new build 2.0

[Psi*]

The last new build thread I started here led me (kudos to Rob for recommendation at the time) to a WC OC-ed i7-920. And, it is still smokin' hot at ~4.4 GHz.

I have had interest in anudder system for several months. I have been very busy with work (self employed & thus reason why I can even think about an additional system) but have had a little time to keep up with CPUs current/predicted as well as a few benchmarks. I had fully planned to get a i7-975/980x/990x but the 920 runs so well that it is not worth it. Actually any of those processors would out perform the 920 but for many reasons specific to the software I use .. just are not that much of a margin of improvement. Part of that is that the software does not run 12 threads 24x7 as it has to wait on me & pre-/post- routines that are single threaded.

However, an additional machine is a new machine. TO THE POINT, do I build around the i7-990x or wait for the more serious Sandy Bridge? Given the recovery rate of the economy, 6 or 8 real cores at max clock rate for $1K is no problem. Like the i7-920, I am good with buying cheap & OC-ing the heck out of it till something better fits in that socket.

So, what do you know? What is in your xtal ball?

 

[Rob Williams]

First and foremost, the i7-920 has been stable at 4.4GHz since the machine was built? And remained so through all of the brutal computations you've streamed through it? If so, then that's great to hear I feel like I can barely get a 4GHz stable overclock for more than a couple of days, let alone 4.4GHz.

You mentioned that your current workloads aren't exactly able to stress a 12 thread CPU, so is that an option even worth looking at? To me, it seems that the better choice would be to go the same route as before. Purchase an i7-930, overclock the heck out of it and put it to good use just like the i7-920.

If the workloads were seriously multi-threaded, I'd almost half-recommend a Xeon platform instead, since 8-core / 16-thread CPUs are available there, but A) They are expensive and B) They are so expensive that you'd end up getting a lower-clocked model, which is unlikely to overclock to 4.0GHz.

According to Intel's roadmap, we should be seeing the first 4+ core Sandy Bridge CPUs this fall (probably around November), which will certainly include 6-core models and possibly 8-core models. If we <em>do</em> see an 8-core model, it'll likely retail for the same $999 pricetag that all enthusiast parts do. But, if an 8-core isn't available at the same time, it'll probably be months before it <em>is</em> available.

I guess it all boils down to... are you in bad need of another machine <em>now</em> (time is money, after all), or is it going to be more of a luxury? If the former, then I'd still do the same route as mentioned earlier... pick up an i7-930 and work your overclocking magic as before. If you'd rather try to get a better solution down the road and don't mind splurging, the Sandy Bridge six-cores @ the same clock speed will absolutely kill a quad-core i7-930.

But again, whether or not SB's improvements will improve your workloads, and whether or not those extra threads will prove beneficial to you, I'm uncertain.

 

[Optix]

I would think that you're still in good shape if what you do isn't stressing your current CPU. You could simply swap out the slowest part of your system and stay current.

I'd say wait for the socket 2011 motherboards and chips. P67 pretty much made 1366 obsolete in my opinion with an overclocked 2500K for $220 beating out a i7 for $1,000.

 

[DarkStarr]

You could go with an 1155 setup but its up to you, I like my folding machine, i7 2600k @ 4.8Ghz its just awesome. That and Ivy chips are supposed to be dropped for 1155 later

 

[TheCrimsonStar]

*cough*

http://www.newegg.com/Product/Product.aspx?Item=N82E16820227516

 

[Psi*]

*cough*

http://www.newegg.com/Product/Product.aspx?Item=N82E16820227516

BUT it is out of stock == disappoint

Yes this i7-920 is the best system ever. It is the one I built in the old beige 10 y/o case (that was water cooled even then). I do get a lot of flak because it is not black. I have another identical case ready to go ... not black either. These cases are nice in that they are roomy & the metal work is already done. In other words the new additional machine will be water cooled also.

The rants I have had in the past about water cooling which were all about cleaning the dust not to mention new life forms developing in the cooling loop are solved and appear to be solved, in that order. For the dust, cans of compressed air are your friend ... with a vacuum catching what blows out. Do this every month. Who knew?

Inside the loop is distilled water, Water Wetter, and a bactericide. I bought the cheapest 1/2" ID tubing at the time. After a 1 1/2 years the tubing has turned a faint yellow brown but the reservoir water is perfectly clear. The loop has only needed about 4 ounces of water added during this time which I think is incredibly good. In the next build I will probably use separable water connectors at strategic locations so sections of the loop can be easily disconnected & removed for whatever reason.

I can wait and was a much easier decision several months ago when I had no $$. I will probably start getting the necessary parts for the loop. I have read *everything* on the web about cooling system performance. So the specific parts are no question. This would be more of trying out the separable water connections. Needed is a method of assembling spacers between the fans (pushing) & the radiator. This is cheap and has been demonstrated to be a measurable improvement in heat exchange.

