Nice!
Cleary stronger (out of errorbars) than Stockfish. With bullettime and a valid Leela-Ratio of 1.0.
After years and years of Stockfish dominating the computerchess, finally a new world's number 1.
pohl4711 wrote: ↑Wed Feb 12, 2020 7:03 am
a valid Leela-Ratio of 1.0.
It's not "valid". It's "arbitrary" at best, and probably very heavily in favor of Leela by any reasonable measure. (e.g. if you follow crem's suggestion, you find it gives way too much hardware power to Leela.)
pohl4711 wrote: ↑Wed Feb 12, 2020 7:03 am
a valid Leela-Ratio of 1.0.
It's not "valid". It's "arbitrary" at best, and probably very heavily in favor of Leela by any reasonable measure. (e.g. if you follow crem's suggestion, you find it gives way too much hardware power to Leela.)
One day, Leela with a standard good graphic card will be stronger than Stockfish on any number of core. So it will be 500$ GPU >>>>> 20K $ CPU. So Leela ratio doesn't really matter anymore.
JJJ wrote: ↑Thu Feb 13, 2020 4:57 am
One day, Leela with a standard good graphic card will be stronger than Stockfish on any number of core. So it will be 500$ GPU >>>>> 20K $ CPU.
Maybe. I don't know. I can't really predict such things.
But this article suggests gamers disagree with you. The author is arguing it's not as bad as they think and it "seems like solid progress to us", which is hardly a ringing endorsement of your prediction.
Something we’ve been hearing a lot this year is that PC gaming is not in great shape, how there’s a lack of development and progress on the hardware front... But are things really that bad?
For example... So at least on the CPU front, things have without question never been better.
Memory and storage pricing is also amazing right now, but what about graphics cards? It’s GPUs that we suspect most gamers are focusing on when complaining about the current state of the PC gaming industry.
We've come to terms with the fact that we’re not going to get 20 to 30% performance gains year on year anymore. Some have come to expect those kind of gains no matter what, but we must wonder where they were during the five years of 28nm GPUs? Let's also consider Nvidia lacked real competition for a few years and naturally that didn't help with pricing or performance.
The performance uplift offered by the GeForce RTX 2060 Super over the ~3.5 year old GTX 1070 isn’t overly impressive, but a 35% performance boost isn’t insignificant either. The Radeon 5700 XT boost is more noteworthy, 48% faster on average when compared to the GTX 1070. That’s a very reasonable performance gain after 3 years.
jp wrote: ↑Thu Feb 13, 2020 9:01 amThe performance uplift offered by the GeForce RTX 2060 Super over the ~3.5 year old GTX 1070 isn’t overly impressive, but a 35% performance boost isn’t insignificant either.
I find it hilarious your example of a small performance uplift is the one that is like 200-300% for lc0 and other AI applications
And you are absolutely right that traditional rasterization approaches is reaching a dead end in games, with diminishing returns. That's why there is a sudden focus on new approaches like ray tracing and AI enhancements like smarter upscaling.
jp wrote: ↑Thu Feb 13, 2020 9:01 amThe performance uplift offered by the GeForce RTX 2060 Super over the ~3.5 year old GTX 1070 isn’t overly impressive, but a 35% performance boost isn’t insignificant either.
I find it hilarious your example of a small performance uplift is the one that is like 200-300% for lc0 and other AI applications
Note that is not my example. I am not the author of that article. As I said in my post, I can't predict such things (not that he is forecasting in his article).