Upcoming graphics cards for Leela

[Werewolf]

According to this site

http://www.hexus.net/tech/news/graphics ... r-roundup/

new cards from Nvidia are on their way. The speed difference is mentioned with existing cards. Unless my maths is off I calculate the following speedups based on an old 1060 = 1

TITAN RTS = 3.40
TITAN V = 2.96
2080 = 2.46
1080Ti = 2.28
2070 = 1.95
1080 = 1.64
1070 = 1.37
2060 = 1.27
1060 = 1

Obviously the measurements they used may not apply neatly to Leela, but it gives an idea. Looks to me Nvidia are starting to slow down a bit at the top end.

[Werewolf]

Here's some better info, based on what's now released and in terms of TFLOPS, rather than FPS or something silly like that.

Titan RTX (??)
Quadro RTX 6000 - 15.9
Titan V - 14.9 (29.8 half P)
2080 Ti - 13.4
Quadro P6000 - 11.7
1080 Ti - 11.3
Quadro RTX 5000 - 10.6
2080 - 10.1
Quadro P5000 - 9.5
1080 - 8.9
2070 - 7.5
1060 - 4.4

[Dann Corbit]

I just bought a 1080 Ti because you can get them under $700 now.

[Nay Lin Tun]

I just bought a 1080 Ti because you can get them under $700 now.

Well, for AI (Leela ). New cards are advertised as 10x faster ??
https://wccftech.com/nvidia-rtx-2080-ti ... y-tracing/

[Dann Corbit]

I just bought a 1080 Ti because you can get them under $700 now.

Well, for AI (Leela ). New cards are advertised as 10x faster ??
https://wccftech.com/nvidia-rtx-2080-ti ... y-tracing/

If Werewolf's numbers are accurate, then 2080 Ti - 13.4 over 1080 Ti - 11.3: gives:
13.4 / 11.3 = 1.18 times faster. 18%.
Cost of 2080 ti is $1199.00

1199.00 / 650 = 1.84 or 84% more.

I get a lot better mips per dollar.
I also bought 4 smaller cards recently (1060? I forget).
I am going to put those in one of my Unix boxes.

[jkiliani]

Here's some better info, based on what's now released and in terms of TFLOPS, rather than FPS or something silly like that.

Titan RTX (??)
Quadro RTX 6000 - 15.9
Titan V - 14.9 (29.8 half P)
2080 Ti - 13.4
Quadro P6000 - 11.7
1080 Ti - 11.3
Quadro RTX 5000 - 10.6
2080 - 10.1
Quadro P5000 - 9.5
1080 - 8.9
2070 - 7.5
1060 - 4.4

I think this is incorrect. While we don't have actual test benchmarks yet, the RTX lineup shares wiith the Titan V one important advantage over the GTX 10x0 cards: It can utilize fp16 inference fully, which gives a factor of 2 to speed in addition to the TFLOPS you posted. Currently, a Titan V can do around triple the Leela NPS that a 1080Ti can do, and the 2080Ti should be very close to that.

[Milos]

Here's some better info, based on what's now released and in terms of TFLOPS, rather than FPS or something silly like that.

Titan RTX (??)
Quadro RTX 6000 - 15.9
Titan V - 14.9 (29.8 half P)
2080 Ti - 13.4
Quadro P6000 - 11.7
1080 Ti - 11.3
Quadro RTX 5000 - 10.6
2080 - 10.1
Quadro P5000 - 9.5
1080 - 8.9
2070 - 7.5
1060 - 4.4

I think this is incorrect. While we don't have actual test benchmarks yet, the RTX lineup shares wiith the Titan V one important advantage over the GTX 10x0 cards: It can utilize fp16 inference fully, which gives a factor of 2 to speed in addition to the TFLOPS you posted. Currently, a Titan V can do around triple the Leela NPS that a 1080Ti can do, and the 2080Ti should be very close to that.

Why this BS-ing when it is also known that neither 2070, nor 2080, nor 2080Ti will have real FP16, i.e. all 3 will have it as 10xx cards at 1:64 rate?
Only Quadro's and Titan RTX will have actual FP16.

[jkiliani]

Why this BS-ing when it is also known that neither 2070, nor 2080, nor 2080Ti will have real FP16, i.e. all 3 will have it as 10xx cards at 1:64 rate?
Only Quadro's and Titan RTX will have actual FP16.

Someone who actually works for Nvidia driver development has stated the exact opposite of this on Leela Discord: "Rtx 2080ti is almost same as titan v for lc0" (by Ankan, https://chessprogramming.wikispaces.com/Ankan+Banerjee). So if you want to call BS on me, you'd better back that up with actual data.

[Milos]

Why this BS-ing when it is also known that neither 2070, nor 2080, nor 2080Ti will have real FP16, i.e. all 3 will have it as 10xx cards at 1:64 rate?
Only Quadro's and Titan RTX will have actual FP16.

Someone who actually works for Nvidia driver development has stated the exact opposite of this on Leela Discord: "Rtx 2080ti is almost same as titan v for lc0" (by Ankan, https://chessprogramming.wikispaces.com/Ankan+Banerjee). So if you want to call BS on me, you'd better back that up with actual data.

Sorry that I believe more the most reputable graphics and hardware in general test and review website than hearsay of some quasi insider especially when it defies common sense.
https://www.techpowerup.com/gpudb/3305/ ... tx-2080-ti

[Nay Lin Tun]

This looks like detail article. (in short, 2080Ti will perform 2 x AI works than 1080Ti)
http://timdettmers.com/2018/08/21/which ... -learning/