Title: . Post by: jjames888 on February 12, 2012, 03:23:47 AM .
Title: Re: Kepler hashing performance Post by: racerguy on February 12, 2012, 06:32:58 PM interesting.
Title: Re: Kepler hashing performance Post by: Kluge on February 12, 2012, 06:42:13 PM Expected by Internet rumor mills to cost $350-400 (ETA: source of first OP article suggests MSRP of $299), TDP of 225W. 6970 costs ~$300-400, TDP of 250W.
... But if ATI loses at Bitcoin mining.... what else is left for it? Will be interesting to see specs when card is released. Title: Re: Kepler hashing performance (Numbers) Post by: BinaryMage on February 13, 2012, 03:16:14 AM Definitely a possibility. Still not in the MH/$ range of 5870s or 5970s, but once the Kepler cards come down in price somewhat, they could become popular.
I think we may be on the verge of the GPU compute wars. Quite exciting, in a way, and disturbing, in another... Title: Re: Kepler hashing performance (Numbers) Post by: bulanula on February 13, 2012, 10:46:14 AM If this is true then I am in heaven ;D
Sadly, I still think we have to put up with AMD's BS drivers and bugs :'( Yeah, come to think of it I still think they need to implement something like bitalign and BFI_INT to be able to compete with 5870s and 7970s etc. I think in this game the more shaders you have the more MHash/s you get. 1536 < 2048 for 7970s >:( Nvidia fail again. ??? Title: Re: Kepler hashing performance (Numbers) Post by: bulanula on February 13, 2012, 03:49:14 PM If this is true then I am in heaven ;D Sadly, I still think we have to put up with AMD's BS drivers and bugs :'( Yeah, come to think of it I still think they need to implement something like bitalign and BFI_INT to be able to compete with 5870s and 7970s etc. I think in this game the more shaders you have the more MHash/s you get. Nvidia fail again. ??? 1536 < 2048 for 7970s >:( Also just the mid end card. The 680 will probably hit that number. We still don't know for sure if it will hash well Don't get me wrong. I am literally praying that Nvidia come out with 2048 shader monster of hashing but without BFI_INT and bitalign I really doubt it. The other thing that could happen is Nvidia owns 7970s and prices come way down for AMD cards. Title: Re: Kepler hashing performance (Numbers) Post by: jake262144 on February 17, 2012, 09:09:13 PM I'm sorry to barge in and spoil your party guys, but did you per chance miss the fact that Kepler's shaders will be running at core speed?
Fermi-based cards run their shaders at twice the core speed. The new kepler gk104 will have 1536 cuda cores. That is 3 times the gtx 580. The gtx 580 gets 140 mhash/s so wouldn't the new kepler card get 420 mhash/s give or take, or at least 6970 territory? Unfortunately, I don't think so.GTX 580's shader clock is 1544 MHz. Let's optimistically assume 1 GHz stock clock speed for Kepler (that translates to 65% of a GTX 580's shader speed). Therefore, a better estimate might be Code: 3 * 140 * 0,65 * a * b Let's assume that Kepler will overclock to 1200 MHz, what results in a = 1.2. Since integer operations efficiency has long played second fiddle to floating-point operations, there is no reason to expect huge gains. Let's assume b = 1.1, i.e. Kepler being 10% better at integer operations than Fermi. A stock Kepler running its 1536 shaders at 1 GHz would achieve 300 MHash/s. Overclocking the card to 1200 MHz would boost the hash rate to 360 MHash/s - that's uncomfortably close to a stock VLIW4-based hd6950. While half of a 7970's hashing speed is far less embarrassing than nVidia's previous generation of GPUs, it's still pretty underwhelming for an expensive and power hungry card a 1536 SP Kelper running at 1 GHz will have to be. Title: Re: Kepler hashing performance (Numbers) Post by: bulanula on February 18, 2012, 03:06:31 PM I'm sorry to barge in and spoil your party guys, but did you per chance miss the fact that Kepler's shaders will be running at core speed? Fermi-based cards run their shaders at twice the core speed. The new kepler gk104 will have 1536 cuda cores. That is 3 times the gtx 580. The gtx 580 gets 140 mhash/s so wouldn't the new kepler card get 420 mhash/s give or take, or at least 6970 territory? Unfortunately, I don't think so.GTX 580's shader clock is 1544 MHz. Let's optimistically assume 1 GHz stock clock speed for Kepler (that translates to 65% of a GTX 580's shader speed). Therefore, a better estimate might be Code: 3 * 140 * 0,65 * a * b Let's assume that Kepler will overclock to 1200 MHz, what results in a = 1.2. Since integer operations efficiency has long played second fiddle to floating-point operations, there is no reason to expect huge gains. Let's assume b = 1.1, i.e. Kepler being 10% better at integer operations than Fermi. A stock Kepler running its 1536 shaders at 1 GHz would achieve 300 MHash/s. Overclocking the card to 1200 MHz would boost the hash rate to 360 MHash/s - that's uncomfortably close to a stock VLIW4-based hd6950. While half of a 7970's hashing speed is far less embarrassing than nVidia's previous generation of GPUs, it's still pretty underwhelming for an expensive and power hungry card a 1536 SP Kelper running at 1 GHz will have to be. I guess we will see. Until now all this is speculation on our part until the official details are announced. I still think they COULD become competitive if they wanted to with a 4608 shader dual GPU moster that is GTX790 but as I said, we will see ... Title: Re: Kepler hashing performance (Numbers) Post by: jake262144 on February 18, 2012, 03:18:56 PM Don't get me wrong, I want to have a choice when it comes to mining card manufacturers.
I hate being tied to AMD and having to deal with each driver update introducing performance loss and craploads of fail. With shader clocks significantly lowered, however, I just don't expect competitive performance from Kepler :( |