You can help me specify a new computer

sje · Post by **sje** » Fri May 22, 2015 7:54 pm

bob wrote:They also run these up to 18 cores and 45mb L3.

We also have a "phi" board in this box. 61 cores of what is basically an intel pentium processor (in order execution). I have not yet done any testing on that as our tech guys apparently broke the software with an upgrade. But Linux runs on it and it looks like a "computer within the computer." I log into this 20 core box, then ssh to the 61 core box, which seems strange, but seems to work.

More later. There might be some opportunities here for additional shenanigans, such as using the phi cores to do evaluations and such sort of "belle-like"

I read about that 18 core Xeon and I thought it was a misprint. My general assumption is that any multicore Intel CPU where the core count is NOT a power of two in really a chip WITH a core count of a power of two but with some cores disabled. Would Intel sell a chip with 14 of 32 cores turned off?

Things have come a long way since I used a CDC 6500. It too was a multicore machine -- a pair of CDC 6400s in twelve racks of circuitry sharing the same central memory, PPU ring, and operator's console.

Zenmastur · Post by **Zenmastur** » Fri May 22, 2015 10:39 pm

Joost Buijs wrote:Most of the lower powered PSU's are also less well designed because they are meant for an average consumer PC. Often they use 2nd grade Chinese electrolytic capacitors which are certain to fail within a couple of years.

I'm not confusing size with quality, I have an education in electronics so I know exactly what you are talking about.

Anyway let everybody decide for themselves where they want to spend their money on.

I'm not trying to tell you or Steve what to buy. There are some people that are interested in computer chess that don't know much about computers and even less about power supplies. These people may read your post and think that over buying the PSU is a requirement or a "good" choice. It's not!

You're doing a dis-service to the less well informed individuals that come here looking for sound advice by making mis-leading statements or not qualifying what you are saying. e.g.

Joost Buijs wrote: I agree with you that in practice you will never reach the calculated load. On the other hand these 'Platinum' PSU's are so good that even at 10% of their maximum load they still have a higher efficiency than those cheap '80+' labeled PSU's.

Joost Buijs wrote:When you build a machine which cost you >$2k you want to have reliable components in it. The $100 difference in cost between a crappy PSU and a very good one really doesn't matter, at least not to me.

No one said anything about buying "crappy" power supplies.

You overlook the fact that they make good quality power supplies that are 80+Platinum and 80+Gold certified in more than one size. Some people are on tight budgets and even $20 in price will make a difference to them. So, blowing an extra $100 that is in noway required to get either power efficiency or reliability makes little sense to them, nor should it!

So, I want to make it clear to those less well informed that they needn't waste money by over-buying the power supply. You can get a reliable and high performing system without a huge power-supply or the added expense that goes with it.

Regards,

Zen

bob · Post by **bob** » Fri May 22, 2015 11:36 pm

sje wrote:
bob wrote:They also run these up to 18 cores and 45mb L3.

We also have a "phi" board in this box. 61 cores of what is basically an intel pentium processor (in order execution). I have not yet done any testing on that as our tech guys apparently broke the software with an upgrade. But Linux runs on it and it looks like a "computer within the computer." I log into this 20 core box, then ssh to the 61 core box, which seems strange, but seems to work.

More later. There might be some opportunities here for additional shenanigans, such as using the phi cores to do evaluations and such sort of "belle-like"
I read about that 18 core Xeon and I thought it was a misprint. My general assumption is that any multicore Intel CPU where the core count is NOT a power of two in really a chip WITH a core count of a power of two but with some cores disabled. Would Intel sell a chip with 14 of 32 cores turned off?

Things have come a long way since I used a CDC 6500. It too was a multicore machine -- a pair of CDC 6400s in twelve racks of circuitry sharing the same central memory, PPU ring, and operator's console.

