the RAM you mentioned in your build won't work.

You need 2000 MHz or faster RAM. Yours was down around 1600.

You can't get 5.0 GHz with that lower cost RAM.

They mentioned another thing about your build spec that would prevent it from reaching 5.0 GHz but I agreed not to publicize that on here, since I would be giving away all of their components.

I appreciate all of your help with this, but sometimes finding the cheapest parts isn't the way to go. If I would have purchased all of what you listed, I would have not been able to overclock the system, and, naturally, I would have thought I did something wrong. I probably would mess it up even with the right parts, but if there was no way to succeed at the start, I'm really disadvantaged now, aren't I? :D 

I'm going to recommend to my manager that he goes with the http://www.LiquidNitrogenOverclocking.com build for now. They've already got the kinks out of the system, and there's probably no way to anticipate some of the things that could go wrong unless you actually try to put one of these things together. The Liquid Nitrogen folks have done it already.

That's not to say I'd love to overclock a system you've already done and can get to 4.0 Ghz, mine is a toad compared to some of the boxes I've read about here!
October 17, 2010 1:11:48 AM

There wrong, flat out wrong about that. When your overclocking with an unlocked processor, most of the work is done by raising the multiplier. That has zero effect on ram speed.

Another thing is the parts are not cheap... They are high quality parts, just not marked up 400 percent. Asus makes some of the best motherboards, corsair makes one of the best power supply's etc. I put together a solid build that would reach 5.0 if you have adequate cooling. The Ram would not be your limiting factor, they either don't know what their talking about, or they lied to you. And that ram isn't cheap... It is by a good name brand company that has been making ram and other memory products for a long time.

Anyway this might be my last post in this thread, I'm starting to feel like I'm fighting you on every point.

So go believe the guy that wants to sell you a 5000 dollar rig.
October 17, 2010 1:25:10 AM

I think this guy works for that company and is just "advertising" this computer.

They probably thought since this is a forum for computer people that they would be drooling about this computer at a ridiculous price.

I admit that it worked. I clicked on the link he mentioned in his last post to check it out. LOL

October 17, 2010 2:17:00 AM

^ Your probably right

Maybe MM can close or even delete this thread.
October 17, 2010 4:33:17 PM

I was just trying to get a discussion going about a rather high-end system that my company was looking into, and once I pointed out a few flaws in the "parts build" that another user posted (these components would not achieve the very high overclock that was sough) then my character was impugned, the thread was locked, and now everyone who reads it is going to imply I am somehow associated with the company that is building the unit that could possibly satisfy all of the criteria.

(sigh)

Is this how it works on here? Disagree with someone over component issues, and you get branded?

I don't think it was fair of the moderation to lock the thread after I was labeled like that.

I don't think it is fair to the Tom's Hardware readers who might pay $2300 for parts in a list on that thread, and the machine they build won't be able to function as implied by the person who offered his opinion on how to build it.

So what to do now?

I still am no closer to a solution, at work tomorrow I will be asked about what to do, it's outside of my area of expertise, the mod suggested I come here, but now will my post be treated in the same manner as elsewhere?

Why is it that everyone on here immediately jumps to the conclusion that you have ulterior motives when you post a link to the outside world where another extremely fast system was built?

I just don't see why everyone is so skeptical.
October 17, 2010 4:39:00 PM

It depends on where you post such links as I said in the PM.
October 17, 2010 4:42:19 PM

This topic has been moved from the section Deals to section Systems by Mousemonkey
a c 140 å Intel
October 17, 2010 4:42:23 PM

If you are referring to this thread http://www.tomshardware.com/forum/262441-29-computers Then it got locked since you already selected the best answer (std practice here) end of discussion, problem solved and so on. It has nothing to do with links or opinions.
October 17, 2010 4:43:26 PM

This next topics has been merged by Mousemonkey
  • (sigh) Still no answer after having been impugned elsewhere
    October 17, 2010 5:26:44 PM

    sportsfanboy said:
    There wrong, flat out wrong about that. When your overclocking with an unlocked processor, most of the work is done by raising the multiplier. That has zero effect on ram speed.


