RAM for chess, 16 or 32 GB ?

[Lion]

Hello all,

Currently choosing the material and I was wondering if 16 GB DDR4 RAM at 3000 MHz is good enough for chess or should I go to the more expensive 32 GB ?

The CPU will be either Intel i9 7900x or AMD Threadripper 1950x.

Rgds

[Leo]

Hello all,

Currently choosing the material and I was wondering if 16 GB DDR4 RAM at 3000 MHz is good enough for chess or should I go to the more expensive 32 GB ?

The CPU will be either Intel i9 7900x or AMD Threadripper 1950x.

Rgds

Someone made this statement: "Using the full 16 cores of a dual Xeon machine, 16 to 32 Gb hash still fills fairly quickly."

[Lion]

But is the QTY important for chess ?

[Leo]

But is the QTY important for chess ?

http://rybkaforum.net/cgi-bin/rybkaforu ... ?tid=29501

[Lion]

Thanks for the link.
There is no obvious answer then ?!

I assume I am OK with fast 16 GB ?

rgds

[Nordlandia]

32 GB or more is good idea if you're going to probe 6-men tablebases.

On long time games, syzygy get catched in RAM until it gets full. Once RAM get full, syzygy get stored on secondary storage systems. They are very slow compared to primary RAM.

For example last season's TCEC server had enough RAM to store 6-men syzygy without causing tablebase bottleneck.

[Lion]

32 GB or more is good idea if you're going to probe 6-men tablebases.

On long time games, syzygy get catched in RAM until it gets full. Once RAM get full, syzygy get stored on secondary storage systems. They are very slow compared to primary RAM.

For example last season's TCEC server had enough RAM to store 6-men syzygy without causing tablebase bottleneck.

Thanks for your response.
Is syzygy the "only" reason for chess why a large QTY is good?
What about the Frequency of the RAM ? Is it important to have I.e. A 3600 MHz instead of a 2133 MHz ?

[Morten Lohne]

Depends on how long you want to analyze for. The amount of hash memory doesn't really matter until the engine runs out of memory, at which point the search becomes slower. On my quad-core 4690K using Stockfish, the hash fills at 2-4GB per minute, depending on the position. On your CPU, 16GB of RAM would presumably give you maximum speed for 2-4 minutes, and 32GB would last 4-8 minutes.

Note that the engines can still keep searching after running out of hash memory, but search efficiency will slowly start to decrease. This decrease is stronger in endgame positions.

[Nordlandia]

This decrease is stronger in endgame positions.

Is it triggered by tablebase catching right?

One clever way to prevent 100% RAM usage during long TC games is to probe 5-men during search and set 6-men for adjudication. Latest CuteChess support 6-men adjudication.

[Lion]

Depends on how long you want to analyze for. The amount of hash memory doesn't really matter until the engine runs out of memory, at which point the search becomes slower. On my quad-core 4690K using Stockfish, the hash fills at 2-4GB per minute, depending on the position. On your CPU, 16GB of RAM would presumably give you maximum speed for 2-4 minutes, and 32GB would last 4-8 minutes.

Note that the engines can still keep searching after running out of hash memory, but search efficiency will slowly start to decrease. This decrease is stronger in endgame positions.

By about which magnitude will it affect the playing strength or slow down once the hash is full?