Read a very interesting Recent post on the Neuromorphic Computer Chip (Intel) and it got me thinking on our current state of Affairs with all this Talk of GPU's vs CPU's etc...did soem basic searching and although way overdue many believe this Neuromorphic approach is the way to go because it is more efficient and uses MUCH LESS Energy than Parallel Systems
I always built my own systems and although it has been a few years since powerful Smartphones and light fast and powerful Laptops are available (a few months back I looked at some micro Rasberry Pi systems) and easy to use so I Figured it would be interesting to take a look and went to look at a Few Biostar motherboards on the net...PCI technology and the SDDR (Flash) systems are totally different....Went on YouTube and.. Wow.. shock on how people are constructing Desktops these days...seems cases are passe as the GPU,s can't fit on the motherboards....systems so strong you might need your own POWER STATION and They Run so hot you need a Freezer and/or a Radiator to keep them from burning your house down...Multiple GPU's that they hang on "Cooling Racks" (Mining Systems?).. And the amount of money you would need to sustain such systems? Well you better have your own Printer to make the stuff...Saw cases as big as refrigerators so you need a large room or space for especially if you also have a massive LCD screen...This Stuff COSTS...but the motherboard is less than the Video (GPU) cards! And the power supplies/cords..Wow...Hate to see my power bills! The speed attained with this equipment is indeed impressive (Great for LC0) But This can't be sustainable in the long term..Let me know your opinion on this important issue. Thx AR
Ozymandias wrote: ↑Thu Jul 18, 2019 12:14 pm
What is it you're interested in computing?
Well.....It was going to be a Desktop with a case and a add on GPU card (inside) that would be used for General purpose stuff and Chess Engine Tourneys...Not too big and not too expensive...total investment (under 600.00) LC0 Ready....
Getting good HW for Lc0 in that spending bracket is though, it will mostly depend on whether you can find a good deal in one of the RTX cards that will be replaced by the new "Super" line. Most specifically the 2070, given the price goal. I've seen it around here, temporally, at a very tempting 269€. Case not included, I could quickly setup a sub-300 configuration at newegg:
Ozymandias wrote: ↑Fri Jul 19, 2019 5:30 pm
Getting good HW for Lc0 in that spending bracket is though, it will mostly depend on whether you can find a good deal in one of the RTX cards that will be replaced by the new "Super" line. Most specifically the 2070, given the price goal. I've seen it around here, temporally, at a very tempting 269€. Case not included, I could quickly setup a sub-300 configuration at newegg:
Checked Ebay and 8 Gb Graphics (Nvidia) Cards are running from 280-600 dollars and that makes the card more expensive than the entire system! Thats Crazy!!!...
supersharp77 wrote: ↑Sat Jul 20, 2019 7:49 amChecked Ebay and 8 Gb Graphics (Nvidia) Cards are running from 280-600 dollars and that makes the card more expensive than the entire system! Thats Crazy!!!...
4 gb cards are not too bad..some under 100.00
If you're interested in Lc0 performance, the GPU is where you want your money at. The rest of the system is only there for you to even be able to access that computing power.
The size of the graphics RAM isn't the metric to look at, but the nps you can get with different graphics chips, and the sweet spot is currently between the 2060 and 2070. Depending on what kind of deal you can find for the latter, I'd go with that one. The RTX 2060 isn't going to be superseded by any of the new "Super" arrivals, so don't expect the price to change for the time being.
Yes, Leela isn't going anywhere, and it doesn't need CPU, so you basically can spend most of the budget in fastest GPU, and use the rest in whatever is capable of using that GPU (and the RAM Leela requires.)
Your beliefs create your reality, so be careful what you wish for.
Also, if you are doing analysis instead of using your engines to play other people's engines, a 2060 instead of a more powerful RTX card is more than good enough for most people.
Even a 2060 will use up all of your RAM fairly quickly, so unless you analyze many positions, extra power may be overkill.
If you are playing games, though, the extra power of the more expensive cards does help.
