You can help me specify a new computer

[sje]

Initial benchmark results

For BusyFEN, Symbolic on the new 3.0 GHz Core i7-5680X octo machine returns a 37.8709 MHz throughput node frequency; this is nearly twice that of the 20.2188 MHz from a year 2011 3.4 GHz Core i7-2600 quad.

perft(7) 0.922 seconds
perft(8) 11.883 seconds
perft(9) 121.699 seconds
perft(10) 1,414.779 seconds

The i7-5690X is one of the highest performing CPU's till date.

In my experience ASUS mainboards are very stable and reliable, sometimes they can be a little picky about the brand of memory you use.
But in all these years I never had an ASUS mainboard failing on me once.
The only thing that fails sometimes is the power supply unit, that is the weakest part.

Actually it is a technical miracle that it works. Modern CPU's are drawing in access of 100 Ampere through these fragile sockets and the printed circuit wiring.

Back in the late 1980s and using a 16 MHz Motorola 68020, I calculated perft(7) in about 36 hours of CPU time with my old bitboard program Spector. Admittedly, Spector had no transposition table assistance in perft() mode. With my new machine, the calculation takes just under one second.

In the early 2000s, I had Spector calculate perft(10) on a 1.0 GHz 32 bit PowerPC G4; this took about two weeks. The new machine uses only 24 minutes.

----

Of all the perft(7) calculations done so far in the perft(14) project, the one with the highest subtotal is:
[d]r1b1kbnr/pppp1pp1/2n1p3/6q1/4Q3/4P3/PPPP1PPP/RNB1KBNR b KQkq - 2 4[/d]
with a path count of 282,695,441,146. The i7-5680X did the calculation in 34 seconds. The colorful log file for this is at: https://dl.dropboxusercontent.com/u/316 ... /p7max.log

----

While the i7-5680X may be the highest performing consumer CPU, at US$1,050 it may also be the highest priced consumer CPU, at least in recent memory. Despite the price, there is still a great demand, apparently coming from the overclocker and gamer communities who also drive the market for ridiculously powerful video cards.

Upon close examination, the i7-5680x is really a server CPU, like its Xeon near-siblings. That explains its support for eight SATA 3 interfaces, 2133 MHz DDR4 RAM, and 40 PCI 3 lanes -- things not usually needed in consumer computers.

http://ark.intel.com/products/82930/Int ... o-3_50-GHz

Using water cooling, the i7-5680X can be overclocked by about 40%, at least with some mainboards like Asus X99 Deluxe and the Asus X99 Rampage. The case I have doesn't have vents for a dual radiator water cooler and I have doubts that a single radiator model would suffice. I could buy a larger case with the extra vents, but the one I have is already rather large to allow for the 800 W power supply. I use an 800 W supply because long experience has shown me that specifying a supply with at least twice the needed maximum wattage means many years of reliable operation, even when running 24/7. I'm not sure how much of an overclock can be had with air cooling and I'm not in a big rush to find out.

----

I suspect that the main reason that a chip with 2,011 pins can eat 100 Amps is that a couple of hundred of those pins are connected to the power rails to distribute the load.

[sje]

I re-ran perft(7) through perft(10) on the new machine and uploaded the logfile at: https://dl.dropboxusercontent.com/u/316 ... p7-p10.log

[Joost Buijs]

Initial benchmark results
Using water cooling, the i7-5680X can be overclocked by about 40%, at least with some mainboards like Asus X99 Deluxe and the Asus X99 Rampage. The case I have doesn't have vents for a dual radiator water cooler and I have doubts that a single radiator model would suffice. I could buy a larger case with the extra vents, but the one I have is already rather large to allow for the 800 W power supply. I use an 800 W supply because long experience has shown me that specifying a supply with at least twice the needed maximum wattage means many years of reliable operation, even when running 24/7. I'm not sure how much of an overclock can be had with air cooling and I'm not in a big rush to find out.