Odd numbers of cores have become the norm. Someone here had a 2x18 box they were posting numbers from a couple of months back when the "NPS scaling" discussion was going on. I've seen circuit layouts and it does not appear to be a case of "disabling pieces" (IE like the infamous 486/487, where the 486 was a chip with an intentionally fried FPU, and the 487 "co-processor" simply disabled the 486 and was a full 486 + co-processor.)

And turbo-boost is changing. This machine is a 2600mhz processor, yet with 20 cores running it seems to run at 2900mhz all the time. And it can go faster than that with fewer cores running.

Joost Buijs · Post by **Joost Buijs** » Sat May 23, 2015 7:15 am

Zenmastur wrote: I'm not trying to tell you or Steve what to buy. There are some people that are interested in computer chess that don't know much about computers and even less about power supplies. These people may read your post and think that over buying the PSU is a requirement or a "good" choice. It's not!

I don't think that buying a 750W power supply for a core i7-5960X machine is over buying. It is just a matter of opinion.

Zenmastur wrote: You overlook the fact that they make good quality power supplies that are 80+Platinum and 80+Gold certified in more than one size. Some people are on tight budgets and even $20 in price will make a difference to them. So, blowing an extra $100 that is in noway required to get either power efficiency or reliability makes little sense to them, nor should it!

I'm not overlooking anything.
The Corsair HX-750i is in fact the smallest Platinum certified PSU you can get. I know there are smaller 80+ Gold labeled ones, but you can't get them for $70, at least not here in Europe.

When you are on a budget that is so tight that you can't afford to spend another $20 it is not wise to build or buy a >$2.5k machine, at least that is what we are talking about in this thread.

sje · Post by **sje** » Sat May 23, 2015 1:56 pm

Here's a picture of the specifications label on the power supply in my new machine:

http://images17.newegg.com/is/image/new ... -Z05?$S640$

I bought this CoolMax supply in part because four years ago I bought a 500 W CoolMax supply which has run flawlessly 24/7 for all of that time. In fact, the 500 W supply is installed in the Core i7-2600 box (with four 4 GiB DDR3 DIMMs on an Intel mainboard) which calculated perft(3) over 14 months. As with the new machine, I selected a wattage rating twice what was needed to reduce component-killing heat.

Over the years, I've lost three computer power supplies because the manufacturer didn't allow enough wattage margin or heat removal capability. Two of these were from a Macintosh Plus from the late 1980s and the third was in a BookPC circa 2000. Replacing the Mac supplies was expensive, and replacing the BookPC supply was impossible as it was no longer manufactured.

Makers of consumer computers have little incentive to include power supplies which can handle 24/7 full load duty for much longer than the basic warranty period. For the gamers, you might see models for sale with much more capable supplies -- but also at higher prices. The same is true of the server market.

Joost Buijs · Post by **Joost Buijs** » Sat May 23, 2015 4:10 pm

sje wrote: Over the years, I've lost three computer power supplies because the manufacturer didn't allow enough wattage margin or heat removal capability. Two of these were from a Macintosh Plus from the late 1980s and the third was in a BookPC circa 2000. Replacing the Mac supplies was expensive, and replacing the BookPC supply was impossible as it was no longer manufactured.

Power supplies are the weakest part of a computer setup.

The last 2 years I had 2 power supplies failing, both made by Seasonic, an X-650 that lasted for 2 years and an X-760 that lasted for 3.5 years.
Seasonic is a very reliable company, both PSU's are 'Gold' certified and both have price tag of over 100 euro.
Fortunately Seasonic has a 5 year warranty on their PSU's so it didn't cost me any money to have them replaced.
The second time I asked the retailer to replace it with a Corsair HX-750i and he had no problem with that. The Corsair has a warranty of 7 years and it is possible that the PSU will outlive me.

sje wrote: Makers of consumer computers have little incentive to include power supplies which can handle 24/7 full load duty for much longer than the basic warranty period. For the gamers, you might see models for sale with much more capable supplies -- but also at higher prices. The same is true of the server market.

