Page 1 of 1

3D XPoint

Posted: Sun Aug 02, 2015 2:16 pm
by Edmund
The announcement of 3D XPoint technology by Intel-Micron last week (http://www.anandtech.com/show/9470/inte ... -than-nand) seems to be very interesting for chess engine developers and users. It is a storage technology in production that has a
lower latency than SSDs but higher than DRAM,
higher endurance than SSDs but lower than DRAM,
lower density than SSDs but higher than DRAM and
more expensive than SSDs, but cheaper than DRAM.
Furthermore alike SSDs but unlike DRAM it doesn't lose information when switching off the power.

I find it groundbreaking for the generation of endgame tablebases. Bit level access, low latency, high endurance and high densities are key here.

For the average chess engine it might help by providing larger hash tables, but here the gain won't be so apparent compared to the slightly slower but also cheaper and larger SSDs. As well as again faster on the fly access to endgame tablebases.

Re: 3D XPoint

Posted: Sun Aug 02, 2015 3:51 pm
by mar
That's really impressive! They claim being 1000x faster than SSD while still persistent.
Thanks for the link Edmund.

Re: 3D XPoint

Posted: Sun Aug 02, 2015 3:51 pm
by stegemma
Edmund wrote:The announcement of 3D XPoint technology by Intel-Micron last week (http://www.anandtech.com/show/9470/inte ... -than-nand) seems to be very interesting for chess engine developers and users. It is a storage technology in production that has a
lower latency than SSDs but higher than DRAM,
higher endurance than SSDs but lower than DRAM,
lower density than SSDs but higher than DRAM and
more expensive than SSDs, but cheaper than DRAM.
Furthermore alike SSDs but unlike DRAM it doesn't lose information when switching off the power.

I find it groundbreaking for the generation of endgame tablebases. Bit level access, low latency, high endurance and high densities are key here.

For the average chess engine it might help by providing larger hash tables, but here the gain won't be so apparent compared to the slightly slower but also cheaper and larger SSDs. As well as again faster on the fly access to endgame tablebases.
It seems that this new technology is more closest to DRAM for read cycles than SSD. This could means that when the technology will evolves it can even substitutes DRAM itself, maybe as a mixed chip, where DRAM will be used as a sort of cache for the XPoint memory. We only have to wait and see what happens...

Re: 3D XPoint

Posted: Mon Aug 03, 2015 1:32 am
by sje
Given all of the advances in hardware, we are the Jetsons.

Given the lack of progress in software, we are the Flintstones.

Re: 3D XPoint

Posted: Tue Aug 04, 2015 1:54 pm
by Daniel Shawul
This will be used in the exa-scale supercomputer that obama announced.

Re: 3D XPoint

Posted: Wed Aug 05, 2015 1:19 am
by Dann Corbit
sje wrote:Given all of the advances in hardware, we are the Jetsons.

Given the lack of progress in software, we are the Flintstones.
A study in the programmer forum showed that the Elo advancement in software for Chess programs is equal to the Elo advancement from hardware advances.