#nvidia 16 #pin RTX 4090 #gpu 's are #melting under their 450 #watt #load https://wccftech.com/native-atx-3-0-16-p.. It turns out nobody considered that the #connector would #fail due to #heat , likely created by the #current or #gpu #temperature itself. #heatproof #plastic doesn't even solve this problem either.
Apparently the #eth #crypto switch from #pow to #pos has killed *all* #gpu #mining of the #coin , and the #miners have been largely #unsuccessful switching to other #coins https://www.pcgamer.com/reasons-to-be-ch.. Apparently all it took was 24 hours!
Oh dear, the #us #government have #banned the #export of A100 and H100 #nvidia #gpu #hardware to #russia and #china https://www.sec.gov/Archives/edgar/data/.. This is the beginning of something quite a lot worse.
#intel have just #accelerated their #raytracing #performance on their #gpu 's with a *single line of code* that #allocates #memory #locally , rather than on the #system https://www.phoronix.com/news/Intel-Vulk.. The change increases performance by *100%!*
People performing their own #ram #upgrade on their #gpu just screams that the #industry needs to allow for #replacement #ram to be #installed https://hackaday.com/2022/01/30/gpu-ram-.. For #amd or #nvidia it would also mean only needing to #manufacture one #pcb and #bom too!
Seems like it has never been easier to write #generic #gpu #kernels https://mikeinnes.github.io/2017/08/24/c.. Very powerful stuff!
#intel #disabling #directx 12 on older #cpu #architectures because they somehow managed to fuck up their #embedded #gpu #api https://www.techpowerup.com/288676/intel..
Would you look at that, #apple have #released a new #cpu for their #device line-up: https://www.apple.com/newsroom/2021/10/i.. Apparently most of the #innovation went into more #gpu and more #video #codecs ...
Very, very impressive #hobbyos called #essence https://essence.handmade.network/ It's #opensource and supports a decent number of #features including #disk #audio #gpu #usb #network and much more: https://gitlab.com/nakst/essence The #design is also very nice - an extremely impressive #os !
Nice discussion on whether #halfprecision #floats are worth the effort or not, specifically using the #gpu https://futhark-lang.org/blog/2021-08-05.. Turns out the #speedup really isn't so much. I suspect it *may* be better when you have large vectors, you can literally keep twice as much in #cache and larger models in #ram ... But still, it's worth checking whether you actually get the expected speed-up or not.
This #laptop is getting slower and slower... I suspect in part it could be a #temperature issue - as cooling down the #gpu allowed it to start the other day... I've just manually turned on the fans and now it is already starting to run faster... I suspect this machine is not longed for this world.