The average consumer power supply is not capable of running 24h/day for more than 2 years, however for normal use of 8 hours/day they can last for something like 5 years.
Here in the Netherlands the prices for the better power supplies are very steep, somewhere between 170 and 200 euro's, recently the value of the euro went down and the 21% FAT that will be added makes it even more expensive.

I'm very happy that I build my new machine 9 months ago because now it will cost at least 20 to 25% more.

Zenmastur · Post by **Zenmastur** » Sat May 23, 2015 10:06 pm

Joost Buijs wrote:I don't think that buying a 750W power supply for a core i7-5960X machine is over buying. It is just a matter of opinion.

I think there are several points that aren't a matter of opinion. When the maximum power draw is ~365 watts and the maximum sustained power draw is ~240 watts then buying a 750-850 watt power supply is a poor decision. Period! The point to spending additional money to buy a 80+Platinum certified power supply is to use energy efficiently. Buying too large of a power supply is therefore counter-productive since it uses energy less efficiently. This is due to the principle operating point being in an off-design operating region where energy efficiency suffers. This is not a matter of opinion.

Zenmastur wrote:You overlook the fact that they make good quality power supplies that are 80+Platinum and 80+Gold certified in more than one size. Some people are on tight budgets and even $20 in price will make a difference to them. So, blowing an extra $100 that is in noway required to get either power efficiency or reliability makes little sense to them, nor should it!

Joost Buijs wrote:I'm not overlooking anything.
The Corsair HX-750i is in fact the smallest Platinum certified PSU you can get. I know there are smaller 80+ Gold labeled ones, but you can't get them for $70, at least not here in Europe.

I don't live in Europe. Here is a 550 watt 80+ Platinum power supply: http://www.newegg.com/Product/Product.a ... 6817371055

It's currently on sale for $89.99 minus 10% discount and comes with a $30 mail-in rebate. So the total price would be $51 or 46.3 Euro's. I would suggest this or a somewhat smaller 80+ Gold certified power supply will be both less costly and more energy efficient with the given load.

Joost Buijs wrote:When you are on a budget that is so tight that you can't afford to spend another $20 it is not wise to build or buy a >$2.5k machine, at least that is what we are talking about in this thread.

I disagree. I've built lots of high performance systems on shoe-sting budgets. Some people want/need a performance machine but don't have large budgets. I don't see anything un-wise about their needs or their decision to acquire such machines.

Regards,

Zen

Joost Buijs · Post by **Joost Buijs** » Sat May 23, 2015 11:28 pm

Zenmastur wrote: I think there are several points that aren't a matter of opinion. When the maximum power draw is ~365 watts and the maximum sustained power draw is ~240 watts then buying a 750-850 watt power supply is a poor decision. Period! The point to spending additional money to buy a 80+Platinum certified power supply is to use energy efficiently. Buying too large of a power supply is therefore counter-productive since it uses energy less efficiently. This is due to the principle operating point being in an off-design operating region where energy efficiency suffers. This is not a matter of opinion.

Yes that is your opinion. When I build a machine I want to have room to grow, maybe I want to add a few video cards or some extra drives in the future and it would be crazy if I have to buy another power-supply in this case.

Zenmastur wrote:
I don't live in Europe. Here is a 550 watt 80+ Platinum power supply: http://www.newegg.com/Product/Product.a ... 6817371055

It's currently on sale for $89.99 minus 10% discount and comes with a $30 mail-in rebate. So the total price would be $51 or 46.3 Euro's.

Well consider yourself lucky that you live in the US, because this same power supply (which I would never buy because it isn't modular) cost's in my country around 100 euro including shipping.

