Intel’s 28-Core Xeon Platinum 8176

[Yar]

Hello !

Not bad from Intel
http://wccftech.com/intel-28-core-xeon- ... enchmarks/

[Vinvin]

Hello !

Not bad from Intel
http://wccftech.com/intel-28-core-xeon- ... enchmarks/

The 8176 runs 28 cores @ 2.1 GHz but the 8168 is probably faster for chess : 24 cores @ 2.7 GHz.

https://en.wikichip.org/wiki/intel/xeon_platinum/8168
https://en.wikichip.org/wiki/intel/xeon_platinum/8176

[Modern Times]

The 8176 runs 28 cores @ 2.1 GHz but the 8168 is probably faster for chess : 24 cores @ 2.7 GHz.


https://en.wikichip.org/wiki/intel/xeon_platinum/8168

TDP of 205 W !!

[Vinvin]

The 8176 runs 28 cores @ 2.1 GHz but the 8168 is probably faster for chess : 24 cores @ 2.7 GHz.


https://en.wikichip.org/wiki/intel/xeon_platinum/8168

TDP of 205 W !!

Code: Select all

Max CPUs	2 (Multiprocessor)

We can put 2 of them in one computer ... 410 W only for CPUs

[jdart]

At over $3000 a processor, that would be a pricey box. Still, the only way to get that core count in one box currently is to go to a quad Xeon configuration and that is even costlier, and currently AFAIK not available in any kind of workstation configuration, just as rack servers.

I think the forthcoming AMD Zen server CPUs also have high core counts, although quite probably not the same level of performance.

I like lots of cores because I can use them to run a lot of processes for testing. But if you are looking at using all cores for one process (like a chess engine) to maximize performance, then a lower core count/higher clock processor may work better.

--Jon

[cma6]

"But if you are looking at using all cores for one process (like a chess engine) to maximize performance, then a lower core count/higher clock processor may work better. "

That has been my experience with the Intel Xeon E5-2686-V3, 2 X 18 = 36 physical cores. Using the system with one process (chess infinite analysis) has been quite disappointing. The time-to-depth, using asmFish, is inferior to that provided by my 6-core Intel 4930k.
Some claim that the dual-Xeon is providing more thorough analysis at a given depth than the Intel 4930k. Perhaps, but even if true, that is little consolation.
With an onslaught of new multi-core chips coming from Intel and AMD, I am hoping that we have TalkChess experts who can find the sweet spot for the # of physical cores that will providee the best infinite analysis.

[Leto]

"But if you are looking at using all cores for one process (like a chess engine) to maximize performance, then a lower core count/higher clock processor may work better. "

That has been my experience with the Intel Xeon E5-2686-V3, 2 X 18 = 36 physical cores. Using the system with one process (chess infinite analysis) has been quite disappointing. The time-to-depth, using asmFish, is inferior to that provided by my 6-core Intel 4930k.
Some claim that the dual-Xeon is providing more thorough analysis at a given depth than the Intel 4930k. Perhaps, but even if true, that is little consolation.
With an onslaught of new multi-core chips coming from Intel and AMD, I am hoping that we have TalkChess experts who can find the sweet spot for the # of physical cores that will providee the best infinite analysis.

It could also be that today's engines don't take full advantage of systems with high core counts because most engine users have less than 8 cores. Perhaps as the number of cores increases over the years more focus will be placed on taking advantage of high core counts.

[Dann Corbit]

"But if you are looking at using all cores for one process (like a chess engine) to maximize performance, then a lower core count/higher clock processor may work better. "

That has been my experience with the Intel Xeon E5-2686-V3, 2 X 18 = 36 physical cores. Using the system with one process (chess infinite analysis) has been quite disappointing. The time-to-depth, using asmFish, is inferior to that provided by my 6-core Intel 4930k.
Some claim that the dual-Xeon is providing more thorough analysis at a given depth than the Intel 4930k. Perhaps, but even if true, that is little consolation.
With an onslaught of new multi-core chips coming from Intel and AMD, I am hoping that we have TalkChess experts who can find the sweet spot for the # of physical cores that will providee the best infinite analysis.

It could also be that today's engines don't take full advantage of systems with high core counts because most engine users have less than 8 cores. Perhaps as the number of cores increases over the years more focus will be placed on taking advantage of high core counts.

Due to SMP loss, you will get less total calculation horsepower using 1 engine with all N cores than with 2 engines at N/2 cores each. However, you will still get a little boost for each additional thread added.

I don't think SMP loss will ever go away completely, so very high core counts will show less and less efficiency as you rise up in the count.

What that means, for instance, is that if you have 1000 EPD records to analyze to some particular depth in plies, you will get done sooner if you analyze 250 records each with four processes, each of which has 1/4 as many cores than if you gave one process all the cores and analyzed the whole pile serially with the one process.

On the other hand, if you just have one position to analyze, you may as well give the process every thread on the machine that can be used (all of them, assuming that the machine is idle) in order to do the best possible job in the allotted time.

[Dann Corbit]

Threadripper machines will be available by July 27th:
https://www.theverge.com/circuitbreaker ... -processor

[syzygy]

Hello !

Not bad from Intel
http://wccftech.com/intel-28-core-xeon- ... enchmarks/

The 8176 runs 28 cores @ 2.1 GHz but the 8168 is probably faster for chess : 24 cores @ 2.7 GHz.

https://en.wikichip.org/wiki/intel/xeon_platinum/8168
https://en.wikichip.org/wiki/intel/xeon_platinum/8176

All-core turbos of 2.7 Ghz for the 8176 and 3.3 Ghz for the 8168. Not bad.

The 24-core chip should be the faster one for essentially any application.