    Say you overclock to the same speed two different ways.

    1) Just use the Multiplier and you reach 5.0 GHz
    2) Overclock the QPI/Front Side Bus to 5.0 GHz

    When you run your benchmarking applications, you'll always have better performance with option 2).

    The folks at "that other place" (God forbid I use the name of the company that has actually succeeded in doing this) use option #2.

    Having said that, RAM is often the limiting factor in such cases.

    Not all kits of 1600 MHz RAM can reach the speeds that are necessary for a good QPI/FSB overclock. And, for those 1600 MHz kits that are able to, you're teetering on the edge of instability, and long overnight runs of Prime95 will often fail between 18 and 24 hours of continuous testing on systems with 8 threads or more.

    The 2000 MHz Corsair Dominator-GT RAM handles the higher overclocks much better.

    sportsfanboy said:
    The Ram would not be your limiting factor, they either don't know what their talking about, or they lied to you.


    I think I answered this sufficiently above.

    sportsfanboy said:
    Asus makes some of the best motherboards, corsair makes one of the best power supply's etc. I put together a solid build that would reach 5.0 if you have adequate cooling.


    Motherboards can become a "personal preference" over time, and some of us lose our objectivity. Some of the newer ASUS mobos are rather good, but you have to factor RMA reputation into your decision making. EVGA is outstanding in this regard, and so is Gigabyte in dealing with boards that have gone out on you. I haven't heard anything about ASUS either way, so I just don't know.

    Your power supply/cooler recommendations: forget about them! There's no way your "180 watt of heat removal" cooling unit and under-powered power supply were going deliver what was needed to cool 6-cores enough to deal with the heat load of running at 5.0 Ghz. Sorry, this had to be said.
    October 18, 2010 12:57:33 AM

    Silence speaks volumes.
    October 18, 2010 1:38:43 AM

    Quite honestly I find it hard to believe that they reached 5.0ghz using only QPI/FSB when the 980x has an unlocked multiplier. Also asus makes some of the best motherboards in every aspect reliability, features, performance and is just as good as gigabyte or evga. Get 2000mhz or higher ram and be done with it. Its not like it costs significantly more when you're already paying $1000 for a processor. I highly doubt that you will need more than a good quality 650w PSU even for that type of overclock as you are using a GPU that could run on a good quality 400-450w PSU. Believe us when we say the corsair is a great unit. It can withstand loads higher than it is rated without failing. The cooling is essentially the same as what they are using as it is a phase change system.
    October 18, 2010 4:43:33 AM

    Interesting thread here.... :pt1cable: 

    Personally I'm of the opinion that there's really no point to overclocking the 980x to 5.0 ghz. Intel obviously knows theres a market for people who need this kind of processing power and they have products accordingly. Dual CPU Mobo is easily the best answer in terms of performance and probably price.
    October 18, 2010 6:51:07 AM

    anonymousdude said:
    Quite honestly I find it hard to believe that they reached 5.0ghz using only QPI/FSB when the 980x has an unlocked multiplier.


    If you read carefully, you will see "using only" QPI/FSB was never claimed. It was part of an example used to explain how, of the two choices to reach the same final frequency, QPI/FSB will net the best performance results.

    Liquid Nitrogen Overclocking used both QPI/FSB and adjusted the multiplier to get a 5.0 Ghz result.

    anonymousdude said:
    Also asus makes some of the best motherboards in every aspect reliability, features, performance and is just as good as gigabyte or evga.


    The quality of the Asus boards was never disputed. In fact, I said their newer stuff was good. The only discussion was the RMA policy regarding the returning of defective boards. Based on threads posted around the internet, on the experiences of having to RMA a defective board, asus was not on top. Gigabyte and evga had plenty of happy campers who were willing to share the positive experiences of the otherwise dreaded RMA process.

    anonymousdude said:
    Get 2000mhz or higher ram and be done with it.


