TalkChess.com

Milos wrote: ↑Mon Feb 18, 2019 4:48 am

M ANSARI wrote: ↑Sun Feb 17, 2019 4:30 pm Actually I also have wondered if SLI is a good idea. Of course I know you don't need it for Lc0, but maybe it will consider that the 2 cards are 1 card and thus might be more efficient. Maybe not.

mwyoung wrote: ↑Mon Feb 18, 2019 2:46 am If you are running 2 rtx 2070 GPUs. I would only connect them in SLI. Or any cards that are able to run in SLI.

Gee, how can you guys be so clueless???
SLI is a parallel processing algorithm for computer graphics. Acronym means Scan-Line Interleave. It has absolutely nothing to do with FMA which is exclusively what GPUs do in machine learning inference.

Thanks for connecting those dots Einstein.

Gee, how can you guys be so clueless???

Its okay to be clueless. Forums like ccc can help with experience, thats one of the reasons to be here.

Giving advice while being clueless is human

C.K.

Milos wrote: ↑Mon Feb 18, 2019 4:48 am Gee, how can you guys be so clueless???
SLI is a parallel processing algorithm for computer graphics. Acronym means Scan-Line Interleave.

https://en.wikipedia.org/wiki/Scalable_Link_Interface

mar wrote: ↑Mon Feb 18, 2019 12:38 pm

Milos wrote: ↑Mon Feb 18, 2019 4:48 am Gee, how can you guys be so clueless???
SLI is a parallel processing algorithm for computer graphics. Acronym means Scan-Line Interleave.

https://en.wikipedia.org/wiki/Scalable_Link_Interface

So Mr. Smartass, how about the link to a page regarding SLI and machine learning applications?

Hugo wrote: ↑Mon Feb 18, 2019 11:59 am

Gee, how can you guys be so clueless???

Its okay to be clueless. Forums like ccc can help with experience, thats one of the reasons to be here.

Giving advice while being clueless is human

C.K.

Being clueless is ofc fine. Lacking simple comprehension skills OTOH is not something you'd be able to get much help with on this forum though.

Chessbase has article about building two RTX (2080?) machine and claim: There is one interesting thing to consider: if I had had this machine around the year 2000 it would have been the most powerful computer in the world! Really?

Jouni wrote: ↑Tue Feb 19, 2019 8:11 pm Chessbase has article about building two RTX (2080?) machine and claim: There is one interesting thing to consider: if I had had this machine around the year 2000 it would have been the most powerful computer in the world! Really?

Apparently so. The fastest machine was the Lawrence Livermore National Laboratory with around 5 TFLOPS. A dual 2080 Ti would be several times that.

However, as the years advance the Supercomputers improve fast

Yeah had another look:

A dual 2080 Ti is about 27 TFLOPS (single precision - I assume that's that the Top 500 is measured in?)

In June 2002 there was a massive jump as Japan came out with its mighty Earth Simulator (35 TFLOPS!!) blowing away its nearest competitor. I remember this coming out and being amazed.

By contrast the world's fastest machine today - Summit - runs at 143,500 TFLOPS, which is 4000x faster than the Earth Simulator (or fast multi GPU box).

I hope it does something more useful than calculate how to nuke other countries...

https://www.top500.org/lists/2018/11/

Thx !
This link is quite intresting and helpfull.
Do you know, how to add results of my own configuration to the spreadsheet?
kind regards, C.K.

Greyguy wrote: ↑Sat Feb 16, 2019 6:34 pm You have some alternatives here:
https://docs.google.com/spreadsheets/d/ ... 1508569046