TalkChess.com

Is anyone fired up about this? No. It's been some time since I was a teenager. For me a computer is a tool to get stuff done (or play games). At some point, I buy the fastest computer parts I can afford (or wish to spend money on), build the machine, and then I use it for 5-8 years. I replace it wh...

... But why shouldn't the SF team double, quadruple, and so on the size of the net until no more Elo falls out of it? ... I think it is a trade-off, bigger nets = less nps = shallower AB searches. Elo gain by bigger nets vs. Elo gain by deeper searches, or alike. With more horse-power from vector u...

My take on this:

http://talkchess.com/forum3/viewtopic.p ... 20#p886539

--
Srdja

Missed one about Apple silicon for the serious Pro models in 2021/2022...

- up to 128 core GPU (8 core on M1 with > 2 TFLOPS peak)
- up to 32 high-performance-core CPU

https://www.tomshardware.com/news/apple ... ics-report

--
Srdja

Their paper on MCGS - "Monte-Carlo Graph Search for AlphaZero" is only 702.4 kB

https://arxiv.org/pdf/2012.11045.pdf

--
Srdja

Ah, I missed one rumor:

- Nvidia RTX 30xx Super series refresh on TSMC 7nm in 2021?

https://www.thefpsreview.com/2020/10/09 ... m-process/

Dunno how this chip shortage thingy will play out in all of this.

--
Srdja

############################################################################### ### GPU ### ############################################################################### - Nvidia Lovelace Either the gamer arch split of Nvidia for 2022, or an in between step before Hopper arch, maybe TSMC 5nm. Doub...

Pondering about Camus and his Absurdism, about Baudrillard and his Hyper-
reality, it seems to me that chess is an isle of sanity in an absurd world,
chess per se makes sense, it might be an own world/dimension for itself, but I
guess that is the beauty in it.

--
Srdja

As far as I got it the M1 has 4 vector 128b NEON pipelines per Firestorm core: https://www.anandtech.com/show/16226/apple-silicon-m1-a14-deep-dive/2 can someone clarify how these are used for NNUE? Does SF utilize only one 128b pipeline, or can they be coupled to 512b? -- Srdja According to SF and ...

Followup: - I mixed up NNUE first layer INT16 and further INT8 weights, so the possible 4x inference speedup holds only if we assume 8-bit vector packed math on gpu. - I was not able to implement an efficient 8-bit 0x88 vector-based board representation on pen n paper, hence no 8-bit speedup for mov...