    Have you read what I wrote about RAM? I've been the advocate for the 2000 Mhz RAM all along. I never intended to get anything else. It was that other dude that was recommending the 1600 stuff.


    anonymousdude said:
    I highly doubt that you will need more than a good quality 650w PSU even for that type of overclock as you are using a GPU that could run on a good quality 400-450w PSU.


    Well, if you were correct, I would agree with you.

    There's just far too much heat at 5.0 Ghz. Each 100 Mhz in addition adds more heat than the prior 100 Mhz added. This thing gets really, really hot! You need a higher capacity cooling solution, and that adds overhead in terms of power needed.

    anonymousdude said:

    Believe us when we say the corsair is a great unit. It can withstand loads higher than it is rated without failing. The cooling is essentially the same as what they are using as it is a phase change system.


    The boxbuilder at Liquid Nitrogen told me he loves Corsair products but the fact was his last build with a 950watt PSU died just 2 weeks into its life cycle and had to be RMA'd. Stuff like this just happens. Corsair took care of that defective PSU very fast, plus they are close to where he lives, which made the exchange that much more expedient.

    Running a PSU at full throttle near the upper bank of its rated power output can cause problems. The most common one is related to the higher signal-to-noise ratio at the upper end of the threshold. You'll get unwanted noise in the output.

    When you use a PSU that is over and above your anticipated worst case load scenario, you don't get unwanted noise.


    By the way, can you tell me please, which Phase Change unit were you talking about ?


    Phase units are built and tuned for a load handling capability (and other things like being able to fit into a case). A unit built like the CRYO-Z which handles the dissipation of up to 180 watts of heat will start to crash when the heat load is higher. And, in this case, the heat load is considerably higher.

    The folks at Liquid Nitrogen Overclocking have used SS Phase, 2 Stage Cascades, and Liquid Nitrogen for several years and know what these units can do.







    You can read the full article about "Buckeye", the builder of the 5.0 GHz Trinity Lightning, here:

    http://www.liquidnitrogenoverclocking.com/news_03.shtml
    October 19, 2010 5:57:07 AM

    If you want get a higher wattage PSU as it can't hurt. The problem with systems like this is that they are charging you for the overclock. You can achieve this yourself as long as you know what you are doing. Another problem lies in the fact that increasing ghz past a certain point will not increase performance to be worth it. Its like going from 2.5ghz to 5ghz. Performance will increase greatly until ~4.0ghz. Once you reach that point it is impractical to push it higher as performance gains will be minimal, you will require a better cooling solution, and you will put too much stress on the chip. I don't even want to think about the voltage they are using to obtain the 5,0ghz oc. I know that it is way too high and has the potential to kill the chip even with the cooling they are using. Also the program you are using is multi threaded in which case you will be better served by a dual socket setup. Two mildly overclocked 6 core xeons or i7's will outperform a single highly clocked gulftown simply because the program to me sounds like it likes more cores rather than higher speeds as it is loading 12 threads and I'm willing to bet it can load more. If you can confirm that it can load more then 12 threads then your best option will be to get dual socket mobo. The best part is that you can reach those overclocks on air if you had to and can easily reach them with water. Two six cores will cost you $2000. Memory will depend on which processor you get as xeon will use ecc memory. mobo another $500 or so. This comes to about $3000-3500 plus case, hdd, etc. $4000 max will get you better performance even if you don't overclock if the program you are using can really load all the threads.
    October 19, 2010 5:00:19 PM

    anonymousdude said:
    If you want get a higher wattage PSU as it can't hurt..


    Agreed, but I'd go one step further based on what was said about the signal/noise ratio. I'd say it was necessary to do so to guarantee there is no interference from the power supply.

    anonymousdude said:
    The problem with systems like this is that they are charging you for the overclock..