I have in a Newegg wish list with the EVGA Classified SR-2 e/w 2 Xeon X5690 hexacore == $3900+ for just those 3 items ... then there is RAM. I like the idea of being able to have 32 GB or 48 GB RAM. Potential work could use that much RAM, but only potential at the moment and it takes more that one project before I make a jump. This amount of RAM still allows for the system to be OC-ed, I think.

Why would I even consider OC-ing a system like that? The software I use is a "package" of single and multi-threaded routines. For some problems the single threaded modules can take a lot of time to finish and I do not have access to the individual routines. Some of these single threaded routines can take hours to run so clock speed is quite important.

Also, not all problems lend themselves to the maximum smoke that using all threads or cores available suggests. In other words, 24 threads might not be that significant of an improvement over the 8 available in the current i7-920. If that were the case ... well ... I would be very sad. :'( Some problems do much better than others tho, but this is the reason to have somewhat less expensive systems versus the mother of all (current) systems.

 

[Kougar]

If it will be running the same workloads, which you've stated can't spawn or fully utilize 12 thread most of the time, then you might as well stick to a Sandy Bridge chip and save yourself some real money. For the cost of a 990X or 975 you might as well go Sandy Bridge, buy a good SSD to pair with it, and spring for a UEFI BIOS type motherboard to take best advantage of the SSD.

Based on my own watercooling experience, Water Wetter is what will cause the brown tinting in the polyvinyl tubing. Not sure if it is the compounds in it or simply the cute neon pink dye itself, but something in it breaks down and stains the tubing. I threw the last of mine in my car where it was intended and quit adding it to watercooling loops, just a little pool or pond rated algaecide & coloring and I've had no problems since.

 

[Psi*]

Good thoughts ... wonder what the memory limit will be for Sandy Bridge. For the i7-920 I have now the RAM cost > $500 for 12 GB. I noticed just yesterday that 24 GB could be plugged in and that will cost ~$335 from 'egg.

Ha! Hadn't occurred to me to *not* use Water Wetter. How long has your loop been running? Any concerns about or any issues about corrosion? Such as in the CPU block?

On other forums I have seen pix of a lot of grunge in the CPU block itself, but I did not have confidence that anything was very well reported about the system(s) that have that problem. I had 2 dual Athlon systems (~10y/o) that were water cooled and they never had *that* issue. Grossly plugged up radiators from dust ... YES!

 

[Kougar]

You can put as much RAM as you can squeeze into the slots, your only limitation is what capacities they manufacture at 16GB is easily possible now...

Half of my loop has been running for two years, had to rework it awhile back so some of it is new. I still have some extremely discolored brown tubing, and it looks even worse next to new, clear tubing, hence why I never take photos of my case. I'm in the middle of transitioning to a new case so I held off on fixing the loop properly.

Never had any corrosion issues so far, but I avoid mixing metal types in my loop to prevent Galvanic reactions ... Water Wetter isn't designed to stop corrosion though anyway.

I do very rarely clean out the blocks if I can see anything in them, but since you use Water Wetter, even if you DON"T use any dyes in your loop I suggest cleaning your blocks next time you service your loop. Dyes in particular like to break down and collect in the CPU waterblock... I've noticed it doesn't seem to occur nearly as much with pond/pool dye though, it's another reason I love the stuff. Never mind that a tiny drop of the stuff is enough for a gallon or two of water...

 

[Psi*]

*now* is not the time to buy anyway. In Q2 the i7-995X is supposed to be released and supposedly for $999. Waiting for people to pay me now anyway.

 

[Kougar]

I've not heard of the 995X... I may be wrong but I think people got confused with the 990X, which Intel just recently released to replace the 980X. Regardless unless the program can make very good use of the additional cores/threads, I'd say just stick to a 2500 or 2600 chip for future builds. Or wait until eight-core processors arrive late this year... LGA1366 is on the way out.

 

[Psi*]

long winded description

I have seen references to both the 990x & 995x ... suspicious but I haven't been watching very close until quite recently.

The software I use does have an option, for more money, of setting up different versions of the same problem via what is described as network distributed computing. The different versions can be run on other network machines running as Solver Servers. This allows for these separate problems (but apart of the base model) to be run simultaneously.

Alternatively each model version runs sequentially. If each version requires 2 hours and there are 16 versions (a current model), then we are warming the office for well over a day. So the more servers that can be setup the quicker the whole model can be finished.

Note that there is performance hit which is model dependent slowing the servers' speed up by using multiple threads. 4 threads do not use 100% of 4 cores ... 8 threads do not use 100% of 4 cores plus hyperthreading. The usage & varies from >95% to ~60%.