There are people running it with water cooling at 4.6 GHz.
On air cooling it is not advisable to run it above 4.2 GHz.
This CPU has a tendency to get very hot when you run it above 4 GHz.

I normally run it at 3.6 GHz. Only for tournaments and such I run it at 4.0 or 4.2 GHz.

The PSU I use at the moment is a Corsair HX-750i, 750 Watts with a very high efficiency of 95%.
My old PSU, a Seasonic X-760 was giving some problems lately, since it has a warranty of 5 years I was able to trade it in for the Corsair.

[sje]

I estimated 200 W for the video, 150 W for the CPU, and 50 W for everything else. Doubling this gives 800 W and that's the rating on CoolMax power supply I picked.

The video is probably too much for what I need as the machine will be run 99% of the time via remote access with its monitor turned off.

I have an older Acer brand computer with a 3.1 GHz Athlon II X4 and its CPU fan often revs up, particularly on warmer days. In the summertime, I tell Symbolic to use only three worker threads instead of four so that one CPU core will be mostly unused and the fan won't run so fast and loud.

----

I managed to get the on-board wireless working and connected to my hidden 802.11ac network. This was a non-trivial task on Linux, mostly because of the non-cooperative attitude of Broadcom, the maker of the wireless hardware.

[Joost Buijs]

You are right, usually I also want to have the power supply running at 50% of it's capabilities. It gives you the highest efficiency.

In my main computer with the 5960X I have a mediocre nVidia video card which uses max. 150W. I calculated CPU 150W. Video 150W. and about 75W. for everything else, I use a WD enterprise drive for secondary storage which takes about 20W extra.

For Open-CL I use an old core i7-980X with 3 AMD R290 cards.
It draws ~1KW. from the wall when it runs it with all GPU's fully occupied. You have to feel the heat of the air coming out on the back of the computer when it is running, it is about 90 deg. C.

[Robert Pope]

You are right, usually I also want to have the power supply running at 50% of it's capabilities. It gives you the highest efficiency.

Really? I thought they were supposed to be more efficient the closer you got to full load.

[bob]

I re-ran perft(7) through perft(10) on the new machine and uploaded the logfile at: https://dl.dropboxusercontent.com/u/316 ... p7-p10.log

Here's a number from a node on our new cluster (2x10cores, looks like 2.9ghz sustained:

time=1:01(94%) nodes=5226911714(5.2B) fh1=87% pred=8 nps=85.1M
checks=144.8M qchecks=280.1M fp=1.8B mcp=264.0M reversible=0
LMReductions: 1/118.5M 2/66.2M 3/27.8M 4/1.6M 5/13.9K 6/9
null-move (R): 3/151.1M 4/8.8M 5/172.2K 6/4.8K 7/48
splits=449.4K aborts=81.1K joins=1.8M data=39% tbhits=0

This is Crafty running 20 threads. It is scaling (NPS) almost perfectly (that 94% number). I am going to run some SMP speedup numbers but that will take at least a week to produce the data.

This is far from the fastest thing Intel makes:

model name : Intel(R) Xeon(R) CPU E5-2660 v3 @ 2.60GHz

I ran 20 threads and it seems to be able to overclock itself to about 2.9 and sustain that, which is not so bad. This will be a bad-a$$ cluster when it is assembled, and also has something like 16 petabytes (and growing) of disk space as well, for those with large files...

I let it play a game last night on ICC to see what the numbers would look like, the above is the first search output in the log file...

[Joost Buijs]

You are right, usually I also want to have the power supply running at 50% of it's capabilities. It gives you the highest efficiency.

Really? I thought they were supposed to be more efficient the closer you got to full load.

The efficiency drops on both ends of the curve, most power supplies have a sweet spot somewhere.
When you look at the power supply I use at the moment it's highest efficiency is exactly at 50% of it's maximum load.

http://www.anandtech.com/show/8835/cors ... y-review/4

[sje]

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

[Vinvin]

I estimated 200 W for the video...

I you don't use 3D, the graphic card use only around 20-30 watts.
http://www.hardware.fr/articles/817-3/resultats.html