    You mean Liquid Nitrogen? I don't think anyone in the computer business does things for charity, do they?

    anonymousdude said:
    You can achieve this yourself as long as you know what you are doing.


    I'd probably be one helluva brain surgeon, if I knew what I was doing. :) 

    The problem is, for this radical overclock, I don't know what I am doing. I've never seen a single stage phase design before. I am still keeping my options open, and waiting for a decision from the management team at the office. I gave them all the caveats about me doing it vs. LNO just selling us one.

    anonymousdude said:
    Another problem lies in the fact that increasing ghz past a certain point will not increase performance to be worth it.


    EVERYONE WHO SAYS THIS IS 100% WRONG AND I WISH YOU WOULD REALIZE SUCH A REMARK IS RIDICULOUS!

    If that was the case, you could say any stock system today would look like a massively-overclocked system from the perspective of 15 years into the past. And systems from 15 years into the past would look like a massive supercomputer from the perspective of 30 years into the past. Line up every 15 year difference in time, month by month, and go through the entire collection of calendars, and what is the result? Every computer ever made, the instant it first came out, would not be worth having, since its "performance gains" wouldn't be worth it from any comparison to a past system that would make it look like an extremely high-performance machine.

    You can't ever say that a system "is too fast", haven't you learned anything from the history of computer evolution?


    anonymousdude said:
    Its like going from 2.5ghz to 5ghz. Performance will increase greatly until ~4.0ghz.Once you reach that point it is impractical to push it higher as performance gains will be minimal


    Based on what facts?

    It seems like the OPPOSITE is true.

    As you remove one bottleneck (the CPU's effective speed) processes that were otherwise CPU-bound are "unclogged." You get better I/O results, even though you didn't do anything to alleviate any of the potentially I/O-bound processes.

    Read the review of this workhouse on the Amazon site:

    http://www.amazon.com/Trinity-Lightning-Vapor-Cooled-I7-980x/product-reviews/B0042R3A2W/ref=sr_1_5_cm_cr_acr_txt?ie=UTF8&showViewpoints=1&qid=1287506159&sr=1-5

    It reads:

    Quote:
    Saw this box at a Liquid Nitrogen Overclocking trade show, end of August 2010. Waited 20 minutes in line to get a seat to test drive it. Couldn't believe the speed! Blitzed through SuperPI, Prime95, and other benchmarks faster than I thought possible, and I overclock boxes to 3.8 and 3.9 Ghz myself. Bought one for myself just this week (10/5) and still can't get over the speed. It is 2.3 times as fast as my Intel Core i7 3.2 Ghz chip, meaning it would be like 7.36 Ghz if it was on that same motherboard and chipset!


    The ratio of the clocks speeds is 5.0/3.2 = 1.5625 but the ratio in that person's performance gain was 2.3. The alleviated bottlenecks resulted in a symbiotically improved overall system. The 5.0 GHz Gulftown behaves like a 7.36 i7, whichever one he has. We don't know what benchmarks the Amazon reviewer ran, but he's a self-proclaimed overclocker, and he mentioned benchmarks any overclocker is familiar with, so we take his word for it that he calculated the performance increase properly.

    anonymousdude said:
    ...you will require a better cooling solution, and you will put too much stress on the chip.


    Liquid Nitrogen has an 18-month warranty on it, if we get it from there. If I do it, there's a 0-month warranty.

    anonymousdude said:
    I don't even want to think about the voltage they are using to obtain the 5,0ghz oc. I know that it is way too high and has the potential to kill the chip even with the cooling they are using.


    You claim you know it is "way too high", but, in fact, you don't know. Why don't you tell us, what is your threshold for "way too high"? Give us an exact voltage. I'll ask them what the voltage is while I wait for your answer. I'll even send it privately to a mod of your own choosing. You do the same. We'll let the mod answer, and tell us if their voltage was higher than yours, or if it was within your limits. That way, you'll know 100% that I didn't just say "it was lower" than whatever voltage you post here publicly.

    anonymousdude said:
    Also the program you are using is multi threaded in which case you will be better served by a dual socket setup. Two mildly overclocked 6 core xeons or i7's will outperform a single highly clocked gulftown simply because the program to me sounds like it likes more cores rather than higher speeds as it is loading 12 threads and I'm willing to bet it can load more.


