Superposition on Linux and Pop!_OS did not work it was looking for qt support and I didnt want to mess up the llama.cpp install or the HPL install to put qt on, so I didnt run it
Thanks for the video! I thought that TOPS is acronym for " Timesharing OPerating System" for DEC PDP-10? What's next, reading VMS as Vendor Management System? Oh, we are already there … 🙄
llama.cpp is tuned to choose the optimal cores for the system which you can see in the start to be 6 in your case, there are command line options to override this if you want, of course. A better test would be not to use the chat interface but to have it generate a fixed count of tokens (like 1000 or so) with the stop token ignored.
you might want to study the linux kernel a bit more, it already does that, so that code is unnecessary, but not only unnecessary, the code in the application could be doing more harm than good, as the Intel code also checks things like energy efficiency, a cold core take more energy to power up. This is why Dennis Ritchie of Unix fame always said " Those you do not understand UNIX are bound to re-invent it....badly". The same is true of Linux.
@ If you think the kernel can manage it you can just give a command line flag (“-t 14”) to make it use all the cores. But you will probably not get any performance improvement because of the way the matrix multiplication code distributes operations over the worker threads. Adding in the efficiency cores would slow the entire computation down, not speed it up.
Superposition on Linux and Pop!_OS did not work it was looking for qt support and I didnt want to mess up the llama.cpp install or the HPL install to put qt on, so I didnt run it
Thanks for the video! I thought that TOPS is acronym for " Timesharing OPerating System" for DEC PDP-10?
What's next, reading VMS as Vendor Management System? Oh, we are already there … 🙄
LOL
electric car real world range is another mystery .
My 30 miles per gallon Acura gets 12 miles per gallon. Give me an electric vehicle any day
I was struck by the 90s Waiting for a Motivational Presentation Music. If you thought TOPS was out of date....
20 TOPS = 20 concurrent instances of top.
llama.cpp is tuned to choose the optimal cores for the system which you can see in the start to be 6 in your case, there are command line options to override this if you want, of course. A better test would be not to use the chat interface but to have it generate a fixed count of tokens (like 1000 or so) with the stop token ignored.
you might want to study the linux kernel a bit more, it already does that, so that code is unnecessary, but not only unnecessary, the code in the application could be doing more harm than good, as the Intel code also checks things like energy efficiency, a cold core take more energy to power up. This is why Dennis Ritchie of Unix fame always said " Those you do not understand UNIX are bound to re-invent it....badly". The same is true of Linux.
@ If you think the kernel can manage it you can just give a command line flag (“-t 14”) to make it use all the cores. But you will probably not get any performance improvement because of the way the matrix multiplication code distributes operations over the worker threads. Adding in the efficiency cores would slow the entire computation down, not speed it up.
Thumbnail, I thought I read: TOPS . . . Exposing The Meth!
Could be...
Cosmic and PopOs +++
eVAX units equivelent Vax780 units
I'm spinning my TOP🪅