To complicate the description, the Solver Servers will only recognize the number of discreeet processors ... so hyperthreading appears to be of no value and in fact the software manufacturer says so & indicates to disable it. However, I have figured out how to setup more than 1 solver server in the same machine ... such as 2. Therefore, the 2nd server can make use of the processors enabled by hyperthreading; actually there is no distinction like all other multithreaded software.

This setup makes total use of multicore CPUs and with hyperthreading enabled. Maybe you remember a thread I started at least a year ago about some kind of system performance degradation (claimed by this software manufacturer) occurring with memory intensive mathematical software saturating the single memory controller of the i7 processors? No one here believed it and I have not been able to measure in anyway a degradation caused by 2 Solver Servers running at the same time in the same box ... not necessarily conclusive.

Soooooooo, my interest in more cores with hyperthreading are that I can setup more solver servers in 1 box, give more threads to a server or decrease & add more servers.

The interest in maximum clock rate is that the above starts after a few single threaded processes.

And, this is why I am interested in Rob's router thread, as there is a lot of large files that get passed back & forth across the network. Everything gets dumped back into the main machine eventually and depending the model ... this can take many minutes. Although, the solver server can take off & start on the next task during this transfer so network thru put may not be that much of an issue. It does look like a hit tho.

 

[Psi*]

Instead of the i7-990X at $1K, what is the next CPU that anyone is aware of OC-ing to as ... high as they get with water.

 

[DarkStarr]

Well.... you could say screw it go SR2 with dual xeon hexcores and water that sucker down or even a pair of quad xeons, it will clock higher than hell under water. Maybe for socket 2011 EVGA will make a SR3?

 

[Psi*]

Decided to take a cold shower & wait a couple of weeks for SB to see what they can do ... OC-ed versus old i7s.

 

[Optix]

SB makes 1366 obsolete if you're overclocking. A $220 chip (2500K) can match a $1,000 last gen i7 chip in gaming and overclocks to around 4.6-4.8 with air. Under water I have read about 5.2 but the voltage is out of this world.

 

[Psi*]

sandy bridge versus which i7?

Those are the claims. At $220 I will have no fear ... crank up the voltage!! Except .... there was some documentation of 1156 sockets literally burning out from over current. THAT could be upsetting

As far as performance comparisons to old i7, this is the most concise. "Sandy Bridge processors are up to 17% faster clock-for-clock compared to Nehalem processors."

Well, that note does cite a reference for that statement to this. Half way down that page is "At the same core frequency,Core i5 2300 has great advantage over Core i5 760 and up to 17%!"

That is not the "old i7s" that I am interested in. Damn! The water in that shower was cold this AM!

 

[Optix]

I'm pretty sure that those issues with the sockets have been resolved but we're talking about 1155, not 1156. No socket issues have cropped up on that platform that I'm aware of. Besides, I'd be more worried about the VRMs popping with massive amounts of voltage instead of the socket.

 

[Psi*]

finally dropped the hammer

I have decided to build an i7-990x /w Asus P6X58D Premium /w 24 GB RAM ... maybe Patriot Viper & haven't decided what brand of RAM just yet. I will make that order next week.

Yesterday the PSU & WC parts were ordered ...

Seasonic 750X &
Switftech Apogee XT from 'egg.

Black Ice Gen Two Extreme 240 radiator,
Feser/Noiseblocker Triebwerk TK-121 NB-Multiframe® 120mm x 55mm Low Speed Fan - 1200 RPM, &
Watercool AB-K maxi PRO Water Tank for Ehiem 1046/1048 from performance-pcs in Florida.

I am re-using another case (beige one identical to the other machine I built a little over a year ago) and a Ehiem 1048 pump. It is so easy to get into $-run away with this stuff. I was drooling over this dual pump for only $300. as well as a bigger (& more $) reservoir. I decided the new fans would be the luxury items.

I also have 1 app of the Indigo Xtreme TIM left from the i7-920 build. This time I will run for a while using Artic Cooling before I switch to I/X ... sort of the recommended process by I/X anyway.

I did get 4 fans as I am upgrading the current i7-920 system. The current fans are some nondescrip type >10 yrs old. Probably adding just a 25 mm spacer to the old ones would have been adequate, but all my tools are still in storage except for a few hand tools.

I will add pictures in the coming days of what I am starting with. The radiator & fans add up to 109 mm thick! But the fans are mounted inside the case with the radiator flush to the outside ... at least that is the current plan just as I did with the 920 box.

I guess I should continue this thread in Modding or Overclocking & Cooling

 

[Doomsday]

i7-990X = Beasty!!