This whole discussion started around the fact that you stated that a power supply has it's highest efficiency at maximum load, which is not true.
And now you are trying to give it a totally different twist.

sje · Post by **sje** » Tue May 26, 2015 8:32 am

sje wrote:Perft(11) (= 2,097,651,003,696,806) on the new box takes just under five hours: https://dl.dropboxusercontent.com/u/316 ... es/p11.log

As mentioned, I started a perft(12) calculation on the new machine. It began at 2015-05-21 23:56:56.496 UTC and is still cranking, taking much more time than I'd estimated. The machine, running Debian 8 Linux, is responding normally and top reports Symbolic running as expected (48+ GiB RAM, 1600% CPU). My guess is that perft(12) has passed the sweet spot of a 2^31 entry transposition table and now excessive churning is causing the long run time.

I'll give it another day or two. If I have to re-run it, then I'll have the program report all intermediate perft(8) results, about one a minute, as they are generated.

BTW: perft(12) = 62,854,969,236,701,747

sje · Post by **sje** » Tue May 26, 2015 9:21 pm

They say that all things come to he who waits.

The new machine finished the perft(12) calculation in 4 days 18 hours 23 minutes with an answer of 62,854,969,236,701,747 using 72 days of hyperthread calculation time. The colorful logfile: https://dl.dropboxusercontent.com/u/316 ... es/p12.log

Excerpts:

Code: Select all

2015-05-26 08&#58;16&#58;54.553 > ME&#58; Na3 2,101,612,201,748,156
2015-05-26 08&#58;50&#58;28.484 > ME&#58; Nc3 2,731,501,636,365,779
2015-05-26 08&#58;51&#58;41.766 > ME&#58; b3 2,407,514,849,528,875
2015-05-26 08&#58;52&#58;45.183 > ME&#58; b4 2,412,357,918,298,534
2015-05-26 08&#58;52&#58;48.128 > ME&#58; f4 2,050,768,802,609,121
2015-05-26 08&#58;52&#58;52.475 > ME&#58; Nf3 2,704,348,041,301,604
2015-05-26 08&#58;52&#58;58.254 > ME&#58; a4 2,572,564,331,526,038
2015-05-26 08&#58;53&#58;30.930 > ME&#58; c3 2,751,675,948,507,059
2015-05-26 08&#58;54&#58;03.489 > ME&#58; c4 3,119,892,147,087,203
2015-05-26 08&#58;54&#58;40.551 > ME&#58; g3 2,498,600,008,341,437
2015-05-26 08&#58;56&#58;16.263 > ME&#58; g4 2,217,762,743,088,597
2015-05-26 08&#58;57&#58;51.948 > ME&#58; h4 2,620,620,274,642,577
2015-05-26 11&#58;44&#58;03.318 > ME&#58; d3 4,588,998,634,450,632
2015-05-26 12&#58;40&#58;56.391 > ME&#58; h3 1,814,268,178,532,771
2015-05-26 12&#58;52&#58;33.257 > ME&#58; Nh3 2,133,059,306,892,947
2015-05-26 13&#58;16&#58;41.087 > ME&#58; f3 1,552,858,858,446,419
2015-05-26 13&#58;44&#58;48.816 > ME&#58; a3 1,825,396,176,881,632
2015-05-26 18&#58;07&#58;55.112 > ME&#58; d4 6,326,899,070,222,383
2015-05-26 18&#58;10&#58;50.637 > ME&#58; e3 7,160,631,171,539,800
2015-05-26 18&#58;19&#58;58.605 > ME&#58; e4 7,263,638,936,690,183

Code: Select all

2015-05-26 18&#58;20&#58;00.028 > Count&#58; 62,854,969,236,701,747   Pt&#58; 72&#58;00&#58;50&#58;20.460   Wt&#58; 4&#58;18&#58;23&#58;00.199   U&#58; 0.944651   161.586 GHz   6.18867 ps

It looks like the machine checks out okay. Seeing the results appear staggered over ten hours has given me an idea on how to use deep splitting vs split-at-root for much better throughput.

You can help me specify a new computer

Re: Another logfile

Re: Wattage

Re: Another logfile

Re: Wattage

More wattage

Re: More wattage

Re: Wattage

Re: Wattage

Perft(12)

And it's done