TCEC has made hardware downgrade

[Jouni]

Previously cores: 88 physical / 176 threads - now "only" 104 threads . Has noobpwnftw stopped support maybe?

[FredFergusonPM]

Previously cores: 88 physical / 176 threads - now "only" 104 threads . Has noobpwnftw stopped support maybe?

Those threads are 20% faster, so no.

[Raphexon]

It's unlikely to be a downgrade.

The increase in single threaded performance should make up for the loss in threads, even with lower total nps.

[Jouni]

In superfinal 21 SF speed was 200 Mnps now 100 Mnps. This is downgrade!

[AndrewGrant]

I would rather have 100 cores at 2mnps each, than 200 cores at 1mnps each. It could be a downgrade, but its not clear whether it is or not. And pointing to the NPS metric is not a way to confirm or refute the change.

[Raphexon]

In superfinal 21 SF speed was 200 Mnps now 100 Mnps. This is downgrade!

It's not.

LazySMP is lossy in the sense that 100 mnps from 100 threads is weaker than 100 mnps from 1 threads.

LazySMP changes the gametree.
If you visualize the gametree with 1 node to 1 billion (or whatever) nodes, it's a pretty narrow tree.
If you visualize the gametree still with 1 billion nodes but with 100 threads it becomes a very wide tree.

Since you can't know in advance the best moves, you can't just have all the other threads search the best nodes before previous ones are looked at.
So you go wider, and look at less promising nodes instead.

This has the advantage of having somewhat linearish NPS gains. But strength gains won't be 1:1 compared to a straight single core speedup.

Even if you assume LazySMP is 90-95% efficient. The TCEC setup is stronger now than it was before.

[Modern Times]

What exactly is the new hardware ?

[Raphexon]

What exactly is the new hardware ?

Current:

2x Xeon 6230R (52 cores/104 threads) 2.1 GHz | 2xA100-PCIE-40GB | 256GiB total RAM (96GiB per engine), 2TB SSD (3-6-piece Syzygy and 6-piece Scorpio bitbases) | OS: CentOS Stream release 8 | All engines in the same machine

Previously it was:

CPUs: 4 x Intel Xeon 4xE5-4669v4
Cores: 88 physical / 176 threads
RAM: 128 GB DDR4 (available to engines)
RAM: 1 TB (available to 6-pieces Syzygy)
HDD: 7 TB total
OS: CentOS Linux release 7.7.1908 (Core)

and

GPUs: 4x V100
CPU: Intel(R) Xeon(R) Platinum 8163 CPU @ 2.50GHz, 32 vcores
RAM: 48GiB (available to engines)
RAM 128GiB (RAM unused by engines is used for caching tablebase files)
SSD: 500GB
6-piece Syzygy and 6-piece Scorpio bitbases
OS: Ubuntu 18.04.4 LTS

[AndrewGrant]

What exactly is the new hardware ?

Current:

2x Xeon 6230R (52 cores/104 threads) 2.1 GHz | 2xA100-PCIE-40GB | 256GiB total RAM (96GiB per engine), 2TB SSD (3-6-piece Syzygy and 6-piece Scorpio bitbases) | OS: CentOS Stream release 8 | All engines in the same machine

Previously it was:

CPUs: 4 x Intel Xeon 4xE5-4669v4
Cores: 88 physical / 176 threads
RAM: 128 GB DDR4 (available to engines)
RAM: 1 TB (available to 6-pieces Syzygy)
HDD: 7 TB total
OS: CentOS Linux release 7.7.1908 (Core)

and

GPUs: 4x V100
CPU: Intel(R) Xeon(R) Platinum 8163 CPU @ 2.50GHz, 32 vcores
RAM: 48GiB (available to engines)
RAM 128GiB (RAM unused by engines is used for caching tablebase files)
SSD: 500GB
6-piece Syzygy and 6-piece Scorpio bitbases
OS: Ubuntu 18.04.4 LTS

Ill be impressed when people finally abandon these shitty Intel servers that have not improved or gone done in price in a decade.

[MikeB]

Can anyone ballpark the estimated cost? (just curious, certainly not buying that setup)