    The program we have licensed on our site does have a multi-threaded limit, but I forget what it is. We could purchase one with even more multiprocessing capability, but you pay "per core" for these licenses, and they do get expensive. The version of the software we have now is even more expensive than the machine we are thinking about purchasing.

    The problem is, we can't "subdivide" the building implosion simulations across multiple machines, we need to have one virtual machine. Linking up many computers to "act like" one is non-trivial, and the consulting rates to bring in a team to customize a solution for our simulation software would be through the roof.

    So, we have two big costs jumps:

    1) The software licensing would go from $7000 per installed instance to about $12,500 (slight "per-core" discount, but still, more cores = more cost and we already have a stock Gulftown).

    2) The consulting costs for asserting that the two 6-core systems tie in and work with the LS-DYNA 12-core version.

    And one little cost increase:

    3) The price of the second 6-core hardware.

    anonymousdude said:
    If you can confirm that it can load more then 12 threads then your best option will be to get dual socket mobo. The best part is that you can reach those overclocks on air if you had to and can easily reach them with water. Two six cores will cost you $2000. Memory will depend on which processor you get as xeon will use ecc memory. mobo another $500 or so. This comes to about $3000-3500 plus case, hdd, etc. $4000 max will get you better performance even if you don't overclock if the program you are using can really load all the threads.


    So, your solution would be $4000 + ($12500 - $7000) plus the unknown cost for item 2) above, let's call that $1000 (it would be higher I would guess) so $10500.

    Hmmm, that looks like $5500 more.
    October 19, 2010 7:18:20 PM

    Hi everyone
    Just thought I would add in here a few thoughts after I have been following this thread.

    My name is Buckeye or just Bill :) 

    I would like to address the power supply questions here and give some of my thoughts on this.

    My friend over at Corsair, Cpt. Planet or Jake is a very good overclocker and recently finished up this blog on what a high end system using a 4 way SLI setup power use would be. Even though our system does not use this many graphics cards it does point to how much power a system can use.
    http://blog.corsair.com/?p=4264

    The highest overclock that I have gotten on a I7 920 like the one Jake had on his machine is 5.3ghz, but was under LN2. With a 920 CPU QPI/FSB overclocking is pretty much all you can do as it has a locked multi, but I did achive a 253.71 speed on that run.
    http://www.hwbot.org/community/submission/856882_buckey...

    As you increase the speed of a CPU its power draw also goes higher and that becomes critical for a nice stable overclock. Running a PSU at max rated output under these conditions is asking for trouble as you cannot anticapte CPU usage spikes that would push the PSU even harder. Using a higher rated PSU at lesser loads gives you head room for those spikes and keeps the power output nice and clean.

    A simple purchase of a Kill-O-Watt metter will give you an idea on what load your rig is pulling from the wall so its pretty easy to check.

    Using a Dual CPU system is pretty costly in terms of hardware. anonymousdude is correct on some parts of his post, the CPU's needed require a Dual QPI Link and are not the same as a 980x. But your CPU cost is a bit low as they can get very costly for each CPU. Also your RAM requirements double in cost on these systems. Add in what a Dual CPU motherboard , power supplies, hard drives, cooling and a case to put this in and I think your estimate would be rather low.

    Many of the guys like Kingpin who bench these systems use 2x 1,200 watt PSU's to supply power to these power hungry systems loaded with GPUs when benching.

    Can your wall socket handle these loads without popping fuses ?

    Back when I was running AMD 965's at above 6.0ghz it was estimated that power draw for just the CPU was well over 500watts, and the guys running TWKR's under LHe at close to 7.0ghz was in the high 600watt loads and higher.

