(02-02-2015, 04:35 AM)DatKid20 Wrote: [ -> ] (02-02-2015, 02:57 AM)AnyOldName3 Wrote: [ -> ] (02-01-2015, 06:03 PM)Sweezy Wrote: [ -> ]I also forgot to mention that I do use Photoshop, Blender, and a few other intensive programs. (The reason why I'm using 16gb of Ram.)
I'm assuming you'll want CUDA for Cycles, then. That kind of forces you to go for an nVidia GPU. It's night and day versus CPU rendering, and might well be worth the cost bump to the 980 if you're using it a lot.
Yep. If it's for Blender I would just get a Nvidia GPU instead of waiting for Cycles/AMD to get better OpenCL support. (Intel actually has the best OpenCL support for whatever reason.)
It's mostly an AMD issue rather than a cycles issue that makes OpenCL Cycles crap. If you use an older build of Blender that had both CUDA and OpenCL, OpenCL would work an anything if the scene was small, but would crash on AMD cards with larger scenes, and on anything if the scene was ridiculous, while CUDA worked with slightly bigger scenes and was a little faster (and was a little less buggy, but we'll pretend bugs are easy to fix for a minute). The reason it was crashy on AMD cards is that they didn't support an OpenCL feature/extension (idk which) that allowed global functions, so due to the design of Cycles, it ended up having duplicated executable code in VRAM for every triangle in a mesh. Intel and nVidia GPUs supported global functions in the right way, so OpenCL worked, but it was always faster to render with CUDA on NV hardware, and to use the CPU renderer for Intel hardware, so as OpenCL was harder to write for, development was abandoned, and no-one's ever picked it up again. It might well be that AMD now support the necessary things, or that someone could design another implementation that didn't need it, but OpenCL's now so far behind CUDA and the CPU renderer that no-one with the skill and time wants to do it.
(02-02-2015, 05:05 AM)combine1237 Wrote: [ -> ]http://siliconlottery.com/collections/frontpage/products/4790k48g?variant=1019267195
This site tests, bins, and sells chips with guaranteed overclocks.
$310 USD for a 4790k that will do 4.8 @ 1.325 vcore 1.92 vccin.
Man I lucked out with my 4790k 4.9@1.29 could probably get 5 if I was willing to put voltage up to 1.325 but that's to high for my tastes I'm right on my comfort zone being a longtime AMD user.
(02-02-2015, 06:20 AM)AnyOldName3 Wrote: [ -> ] (02-02-2015, 04:35 AM)DatKid20 Wrote: [ -> ] (02-02-2015, 02:57 AM)AnyOldName3 Wrote: [ -> ] (02-01-2015, 06:03 PM)Sweezy Wrote: [ -> ]I also forgot to mention that I do use Photoshop, Blender, and a few other intensive programs. (The reason why I'm using 16gb of Ram.)
I'm assuming you'll want CUDA for Cycles, then. That kind of forces you to go for an nVidia GPU. It's night and day versus CPU rendering, and might well be worth the cost bump to the 980 if you're using it a lot.
Yep. If it's for Blender I would just get a Nvidia GPU instead of waiting for Cycles/AMD to get better OpenCL support. (Intel actually has the best OpenCL support for whatever reason.)
It's mostly an AMD issue rather than a cycles issue that makes OpenCL Cycles crap. If you use an older build of Blender that had both CUDA and OpenCL, OpenCL would work an anything if the scene was small, but would crash on AMD cards with larger scenes, and on anything if the scene was ridiculous, while CUDA worked with slightly bigger scenes and was a little faster (and was a little less buggy, but we'll pretend bugs are easy to fix for a minute). The reason it was crashy on AMD cards is that they didn't support an OpenCL feature/extension (idk which) that allowed global functions, so due to the design of Cycles, it ended up having duplicated executable code in VRAM for every triangle in a mesh. Intel and nVidia GPUs supported global functions in the right way, so OpenCL worked, but it was always faster to render with CUDA on NV hardware, and to use the CPU renderer for Intel hardware, so as OpenCL was harder to write for, development was abandoned, and no-one's ever picked it up again. It might well be that AMD now support the necessary things, or that someone could design another implementation that didn't need it, but OpenCL's now so far behind CUDA and the CPU renderer that no-one with the skill and time wants to do it.
Well the Omega Drivers seems to have fixed some of the problems but the performance is still behind CUDA and so is Accuracy. AMD is the main one pushing OpenCL yet they don't even support all of the functions correctly.
A little update:
Finally bought all of the parts, I ended up buying the EVGA GTX 980 Superclocked and went with an Asus Maximus VII Hero Z97 to go with my 4790k.
Nice. Good look with your new build. Did you buy a 4k monitor?
(02-05-2015, 01:43 PM)DatKid20 Wrote: [ -> ]Nice. Good look with your new build. Did you buy a 4k monitor?
Yes, I bought
this 4K monitor. Pretty nice for the price.
I'm so jealous of all you people with your tax refunds going towards PCs(assuming that the timing isn't just coincidence lol).
Mine is getting blown on stuff that I actually need....and vape shit lol
That's a sexy monitor though.
People can be financing it too FYI if they buy from certain places and have a certain kind of account, they can get it on a credit line and have X amount of months to pay it off without interest
(02-05-2015, 06:58 PM)BONKERS Wrote: [ -> ]People can be financing it too FYI if they buy from certain places and have a certain kind of account, they can get it on a credit line and have X amount of months to pay it off without interest
I know, but this is the time of year that new PCs get built the most. I mean, Micro Center has named February "Build your own PC month" for a reason
