I post from mobile because I post from bed
this shit is cancer
https://files.facepunch.com/forum/upload/1755/62939e23-c952-493b-9ab2-56cc059d88ed/image.png
http://i.heykidwannayiff.com/Taskmgr_2018-09-24_19-59-08.png
I think my desktop needs an exorcism
or more ram
probably just more ram
what am I looking at
Never used Near Share?
is that a bluetooth thing
"Maybe I'll switch back to Chrome for a while..."
https://i.imgur.com/OttaQvE.png
"Oh, maybe not."
microsoft finally made their own AirDrop after 7 years, eh
I put the linebreaks because of the habit of not exceeding 80 chars, thanks for letting me know, I suppose I should just type in long lines and let the browser decide where they should break.
You don't need a copy of the scene for every pixel but duplicating the scene is the only way to parallelize it without the memory becoming a bottleneck since I don't see clock speeds increasing soon.
For every pixel you send one or several rays, with little or no spread. Each of those rays hits a surface where it can spawn more rays with a more random direction the more diffuse the material is. Say you're in a room with no lights except some holes in a wall where sunlight can come in; you're looking at a table that is opposite to this wall yet light coming from it is visible; rays sent from the camera had to bounce quite a bit before finding the hole in the walls.
The only way to parallelize that without making memory a bottleneck is to have a copy of the scene for each unit that works on each block/tile since caches don't work(do they?). 1 unit/px is probably too far though yeah.
heya cipwttkt what's a good 2.5 inch 2tb drive I can stick in my laptop that won't make like a deathstar thx ily
Use a Surface?
That logic doesn't make sense. "Memory bandwidth is precious, so let's copy everything a hundred times and make each unit use a different copy so we have a hundred times as much data to try to squeeze into our already limited cache". What exactly do you think is being gained by duplicating the data? It's a read-only scene (within the context of a single frame). You can have a read-only line in multiple caches - read up on MESI cache-coherency protocols.
The way optimized ray-tracing works is to discard parts of the scene we know the ray can't hit, usually by using some sort of tree structure (the exact details are really complicated and, in Nvidia's case, proprietary). This not only cuts down on the number of intersection checks needed, but makes caching much more effective - only the high-level nodes need to be fully in cache, from there we can swap limited subsets of the scene into cache. Cache hit rate is still pretty low between rays - unlike you claim, the big problem is that successive rays diverge, rather than be parallel - but this at least makes the best possible use of cache within a single raytrace.
https://pcpartpicker.com/product/fxfhP6/western-digital-blue-mobile-2tb-25-5400rpm-internal-hard-drive-wd20spzx
Moving Windows from drive to drive is usually more trouble than it's worth, it's easier to reinstall clean and copy files over from the old drive
Agreed. My Surface has become my "bed" computer. Mostly watching Youtube videos to fall asleep, but I do shitpost before putting on pants some days.
Likewise, my laptop has become my "couch" computer, since the battery is basically dead and lasts maybe five minutes before dying.
I might have too many computers.
Oh I get it, you're thinking I mean copy the same scene to different parts of the same chip and have all the units trying to get it from there and thrashing the caches.
I'm just thinking how would you even optimize something that's dependent on memory (latency more than bandwidth?), so I thought that by having each unit with it's own dedicated memory chip that need for access/bandwidth wouldn't be an issue.
A further optimization of that would be to have writes selecting all chips such that for example, writing value x to address y woul do that for every memory chip yet reads from each units would be independent.
Wouldn't that tree optimization be dependent on limiting the max number of bounces?, since any hole in the geometry means rays have a chance to go in, even if small. Though I guess that'd be such a small contribution to the pixels it would be fine to discard them.
What? Cloning a W10 OS drive works fine.
should I look for a drive with a larger cache or is that dumb if it't not even my boot disc?
Things a damn kilogram, no way
It's much more efficient to have a single wide memory bus shared among all units than multiple narrow buses, each dedicated to one core or execution unit.
Imagine a card with 32 execution units and a 256-bit memory bus (I tried using a real card but the numbers kept going wonky). If you assigned each EU its own memory interface, that would be just 8-bit-wide per EU - long memory reads and writes would take longer, adding to latency, and what happens if that EU is the only one doing work? Your total capacity for throughput is equal, but you'll be wasting much more of it whenever every single EU isn't 100% bottlenecked on memory throughput, or if a single EU needs more in memory than will fit in its fixed space. Far better to have a wide bus going out to memory, and an on-chip bus to transfer data between it and whatever EUs need it - and to the PCIe interface, while we're at it.
wat
It's the lightest device I've ever put on my lap besides my phone. You're going to be hard pressed to find something lighter that isn't a shitty tablet.
Is still 910 grams
My switch is half the weight and has no web browser.
A phone is perfect for using lying down, because you don't have the same amount of lift power in this pose
I guess the Switch weight it irrelevant since it can't browse the web. But FP mobile sucks, and so does phone browsing in bed.
Auto-rotate always fucks up when you're laying down.
Keyboards are garbage for long posts.
The reply box on FP doesn't work with backspace on 99% of keyboards.
Power cords.
Laying on your phone because you fell asleep while shitposting, and then somehow managing to send a GIF of a metal rock dude to everyone on Facebook Messenger that you haven't talked to in 6 years.
Arms fall asleep.
The screen doesn't have an option to go between no back light and luminescence of 10,000 suns.
Cold arms because you can't keep them under the blankets.
Overheating because your arms got cold and you had to cover your head with the blanket so you can see your phone.
There's just so much.
this is the only issue, all the rest aren't my problem
You do know you can lock auto rotate... Right? Right...?
Yeah, but you have to toggle it on and off to have YouTube go fullscreen automatically. Same with a few other video watching instances across apps.
Why does it need to be automatic? just hit the fullscreen button.
I never use rotation lock, I just know how to hold a phone correctly
fuck
time for a bit lot of epoxy.
https://files.facepunch.com/forum/upload/237756/648e52d9-cbc3-4508-9caf-4c62460cc132/IMG_20180925_233035.jpg
Moving between SSD sizes is a pain in the ass, especially if you're going from larger -> smaller.
Yeah if you go small -> big it's easier but still possibly messy if you use GParted
Sorry, you need to Log In to post a reply to this thread.