best performance hints please

[OldChap]

Is core or mem critical at all is it just shaders that gives performance?

Definitive answer needed

[shadowwind]

ive noticed a big difference in the speed of work units with mem and core oced ive tried it
at stock and only oced the shader. then compared it to mem and core and shader and i think its about a 14% gain but to do it right i would need to do it over again when i first compared them back in july things have changed since then.:yepp:

[OldChap]

well for weeks my 98gtx has been at 770,1944,1150...I've just moved it to 850,1944,1200 overnight it was 1999 shaders :)

which stats are best for watching this?

[shadowwind]

i was watching the times of all the wu's and writing them down from start to finish based on the bionic times most units will finish earlier than the bionic time giving so its a waiting game. then the next time you do that work unit you should have the correct time.
say bionic says the work unit is 12 hour you do it in 8 hour so the next time you do that work unit it should say 8 hour. then just average the time out and do the math.:up:
an easier way to do the time is on the gpugrid view tasks it give the times wished i found my way to the gpugrid earlier i could of saved a lot of time.

:rofl:wow i guess i had a lot of time on my hands i must be a geek.:shocked: i even spent time doing it.

[Chumbucket843]

CUda uses the shaders to do all of the computation because they are where the real horspwer is at and they are programmable. the core of the gpu is the ROP's ,fragment shader etc. which are fixed function and dont do anything unless gaming.memory size and bandwidth are very important. Bandwidth is one of the most important gating factors for performance in CUDA. the memory acts like a cache for your gpu and when it misses it has to go through pci express to your main memory which has much less bandwidth.

[DAK1640]

I will confirm shader dependent. I put all my GPU's back to stock shader and note my output decrease...

[shadowwind]

im understanding that. but CUda isn't running at full speed it is only using 16 to 36 to 68 processors out of 216 wouldn't that cause a bottle neck some where down the road that where core and mem come into play i would think..and that why im thinking some cards kick down to 2d because it is crippled from the start i could be wrong but im not a video expert im just an average joe just grasping at straws at this point and time.:shrug:but i do know core and mem cuts time off the work units..

`

[shadowwind]

I will confirm shader dependent. I put all my GPU's back to stock shader and note my output decrease...

i did also but i notice when you oc core and mem it gets a boost all so when shader is oced.

[Chumbucket843]

the memory controller is most likely in the core that might be why.

[shadowwind]

i should spend more time in the video card section of XS but those guys are talking over my head most of the time its in english and in tec talk i got lost in the tec .:eek:

[DAK1640]

Let's get the Vid Card people to enlighten us...:up:

[PoppaGeek]

Best performance tweak I have found is getting the damn thing to work! :shakes:Before 0 ppd, after 15k ppd.:rofl:

[shadowwind]

Best performance tweak I have found is getting the damn thing to work! :shakes:Before 0 ppd, after 15k ppd.:rofl:

i got a good tweak 4 shots of johnny walker and just sit there and look at it and say NOW WHAT.:rofl::rofl:errors what errors:ROTF:

[INFRNL]

I will should have an answer for you tomorrow. Just found out last night that one of my gpus defaulted back to stock for some reason. I was wondering why ppd was lower on that card. I just increased the shader this time. So far it looks like it may be doing just as good with just a shader oc than full oc. Its a couple grand less, but that could be from the wus ran.

My main rig with dual gpus is back to 6.6.36 and 186.18 drivers, but I am still only getting half output for some reason. Will fix that tomorrow with fresh windows install and fresh boinc/driver install. Need to get this rig back on track.

[INFRNL]

I think I have pretty much concluded that raising the core and mem clocks has NO benefits for the grid! I have 1 card at stock minus the shader and its still producing just as much as when i bumped up the core and mem clocks.

Now the question is does bumping up the core and mem clocks raise the temps at all? I do not think it makes much of a difference if any at all, but I will have a looksy.