AMD's Next-Gen Ryzen CPU Benchmarked Against Intel's Core i7

[Laskos]

Could it be the benchmark was made using only 4 threads for both cpus?

Alvaro

On 16 (hyper-)threads versus 8 (hyper-)threads of i7-7700k, pretty lame.

[Cardoso]

Thanks Kai,
Then it looks AMD's Ryzen is not on par with Intel as I thought it was.

Alvaro

[Modern Times]

On 16 (hyper-)threads versus 8 (hyper-)threads of i7-7700k, pretty lame.

Depends on the price, it is all about performance per dollar and value for money. And of course power consumption and running costs.

[Ras]

The question is, what is the net (!) speedup from 12 to 24 with SF? Is that enough to over-compensate the loss of performance per process? The measure isn't NPS because things are calculated redundantly.

Second, comparing a two-CPU machine against a single CPU machine is misleading at best.

[Ras]

That's why I build up my system with an AMD Phenom-2-x6 1090T (changed later on for a 1100T when I built a second system). Back then, the only Intel chip that offered more performance was the 980X. But while it offered 50% more beat, it would have cost thrice the price, 1000 EUR instead of 300 EUR.

[Milos]

Yes this prototypical, ancient-old hyperthreading is bad for chess argument is getting tiring. If it's so bad then these guys should prove it. Hyperthreading DOES make sense for chess. More performance is more performance. If you could get enough nps to offset the doubling of threads = search inefficiency then you should be fine. I've seen no evidence in any of my games that hyperthreading is bad for chess. Absolutely none. In fact with about a 32 percent increase in nps, I should be gaining a couple of ELO realistically. And I'm talking about 72 logical processors.

It's been explained and proven many times that beyond 4 physical (8 threads) at best hyperthreading doesn't work, but as usual you have no capacity to understand that...
You don't understand what parallelization of alpha-beta means, that there are diminishing returns, what hyperthreading means for chess, LazySMP or any SMP implementation for the matter. You have no clue what is Amdahl's Law nor any interest to figure it out. You only come out with ridiculous claims and statements.

[Milos]

Your system seems to be a bit slower, you should see around 11500.

Probably, the ECC RAM has some additional delay - it's more like a workstation that I got here.

I don't get it. You don't have a server motherboard but use ECC RAM?
Usually reg ECC RAM on server boards is faster than non-ECC, even in case of dual-channel systems.

[Milos]

Could it be the benchmark was made using only 4 threads for both cpus?

Alvaro

On 16 (hyper-)threads versus 8 (hyper-)threads of i7-7700k, pretty lame.

That's really quite lame. On E5-2670 on 8 real cores I get almost 20k Fritz benchmark. With HT on 16 threads I get 25k. And E5-2670 is like 3 generations old CPU.

[jdart]

No idea of the price. But high-end Xeons are crazy expensive. It would be good for Intel to have some competition. I have a dual Opteron server, but the per core performance is way behind even an older Xeon.

-Jon

[Milos]

No idea of the price. But high-end Xeons are crazy expensive. It would be good for Intel to have some competition. I have a dual Opteron server, but the per core performance is way behind even an older Xeon.

Used 3-4 years old Xeons like that E5-2670 are crazy cheap (under 100$ on ebay or aliexpress) and performance-wise, especially for chess, they are great. You can build a dual CPU server for 600-800$.