You can help me specify a new computer

[sje]

Geekbench unit conversion

Very roughly, one Geekbench 3 MT point equals 1.75 KHz on Symbolic's BusyFEN node throughput rate benchmark.

So I expect to see a node frequency of about 40.2 MHz on the new machine. I'll find out for sure in a few days.

BusyFEN base position:
[d]r4rk1/1pp1qppp/p1np1n2/2b1p1B1/2B1P1b1/P1NP1N2/1PP1QPPP/R4RK1 w - - 0 10[/d]

[MikeB]

Geekbench unit conversion

Very roughly, one Geekbench 3 MT point equals 1.75 KHz on Symbolic's BusyFEN node throughput rate benchmark.

So I expect to see a node frequency of about 40.2 MHz on the new machine. I'll find out for sure in a few days.

BusyFEN base position:
[d]r4rk1/1pp1qppp/p1np1n2/2b1p1B1/2B1P1b1/P1NP1N2/1PP1QPPP/R4RK1 w - - 0 10[/d]

Please expand on the concept of node frequency - not sure what you mean, but it sounds intersting.

[sje]

Please expand on the concept of node frequency - not sure what you mean, but it sounds intersting.

Throughput node frequency is the count of position nodes processed through a unit of time. It is the inverse of node period.

For the above, the definition of processed is: For each node, the work done generating the move leading to the node, executing the move, and then retracting the move.

In Symbolic, the execution and retraction means a complete update or regeneration of all position attributes needed for static evaluation, transposition table key(s), tablebase signatures, a few counters, check/double-check status, a target tracker, pinned/frozen bitboards, attack by color bitboards, and complete bitboard attack matrices (from each square and to each square). An instance of the program's Position class is about 2 KiB in length, not counting the lists used to store variable sized history data.

The processing does not include any actual evaluation (aside from a rough material balance), nor does it include transposition table access or move ordering operations.

For Oscar, the definition is much the same except that instead of 64-bit bitboards there are 32-bit piece sets.

More succinctly, the throughput node frequency is an upper bound on what would be seen for a traditional search node frequency. In practice, the search node frequency would be about one half to one quarter of the throughput node frequency.

On the other hand, the perft() bulk counting node frequency is about 15 times the throughput node frequency.

[bob]

Geekbench unit conversion

Very roughly, one Geekbench 3 MT point equals 1.75 KHz on Symbolic's BusyFEN node throughput rate benchmark.

So I expect to see a node frequency of about 40.2 MHz on the new machine. I'll find out for sure in a few days.

BusyFEN base position:
[d]r4rk1/1pp1qppp/p1np1n2/2b1p1B1/2B1P1b1/P1NP1N2/1PP1QPPP/R4RK1 w - - 0 10[/d]

Please expand on the concept of node frequency - not sure what you mean, but it sounds intersting.

Instead of saying 42.0M nps, he uses the terminology 42.0mhz. IE converting nodes to hz 1:1. I think Steven has done that since I've known him. Going WAY back

[melajara]

If you are not in a hurry, I would wait for Q4 with the general introduction of Skylake (6th gen core i7) CPUs from Intel.
This is worth waiting for IMHO as Broadwell is still not deployed to higher clock rates.
So, relatively to current Haswell processors, you will benefit both from a tick and a tock update, and that just for a few months of patience.

[sje]

If you are not in a hurry, I would wait for Q4 with the general introduction of Skylake (6th gen core i7) CPUs from Intel.

I put the order in last week and should get the box in a couple of days.

Intel has been rather slow as of late with delivering significant CPU upgrades. Further, each new chip family also needs a chipset, and the mainboard manufacturers need these to design and and build the boards. Then there's the time for debugging those boards and also for getting Linux to run properly on the new hardware. This takes about eight to twelve months depending on what's changed with the new CPU family.

[sje]

It's here and it can boot the Debian 8 AMD64 Linux installation DVD. I'll have more information later.

----

I learned programming back when a single bit of magnetic core memory cost about US$0.10 in today's money. My new system has 64 GiB RAM which would cost fifty five billion dollars plus a small power plant if it weren't for the advances in memory technology over the past 40+ years.

[sje]

I've never seen an Asus X99 Deluxe without WiFi support here in Europe.
The latest one is X99 v3.1 with USB 3.1, Bluetooth and WiFi support.

Indeed, that is what I have. The board presents three coaxial connectors on the back panel which attach to the supplied external antenna module for WiFi and Bluetooth.

There are 10 USB 3.0 jacks on the back and two USB 3.1 jacks on the front. Also, a couple of USB 2 jacks on the front for a keyboard and a mouse.

There is a Thunderbolt jack, but I believe it requires an optional PCI card to work.

[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

[Joost Buijs]

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.