    If you don't have clean stable power with reserves to handle high CPU loads your overclocks might become unstable which is bad.


    Phase units are made with several factors determined by the build of the unit and can have a different target heat load depending on what it was made for. Older units like teh CRYO-Z have a limited heat load capability, ~180watts. I tested a CRYO-Z on a QX9770 and the CRYO-Z could not handle it even at idle, it would crash the CRYO-Z and bring it into positive temps.

    This unit here that I ran that same QX9770 for over a year at 4.8ghz was able to handle this high heat load and keep it at -30c temps even under load.

    SS PHase unit on the bottom was for the CPU. Top unit was a 500watt load Phase Chiller for GPU's. I never really did run the top chiller as power requirements was pretty high.


    Really bad picture here but this was this beast running back in its day.
    QX9770 @ 4.8ghz, 8x 32gb MTRON PRO SSD Raid 0 with a ARC-1231ML controller which has ~950mb/s bandwidth. This Raid still runs today on my main rig.


    I would like to close with this last comment on these rigs. The target for this build is not a gamer rig, tho it could be by adding higher end GPU's. These could also be added if the target apps needed cuda support for even more horse power.

    I have seen plenty of people using gaming rigs like the 5.0ghz machine tho, including myself backed with a monster SSD Raid 0. Its very sweet to have things run in an instant, games load extremely fast. Yes it could play Crysis on max settings LOL :) 

    Another system we just finished up was based on a I7-870 @ 4.44ghz and the target app was Matlib. The lenght of time to run test code on the clients current machine was 14 hours for 1 instance. Our machine was able to complete the same code in 7 hours and it was able to run 2 instances at the same time. This one machine cut production time by 1/4.

    In our work we do not simply build machines, box them up and out the door they go. We give a huge amount of support to our clients and tune our machines to client apps if needed so the overclocks not only work, but are stable for the client.
    October 20, 2010 5:59:02 AM

    Nice to meet you Buckeye. I do have some questions about some of your rigs.

    What is the voltage used to get the 5.0ghz OC?

    What PSU is in the system?

    What is the system pulling from the wall?

    What were the specs of the system that you i7-870 build replaced?


    @AllanCameron

    I am sorry if I had made a bad first impression. My only intention is to help you make the best decision.

    Quote:
    EVERYONE WHO SAYS THIS IS 100% WRONG AND I WISH YOU WOULD REALIZE SUCH A REMARK IS RIDICULOUS!

    If that was the case, you could say any stock system today would look like a massively-overclocked system from the perspective of 15 years into the past. And systems from 15 years into the past would look like a massive supercomputer from the perspective of 30 years into the past. Line up every 15 year difference in time, month by month, and go through the entire collection of calendars, and what is the result? Every computer ever made, the instant it first came out, would not be worth having, since its "performance gains" wouldn't be worth it from any comparison to a past system that would make it look like an extremely high-performance machine.

    You can't ever say that a system "is too fast", haven't you learned anything from the history of computer evolution?



    I worded my statement very poorly and you are right it is incorrect the way it was worded. In consumer use going past ~4.0ghz yields minimal gains and puts too much stress and heat on the chip to be worth it. This being said it will also depend on the architecture and the ability of the program to use the speed. Throw in number of cores and things just become a mess. I can't say that for your business application as I don't know the effects a higher clock speed will have on it or at what point it becomes pointless to OC it anymore like I know for consumer rigs. What I do know from your posts is that it is multithreaded, so it will respond to more cores.

    The amazon review I will not touch on as neither you or I can dispute his claims without hands on experience with the trinity and his system.

    We come to the voltage issue. My personal preference for 32nm is to not let it get over 1.35v mostly for the sake of stability and temperature. I might push it a bit higher if I ever decide to bench. Intel itself rates it max safe voltage as 1.4v. Look at Page 23

    http://download.intel.com/design/processor/datashts/323...

