• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Anyone know when Kepler support will be working?

Joined
Sep 8, 2005
Messages
580
I didn't see any other threads on this on [H], I recently installed a GTX680 and after doing some research I found out Kepler doesn't work yet (my 460 is working fine still).... Any word on when the client will be updated?

Thanks,
-SR
 
umm.. given the rate GP works and the fact that nvidia completely changed their shader architecture to be a lot more similar to AMD's older shader architecture i'd say 1-2 months before theres an alpha core and 2-3 months for beta and final release core. just hope and pray they don't switch to openCL with the gtx 680.
 
Last edited:
I asked in the DAB, we will see what kind of answer I get.
 
umm.. given the rate GP works and the fact that nvidia completely changed their shader architecture to be a lot more similar to AMD's older shader architecture i'd say 1-2 months before theres an alpha core and 2-3 months for beta and final release core. just hope and pray they don't switch to openCL with the gtx 680.

They really didnt change the architecture much at all. They essentially just doubled the shaders and dropped the clock by half (and then added some more per generational upgrade.) They are the same shaders as fermi...
 
I thought that was their long term goal to switch it all to openCL or was this only for AMD cards?

it probably is their goal eventually but openCL cuda isn't really that good and it would be a bad idea for nvidia to let F@H switch to openCL only.

They really didnt change the architecture much at all. They essentially just doubled the shaders and dropped the clock by half (and then added some more per generational upgrade.) They are the same shaders as fermi...

switching to a 192 shader cluster and dropping the shaders 1:1 to the GPU, sure they make not have changed the underlying architecture of the shader but they removed the one advantage in GPGPU they had and that was the high clock speeds. if the gtx 680 out performs the gtx 580, some one better rip that core apart and figure out what they did, thats all i'm going to say because theres no way in heck the 680 should out perform the 580 in F@H.
 
Last edited:
Probably when no one cares anymore because the 8 series will be out. They aren't exactly known for their speed.
 
I got a responce back on my question in the DAB.

They are currently working on it, but do not have an ETA.

I will ask them to give us a heads up when they get close to release.
 
They really didnt change the architecture much at all. They essentially just doubled the shaders and dropped the clock by half (and then added some more per generational upgrade.) They are the same shaders as fermi...

Um no... The SP are much simpler going to an amd 5k approach.
They tripled the SP count ,
locked the shaders to core clock,
increased the core clock by 300mhz 40%...

overall per spec if they SP weren't stupid it should be multiples faster...

I got a response back on my question in the DAB.

They are currently working on it, but do not have an ETA.

They never give out ETAs... and they seem proud of it...
 
Read the anandtech review, it explains it pretty well. There are more shaders but at quite a bit lower clockspeed than previously so it takes quite a bit more.

Explained pretty well here: http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/2

They are definitely not VLIW shaders. They did simplify the front end, it has no hardware scheduling capabilities (that bit is more like AMD 2K-6K as AMD's VLIW front end depended on the compiler to order instructions optimally as well vs GCN and Fermi which both have hardware scheduling), but really the shaders are very very similar, They just run at a lower clockspeed, but there is more of them. This leaves the future possibility of raising the clockspeed with lots of shaders for an even faster product.
 
The Kepler shaders are still worlds more capable than AMD's VLIW hash engine shaders. Simply halve the core count of a kepler part and you should have a rough point to point comparison in compute power relative to a fermi at the same clockspeed.

The real unknown is the lock of hardware scheduler on the GK104, if Stanford can code around it great however I am afraid the dynamic nature of WU's could leave performance below a 580.

GK110 will be the real compute oriented Kepler and what I am eagerly awaiting. :)
 
Yup, just tweaked the v7 controls to allow the 680 to fold. Wipe the data in the work/# folder first, then in the configure/slots page, select your 680, go to "Extra Slots Options", add a new key, Name "client-type", value "beta". Oh, and add the gpus.txt file from http://folding.typepad.com/news/201...gpus-such-as-kepler-in-the-v7-fah-client.html of course before you start the client up again.

Estimate PPD's are kinda sucky for the 680 right now, about 12k, which is about where my 560Ti is. However, about 10C cooler.
 
Estimate PPD's are kinda sucky for the 680 right now, about 12k, which is about where my 560Ti is. However, about 10C cooler.

wow....

Either divers better get better quick or NV isn't looking good
 
The Kepler shaders are still worlds more capable than AMD's VLIW hash engine shaders.

Estimate PPD's are kinda sucky for the 680 right now, about 12k, which is about where my 560Ti is. However, about 10C cooler.

huh yeah slightly better than VLIW 9-10k off a 5870 at 850...
 
Part of the problem might be that the client is still reading the 680 as a 560 (GPU:0::GF114 [GEForce GTX 680])

I'm debating pulling the 560Ti out of the case, just to see if it's mucking up the 680's numbers. Plus I'm tired of my UPS bitching because I'm drawing over 600 watts! :)
 
huh yeah slightly better than VLIW 9-10k off a 5870 at 850...

The 680's current performance is not the fault of the shaders, without a hardware scheduler it requires specifically optimized client software to be effective.
 
Back
Top