    Any higher and you risk frying the processor.
    October 20, 2010 4:48:54 PM

    They do offer an 18-month warranty on all of their hardware, parts and labor. Whatever their voltage is set at, if it won't fry the processor for 18 months, I think it's a good setting.
    October 20, 2010 5:18:05 PM

    i would imagine somethign like 1.6 vcore
    October 20, 2010 5:18:20 PM

    Intel lists max safe volagte settings for its CPU's using Intels stock cooler that comes with the processor.

    Feel free to search the web for settings used by many overclockers, its not that hard to do.

    Back in the day when I was doing review work I had no problem showing settings and debating this and that. But I think it would be unfair to clients of ours, our company and myself to just list what we have put a large amount of R&D work and $'s into on a forum so that it could just be copied.

    Many of the myths out there about CPU's frying by going above Intels spec's are just that, myths. Many of the guys doing extreme benching are doing just that, extreme overclocks, so yes they do fry CPU's, GPU's and other things from time to time. But when pushing a CPU at -196c and 7.0ghz with 1.9+ vcore stuff happens.

    When running on a air setup 1.4 vcore is prolly a good idea to not exceed, but going subzero temps things change a lot.

    If your the type of overclocker that simply jacks vcore up to achive your goal then you will in all good time fry cpu's.

    Put a few years into subzero overclocking on many different types of CPU's and setups then report back on that limit of vcore you just spoke about. Not one CPU of mine out of a large number have ever been "fried".
    October 20, 2010 9:47:48 PM

    BuckeyeBH said:


    Feel free to search the web for settings used by many overclockers, its not that hard to do.



    Every single one I found for very high overclocks was over 1.55 V, so I'd have to say whoever said 1.35 was a bit misinformed. From my very limited recent exposure, I am concluding that the voltage is more of a function of the temperature you can hold during full load and how fast you intend to drive. Warmer temps and/or slower speeds almost always correlated to lower voltages, higher voltages were for very cold "cascade phase change" setups, whatever that is.
    October 21, 2010 3:21:00 AM

    BuckeyeBH said:
    Intel lists max safe volagte settings for its CPU's using Intels stock cooler that comes with the processor.

    Feel free to search the web for settings used by many overclockers, its not that hard to do.

    Back in the day when I was doing review work I had no problem showing settings and debating this and that. But I think it would be unfair to clients of ours, our company and myself to just list what we have put a large amount of R&D work and $'s into on a forum so that it could just be copied.

    Many of the myths out there about CPU's frying by going above Intels spec's are just that, myths. Many of the guys doing extreme benching are doing just that, extreme overclocks, so yes they do fry CPU's, GPU's and other things from time to time. But when pushing a CPU at -196c and 7.0ghz with 1.9+ vcore stuff happens.

    When running on a air setup 1.4 vcore is prolly a good idea to not exceed, but going subzero temps things change a lot.

    If your the type of overclocker that simply jacks vcore up to achive your goal then you will in all good time fry cpu's.

    Put a few years into subzero overclocking on many different types of CPU's and setups then report back on that limit of vcore you just spoke about. Not one CPU of mine out of a large number have ever been "fried".


    I realize that the chip at subzero or even on air and water can handle higher voltages. My main concern with the higher voltage is that many of the overclockers are only doing benching runs not using it for normal use. Even under your cooling solution 24/7 stability is the only concern I have left as you are very knowledgeable. If it is stable then I have no problem. You are correct I have little experience with subzero overclocking and I don't know what the cpu's can handle. Though I would like to I don't have the time or money to invest in it.

    @AllanCameron

    High voltage=high heat
    Lower voltage=lower heat

    Its as simple as that. The reason people need to overvolt their chips is so that they can achieve a stable high OC. So in a sense the higher voltages were designed for subzero temps as the heat generated will need the subzero temps to even turn on. Some people even undervolt just to save power. Ultimately the final decision is up to your superiors and they should have enough information to make a good decision.
    Related resources
    !